Add files using upload-large-folder tool

peer_reviews/df8d6d6325e463d976feb5f3c2a491e81edbbc83f1a51129123d40ceade37bf1/supplementary_0_Peer Review file/images_list.json

@@ -0,0 +1,122 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Fig. 3 The effects of frosts on net primary productivity (NEP) during the non-treatment periods. (a) Main effects of frosts on net primary productivity (NEP), (b) and the pairwise comparisons among treatment levels. In this study, the growing seasons excluded the frost measurement periods in both spring and autumn. All statistical results were examined by linear mix-effect models with significance difference was denoted by an asterisk at \\(P < 0.05\\) . Data were presented as mean \\(\\pm\\) standard error, \\(n = 6\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        235,
+        207,
+        757,
+        570
+      ]
+    ],
+    "page_idx": 19
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Appendix Fig. 5 The main effects of frosts on annual mean ecosystem respiration (ER, a) and gross ecosystem productivity (GEP, b) during the growing seasons excluding frost measurement periods. Insets refer to the mean annual main effects across from 2018 to 2023. Values were shown by mean and standard error (n = 6).",
+    "footnote": [],
+    "bbox": [
+      [
+        228,
+        98,
+        760,
+        458
+      ]
+    ],
+    "page_idx": 20
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_6.jpg",
+    "caption": "Appendix Fig. 6 Monthly main effects of frosts on net ecosystem productivity (NEP, a), ecosystem respiration (ER, b) and gross ecosystem productivity (GEP, c) during the growing seasons excluding frost measurement periods. Values were shown by mean and standard error (n = 6).",
+    "footnote": [],
+    "bbox": [
+      [
+        241,
+        88,
+        750,
+        485
+      ]
+    ],
+    "page_idx": 21
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Fig. 1 Residual Plot for testing heteroscedasticity for NEP. The left figure shows the original model (NEP \\(\\sim\\) Date \\* S \\* A, random = \\~1|Plot/Year/Date) without incorporating the heteroscedasticity module, the right figure shows the current model (NEP \\(\\sim\\) Date \\* S \\* A, random = \\~1|Plot/Year, weights = varIdent(form = \\~1 | Dr)) with the heteroscedasticity module.",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        135,
+        842,
+        370
+      ]
+    ],
+    "page_idx": 35
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Fig. 3 The effects of frosts on net primary productivity (NEP) during the non-treatment periods. (a) Main effects of frosts on net primary productivity (NEP, unit: \\(\\mu \\mathrm{mol} \\mathrm{m}^{-2} \\mathrm{s}^{-1}\\) ), (b) and the pairwise comparisons among treatment levels. In this study, the growing seasons excluded the frost measurement periods in both spring and autumn. All statistical results were examined by linear mix-effect models with significance difference was denoted by an asterisk at \\(P < 0.05\\) . Main effect of autumn frosts on NEP across years were marginally significant \\((P = 0.057)\\) . The main effect values were presented by the marginal means of the treatment, which are derived from the linear mixed effects models. The standard error measures the uncertainty of the estimated marginal means.",
+    "footnote": [],
+    "bbox": [
+      [
+        152,
+        144,
+        837,
+        611
+      ]
+    ],
+    "page_idx": 36
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_4.jpg",
+    "caption": "Supplementary Fig. 4 Effects of frosts on net primary productivity (NEP) during the spring frost periods. One measurement period included 7 days before, 7 days during and 7 days after the frost treatment. Asterisks (\\*) denoted significant differences \\((P< 0.05)\\) observed in pairwise comparisons. Values were shown by mean and standard error \\((n = 6)\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        90,
+        842,
+        734
+      ]
+    ],
+    "page_idx": 39
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_5.jpg",
+    "caption": "Supplementary Fig. 5 Effects of frosts on net ecosystem productivity (NEP) during the autumn frost periods. One measurement period included 7 days before, 7 days during and 7 days after the frost treatment. Asterisks (\\*) denoted significant differences \\((P< 0.05)\\) observed in pairwise comparisons. Values were shown by mean and standard error \\((n = 6)\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        152,
+        90,
+        777,
+        767
+      ]
+    ],
+    "page_idx": 40
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Appendix Fig. 1 The main effects of frosts on annual mean ecosystem respiration (ER, a) and gross ecosystem productivity (GEP, b) during the growing seasons excluding frost measurement periods. Insets refer to the mean annual main effects across from 2018 to 2023. The main effect values were presented by the marginal means of the treatment, which are derived from the linear mixed effects models. The standard error measures the uncertainty of the estimated marginal means.",
+    "footnote": [],
+    "bbox": [
+      [
+        156,
+        87,
+        840,
+        599
+      ]
+    ],
+    "page_idx": 41
+  }
+]

peer_reviews/df8d6d6325e463d976feb5f3c2a491e81edbbc83f1a51129123d40ceade37bf1/supplementary_0_Peer Review file/supplementary_0_Peer Review file.mmd

@@ -0,0 +1,941 @@
+
+# nature portfolio  
+
+# Peer Review File  
+
+# Coinciding spring and autumn frosts have a limited impact on carbon fluxes in a grassland ecosystem  
+
+Corresponding Author: Professor Shiqiang Wan  
+
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+Version 0:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+The manuscript by Han et al. reports on the results of a six- year artificial frost experiment on net ecosystem productivity (NEP), gross ecosystem productivity (GEP), and ecosystem respiration (ER) in a semi- arid grassland ecosystem in Mongolia. The experiment was set up as a randomized complete block design with six replicates. The authors implemented four treatments: spring frost, autumn frost, spring + autumn frost, and a no- treatment control. Open- top pentagonal chambers were placed on top of each \(2.5 \times 2.5 \text{m}\) plot, and frost treatments were implemented by cooling the air in the chambers by \(8^{\circ} \text{C}\) below the ambient air temperature for seven days.  
+
+The authors describe that spring frost increased NEP, while autumn frost decreased NEP, and that the interaction treatment including both spring and autumn frost did not show an effect. Over the observational period, the authors suggest that treatment- induced changes in ecophysiology may be behind this difference in NEP, while in later years, changes in community structure caused changes in NEP. The authors conclude that more frequent frost events in either spring or fall will not increase carbon emissions from grasslands.  
+
+The experiment represents a considerable effort, and the research question is important. However, the manuscript left me with open questions and doubts about the analysis, interpretation, and validity of the claims. I outline these issues below:  
+
+- The conclusion that more frequent frosts will not result in increased carbon emissions from grasslands is not supported by the experiment. In this experiment, the frost treatment is confounded with cooling because frost was simulated by cooling the air within the chambers by \(8^{\circ} \text{C}\). This relative definition of frost means that actual frost only occurred if the ambient air temperature was sufficiently low. The temperature records provided in figure S1 show temperatures above \(30^{\circ} \text{C}\) during the spring treatment period in 2020, 2021, and 2023. Therefore, the treatment may have resulted in cooling rather than frost, or a combination of both. A cooling effect may as well explain the increased NEP due to the spring frost treatment. 
+- The link to phenology is missing. This means treatments may have been applied during periods when frost hardiness was still high (spring) or already high (autumn). 
+- Data quality: It is unclear how the authors dealt with missing data. There are also unexplained abnormalities in the data. For example, in Figure S3, panels g and h, NEP values are surprisingly close to zero. 
+- Increased NEP due to the cooling treatment may render the conclusions invalid. 
+- The data analysis appears convoluted. The reason for using different models is not explained, and it is unclear why the data was aggregated for different analyses (i.e., piecewise structural equation model). 
+- There is high variability in the data, with few significant differences between spring and autumn frost treatments, and very few between control and frost treatments. 
+- The main effects in figure 3 appear surprisingly large and are missing error bars.  
+
+- The main effects in figure 3 appear surprisingly large and are missing error bars.  
+
+Other Points:  
+
+Lines 133- 134: The meaning is unclear; please rephrase. Lines 143- 144: This statement does not match the significance indicated in figure S4. Line 157: See above. Line 204: It is unclear what climate change factors the authors mean here. Line 223: Do you mean species richness? Please check terminology.
+
+<--- Page Split --->
+
+Line 452: As this is a randomized complete block design, why is 'block' not included in the models?  
+
+Lines 500- 505: Why are two separate models necessary?  
+
+Line 534: What tests specifically?  
+
+Fig 3: Captions do not define what the error bars represent.  
+
+Fig 5: Was there anything specific about the baseline species richness at the test site?  
+
+Line 580: Was this analysis done with fixed effects only on aggregated data? Because this is a randomized complete block design, a straight average may be biased. Random effects should be included here.  
+
+Line 581: The model fit is only marginally acceptable.  
+
+Line 581: Why only 2021 to 2023?  
+
+Fig S1: Does this refer to data recording only, or was the cooling system down?  
+
+Fig S3: Confusing because the y- axes are scaled differently.  
+
+(Remarks on code availability)  
+
+Code not available.  
+
+Reviewer #2  
+
+(Remarks to the Author)  
+
+Under continuously intensified global warming scenarios, frost has been predicted more frequently. However, the effects of frosts on ecosystem functions remain far from sufficient. This study focuses on the impact of frost events on ecosystem carbon balance, an area that has not yet been explored. Therefore, the study is highly original and holds great significance. In addition, the statistical analyses are reasonable and the conclusions are clear and compelling, making it worthy of publication in Nature Communications.  
+
+However, more details are needed in M&M and Results sections in the revised manuscript, I suggest major revisions.  
+
+Major comments:  
+
+1. One key result is that spring frosts increase NEP, while autumn frosts decrease NEP. To examine the main treatment effects induced by frosts, linear mixed-effect models were used to analyze GEP, ER, and NEP data on a monthly, yearly, and overall study period basis. These results of linear mixed-effect models robustly illustrate the directions and magnitudes of the opposing contributions of spring and autumn frosts on NEP and demonstrate reasons from the perspectives of the two components of NEP, i.e., GEP and ER.  
+
+1) However, these opposing responses of NEP to spring and autumn frosts were found in first experimental year. Thus, I suspect that the changes in NEP may be due to the plot setup itself. It is recommended to list the baseline carbon balance values, plant community structure, and soil properties before the experiment was established.  
+
+2) Both the main text and Table S2 mentioned that during the 7-day autumn frost periods, autumn frost enhanced GEP \((P = 0.026)\) . However, this is not labelled in Fig. 2. Please add this information.  
+
+3) In explaining the opposing frost effects, no reasons were found in the first three years; instead, possible speculations were provided. It is recommended to attempt to give more reasonable explanations.  
+
+2. Regarding the frost experiment, many details were missing in this edition. These details are crucial for determining whether the frost treatments were appropriate and non-frost-related factors were introduced.  
+
+1) As far as I know, open-top chambers (OTCs) will significantly elevate air temperature. In addition, OTC could hinder seed dispersal and impact plant phenology. How are OTCs managed during frost and non-frost periods?  
+
+2) In this experiment, which served as the control plots: the ambient conditions or the OTCs frames that without the plastic plates?  
+
+3) To provide readers with a more intuitive understanding of the frost operation, it is recommended to add a conceptual diagram in the supporting information.  
+
+Minor comments:  
+
+The introduction section lays a robust foundation for the present study. However, some literature review is somewhat extensive. I suggest condensing these sections to emphasize the innovation and central scientific questions. Such as, L74- L82.  
+
+(Remarks on code availability)  
+
+Reviewer #3  
+
+(Remarks to the Author)  
+
+This study analyzed the impacts of spring and autumn frosts on ecosystem carbon fluxes in a temperate grassland, and suggests that spring frosts enhance carbon sequestration, while autumn frosts promote carbon release, and their combination exerts an additive effect. The topic is interesting, but I still have following concerns. So my suggestion is "major revision".  
+
+1. Line 111 to 113, the hypothesis should be moved to the beginning of this paragraph.  
+
+2. Line 132, the timing of frosts may be important metric that influences carbon exchange, e.g. the frost treatment periods in
+
+<--- Page Split --->
+
+the beginning or the ending of spring may have different effects. This important information has been ignored in the analysis. 3. Line 134, no effect is too ambiguous, how the no effects on ER and GEP produced the NEP change? This finding stands for both spring and autumn? 4. Line 157, annual NEP? It should be clarified in this part of analysis. 5. Line 158- 160, what's the difference between this sentence and previous one? 6. Line 172- 174, maybe some explanation should be provided for the observed contrast findings. 7. Line 182- 187, I think this section should be moved to section of additive effects of spring and autumn frosts; additionally, what's additive effects? 8. Line 191, please provide a brief introduction of the calculation of "theoretical NEP". 9. Line 216, are you sure the autumn should be here? 10. What's the implication of finding from this study?  
+
+(Remarks on code availability)  
+
+Version 1:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+The paper presents results that are noteworthy, but is based on a small experiment (single site with limitations in the experimental design). While it doesn't outweigh these limitations, the dataset covers repeated measures over multiple years.  
+
+Thank you for revising your manuscript and for thoughtfully responding to the points raised. I still have concerns about the calculation of the main effects and the lack of an indication of uncertainty. I understand that you used raw treatment means because treatments and replications could not be matched one- to- one, and hence, no error bars can be provided for figure 3a. However, it is unclear why this issue would also apply to the statistical analysis with mixed models. This should be clarified and it should be possible to provide error bars to avoid misinterpretation of the main effects.  
+
+Thank you also for providing the analysis code for this review. I noticed that NEP exhibits a strong seasonal trend. Are you sure that your model accurately represents this structure? In figure, the main effect of AU is marked as significant, which doesn't fit with the results obtained with the code provided.  
+
+The use of separate models for each individual year (rather than testing all levels of year within one model) is increasing the risk for false positives.  
+
+(Remarks on code availability) The code documentation could be more detailed.  
+
+I could not reproduce one of the significances reported, which might be an error in the manuscript (see review).  
+
+Reviewer #2  
+
+(Remarks to the Author) The reviewer has well resolved my previous concern, and I have no further comment on the revised manuscript.  
+
+(Remarks on code availability)  
+
+Reviewer #3  
+
+(Remarks to the Author) I am satisfied with the authors' response to my comments, and the manuscript has been greatly improved. The manuscript can be accept in current state.  
+
+(Remarks on code availability)  
+
+Version 2:
+
+<--- Page Split --->
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+Comment 1:  
+
+The paper presents results that are noteworthy, but is based on a small experiment (single site with imitations in the experimental design). While it doesn't outweigh these limitations, the dataset covers repeated measures over multiple years.  
+
+Author response: We fully understand and appreciate the importance of scaling up experimental research findings, as you have highlighted. However, due to financial constraints and measurement pressures, we only had one experiment representing the single site. Nevertheless, the long- term data can offer reliable first- hand information for frost research and provide valuable insights for future studies on a regional and even global scale. In the future, verifying the findings from single site may require the collective efforts of scientists from related fields around the world. We look forward to more in- depth reports on this subject.  
+
+Revisions in MS: Consequently, this study offers insights into predicting the potential risks of frost- induced carbon emissions in a warming climate, and the additional frost experiments are needed to further validate these findings (Lines 235- 237).  
+
+Reviewer response: Thank you for your reply. I recommend to acknowledge with a more specific title. E.g.:"Coincidence of spring and autumn frosts has a limited impact on carbon fluxes in a grassland ecosystem"  
+
+## Comment 2:  
+
+Thank you for revising your manuscript and for thoughtfully responding to the points raised. I still have concerns about the calculation of the main effects and the lack of an indication of uncertainty. I understand that you used raw treatment means because treatments and replications could not be matched one- to- one, and hence, no error bars can be provided for figure 3a. However, it is unclear why this issue would also apply to the statistical analysis with mixed models. This should be clarified and it should be possible to provide error bars to avoid misinterpretation of the main effects.  
+
+Author response: Thank you very much for your suggestion. Since our experiment follows a completely randomized design without blocks, we can only create pseudo- blocks by grouping each C, S, A, and SA that are located nearby each other. As a result, we had a total of 6 blocks because there are 6 replications for each treatment. We calculated the main effect of both spring frosts and autumn frosts within each block, resulting in 6 main effect values (replicate = 6, treatment = 2), from which we derived the mean and standard error of the main effect of spring frosts and autumn frosts. Additionally, when we experimented with randomly grouping C, S, A, and SA without considering plot proximity, and found that regardless of how we formed the 6 blocks, the variation in the mean and standard error of the main effects was minimal. Therefore, it is indeed necessary to add standard errors to the main effect values in a completely randomized design. Once again, thank you for your valuable suggestions, which have significantly enhanced our manuscript.  
+
+In this manuscript, all interaction effects in the linear mixed- effects model were found to be insignificant. Therefore, the main effect can be interpreted as the impact of an independent variable on the dependent variable, averaged across the levels of other variables in the model. It represents the direct relationship between that independent variable and the response variable.  
+
+Revisions in MS: We have added the standard error for main effects on NEP in fig. 3(a) in main text (please see below), and on ER and GEP in Appendix Fig. 5- 6 in other related materials (please see below). We also have provided the data for standard error of main effects in DATA, which have been deposited in figshare, https://doi.org/10.6084/m9.figshare.27452835. ly understand and appreciate the importance of scaling up experimental research findings, as you have highlighted. However, due to financial constraints and measurement pressures, we only had one experiment representing the single site. Nevertheless, the long- term data can offer reliable first- hand information for frost research and provide valuable insights for future studies on a regional and even global scale. In the future, verifying the findings from single site may require the collective efforts of scientists from related fields around the world. We look forward to more in- depth reports on this subject.  
+
+Reviewer comment: Thank you for your revisions and for including error bars in Figure 3a. However, the rationale for post- hoc blocking to calculate standard errors is unclear. Mixed models are suitable to analyze completely randomized designs and can provide estimates and SEs for main effects. Using these model outputs would avoid introducing artificial groupings and more accurately reflect variability.  
+
+Comment 3: The use of separate models for each individual year (rather than testing all levels of year within one model) is increasing the risk for false positives.  
+
+Author response: In this article, we initially applied a linear mixed- effects model to the entire dataset spanning seven years to investigate how frost impacts carbon flux in grassland ecosystems. Subsequently, we employed the same type of model to analyze the data on a year basis. As you mentioned, examining the data separately for each of the seven consecutive years might introduce a risk of false positives due to potential autocorrelation between the annual datasets. To address this, we tested the autocorrelation among years.  
+
+We treated the seven years of data collectively, assuming there might be autocorrelation between them, and constructed a linear mixed- effects model incorporating an autocorrelation component ("correlation = corAR1(form = ~ Year")). We also developed a model without the autocorrelation component. By using a likelihood ratio test ("anova()"function), we compared these two models (specific R code is provided below). The results indicated that the inclusion of the autocorrelation component did not significantly improve the model fit (P > 0.05, red in bold), suggesting that there is no significant
+
+<--- Page Split --->
+
+autocorrelation between the years. Therefore, using separate models for each individual year is appropriate.  # # # # # Auto- correlation test for NEP:  
+
+\(>\) model1 \(< -\) lme(NEP \(\sim\) date\\* S \\* A, dat, random \(= \sim 1\) |plot/year/date, method \(=\) "ML")  
+
+Reviewer comment: My concern was about the increased Type I error rate introduced by multiple testing. Analyzing each year separately increases the number of statistical tests, which increases the likelihood of false positives. The reason for the chosen approach instead of analyzing all years within a single model (including Year as a fixed effect) is unclear. A single model approach would avoid the issue of multiple testing and improve transparency.  
+
+Comment 4: The code documentation could be more detailed.  
+
+Author response: We have added more explanations in R code.  
+
+Reviewer comment: The revised R- code is provided as printed R- console output in pdf format and not easily accessible.  
+
+Comment 5: Please clarify why Maximum Likelihood (ML) was used. Restricted Maximum Likelihood (REML) is a better method for estimating mixed models (due to unbiased estimates of variance components). Variance estimates with ML can be biased. Hence, I recommend analyzing your models using REML for more accurate variance component estimates. If ML was used intentionally, please clarify in the manuscript.  
+
+(Remarks on code availability)  See comments to authors #4.  
+
+Version 3:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  My concerns have been addressed in the revisions. See below for issue with DOI link.  
+
+(Remarks on code availability)  DOI link did not work at the time of review. Code was provided in pdf format.  
+
+Open Access This Peer Review 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 cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+The images or other third party material in this Peer Review File are included in the article's Creative Commons license,
+
+<--- Page Split --->
+
+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 https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+
+## Reviewer #1 (Remarks to the Author):  
+
+The manuscript by Han et al. reports on the results of a six- year artificial frost experiment on net ecosystem productivity (NEP), gross ecosystem productivity (GEP), and ecosystem respiration (ER) in a semi- arid grassland ecosystem in Mongolia. The experiment was set up as a randomized complete block design with six replicates. The authors implemented four treatments: spring frost, autumn frost, spring + autumn frost, and a no- treatment control. Open- top pentagonal chambers were placed on top of each \(2.5 \times 2.5 \mathrm{~m}\) plot, and frost treatments were implemented by cooling the air in the chambers by \(8^{\circ} \mathrm{C}\) below the ambient air temperature for seven days.  
+
+The authors describe that spring frost increased NEP, while autumn frost decreased NEP, and that the interaction treatment including both spring and autumn frost did not show an effect. Over the observational period, the authors suggest that treatment- induced changes in ecophysiology may be behind this difference in NEP, while in later years, changes in community structure caused changes in NEP. The authors conclude that more frequent frost events in either spring or fall will not increase carbon emissions from grasslands.  
+
+The experiment represents a considerable effort, and the research question is important. However, the manuscript left me with open questions and doubts about the analysis, interpretation, and validity of the claims. I outline these issues below:  
+
+- The conclusion that more frequent frosts will not result in increased carbon emissions from grasslands is not supported by the experiment. 1. In this experiment, the frost treatment is confounded with cooling because frost was simulated by cooling the air within the chambers by \(8^{\circ} \mathrm{C}\). This relative definition of frost means that actual frost only occurred if the ambient air temperature was sufficiently low. The temperature records provided in figure S1 show temperatures above \(30^{\circ} \mathrm{C}\) during the spring treatment period in 2020, 2021, and 2023. Therefore, the treatment may have resulted in cooling rather than frost, or a combination of both. 2. A cooling effect may as well explain the increased NEP due to the spring frost treatment.  
+
+Responses: Thank you for your time, effort, and valuable insights. We have thoroughly reviewed and understood your suggestions. We greatly appreciate each question you raised and have made the corresponding changes based on your feedback. These modifications have indeed greatly improved the quality of our manuscript. We hope that we have suitably addressed all concerns below:  
+
+1. 1) Your first concern regarding the frost treatment is confounded with cooling is reasonable. By definition, frost generally refers to the occurrence of air temperatures below freezing during the spring and autumn seasons, typically at night. Thus, when the hourly temperature of the day drops below zero, we consider that day to have experienced a frost event. In this manuscript, we employ real-time air temperature reduction by \(8^{\circ} \mathrm{C}\) as a method of frost treatment, which indeed results in a combination of cooling and frost effects. However, to ensure consistent frost treatment effects between spring and autumn, as well as across different years,
+
+<--- Page Split --->
+
+reducing the temperature by a fixed amount is a straightforward and easily implementable approach, especially since air temperatures in the temperate semi- arid regions during spring and autumn often fluctuate violently (ambient air temperatures, black line, Fig. S2). More importantly, when the natural frost events are imminent, they often lead to a certain degree of cooling throughout the day; hence, the occurrence of frost events is inseparable from cooling events. 2) In addition, selecting an appropriate temperature reduction is crucial. An \(8^{\circ}\mathrm{C}\) drop was calculated based on the probability distribution of daily minimum temperatures during the spring green- up period (mid- April to mid- May) and the autumn senescence period (mid- September to mid- October) from 1954 to 2014 in the experimental site (Duolun County). A 7.3- degree reduction at night during the spring green- up period can ensure that the majority of nights have minimum temperatures below freezing, and a 7.5- degree reduction during the autumn senescence period can achieve the same. Therefore, this experimental treatment meets the requirements for frost.  
+
+2. Your second concern is that the enhancement in NEP may be attributed to the spring cooling effect rather than the spring frost itself. I suspect that you are considering the impact of spring cooling on NEP from the perspectives of accumulated temperature and vernalization. Regarding accumulated temperature, a few days of cooling during the spring green-up period may delay plant growth subsequently, potentially lowering NEP due to the reduced temperature accumulation. From the vernalization standpoint, the cooling caused by frost occurs in spring rather than winter, it does not sufficiently promote plant growth or increase NEP in the later months. Consequently, these two factors can likely be discounted as explanations. In addition, cooling effects may influence plants and soil during the 7-day frost periods but may have limited impact beyond this phase. In contrast, temperatures dropping below freezing (i.e., frost effect) act as a physiological threshold for plants and soil microorganisms, significantly altering biochemical reactions at the cellular level, and this effect may extend to the non-treatment periods. Therefore, our key finding—that spring frosts enhance NEP while autumn frosts reduce it—primarily pertains to the non-treatment period and is mainly attributed to frost rather than cooling.  
+
+Revisions in MS: 1) to highlight the effects of the simulated frosts over the 7- day periods in both spring and autumn, the original subplots in Fig. 1b- c have been updated by new subplots (Fig. 1b- c, related statements refer to Line 105- 110), the new subplots described the mean increase of 7.17 frost hours and 9.70 frost hours per day induced by simulated frosts during the 7- day spring and autumn frost periods, respectively. 2) to emphasize the critical threshold of frost, we have added a critical point line at the \(0^{\circ}\mathrm{C}\) mark in Supplementary Fig. 2.  
+
+We found that during the 7- day frost simulation periods, the real- time air temperature \((T_{\mathrm{air}})\) in the frost- treated plots was significantly lower than the ambient temperature (Fig.1, Supplementary Fig. 2), resulting in an average of 8.38 hours/day (7.17 and 9.70 hours/day in spring and autumn vs. 2.37 hours/day in ambient) of frost with an average frost temperature of - 4.99 \(^\circ \mathrm{C}\) (- 5.54 \(^\circ \mathrm{C}\) and - 4.45 \(^\circ \mathrm{C}\) in spring and autumn vs. - 2.09 \(^\circ \mathrm{C}\) in ambient) (Lines 105- 110).
+
+<--- Page Split --->
+
+- The link to phenology is missing. This means treatments may have been applied during periods when frost hardiness was still high (spring) or already high (autumn).  
+
+Responses: Reviewer 3 also mentioned this issue, thank you for your insightful comments. In this manuscript, we indeed did not initially provide the phenological rationale for the timing of the spring and autumn frost events, which is very crucial. Consequently, we have now incorporated this information into the MS.  
+
+Revisions in MS: The timing for simulating spring and autumn frost events were determined based on long- term phenological observations of dominant species at the Duolun Experimental Station, which employ after the majority of plants have flashed in spring and when the plants have started to senescence in autumn (Lines 263- 266).  
+
+- Data quality: It is unclear how the authors dealt with missing data. There are also unexplained abnormalities in the data. For example, in Figure S3, panels g and h, NEP values are surprisingly close to zero.  
+
+Responses: This figure indeed seems to be missing data, however, in this manuscript, the data for NEP, ER, and GEP are complete without missing values. NEP is a measure of ecosystem carbon flux ( \(\mu \mathrm{mol}^{- 1} \mathrm{m}^{- 2} \mathrm{s}^{- 1}\) ) and represents the instantaneous rate of carbon fixation. It is calculated as the difference between the rate of plant photosynthesis (GEP) and the rate of ecosystem respiration (ER), hence some values are close to zero.  
+
+- Increased NEP due to the cooling treatment may render the conclusions invalid.  
+
+Responses: Our main finding that "spring frost promotes NEP" is specific to the non- treatment periods across the entire growing season, excluding the two 21- day frost measurement periods. Additionally, cooling is an accompanying event of frost events because naturally occurring frost phenomena not only lower air temperatures below freezing for certain periods (usually in night) but also result in reduced air temperatures that above \(0^{\circ}\mathrm{C}\) in day (most in daylight). Thus, cooling is an accompanying event of frost, and we do not need to differentiate between them. Moreover, the impact and duration of cooling on vegetation and soil may be less significant than that of frost, especially for non- treatment period. Frost temperatures dropping below freezing present challenges for plants and microorganisms, resulting in substantial, long- lasting effects that persist beyond the treatment periods. These effects include changes in physiological activity, biological growth, and even cellular alterations. Therefore, in this MS, the increased NEP was mainly due to the frost treatment.  
+
+Revisions in MS: Notably, our frost treatment was employed by reducing the \(T_{\mathrm{air}}\) in real- time by \(8^{\circ}\mathrm{C}\) . As a result, during most of the daytime, the \(T_{\mathrm{air}}\) was lower than the ambient environment but did not reach freezing, and thus, frost and cooling occurred alternatively during the 7- day frost treatment period. This phenomenon is
+
+<--- Page Split --->
+
+also very common in natural frost events, when such events are approaching, a significant drop in \(T_{\text{air}}\) (cooling effect) is observed (Lines 285- 290).  
+
+- 1. The data analysis appears convoluted. The reason for using different models is not explained, 2. and it is unclear why the data was aggregated for different analyses (i.e., piecewise structural equation model).  
+
+Responses: 1. I guess that the "different models" you mentioned likely refer to linear mixed-effects models. In this MS, linear mixed-effects models are used to examine the impacts of frosts on NEP, ER, and GEP during the frost measurement periods (21-day intervals) and non-treatment periods (spanning several months). Subsequently, post hoc pairwise comparisons are then conducted to determine whether the differences between various treatment levels and the control are statistically significant. To explain the opposite effects of spring and autumn frost on NEP during the non-treatment periods, we included the variables regarding vegetational community in a piecewise Structural Equation Modeling (SEM).  
+
+2. Within the piecewise SEM, since NEP is governed by both plant photosynthesis and ecosystem respiration, we used vegetational community-level cover and height to quantify the leaf area through which photosynthesis and respiration occur. Species richness was used as a metric for the photosynthetic capacity of the herbaceous plants due to diverse plants. In addition, vegetation investigation are typically measured once per year during the peak growing season (July-August), while NEP is measured multiple times throughout the growing season, we averaged the NEP to an annual mean to ensure a one-by-one point with vegetation investigation. In this model, these variables (spring frosts or not, autumn frosts or not, species richness, vegetational community-level height and cover) are treated as fixed factors, while the year, nested within the plots, is incorporated as a random factor in the mixed-effects models. To present all steps of this statistics, we had uploaded the relevant data and R code (R code for fig. 5) in figshare website.  
+
+Revisions in MS: Given that frosts affect NEP indirectly by altering plant photosynthesis and ecosystem respiration, we included the vegetational cover and plant height at community-level into the SEM hypothetical model, since these terms reflect the total amount of plants that are capable of photosynthesis and respiration. Species richness was also included in the hypothetical model because it can quantify the maximum photosynthetic potential arising from diverse plant. To match the plant variables in August each year, the repeated measurements of NEP were averaged to annual values. Linear mixed-effect models were used to building the piecewise SEM, spring frosts, autumn frosts, species richness, vegetational cover and plant height at community-level were considered as fixed factors, the year which nested within the plots was incorporated as a random factor. After screening by goodness of fit ( \(\chi^2\) , df, P values), we developed the final model (Lines 359- 370).  
+
+- There is high variability in the data, with few significant differences between spring
+
+<--- Page Split --->
+
+and autumn frost treatments, and very few between control and frost treatments.  
+
+Responses: In this MS, the effects of frosts on carbon fluxes was indeed limited. Prior to conducting the linear mixed- effects models, we performed normality transformations on the data if necessary (see description in M & M, Line 321). Additionally, we have re- examined the statistical analyses to ensure their appropriateness and the accuracy of the results.  
+
+- The main effects in figure 3 appear surprisingly large and are missing error bars.  
+
+Responses: The figure 3a did no have error bars, this is because these bars refer to the main effects. Following are the formulas:  
+
+\[\mathrm{Main~effect~of~spring~frosts~} = \frac{\frac{S + SA}{2} - \frac{A + C}{2}}{\frac{A + C}{2}}\times 100\%\]  
+
+\[\mathrm{Main~effect~of~autumn~frosts~} = \frac{\frac{A + SA}{2} - \frac{S + C}{2}}{\frac{S + C}{2}}\times 100\%\]  
+
+Here, S, A, SA and C refer to the mean values of C fluxes without replications (reps \(= 6\) ). There were no replications because the randomized experimental design makes it challenging to match the replications one- to- one between treatment levels.  
+
+Revisions in the MS: The formulas for the main effects had added (Lines 388- 343).  
+
+Other Points:  
+
+Lines 133- 134: The meaning is unclear; please rephrase.  
+
+Responses: Thank you for your thorough review. We recognize that the illustration for this paragraph was unclear, including the sentences you mentioned, and therefore, we have restructured the entire paragraph.  
+
+Revisions in MS: In this study, frost measurement periods were defined as the 7- day frost treatment periods and the 7- day before and after the frosts. In spring frost measurement period, we found spring frosts had no effects on GEP and ER during any 7- day intervals (all \(P > 0.05\) , Supplementary Table 3). In autumn frost measurement period, ongoing autumn frosts reduced GEP ( \(P = 0.026\) , Supplementary Table 3) while had no effects during the 7- day before and after the frosts; Additionally, autumn frosts did not impact ER during any 7- day intervals (Lines 122- 128).  
+
+We also revised the related statements in abstract: we found that ongoing frosts, whether occurring in spring or autumn, had limited effects on gross ecosystem productivity (GEP), ecosystem respiration (ER) and their differences (i.e., net ecosystem productivity, NEP) during frost measurement periods (Lines 20- 25).  
+
+Lines 143- 144: This statement does not match the significance indicated in figure S4.  
+
+Responses: We modified this statement.  
+
+Revisions in MS: In contrast, ongoing autumn frosts reduced NEP ( \(P = 0.001\) ),
+
+<--- Page Split --->
+
+with consistent reductions observed in three years (2018, 2021 and 2022) and extending to the early stages of plant green- up in the following years ( \(P = 0.011\) , Supplementary Table 3, Supplementary Fig. 5) (Lines 133- 136).  
+
+Line 157: See above.  
+
+Responses: We revised this statement, thanks for your suggestions.  
+
+Revisions in MS: Regarding NEP, we further calculated the main effect values both annually and across multiple years. Across from 2018 to 2023, we found that spring frosts increased NEP while autumn frosts decreased NEP significantly (Fig. 3a insert). At the annual scales, annual main effect of spring frosts on NEP tended to be positive, whereas autumn frosts' tends to be negative (Fig. 3a), although the main effects did not reach statistical significances in some years (Lines 146- 151).  
+
+Line 204: It is unclear what climate change factors the authors mean here.  
+
+Responses: We had revised it.  
+
+Revisions in MS: Most are driven by climate change, such as climate warming, changes in precipitation pattern, greenhouse gas emission (Lines 194- 195)  
+
+Line 223: Do you mean species richness? Please check terminology.  
+
+Responses: Based on your comment, we have revised the manuscript to use "species richness" instead of "plant richness." Thank you for your suggestions.  
+
+Revisions in MS: For example: In these models, species richness, vegetational cover, and plant height at the community level were included. Species richness indicates the photosynthetic potential arising from diverse vegetation, while vegetational cover and plant height in together represent the total biomass involved in photosynthesis and respiration (Lines 203- 207).  
+
+Line 452: As this is a randomized complete block design, why is 'block' not included in the models?  
+
+Responses: I apologized for inaccuracies in this description. The experiment followed a completely randomized design without blocks. Consequently, in all the linear mixed- effects models, the plot rather than block was treated as a random variable.  
+
+Revisions in MS: we modified it (Line 267)  
+
+Lines 500- 505: Why are two separate models necessary?  
+
+Responses: In this case, to examine the overall treatment effect and the specific effects for each year, linear mixed- effects models need to be applied both across all years and individually for each year.
+
+<--- Page Split --->
+
+Line 534: What tests specifically?  
+
+Responses: goodness of fit for piecewise SEM: \(\chi^2\) , df, \(P\) values.  
+
+Revisions in MS: After screening by goodness of fit ( \(\chi^2\) , df, P values), we developed the final model (Lines 369- 370).  
+
+Fig 3: Captions do not define what the error bars represent.  
+
+Responses: We have revised it.  
+
+Revisions in MS: Data were presented as mean \(\pm\) standard error, \(\mathrm{n} = 6\) (Line 576).  
+
+Fig 5: Was there anything specific about the baseline species richness at the test site?  
+
+Response: Thank you for your suggestion. Indeed, the background values, including species richness, prior to the building this experiment are crucial. Notably, Reviewer 2 had also recommended their inclusion.  
+
+Revisions in MS: We had added a Table for background values (Supplementary Table 2).  
+
+Line 580: Was this analysis done with fixed effects only on aggregated data? Because this is a randomized complete block design, a straight average may be biased. Random effects should be included here.  
+
+Responses: In this piecewise SEM, each regression model is constructed using linear mixed- effects models, with Plot/Year as a random variable. we have added detailed explanations for piecewise SEM.  
+
+Revisions in MS: Given that frosts affect NEP indirectly by altering plant photosynthesis and ecosystem respiration, we included the vegetational cover and plant height at community- level into the SEM hypothetical model, since these terms reflect the total amount of plants that are capable of photosynthesis and respiration. Species richness was also included in the hypothetical model because it can quantify the maximum photosynthetic potential arising from diverse plant. To match the plant variables in August each year, the repeated measurements of NEP were averaged to annual values. Linear mixed- effect models were used to building the piecewise SEM, spring frosts, autumn frosts, species richness, vegetational cover and plant height at community- level were considered as fixed factors, the year which nested within the plots was incorporated as a random factor. After screening by goodness of fit, we developed the final model (Lines 359- 370).  
+
+Line 581: The model fit is only marginally acceptable.  
+
+Responses: You are absolutely right. Based on your insightful suggest, we logged
+
+<--- Page Split --->
+
+models were used to analyze GEP, ER, and NEP data on a monthly, yearly, and overall study period basis. These results of linear mixed- effect models robustly illustrate the directions and magnitudes of the opposing contributions of spring and autumn frosts on NEP and demonstrate reasons from the perspectives of the two components of NEP, i.e., GEP and ER.  
+
+1) However, these opposing responses of NEP to spring and autumn frosts were found in first experimental year. Thus, I suspect that the changes in NEP may be due to the plot setup itself. It is recommended to list the baseline carbon balance values, plant community structure, and soil properties before the experiment was established.  
+
+Responses: Thank you for your suggestions.  
+
+Revisions in MS: We have added a Table for background values (Supplementary Table 2).  
+
+2) Both the main text and Table S2 mentioned that during the 7-day autumn frost periods, autumn frost enhanced GEP \((P = 0.026)\) . However, this is not labelled in Fig. 2. Please add this information.  
+
+Responses: The statistical result of autumn frosts enhance GEP \((P = 0.026)\) represents the main effect. Upon further examination with pairwise comparison tests, no significant differences were found among the pairs. Consequently, these differences were not highlighted in Figure 2.  
+
+3) In explaining the opposing frost effects, no reasons were found in the first three years; instead, possible speculations were provided. It is recommended to attempt to give more reasonable explanations.  
+
+Responses: In the first three years, we indeed did not find the reasons for the contrasting effects of frost, but through long-term field observations, we find some clues for spring frosts. By integrating these with previous literature, we have speculated the potential reasons. Here are the revised explanations.  
+
+Revisions in MS: We found that no plant community metrics correlated with frost- induced changes in NEP during the early years (2018- 2020). Despite the lack of conclusive evidence supporting the differential impacts of frosts over these years, field observations have provided some insights into vegetation changes associated with spring frosts. The gramineous plants, which are dominant in this temperate semi- arid grassland, subjected to spring frosts typically develop more tillers in the following months. If these tillers emerging after spring frosts exhibit elevated chlorophyll content and increased bud growth rates<sup>48</sup>, it can be inferred that spring frosts may contribute to an enhancement of NEP (Lines 208- 216).  
+
+2. Regarding the frost experiment, many details were missing in this edition. These details are crucial for determining whether the frost treatments were appropriate and non-frost-related factors were introduced.
+
+<--- Page Split --->
+
+1) As far as I know, open-top chambers (OTCs) will significantly elevate air temperature. In addition, OTC could hinder seed dispersal and impact plant phenology. How are OTCs managed during frost and non-frost periods?  
+
+Responses: Thank you for your good advice, we had added it.  
+
+Revisions in MS: To avoid the warming effect caused by the plastic plates of OTCs, we initially installed only half of the plastic panels in each OTC in one day before the frost treatment began, and then added the rest once the refrigeration system was operating efficiently. On the day the frost treatment ended, we shut down the refrigeration system only after removing all the panels (Lines 279- 284).  
+
+2) In this experiment, which served as the control plots: the ambient conditions or the OTCs frames that without the plastic plates?  
+
+Responses: OTCs frames without the plastic plates were served as the control plots.  
+
+Revisions in MS: see Line 297  
+
+3) To provide readers with a more intuitive understanding of the frost operation, it is recommended to add a conceptual diagram in the supporting information.  
+
+Responses: Good suggestions.  
+
+Revisions in MS: We have added a conceptual diagram for frost operation (Supplementary Fig. 1).  
+
+Minor comments:  
+
+The introduction section lays a robust foundation for the present study. However, some literature review is somewhat extensive. I suggest condensing these sections to emphasize the innovation and central scientific questions. Such as, L74- L82.  
+
+Responses: We have proofread the introduction section and modified the inaccurate citation details and improper related statements accordingly (Lines 67- 70, Lines 78- 80, Lines 84- 86, Lines 90- 93).  
+
+  
+
+## Reviewer #3 (Remarks to the Author):  
+
+This study analyzed the impacts of spring and autumn frosts on ecosystem carbon fluxes in a temperate grassland, and suggests that spring frosts enhance carbon sequestration, while autumn frosts promote carbon release, and their combination exerts an additive effect. The topic is interesting, but I still have following concerns. So my suggestion is "major revision".  
+
+Responses: We are very grateful for your positive comments and the great efforts made in enhancing the overall logical framework of this manuscript. Your insights have markedly enhanced the quality of our manuscript, and we've made all the changes you suggested with great care.
+
+<--- Page Split --->
+
+1. Line 111 to 113, the hypothesis should be moved to the beginning of this paragraph.  
+
+Responses: According to your comments, we had move the hypothesis to the beginning of this paragraph.  
+
+Revisions in MS: To discern whether the spring frosts and autumn frosts have different effects on ecosystem carbon fluxes, a manipulative experiment that simulating frost events occurring in spring (around May 1st) and autumn (around October 1st) was conducted in a temperate grassland of Inner Mongolia (Lines 90- 93).  
+
+2. Line 132, the timing of frosts may be important metric that influences carbon exchange, e.g. the frost treatment periods in the beginning or the ending of spring may have different effects. This important information has been ignored in the analysis.  
+
+Responses: Good suggestions, reviewer 1 also suggest adding the timing of frosts.  
+
+Revisions in MS: The timing for simulating spring and autumn frost events were determined based on long-term phenological observations of dominant species at the Duolun Restoration Ecology Station, which employ frost treatment after the majority of plants have flashed in spring and when the plants have started to senescence in autumn (Lines 263- 266).  
+
+3. Line 134, no effect is too ambiguous, how the no effects on ER and GEP produced the NEP change? This finding stands for both spring and autumn?  
+
+Responses: The statistical expressions for this paragraph were indeed incorrect, thank you for bringing this to our attention so carefully.  
+
+Revisions in MS: In this study, frost measurement periods were defined as the 7- day frost treatment periods and the 7- day before and after the frosts. In spring frost measurement period, we found spring frosts had no effects on GEP and ER during any 7- day intervals (all \(P > 0.05\) , Supplementary Table 3). In autumn frost measurement period, ongoing autumn frosts reduced GEP ( \(P = 0.026\) , Supplementary Table 3) while had no effects during the 7- day before and after the frosts; Additionally, autumn frosts did not impact ER during any 7- day intervals (Lines 122- 128).  
+
+4. Line 157, annual NEP? It should be clarified in this part of analysis.  
+
+Responses: The sentence indeed fails to mention statistical analysis and does not offer a smooth transition from last paragraph. We had modified this sentence, and had added the descriptions for main effects in M&M section.  
+
+Revisions in MS: Regarding NEP, we further calculated the main effect values both annually and across multiple years. Across from 2018 to 2023, we found that spring frosts increased NEP while autumn frosts decreased NEP significantly (Fig. 3a insert) (Lines 146- 148).  
+
+The formulas for main effects see M&M section, Lines 388- 343.
+
+<--- Page Split --->
+
+5. Line 158-160, what's the difference between this sentence and previous one?  
+
+Responses: The previous ones described the mean main effects across multiple years, while this sentence emphasized the annual main effects. We had modified these sentences.  
+
+Revisions in MS: Regarding NEP, we further calculated the main effect values both annually and across multiple years. Across from 2018 to 2023, we found that spring frosts increased NEP while autumn frosts decreased NEP significantly (Fig. 3a insert). At the annual scales, annual main effect of spring frosts on NEP tended to be positive, whereas autumn frosts' tends to be negative (Fig. 3a), although the main effects did not reach statistical significances in some years (Lines 146- 151).  
+
+6. Line 172-174, maybe some explanation should be provided for the observed contrast findings.  
+
+Responses: The explanations for contrasting findings were indeed very necessary. While, we had placed the explanations in the section titled "The Mechanism Underlying the Divergent Impacts of Frosts" in revised version. To clearly elucidate the reasons behind the contrasting findings, we have explored this issue through two avenues: data analysis and a review of existing literature.  
+
+Revisions in MS: See the section titled "The Mechanism Underlying the Divergent Impacts of Frosts" (Lines 200- 229).  
+
+7. Line 182-187, I think this section should be moved to section of additive effects of spring and autumn frosts; additionally, what's additive effects?  
+
+Responses: Thank you for your careful review. Based on your comments, we had moved this paragraph to the section of additive effects, and also added the definition for additive effects.  
+
+Revisions in MS: We further found that, when treatment combined, the increases by spring frosts and decreases by autumn frosts may offset each other. Strong evidence for this was the insignificant interactions between spring and autumn frosts in all mixed- effects models ( \(S \times A\) : all \(P > 0.05\) , Supplementary Table 1). These insignificant interactions suggest an additive effect, where two factors contribute independently, resulting in a combined impact that equals the sum of their individual effects, meaning that the observed values match the expected ones statistically (Lines 171- 177).  
+
+8. Line 191, please provide a brief introduction of the calculation of "theoretical NEP".  
+
+Responses: We had added a introduction in M&M section, see Lines 348- 351.
+
+<--- Page Split --->
+
+9. Line 216, are you sure the autumn should be here?  
+
+Responses: I am sorry for using the term "at early stages", this word brings a misunderstanding. We had used "in the early years" instead of it (Lines 208- 209, 217).  
+
+10. What's the implication of finding from this study?  
+
+Responses: According to the suggestions from you and reviewer 2, I had rewrote the implication part, in this version, we strengthened the implications of finding.  
+
+Revisions in MS: In summary, our study provides the first evidence of the impacts of spring and autumn frosts on ecosystem carbon fluxes in a temperate grassland. Furthermore, it highlights the long-lasting and seasonally unique effects of frosts. As climate change extends the growing season and potentially leads to simultaneous spring and autumn frost events, the overall impact of frosts on ecosystem carbon budgets might be neutralized. Consequently, this study provides a hint for predicting the potential risks of frost-induced carbon emissions in a warming climate (Lines 232- 238).
+
+<--- Page Split --->
+
+## Reviewer #1 (Remarks to the Author):  
+
+The paper presents results that are noteworthy, but is based on a small experiment (single site with limitations in the experimental design). While it doesn't outweigh these limitations, the dataset covers repeated measures over multiple years.  
+
+Responses: We fully understand and appreciate the importance of scaling up experimental research findings, as you have highlighted. However, due to financial constraints and measurement pressures, we only had one experiment representing the single site. Nevertheless, the long- term data can offer reliable first- hand information for frost research and provide valuable insights for future studies on a regional and even global scale. In the future, verifying the findings from single site may requires the collective efforts of scientists from related fields around the world. We look forward to more in- depth reports on this subject.  
+
+Revisions in MS: Consequently, this study offers insights into predicting the potential risks of frost- induced carbon emissions in a warming climate, and the additional frost experiments are needed to further validate these findings (Lines 235- 237).  
+
+Thank you for revising your manuscript and for thoughtfully responding to the points raised. I still have concerns about the calculation of the main effects and the lack of an indication of uncertainty. I understand that you used raw treatment means because treatments and replications could not be matched one- to- one, and hence, no error bars can be provided for figure 3a. However, it is unclear why this issue would also apply to the statistical analysis with mixed models. This should be clarified and it should be possible to provide error bars to avoid misinterpretation of the main effects.  
+
+Responses: Thank you very much for your suggestion. Since our experiment follows a completely randomized design without blocks, we can only create pseudoblocks by grouping each C, S, A, and SA that are located nearby each other. As a result, we had a total of 6 blocks because there are 6 replications for each treatment. We calculated the main effect of both spring frosts and autumn frosts within each block, resulting in 6 main effect values (replicate = 6, treatment = 2), from which we derived the mean and standard error of the main effect of spring frosts and autumn frosts. Additionally, when we experimented with randomly grouping C, S, A, and SA without considering plot proximity, and found that regardless of how we formed the 6 blocks, the variation in the mean and standard error of the main effects was minimal. Therefore, it is indeed necessary to add standard errors to the main effect values in a completely randomized design. Once again, thank you for your valuable suggestions, which have significantly enhanced our manuscript.  
+
+In this manuscript, all interaction effects in the linear mixed- effects model were found to be insignificant. Therefore, the main effect can be interpreted as the impact of an independent variable on the dependent variable, averaged across the levels of other variables in the model. It represents the direct relationship between that independent variable and the response variable.
+
+<--- Page Split --->
+
+Revisions in MS: We have added the standard error for main effects on NEP in fig. 3(a) in main text (please see below), and on ER and GEP in Appendix Fig. 5- 6 in other related materials (please see below). We also have provided the data for standard error of main effects in DATA, which have been deposited in figshare, https://doi.org/10.6084/m9.figshare.27452835.  
+
+![](images/Figure_3.jpg)
+
+<center>Fig. 3 The effects of frosts on net primary productivity (NEP) during the non-treatment periods. (a) Main effects of frosts on net primary productivity (NEP), (b) and the pairwise comparisons among treatment levels. In this study, the growing seasons excluded the frost measurement periods in both spring and autumn. All statistical results were examined by linear mix-effect models with significance difference was denoted by an asterisk at \(P < 0.05\) . Data were presented as mean \(\pm\) standard error, \(n = 6\) . </center>
+
+<--- Page Split --->
+![](images/Figure_5.jpg)
+
+<center>Appendix Fig. 5 The main effects of frosts on annual mean ecosystem respiration (ER, a) and gross ecosystem productivity (GEP, b) during the growing seasons excluding frost measurement periods. Insets refer to the mean annual main effects across from 2018 to 2023. Values were shown by mean and standard error (n = 6). </center>
+
+<--- Page Split --->
+![](images/Figure_6.jpg)
+
+<center>Appendix Fig. 6 Monthly main effects of frosts on net ecosystem productivity (NEP, a), ecosystem respiration (ER, b) and gross ecosystem productivity (GEP, c) during the growing seasons excluding frost measurement periods. Values were shown by mean and standard error (n = 6). </center>  
+
+Thank you also for providing the analysis code for this review. I noticed that NEP exhibits a strong seasonal trend. Are you sure that your model accurately represents this structure?  
+
+Responses: According to your comments, we rebuilt the linear mixed- effect models. The updated Model spanning from 2018 to 2023: NEP \(\sim\) date\*S\*A, random= \(\sim\) 1|plot/year/date. In this fixed effect model, "date" (format: yyyy/mm/dd) was used as a fixed factor to represent seasonal dynamics. Additionally, "date" within "year" within "plot" was used as a random effect model (random= \(\sim\) 1|plot/year/date), where, each plot had its own random intercept, capturing variation among plots, "year" was nested within "plot", meaning that within each plot, there were observations from different years. "date" was nested within "year" (which is within "plot"), indicating that within each year, there were measurements taken on different dates. The updated model for each year: NEP \(\sim\) date\*S\*A, random= \(\sim\) 1|plot. It's worthy to note that the statistical results of this new model were identical to those of the original version, with only minor differences in the numerical values of the \(P\) -
+
+<--- Page Split --->
+
+values.
+
+**Revisions in MS:** We have revised Supplementary Table 1, Supplementary Table 3 (please see below), Fig. 3b and Appendix Fig. 5 (please see above).
+
+**Supplementary Table 1** Statistical results (P-values) of linear mixed-effects models on the effects of date, spring frosts (S), autumn frosts (A) and their interactions on soil temperature (Tsoil), volumetric water content (VWC), net ecosystem productivity (NEP), ecosystem respiration (ER) and gross ecosystem productivity (GEP) throughout the growing seasons from 2017 to 2022. The growing season excluded frost measurement periods in this study. P-values displayed in bold indicate the significance at \(P<0.05.\) 
+
+<table><tr><td></td><td>Tsoil</td><td>VWC</td><td>NEP</td><td>ER</td><td>GEP</td></tr><tr><td>Date</td><td>0.046</td><td>0.019</td><td>&lt;0.001</td><td>&lt;0.001</td><td>&lt;0.001</td></tr><tr><td>S</td><td>0.124</td><td>0.996</td><td>0.022</td><td>0.803</td><td>0.258</td></tr><tr><td>A</td><td>0.236</td><td>0.373</td><td>0.019</td><td>0.540</td><td>0.320</td></tr><tr><td>DatexS</td><td>0.479</td><td>0.760</td><td>0.359</td><td>0.788</td><td>0.766</td></tr><tr><td>DatexA</td><td>0.920</td><td>0.871</td><td>0.042</td><td>0.616</td><td>0.563</td></tr><tr><td>SxA</td><td>0.554</td><td>0.877</td><td>0.417</td><td>0.604</td><td>0.952</td></tr><tr><td>DatexSxA</td><td>0.747</td><td>0.801</td><td>0.449</td><td>0.893</td><td>0.764</td></tr></table>
+
+**Supplementary Table 3** Statistical results (P-values) of linear mixed-effects models on the effects of date (D), spring frosts (S), autumn frosts (A) and their interactions on net ecosystem productivity (NEP), ecosystem respiration (ER) and gross ecosystem productivity (GEP) during the 7-day frost treatment periods (during) and the 7 days before (before) and after the frosts periods (after) across from 2017 to 2023. P-values displayed in bold indicated the significance at \(P<0.05.\) 
+
+<table><tr><td rowspan="2"></td><td colspan="3">NEP</td><td colspan="3">ER</td><td colspan="3">GEP</td></tr><tr><td>Before</td><td>During</td><td>After</td><td>Before</td><td>During</td><td>After</td><td>Before</td><td>During</td><td>After</td></tr><tr><td>Spring</td><td>D</td><td>0.374</td><td>0.010</td><td>&lt;0.001</td><td>&lt;0.001</td><td>&lt;0.001</td><td>&lt;0,001</td><td>&lt;0.001</td><td>&lt;0.001</td></tr><tr><td></td><td>S</td><td>0.751</td><td>0.263</td><td>0.429</td><td>0.993</td><td>0.708</td><td>0.993</td><td>0.887</td><td>0.434</td></tr><tr><td></td><td>A</td><td>0.016</td><td>0.670</td><td>0.458</td><td>0.371</td><td>0.480</td><td>0.556</td><td>0.684</td><td>0.551</td></tr><tr><td></td><td>DxS</td><td>0.834</td><td>0.279</td><td>0.736</td><td>0.725</td><td>0.706</td><td>0.975</td><td>0.655</td><td>0.445</td></tr><tr><td></td><td>DxA</td><td>0.011</td><td>0.641</td><td>0.792</td><td>0.622</td><td>0.786</td><td>0.733</td><td>0.444</td><td>0.699</td></tr><tr><td></td><td>SxA</td><td>0.170</td><td>0.477</td><td>0.396</td><td>0.552</td><td>0.992</td><td>0.929</td><td>0.894</td><td>0.722</td></tr><tr><td></td><td>DxSxA</td><td>0.451</td><td>0.230</td><td>0.579</td><td>0.893</td><td>0.922</td><td>0.997</td><td>0.642</td><td>0.498</td></tr><tr><td></td><td></td><td>Before</td><td>During</td><td>After</td><td>Before</td><td>During</td><td>After</td><td>Before</td><td>During</td></tr><tr><td>Autumn</td><td>D</td><td>&lt;0.001</td><td>&lt;0.001</td><td>&lt;0.001</td><td>&lt;0,001</td><td>&lt;0.001</td><td>&lt;0.001</td><td>&lt;001</td><td>&lt;0.001</td></tr><tr><td></td><td>S</td><td>0.645</td><td>0.620</td><td>0.628</td><td>0.628</td><td>0.902</td><td>0.727</td><td>0.952</td><td>0.778</td></tr><tr><td></td><td>A</td><td>0.689</td><td>0.011</td><td>0.333</td><td>0.338</td><td>0.706</td><td>0.389</td><td>0.843</td><td>0.099</td></tr><tr><td></td><td>DxS</td><td>0.754</td><td>0.802</td><td>0.920</td><td>0.677</td><td>0.869</td><td>0.840</td><td>0.999</td><td>0.809</td></tr></table>
+
+<--- Page Split --->
+
+
+<table><tr><td>D×A</td><td>0.735</td><td>0.192</td><td>0.339</td><td>0.528</td><td>0.992</td><td>0.729</td><td>0.932</td><td>0.375</td><td>0.690</td></tr><tr><td>S×A</td><td>0.897</td><td>0.938</td><td>0.365</td><td>0.767</td><td>0.997</td><td>0.486</td><td>0.950</td><td>0.956</td><td>0.828</td></tr><tr><td>D×S×A</td><td>0.764</td><td>0.634</td><td>0.676</td><td>0.924</td><td>0.745</td><td>0.710</td><td>0.899</td><td>0.646</td><td>0.939</td></tr></table>  
+
+In figure, the main effect of AU is marked as significant, which doesn't fit with the results obtained with the code provided.  
+
+Responses: Thank you very much for your careful review. Indeed, there was a main effect for AU with \(P = 0.0537\) for 2020. We have removed this asterisk notation in new version.  
+
+Revisions in MS: we have revised in fig. 3b (please see above).  
+
+The use of separate models for each individual year (rather than testing all levels of year within one model) is increasing the risk for false positives.  
+
+Responses: In this article, we initially applied a linear mixed- effects model to the entire dataset spanning seven years to investigate how frost impacts carbon flux in grassland ecosystems. Subsequently, we employed the same type of model to analyze the data on a year basis. As you mentioned, examining the data separately for each of the seven consecutive years might introduce a risk of false positives due to potential autocorrelation between the annual datasets. To address this, we tested the autocorrelation among years.  
+
+We treated the seven years of data collectively, assuming there might be autocorrelation between them, and constructed a linear mixed- effects model incorporating an autocorrelation component ("correlation \(=\) corAR1(form \(= \sim\) Year")). We also developed a model without the autocorrelation component. By using a likelihood ratio test ("anova()"function), we compared these two models (specific R code is provided below). The results indicated that the inclusion of the autocorrelation component did not significantly improve the model fit ( \(P > 0.05\) , red in bold), suggesting that there is no significant autocorrelation between the years. Therefore, using separate models for each individual year is appropriate.  
+
+\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*Auto- correlation test for NEP: > model1 <- lme(NEP \(\sim\) date\\* S \\* A, dat, random \(= \sim 1\) |plot/year/date, method \(=\) "ML") > model1a <-lme(NEP \(\sim\) date\\* S \\* A, dat, random \(= \sim 1\) |plot/year/date, correlation \(=\) corAR1(form \(= \sim\) Year), method \(=\) "ML") > anova(model1, model1a) Model df AIC BIC logLik Test L.Ratio p- value model1 1 12 8757.851 8820.075 - 4366.925 model1a 2 13 8759.851 8827.261 - 4366.925 1 vs 2 9.627911e- 09 0.9999  
+
+\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*Auto- correlation test for ER:
+
+<--- Page Split --->
+
+> model2 <- 1me(ER ~ date\* S \* A, dat, random = \~1 |plot/year/date, method = "ML") > model2a <- 1me(ER ~ date\* S \* A, dat, random = \~1 |plot/year/date, correlation = corAR1(form = \~ Year), method = "ML") > anova(model2, model2a) Model df AIC BIC logLik Test L.Ratio p-value model2 1 12 8565.993 8628.217 -4270.996 model2a 2 13 8567.993 8635.403 -4270.996 1 vs 2 7.695598e-08 0.9998  
+
+#### Auto- correlation test for GEP: > model3 <- 1me(GEP ~ date\* S \* A, dat, random = \~1 |plot/year/date, method = "ML") > model3a <- 1me(GEP ~ date\* S \* A, dat, random = \~1 |plot/year/date, correlation = corAR1(form = \~ Year), method = "ML") > anova(model3, model3a) Model df AIC BIC logLik Test L.Ratio p-value model3 1 12 10070.1 10132.32 -5023.05 model3a 2 13 10072.1 10139.51 -5023.05 1 vs 2 2.056549e-08 0.9999  
+
+## Reviewer #1 (Remarks on code availability):  
+
+The code documentation could be more detailed. Responses: We have added more explanations in R code.  
+
+I could not reproduce one of the significances reported, which might be an error in the manuscript (see review).  
+
+Responses: We had revised the related R codes.
+
+<--- Page Split --->
+
+Reviewer #1 (Remarks to the Author):  
+
+## Comment 1:  
+
+The paper presents results that are noteworthy, but is based on a small experiment (single site with imitations in the experimental design). While it doesn't outweigh these limitations, the dataset covers repeated measures over multiple years.  
+
+Author response: We fully understand and appreciate the importance of scaling up experimental research findings, as you have highlighted. However, due to financial constraints and measurement pressures, we only had one experiment representing the single site. Nevertheless, the long- term data can offer reliable first- hand information for frost research and provide valuable insights for future studies on a regional and even global scale. In the future, verifying the findings from single site may requires the collective efforts of scientists from related fields around the world. We look forward to more in- depth reports on this subject.  
+
+Revisions in MS: Consequently, this study offers insights into predicting the potential risks of frost- induced carbon emissions in a warming climate, and the additional frost experiments are needed to further validate these findings (Lines 235- 237).  
+
+Reviewer response: Thank you for your reply. I recommend to acknowledge with a more specific title. E.g.:"Coincidence of spring and autumn frosts has a limited impact on carbon fluxes in a grassland ecosystem"  
+
+Author further responses: We revised the title based on your suggestion, and the new version more accurately reflects the significance of the research findings. Thank you for your valuable suggestion.  
+
+Further revisions in MS: We revised the title to: Coinciding Spring and Autumn Frosts Have a Limited Impact on Carbon Fluxes in a Grassland Ecosystem  
+
+## Comment 2:  
+
+Thank you for revising your manuscript and for thoughtfully responding to the points raised. I still have concerns about the calculation of the main effects and the lack of an indication of uncertainty. I understand that you used raw treatment means because treatments and replications could not be matched one- to- one, and hence, no error bars can be provided for figure 3a. However, it is unclear why this issue would also apply to the statistical analysis with mixed models. This should be clarified and it should be possible to provide error bars to avoid misinterpretation of the main effects.
+
+<--- Page Split --->
+
+Author response: Thank you very much for your suggestion. Since our experiment follows a completely randomized design without blocks, we can only create pseudo- blocks by grouping each C, S, A, and SA that are located nearby each other. As a result, we had a total of 6 blocks because there are 6 replications for each treatment. We calculated the main effect of both spring frosts and autumn frosts within each block, resulting in 6 main effect values (replicate \(= 6\) , treatment \(= 2\) ), from which we derived the mean and standard error of the main effect of spring frosts and autumn frosts. Additionally, when we experimented with randomly grouping C, S, A, and SA without considering plot proximity, and found that regardless of how we formed the 6 blocks, the variation in the mean and standard error of the main effects was minimal. Therefore, it is indeed necessary to add standard errors to the main effect values in a completely randomized design. Once again, thank you for your valuable suggestions, which have significantly enhanced our manuscript.  
+
+In this manuscript, all interaction effects in the linear mixed- effects model were found to be insignificant. Therefore, the main effect can be interpreted as the impact of an independent variable on the dependent variable, averaged across the levels of other variables in the model. It represents the direct relationship between that independent variable and the response variable.  
+
+Revisions in MS: We have added the standard error for main effects on NEP in fig. 3(a) in main text (please see below), and on ER and GEP in Appendix Fig. 5- 6 in other related materials (please see below). We also have provided the data for standard error of main effects in DATA, which have been deposited in figshare, https://doi.org/10.6084/m9.figshare.27452835.  
+
+Reviewer comment: Thank you for your revisions and for including error bars in Figure 3a. However, the rationale for post- hoc blocking to calculate standard errors is unclear. Mixed models are suitable to analyze completely randomized designs and can provide estimates and SEs for main effects. Using these model outputs would avoid introducing artificial groupings and more accurately reflect variability.  
+
+Author further response: We sincerely appreciate the reviewer's insightful suggestions, which have provided us with a more efficient and rigorous method for calculating the main effects and their uncertainty (SE). In this revision, we applied linear mixed- effect models and analyzed all six years of data collectively to assess the main and interactive effects of frost treatments and year. Since successive droughts led to a significant shift in variance between the first three years and the following three years (with variance ranging from approximately 70- 80 in the first three years and dropping to 0- 10 in the last three years), we incorporated this changing variance into the linear mixed- effects model using a weights structure. The specific model was
+
+<--- Page Split --->
+
+as follows:  
+
+\(\mathrm{Y}\sim \mathrm{Year}\times \mathrm{S}\times \mathrm{A}\) , random \(= \sim 1|\mathrm{plot}\) , weights \(=\) varIdent(form \(= \sim 1|\mathrm{Dr})\) In this model, Year was treated as a categorical variable to obtain year- specific statistical results. Dr was used to quantify the changes due to interannual variations in temperature and precipitation, where the first three years were labeled as "A", and the last three years as "B".  
+
+Following these models, the main effects and SE were obtained using the marginal means approach with the emmeans() function.  
+
+Below is an exemplar R code for NEP and the output is attached at the end of our responses.  
+
+## weights structure measures the different variance control. model2 <- lme(NEP \(\sim\) Year \\* S \\* A, random \(= \sim 1\) | Plot, data \(=\) dat1, weights \(=\) varIdent(form \(= \sim 1\) | Dr), method="REML")  
+
+library(emmeans)  
+
+Main effects across years, fig 3a insert figure lmeans_s<- emmeans(model2, pairwise \(\sim \mathrm{S}\) ); lmeans_s lmeans_A<- emmeans(model2, pairwise \(\sim \mathrm{A}\) ); lmeans_A  
+
+# Yearly main effects for fig.3a: lmeans_year_s <- emmeans(model2, pairwise \(\sim \mathrm{S}\) | Year); lmeans_year_s  
+
+lmeans_year_A <- emmeans(model2, pairwise \(\sim \mathrm{A}\) | Year); lmeans_year_A  
+
+Additionally, given the heteroscedasticity issue identified in all similar analyses, we examined all the linear mixed- effects models throughout our manuscript and found that heteroscedasticity was occurred in all original models. To address this, we improved all models by incorporating a weighting structure (weights \(=\) varIdent(form \(= \sim 1\) | Dr)). This modification also addressed the reviewer's concern from the second round of review: "I noticed that NEP exhibits a strong seasonal trend. Are you sure that your model accurately represents this structure?".  
+
+The original model in second round was:  
+
+lme(Y \(\sim\) Date \(\times \mathrm{S}\times \mathrm{A}\) , random \(= \sim 1|\mathrm{Plot / Year / Date}\) , data, method \(=\) "ML")  
+
+And the new model was:  
+
+lme(Y \(\sim\) Date \(\times \mathrm{S}\times \mathrm{A}\) , random \(= \sim 1|\mathrm{Plot / Year}\) , weights \(=\) varIdent(form \(= \sim 1|\mathrm{Dr})\) data, method \(=\) "REML").  
+
+In this new version, the change did not influence the effect estimations but reduced model errors, leading to improved model performance.  
+
+Further revision in MS: Based on the updated statistical results from the linear
+
+<--- Page Split --->
+
+mixed- effects models, we have revised Fig. 3, Supplementary Table 1, Supplementary Table 3, Supplementary Tables 4- 5, and Appendix Fig. 1 (attached at the end). The corresponding descriptions in the text have also been updated (Lines 123- 132, Lines 139- 149, and Lines 317- 344).  
+
+## Comment 3:  
+
+The use of separate models for each individual year (rather than testing all levels of year within one model) is increasing the risk for false positives.  
+
+Author response: In this article, we initially applied a linear mixed- effects model to the entire dataset spanning seven years to investigate how frost impacts carbon flux in grassland ecosystems. Subsequently, we employed the same type of model to analyze the data on a year basis. As you mentioned, examining the data separately for each of the seven consecutive years might introduce a risk of false positives due to potential autocorrelation between the annual datasets. To address this, we tested the autocorrelation among years.  
+
+We treated the seven years of data collectively, assuming there might be autocorrelation between them, and constructed a linear mixed- effects model incorporating an autocorrelation component ("correlation \(=\) corAR1(form \(= \sim\) Year")). We also developed a model without the autocorrelation component. By using a likelihood ratio test ("anova()"function), we compared these two models (specific R code is provided below). The results indicated that the inclusion of the autocorrelation component did not significantly improve the model fit ( \(\mathrm{P} > 0.05\) , red in bold), suggesting that there is no significant autocorrelation between the years. Therefore, using separate models for each individual year is appropriate.  
+
+\\*\\*\\*\\*\\*\\*\\*\\*\\*\\*\\*\*\\*\\*\\*\\*\\*\\*\\*\\*\\*Auto- correlation test for NEP:  
+
+\(> \mathrm{model1} < - \mathrm{lme(NEP} \sim \mathrm{date}* \mathrm{S} * \mathrm{A}\) , dat, random \(= \sim 1\) |plot/year/date, method \(=\) "ML")  
+
+Reviewer comment: My concern was about the increased Type I error rate introduced by multiple testing. Analyzing each year separately increases the number of statistical tests, which increases the likelihood of false positives. The reason for the chosen approach instead of analyzing all years within a single model (including Year as a fixed effect) is unclear. A single model approach would avoid the issue of multiple testing and improve transparency.  
+
+Author further response: As described above, in this revision, we analyzed data from all years using a single linear mixed- effect model, treating year as a fixed effect. The R code for performing multiple comparisons for each year is provide below, with
+
+<--- Page Split --->
+
+the corresponding output attached at the end:  
+
+## weights structure measures the different variance control. model2 <- lme(NEP \\~Year \\* S \\* A, random = \\~1 | Plot, data = dat1, weights = varIdent(form = \\~1 | Dr), method="REML")  
+
+library(emmeans)  
+
+#Multiple comparisons across years:  
+
+lmeans_SA<- emmeans(model2, pairwise \(\sim\) S\\*A);lmeans_SA  
+
+#Multiple comparisons for each year, fig. 3b  
+
+lmeans_SA1<- emmeans(model2, pairwise \(\sim\) S\\*A|Year);lmeans_SA1  
+
+Further revision in MS: We revised Fig. 3b (attach at the end).  
+
+## Comment 4:  
+
+The code documentation could be more detailed.  
+
+Author response: We have added more explanations in R code.  
+
+Reviewer comment: The revised R- code is provided as printed R- console output in pdf format and not easily accessible.  
+
+Author further response: We have updated all the R codes and uploaded them to the Figshare website in ".r" format (https://doi.org/10.6084/m9. figshare.27452835). We also updated R codes in PDF format.  
+
+## Comment 5:  
+
+Please clarify why Maximum Likelihood (ML) was used. Restricted Maximum Likelihood (REML) is a better method for estimating mixed models (due to unbiased estimates of variance components). Variance estimates with ML can be biased. Hence, I recommend analyzing your models using REML for more accurate variance component estimates. If ML was used intentionally, please clarify in the manuscript.  
+
+Author response: We changed the methods in all linear mixed- effects models from "ML" to "REML". Thank you for your suggestion.  
+
+Reviewer #1 (Remarks on code availability): See comments to authors #4.  
+
+#R code for Fig. 3:
+
+<--- Page Split --->
+
+##dat1:dataset  
+
+> dat1\$year<- as.factor(dat1\$year)  
+
+> variances <- ddply(dat1,c("Year"),summarise, var.NEP=var(NEP));vari ances  
+
+Year var.NEP 1 2018 72.513015 2 2019 80.849832 3 2020 71.120523 4 2021 0.769025 5 2022 6.257297 6 2023 8.412060  
+
+> dat1\$Dr[dat1\$Year \(= =\) "2018"|dat1\$Year \(= =\) "2019"|dat1\$Year \(= =\) "2020"]<- "A" > dat1\$Dr[dat1\$Year \(= =\) "2021"|dat1\$Year \(= =\) "2022"|dat1\$Year \(= =\) "2023"]<- "B"  
+
+> ## weights structure measures the different variance control.  
+
+> model2 <- lme(NEP \\~Year \\* S \\* A, random = \\~1 | Plot, + weights = varIdent(form = \\~1 | Dr), data = dat1, method="R EML")  
+
+##Main effects of S on NEP across years:  
+
+> lsmeans_S<- emmeans(model2, pairwise \(\sim\) S);lsmeans_S  
+
+\$emmeans  
+
+S emmean SE df lower.CL upper.CL  
+
+0 4.84 0.259 20 4.30 5.38  
+
+1 6.16 0.259 20 5.62 6.70  
+
+Results are averaged over the levels of: Year, A Degrees- of- freedom method: containment Confidence level used: 0.95  
+
+\$contrasts  
+
+contrast estimate SE df t.ratio p.value  
+
+S0 - S1 -1.32 0.366 20 - 3.604 0.0018  
+
+Results are averaged over the levels of: Year, A Degrees- of- freedom method: containment  
+
+##Main effects of A on NEP across years:  
+
+> lsmeans_A<- lsmeans(model2, pairwise \(\sim\) A);lsmeans_A \$lsmeans  
+
+A lsmean SE df lower.CL upper.CL  
+
+0 6.17 0.259 20 5.63 6.71  
+
+1 4.83 0.259 20 4.29 5.37
+
+<--- Page Split --->
+
+Results are averaged over the levels of: Year, S Degrees- of- freedom method: containment Confidence level used: 0.95  
+
+\$contrasts contrast estimate SE df t.ratio p.value A0 - A1 1.34 0.366 20 3.661 0.0016  
+
+Results are averaged over the levels of: Year, S Degrees- of- freedom method: containment  
+
+##Main effect of S on NEP for each year:  
+
+\$ \# \text{smeans section will not be presented here} \$ > \text{lmeans\_year\_S <- emmeans(model2, pairwise ~ S | Year); lmeans\_year\_S} \$  
+
+\$contrasts\$  
+
+Year = 2018: contrast estimate SE df t.ratio p.value S0 - S1 -2.036 1.234 20 -1.649 0.1147  
+
+Year = 2019:  
+
+contrast estimate SE df t.ratio p.value S0 - S1 -2.014 1.164 20 -1.729 0.0991  
+
+Year = 2020:  
+
+contrast estimate SE df t.ratio p.value S0 - S1 -1.564 1.010 20 -1.548 0.1373  
+
+Year = 2021:  
+
+contrast estimate SE df t.ratio p.value S0 - S1 -0.491 0.460 20 -1.065 0.2994  
+
+Year = 2022:  
+
+contrast estimate SE df t.ratio p.value S0 - S1 -0.843 0.324 20 -2.599 0.0172  
+
+Year = 2023:  
+
+contrast estimate SE df t.ratio p.value S0 - S1 -0.966 0.338 20 -2.860 0.0097  
+
+Results are averaged over the levels of: A Degrees- of- freedom method: containment  
+
+##Main effect of A on NEP for each year: \$ \# \text{smeans section will not be presented here\$
+
+<--- Page Split --->
+
+> !smeans_year_A <- emmeans(model12, pairwise ~ A | Year); !smeans_year_A $contrasts  
+
+Year \(= 2018\)  
+
+contrast estimate SE df t.ratio p.value A0 - A1 2.074 1.234 20 1.680 0.1085  
+
+Year \(= 2019\)  
+
+contrast estimate SE df t.ratio p.value A0 - A1 2.867 1.164 20 2.463 0.0230  
+
+Year \(= 2020\)  
+
+contrast estimate SE df t.ratio p.value A0 - A1 2.526 1.010 20 2.499 0.0213  
+
+Year \(= 2021\)  
+
+contrast estimate SE df t.ratio p.value A0 - A1 0.268 0.460 20 0.581 0.5675  
+
+Year \(= 2022\)  
+
+contrast estimate SE df t.ratio p.value A0 - A1 0.412 0.324 20 1.269 0.2189  
+
+Year \(= 2023\)  
+
+contrast estimate SE df t.ratio p.value A0 - A1 -0.109 0.338 20 -0.323 0.7498  
+
+Results are averaged over the levels of: S Degrees- of- freedom method: containment  
+
+##Multiple comparisons for each year:  
+
+# #smeans section will not be presented here  
+
+> !smeans_year_SA <- emmeans(model12, pairwise ~ S \* A | Year); !smeans_year_SA  
+
+$contrasts  
+
+Year \(= 2018\)  
+
+contrast estimate SE df t.ratio p.value S0 A0 - S1 A0 -1.318 1.745 20 -0.755 0.8734  
+
+S0 A0 - S0 A1 2.791 1.745 20 1.599 0.4016  
+
+S0 A0 - S1 A1 0.038 1.745 20 0.022 1.0000  
+
+S1 A0 - S0 A1 4.109 1.745 20 2.354 0.1191  
+
+S1 A0 - S1 A1 1.356 1.745 20 0.777 0.8639  
+
+S0 A1 - S1 A1 -2.753 1.745 20 -1.577 0.4133  
+
+Year \(= 2019\)
+
+<--- Page Split --->
+
+<table><tr><td>contrast</td><td>estimate</td><td>SE df t.ratio p.value</td></tr><tr><td>S0 A0 - S1 A0</td><td>-0.912</td><td>1.647 20 -0.554 0.9444</td></tr><tr><td>S0 A0 - S0 A1</td><td>3.969</td><td>1.647 20 2.410 0.1073</td></tr><tr><td>S0 A0 - S1 A1</td><td>0.854</td><td>1.647 20 0.518 0.9537</td></tr><tr><td>S1 A0 - S0 A1</td><td>4.881</td><td>1.647 20 2.964 0.0355</td></tr><tr><td>S1 A0 - S1 A1</td><td>1.766</td><td>1.647 20 1.072 0.7098</td></tr><tr><td>S0 A1 - S1 A1</td><td>-3.115</td><td>1.647 20 -1.892 0.2629</td></tr></table>
+
+Year = 2020: 
+
+<table><tr><td>contrast</td><td>estimate</td><td>SE df t.ratio p.value</td></tr><tr><td>S0 A0 - S1 A0</td><td>-0.662</td><td>1.429 20 -0.463 0.9662</td></tr><tr><td>S0 A0 - S0 A1</td><td>3.428</td><td>1.429 20 2.399 0.1096</td></tr><tr><td>S0 A0 - S1 A1</td><td>0.961</td><td>1.429 20 0.673 0.9062</td></tr><tr><td>S1 A0 - S0 A1</td><td>4.090</td><td>1.429 20 2.862 0.0439</td></tr><tr><td>S1 A0 - S1 A1</td><td>1.623</td><td>1.429 20 1.136 0.6723</td></tr><tr><td>S0 A1 - S1 A1</td><td>-2.467</td><td>1.429 20 -1.726 0.3369</td></tr></table>
+
+Year = 2021: 
+
+<table><tr><td>contrast</td><td>estimate</td><td>SE df t.ratio p.value</td></tr><tr><td>S0 A0 - S1 A0</td><td>-0.588</td><td>0.651 20 -0.903 0.8032</td></tr><tr><td>S0 A0 - S0 A1</td><td>0.170</td><td>0.651 20 0.261 0.9936</td></tr><tr><td>S0 A0 - S1 A1</td><td>-0.223</td><td>0.651 20 -0.342 0.9858</td></tr><tr><td>S1 A0 - S0 A1</td><td>0.758</td><td>0.651 20 1.164 0.6553</td></tr><tr><td>S1 A0 - S1 A1</td><td>0.365</td><td>0.651 20 0.561 0.9424</td></tr><tr><td>S0 A1 - S1 A1</td><td>-0.393</td><td>0.651 20 -0.603 0.9298</td></tr></table>
+
+Year = 2022: 
+
+<table><tr><td>contrast</td><td>estimate</td><td>SE df t.ratio p.value</td></tr><tr><td>S0 A0 - S1 A0</td><td>-0.915</td><td>0.459 20 -1.995 0.2230</td></tr><tr><td>S0 A0 - S0 A1</td><td>0.340</td><td>0.459 20 0.741 0.8796</td></tr><tr><td>S0 A0 - S1 A1</td><td>-0.431</td><td>0.459 20 -0.941 0.7837</td></tr><tr><td>S1 A0 - S0 A1</td><td>1.255</td><td>0.459 20 2.736 0.0569</td></tr><tr><td>S1 A0 - S1 A1</td><td>0.484</td><td>0.459 20 1.054 0.7201</td></tr><tr><td>S0 A1 - S1 A1</td><td>-0.771</td><td>0.459 20 -1.681 0.3591</td></tr></table>
+
+Year = 2023: 
+
+<table><tr><td>contrast</td><td>estimate</td><td>SE df t.ratio p.value</td></tr><tr><td>S0 A0 - S1 A0</td><td>-0.953</td><td>0.478 20 -1.995 0.2232</td></tr><tr><td>S0 A0 - S0 A1</td><td>-0.096</td><td>0.478 20 -0.201 0.9970</td></tr><tr><td>S0 A0 - S1 A1</td><td>-1.075</td><td>0.478 20 -2.251 0.1438</td></tr><tr><td>S1 A0 - S0 A1</td><td>0.857</td><td>0.478 20 1.794 0.3054</td></tr><tr><td>S1 A0 - S1 A1</td><td>-0.122</td><td>0.478 20 -0.256 0.9939</td></tr><tr><td>S0 A1 - S1 A1</td><td>-0.979</td><td>0.478 20 -2.050 0.2037</td></tr></table>
+
+<--- Page Split --->
+
+Degrees- of- freedom method: containment P value adjustment: tukey method for comparing a family of 4 estimates  
+
+##R code for Supplementary Table 1 and Supplementary Table 3:  
+
+#Similar code for ER, GEP, Ts, VWC  
+
+> model1 <- lme(sqrt(NEP+28)~ Date \* S \* A, dat, random = ~1 | Plot/Year, weights = varIdent(form = ~1 |Dr), method = "REML")  
+
+> anova(model1) # Perform ANOVA to assess the significance of the model terms  
+
+numDF denDF F- value p- value (Intercept) 1 1172 109831.79 <.0001 Date 1 1172 45.83 <.0001 S 1 20 7.58 0.0123 A 1 20 4.10 0.0565 Date:S 1 1172 0.53 0.4658 Date:A 1 1172 4.78 0.0290 S:A 1 20 0.31 0.5836 Date:S:A 1 1172 0.76 0.3835  
+
+> lemmas(mean) # Load emmeans package for estimated marginal means > emmeans(model1, pairwise \(\sim\) S\*A) # Compare least-squares means for in  
+
+teractions  
+
+Smeans  
+
+S A emmean SE df lower.CL upper.CL 0 0 5.72 0.0371 23 5.64 5.80 1 0 5.79 0.0371 20 5.72 5.87 0 1 5.59 0.0371 20 5.51 5.67 1 1 5.72 0.0371 20 5.65 5.80  
+
+Degrees- of- freedom method: containment Results are given on the sqrt(mu + 28) (not the response) scale. Confidence level used: 0.95  
+
+$contrasts  
+
+contrast estimate SE df t.ratio p.value S0 A0 - S1 A0 - 0.07291 0.0525 20 - 1.389 0.5201 S0 A0 - S0 A1 0.13257 0.0525 20 2.526 0.0860 S0 A0 - S1 A1 - 0.00372 0.0525 20 - 0.071 0.9999 S1 A0 - S0 A1 0.20548 0.0525 20 3.916 0.0044 S1 A0 - S1 A1 0.06919 0.0525 20 1.318 0.5624 S0 A1 - S1 A1 - 0.13629 0.0525 20 - 2.597 0.0749  
+
+Note: contrasts are still on the sqrt(mu + 28) scale Degrees- of- freedom method: containment
+
+<--- Page Split --->
+![](images/Figure_1.jpg)
+
+<center>Fig. 1 Residual Plot for testing heteroscedasticity for NEP. The left figure shows the original model (NEP \(\sim\) Date \* S \* A, random = \~1|Plot/Year/Date) without incorporating the heteroscedasticity module, the right figure shows the current model (NEP \(\sim\) Date \* S \* A, random = \~1|Plot/Year, weights = varIdent(form = \~1 | Dr)) with the heteroscedasticity module. </center>
+
+<--- Page Split --->
+
+The following were the updated figures and tables in new version of MS:  
+
+![](images/Figure_3.jpg)
+
+<center>Fig. 3 The effects of frosts on net primary productivity (NEP) during the non-treatment periods. (a) Main effects of frosts on net primary productivity (NEP, unit: \(\mu \mathrm{mol} \mathrm{m}^{-2} \mathrm{s}^{-1}\) ), (b) and the pairwise comparisons among treatment levels. In this study, the growing seasons excluded the frost measurement periods in both spring and autumn. All statistical results were examined by linear mix-effect models with significance difference was denoted by an asterisk at \(P < 0.05\) . Main effect of autumn frosts on NEP across years were marginally significant \((P = 0.057)\) . The main effect values were presented by the marginal means of the treatment, which are derived from the linear mixed effects models. The standard error measures the uncertainty of the estimated marginal means. </center>
+
+<--- Page Split --->
+
+**Supplementary Table 1** Statistical results (P-values) of linear mixed-effects models on the effects of date, spring frosts (S), autumn frosts (A) and their interactions on soil temperature (Tsoil), volumetric water content (VWC), net ecosystem productivity (NEP), ecosystem respiration (ER) and gross ecosystem productivity (GEP) throughout the growing seasons from 2017 to 2022. The growing season excluded frost measurement periods in this study. P-values displayed in bold indicated the significance at \(P<0.05\) . 
+
+<table><tr><td></td><td>Tsoil</td><td>VWC</td><td>NEP</td><td>ER</td><td>GEP</td></tr><tr><td>Date</td><td>0.456</td><td>0.079</td><td>&lt;0.001</td><td>&lt;0.001</td><td>&lt;0.001</td></tr><tr><td>S</td><td>0.143</td><td>0.944</td><td>0.012</td><td>0.990</td><td>0.237</td></tr><tr><td>A</td><td>0.269</td><td>0.406</td><td>0.057</td><td>0.600</td><td>0.427</td></tr><tr><td>Date×S</td><td>0.502</td><td>0.806</td><td>0.466</td><td>0.748</td><td>0.964</td></tr><tr><td>Date×A</td><td>0.875</td><td>0.935</td><td>0.029</td><td>0.630</td><td>0.657</td></tr><tr><td>S×A</td><td>0.552</td><td>0.902</td><td>0.584</td><td>0.580</td><td>0.915</td></tr><tr><td>Date×S×A</td><td>0.744</td><td>0.778</td><td>0.384</td><td>0.950</td><td>0.723</td></tr></table>
+
+<--- Page Split --->
+
+**Supplementary Table 3** Statistical results (P-values) of linear mixed-effects models on the effects of date (D), spring frosts (S), autumn frosts (A) and their interactions on net ecosystem productivity (NEP), ecosystem respiration (ER) and gross ecosystem productivity (GEP) during the 7-day frost treatment periods (during) and the 7 days before (before) and after the frosts periods (after) across from 2017 to 2023. \(P\) -values displayed in bold indicated the significance at \(P<0.05\) . 
+
+<table><tr><td rowspan="2"></td><td rowspan="2"></td><td colspan="3">NEP</td><td colspan="3">ER</td><td colspan="3">GEP</td></tr><tr><td>Before</td><td>During</td><td>After</td><td>Before</td><td>During</td><td>After</td><td>Before</td><td>During</td><td>After</td></tr><tr><td rowspan="5">Spring</td><td>D</td><td>0.400</td><td>&lt;0.001</td><td>0.003</td><td>&lt;0.001</td><td>&lt;0.001</td><td>&lt;0.001</td><td>&lt;0,001</td><td>&lt;0.001</td><td>&lt;0.001</td></tr><tr><td>S</td><td>0.741</td><td>0.651</td><td>0.422</td><td>0.446</td><td>0.611</td><td>0.831</td><td>0.824</td><td>0.457</td><td>0.114</td></tr><tr><td>A</td><td>0.023</td><td>0.754</td><td>0.795</td><td>0.095</td><td>0.417</td><td>0.549</td><td>0.752</td><td>0.549</td><td>0.975</td></tr><tr><td>D×S</td><td>0.846</td><td>0.036</td><td>0.695</td><td>0.905</td><td>0.747</td><td>0.942</td><td>0.668</td><td>0.415</td><td>0.765</td></tr><tr><td>D×A</td><td>0.013</td><td>0.247</td><td>0.428</td><td>0.695</td><td>0.752</td><td>0.684</td><td>0.436</td><td>0.653</td><td>0.198</td></tr><tr><td></td><td>S×A</td><td>0.200</td><td>0.391</td><td>0.621</td><td>0.031</td><td>0.905</td><td>0.795</td><td>0.964</td><td>0.819</td><td>0.412</td></tr><tr><td></td><td>D×S×A</td><td>0.455</td><td>0.143</td><td>0.512</td><td>0.697</td><td>0.910</td><td>0.963</td><td>0.658</td><td>0.477</td><td>0.639</td></tr><tr><td rowspan="5">Autumn</td><td>D</td><td>&lt;0.001</td><td>&lt;0.001</td><td>&lt;0.001</td><td>&lt;0,001</td><td>&lt;0.001</td><td>&lt;0.001</td><td>&lt;1.001</td><td>&lt;0.001</td><td>&lt;0.001</td></tr><tr><td>S</td><td>0.679</td><td>0.592</td><td>0.530</td><td>0.707</td><td>0.860</td><td>0.875</td><td>0.957</td><td>0.773</td><td>0.582</td></tr><tr><td>A</td><td>0.698</td><td>0.011</td><td>0.385</td><td>0.476</td><td>0.731</td><td>0.395</td><td>0.880</td><td>0.111</td><td>0.983</td></tr><tr><td>D×S</td><td>0.798</td><td>0.800</td><td>0.848</td><td>0.722</td><td>0.802</td><td>0.958</td><td>0.986</td><td>0.812</td><td>0.961</td></tr><tr><td>D×A</td><td>0.790</td><td>0.238</td><td>0.342</td><td>0.620</td><td>0.952</td><td>0.900</td><td>0.931</td><td>0.450</td><td>0.675</td></tr><tr><td></td><td>S×A</td><td>0.930</td><td>0.967</td><td>0.306</td><td>0.831</td><td>0.919</td><td>0.532</td><td>0.938</td><td>0.991</td><td>0.757</td></tr><tr><td></td><td>D×S×A</td><td>0.809</td><td>0.648</td><td>0.759</td><td>0.975</td><td>0.850</td><td>0.751</td><td>0.924</td><td>0.682</td><td>0.994</td></tr></table>
+
+<--- Page Split --->
+![](images/Supplementary_Figure_4.jpg)
+
+<center>Supplementary Fig. 4 Effects of frosts on net primary productivity (NEP) during the spring frost periods. One measurement period included 7 days before, 7 days during and 7 days after the frost treatment. Asterisks (\*) denoted significant differences \((P< 0.05)\) observed in pairwise comparisons. Values were shown by mean and standard error \((n = 6)\) . </center>
+
+<--- Page Split --->
+![](images/Supplementary_Figure_5.jpg)
+
+<center>Supplementary Fig. 5 Effects of frosts on net ecosystem productivity (NEP) during the autumn frost periods. One measurement period included 7 days before, 7 days during and 7 days after the frost treatment. Asterisks (\*) denoted significant differences \((P< 0.05)\) observed in pairwise comparisons. Values were shown by mean and standard error \((n = 6)\) . </center>
+
+<--- Page Split --->
+![](images/Figure_1.jpg)
+
+<center>Appendix Fig. 1 The main effects of frosts on annual mean ecosystem respiration (ER, a) and gross ecosystem productivity (GEP, b) during the growing seasons excluding frost measurement periods. Insets refer to the mean annual main effects across from 2018 to 2023. The main effect values were presented by the marginal means of the treatment, which are derived from the linear mixed effects models. The standard error measures the uncertainty of the estimated marginal means. </center>
+
+<--- Page Split --->

peer_reviews/df915f8909e53f3f2d709aa751c4ec3e7bb52f81abb1679add07da6037e2ba76/supplementary_0_Peer review file/images_list.json

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

peer_reviews/df915f8909e53f3f2d709aa751c4ec3e7bb52f81abb1679add07da6037e2ba76/supplementary_0_Peer review file/supplementary_0_Peer review file.mmd

@@ -0,0 +1,371 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Common anti- cancer therapies induce somatic mutations in stem cells of healthy tissue  
+
+![PLACEHOLDER_0_0]
+  
+
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+
+Reviewer #1 (Remarks to the Author): Expert in somatic mutational signature analysis and genomics  
+
+Kuijk et al., have studied the effect of CapOx (5- FU and Oxaliplatin) and/or radiotherapy on normal stem cells of colon and liver tissues from individuals with either colorectal and or metastatic liver cancers. Their results highlight the tissue- specific respond to chemotherapy treatment across non- cancerous cells. They suggest colon is more vulnerable to chemo treatments, especially platinum- based treatments (e.g., SBS35) compared to liver which seem to be more protected. This vulnerability in turn can increase the risk of secondary cancer incident in colonic tissues.  
+
+Some recent studies of somatic mutations in normal cells using laser capture microdissection (e.g., Moore, et. al., 2021 and Lee- Six et al., 2019) or single cell derived colonies (e.g., Pich et al., 2021) have highlighted the footprints of chemotherapy on mutational landscape and clonal dynamics of normal cells. However, more work in this area is certainly required to systematically measure the effect of chemo on normal cells and to assess the risk of further cancer transformation. Therefore, this work is certainly relevant and can be interesting to the readers of this journal. However, I have listed some comments below to be clarified by the authors:  
+
+- What is the average number of treatment cycles for liver metastases? Biopsies from CRC patients were collected around \(\sim 1.5\) months after latest CapOx, while for liver metastases samples were biopsied \(\sim 15\) months post treatment. How variation to the time of biopsy may explain some of the variation in effect of chemo they have reported between these two groups? The authors indeed have suggested that damaged cells may have been removed from the liver due to longer time from the treatment to the biopsy collection. However, liver stem cell turnover is slower compared to colonic crypts stem cells. Hence, I wonder other factors might be at play such as variation in clonal dynamics between the two tissues?  
+
+- It is quite surprising that the 58 and 76 yrs-old CapOx-treated donors as well as the 5-FU-only treated donor did not show any elevation in SBS burden. How the authors explain this observation?  
+
+- Is there any corelation between SBS35 load and duration of the treatment and or different sections of colon that were biopsied?  
+
+- They have shown that the effect of 5-FU is more heterogenous across the individuals studied. Similar observation was also made in AML cases by Pich et al., 2021. Does the dose of 5-FU vary between individuals and if so, is there any correlation between the dose and SBS17 burden?
+
+<--- Page Split --->
+
+- The observation of sub-clonal platinum-induced mutations is quite unexpected. Although the authors suggested alternative possibility for this observation, yet it highlights that some of the chemo effect might be missed with their method. Can the authors give an estimate of the fraction of chemo effect that might have missed due to their sub-clonality nature?  
+
+- "Our findings reveal that both CapOx chemotherapy and radiotherapy are mutagenic in colorectal ASCs and significantly increase the mutational burden in normal non-cancerous cells beyond typical age-related mutation accumulation." It would be great if the authors can expand on this observation by providing more insights into the type of mutations with respect to their functional consequences.  
+
+Minor comments:  
+
+- Overall, the way the experimental design has described is slightly confusing. It is unclear to me how many colonies were sequenced per individual. A simple summary table to include this information as well as, type of treatment, dosage, length of treatment and the time interval from the last treatment to sample collection would be very helpful for the readers.  
+
+- Fig. 2 left and write panels are not labelled so I assumed the left panel is colon and right panel is liver. In Fig. 2f, the anti-correlation of indel with age even in healthy individuals is unexpected.  
+
+Reviewer #2 (Remarks to the Author): Expert in colorectal cancer organoids, therapy, and stem cells  
+
+The manuscript by Kuijk et al. identifies the mutational landscape in normal colonic and liver cells of CRC patients after being treated with several cycles of therapy (chemo and/or radiotherapy). This is an relevant study that addresses the key question of how normal tissue is damaged after treatment. The authors use whole genome sequencing to demonstrate that the mutational pattern in the liver does not change significantly after chemo but intestinal cells are affected (mainly by Oxaliplatin).  
+
+The work is original as the mutational landscape by NGS has not been well reported in normal intestinal and liver cells. It is an interesting study, with a good experimental set up. The group has an excellent
+
+<--- Page Split --->
+
+background in NGS/cancer genetics and in identifying mutational landscapes of different cell types. The results are well interpreted, however a more exhaustive report would need to be addressed.  
+
+Major points:  
+
+1. My main concern is the low number of CRC samples analysed and the poor description in the text about the samples/crypts that were sequenced. Are the results statistically significant? Can they do further analyses to be sure that the samples are representative? If they do power calculations/sample size, what would be the result? The liver data looks convincing but the authors would need to be sure that the rest shown is also representative.  
+
+2. Why the authors discuss the 5FU influence as it has an impact, when 5 out of 7 show no effect and another individual only in one of the samples?  
+
+3. Colorectal cancer histology/pathology is not shown/described however, which part of the colon the samples were resected from? It is important to know because may add more variability to the N analyzed (in addition to the age factor which is cover a big range). The text does not address it and this may have an impact. Did the authors see any difference in SBS and ID mutations by anatomical site in the colon and rectum? this may need to be discussed as a potential limitation.  
+
+4. In my opinion, the telomere section is roughly described and data are not included in the abstract/main figures. Makes sense to leave it? I would suggest to leave it for a better analysis.  
+
+5. The irradiation findings are not so novel as have already been described in similar approaches/cancers (e.g. 10.1038/ncomms12605). Authors refer to this in the introduction but do not discuss it later on.  
+
+6. The crypts were expanded from few cells, but as confirmed by the authors the results show that there could be more than one stem cell in each sample before expansion. Did they analyse this in depth? Using stem cell markers, flow cytometry and quantification? through-out the text is written continuously that they are working with single adult stem cells, is this always true? For example, figure legend states "single adult SCs solutions", I think this can take the reader to confusion.  
+
+7. A similar set up was done by Halazonetis group (Switzerland) where single crypts were expanded into organoids from mouse samples (AKP) before sequencing. A previous manuscript from the same group using APC min also revealed the mutation change when crypts were cultured more than 4 months.
+
+<--- Page Split --->
+
+Authors should mention and compare results. The authors should also look at the AKP manuscript to do a scheme similar to show the number of crypts expanded/sequenced from each sample.  
+
+Minor points:  
+
+1. Graphs in figure 3 and 4 are too small. Supplementary data should also be shown with a bigger size.  
+
+2. The statistics section is poorly described. The normal and non-normal distribution should be better study than doing "assumptions" or "likely".  
+
+3. It should be written in the abstract how many individuals for each sample type were used to do the study.  
+
+4. Methods section: "Sorbitol and 0.5mM dithiothreitol (all purchased from Merck) until the supernatant was." (words are lacking here).  
+
+Reviewer #3 (Remarks to the Author): Expert in colorectal and hepatic organoids, stem cells  
+
+Kuijk et al. analyzed mutation patterns in chemotherapy- treated colon/liver tissue using organoid technology and estimated the number of mutations induced by chemotherapies. This is a fascinating study based on a valuable dataset from clinical specimens. The finding is novel in that chemotherapy- induced mutations in a tissue- specific manner. I would like to recommend its publication in Nature Communications, but several points are to be addressed as follows.  
+
+1. The "liver organoids" used in this study are actually derived from intrahepatic cholangiocytes and thus should be referred to as intrahepatic cholangiocyte organoids (ICO) (Marsee et al. Cell Stem Cell 2021).  
+
+2. As described in the discussion, the weak point of this paper is a limited number of specimens, and it is unclear why only a fraction of organoid clones showed the 5-FU signature. Are the two lines of
+
+<--- Page Split --->
+
+organoids that showed the 5- FU signature associated with radiation therapy (or rectum origin)? I think the authors should provide experimental evidence that  
+
+3. ICOs were devoid of platinum signatures. The authors provided an interesting explanation that liver ASCs have more effective inherent mechanism to protect against chemo-induced mutagenesis. However, there is an alternative explanation that all liver ASCs that gained platinum signature died out or underwent senescence in vivo (thus, they cannot be recovered as organoids). In addition, liver ASCs are non-cycling, and non-cycling cells may gain fewer mutations by platinum drugs? To exclude these possibilities, the authors should treat ICO with platinum and determine whether ICO can gain platinum signature when proliferating in vitro.  
+
+4. The authors excluded escaping from drugs in liver ASCs. However, oxaliplatin is a renal excretion drug. Why didn't they consider this possibility?  
+
+## Minor points  
+
+1. Fig.1 is not very self-explanatory. I think it is better to label colons and cholangiocytes in the figure. The color codes for 5-FU+platinum+radiation and 5-FU+platinum are very similar, and I cannot distinguish the two colors. I cannot see which plots are significant in the figure.
+
+<--- Page Split --->
+
+## Response to REVIEWER COMMENTS  
+
+## Reviewer #1 (Remarks to the Author): Expert in somatic mutational signature analysis and genomics  
+
+Kuijk et al., have studied the effect of CapOx (5- FU and Oxaliplatin) and/or radiotherapy on normal stem cells of colon and liver tissues from individuals with either colorectal and or metastatic liver cancers. Their results highlight the tissue- specific respond to chemotherapy treatment across noncancerous cells. They suggest colon is more vulnerable to chemo treatments, especially platinum- based treatments (e.g., SBS35) compared to liver which seem to be more protected. This vulnerability in turn can increase the risk of secondary cancer incident in colonic tissues.  
+
+Some recent studies of somatic mutations in normal cells using laser capture microdissection (e.g., Moore, et. al., 2021 and Lee- Six et al., 2019) or single cell derived colonies (e.g., Pich et al., 2021) have highlighted the footprints of chemotherapy on mutational landscape and clonal dynamics of normal cells. However, more work in this area is certainly required to systematically measure the effect of chemo on normal cells and to assess the risk of further cancer transformation. Therefore, this work is certainly relevant and can be interesting to the readers of this journal.  
+
+However, I have listed some comments below to be clarified by the authors:  
+
+- What is the average number of treatment cycles for liver metastases? Biopsies from CRC patients were collected around \(\sim 1.5\) months after latest CapOx, while for liver metastases samples were biopsied \(\sim 15\) months post treatment. How variation to the time of biopsy may explain some of the variation in effect of chemo they have reported between these two groups? The authors indeed have suggested that damaged cells may have been removed from the liver due to longer time from the treatment to the biopsy collection. However, liver stem cell turnover is slower compared to colonic crypts stem cells. Hence, I wonder other factors might be at play such as variation in clonal dynamics between the two tissues?  
+
+Clinical details are described in Suppl table 1 including the number of treatment cycles. We have also included this information in figure 1 panel A. The liver patients included in this study have undergone more CAPOX treatment cyclic than the colon patients (CAPOX treatment cyclic of \(5.8 \pm 3.5\) for liver patients and \(3.2 \pm 1.7\) for colon patients; mean \(\pm\) standard deviation [SD]) and thus the lack of treatment- related mutations in liver ASCs cannot be explained by a lower number of treatment cycles. We have now included this sentence in the main text.  
+
+On average, there is indeed more time between the end of CAPOX treatment regime and surgical removal of tumor and surrounding healthy tissue (= time of biopsy) for liver patient than colon patients. However, healthy tissue of 2 liver donors (Donor 8, 10) was collected 1 month after the end of CAPOX treatment, which is below the average of 1.5 month for colon donors. All 5 liver ASCs of these 2 liver donors showed a complete absence of CAPOX induced mutations as well, indicating that time between treatment and biopsy is likely not the cause for the lack of CAPOX related mutations for liver ASCs. However, we cannot formally exclude the possibility of a highly efficient clearance of treatment damaged healthy liver ASCs. We consider this hypothesis unlikely because 5-
+
+<--- Page Split --->
+
+FU or platinum therapies rarely induce liver damage<sup>1</sup>. We have added this text to the discussion section of the manuscript.  
+
+Given the very slow liver stem cell turnover rate, we expect that healthy liver tissue harbour different clonal compositions than colon tissue. Sequencing a clonal group of nearby cells that have recently derived from a single (adult stem) cell (e.g. by micro laser biopsy sampling as conducted in Moore et al 2021) could shed more light on clonal dynamics of treatment- induced mutations. However, we used a single- cell based whole genome sequencing methodologies that integrate clonal expansion of a single cell and thus clonal dynamics cannot be studied with this approach.  
+
+Although it has been demonstrated that treatment changes the clonal architecture in cancer tissue due to selection of treatment driver resistance mechanisms (Martinez- Jimenez et al., 2022 BioRxiv), it may induce similar effects in healthy tissue. However, we have not found a different driver landscape between treated and untreated tissue in either liver or colon adult stem cells. Nevertheless, the number of samples we have sequenced here are not sufficient to be fully conclusive, but we also believe that studying clonal dynamics, although very interesting, would require different experimental approaches that are beyond the scope of this study.  
+
+- It is quite surprising that the 58 and 76 yrs-old CapOx-treated donors as well as the 5-FU-only treated donor did not show any elevation in SBS burden. How the authors explain this observation?  
+
+There is natural variation in the correlation between the age and the number of mutations, which increases with age. This means that the older the patients are, the more difficult it is to get significant effects of increases in mutational burden as illustrated by higher standard deviation at older age in Figure 2a and 2b. Thus, while we can clearly identify treatment- induced mutations in these donors by mutational signature analysis, the effect of the increase in mutations as a result of the treatment is not large enough to become apparent by mutational burden analysis alone.  
+
+We have clarified this by changing the results section as follows: "The SBS mutation burden of the 58 and 76 year old CapOx- treated donors, as well as the 5- FU- only treated donor, was within the natural variation in SBS mutation burden in the correlation with the age"  
+
+- Is there any correlation between SBS35 load and duration of the treatment and or different sections of colon that were biopsied?  
+
+There was no correlation between SBS35 mutation load and treatment duration. See comments on reviewer 2 related to colon anatomical site in the colon and rectum.  
+
+- They have shown that the effect of 5-FU is more heterogenous across the individuals studied. Similar observation was also made in AML cases by Pich et al., 2021. Does the dose of 5-FU vary between individuals and if so, is there any correlation between the dose and SBS17 burden?  
+
+Unfortunately, treatment dose information was not available for the current study. Nevertheless, only 5 of 28 colon ASCs showed 5-FU related mutations. Of these 5 colon ASCs, 4 (of 4) samples
+
+<--- Page Split --->
+
+came from the 24- year old donor and 1 (of 3) of the 66- year old donor. Thus, even if the 5- FU dose of each patient was shared, we would not observe any relationship between the given 5- FU dose and SBS- 17 mutation burden because \(\sim 85\%\) of the 5- FU exposed colon ASC show a complete absence of 5- FU related mutations. Moreover, internal 5- FU dose is also heavily influenced by pharmacogenetics as patients with (partial) DPD deficiency are less efficient in 5- FU clearance and typically display 5- FU toxicity during treatment (doi: 10.1016/j.ejca.2003.12.004). Finally, we observe some variability within the 24 year old patient, indicating that the variation in mutation accumulation is independent of 5- FU dosage.  
+
+- The observation of sub-clonal platinum-induced mutations is quite unexpected. Although the authors suggested alternative possibility for this observation, yet it highlights that some of the chemo effect might be missed with their method. Can the authors give an estimate of the fraction of chemo effect that might have missed due to their sub-clonality nature?  
+
+This information can be depicted from Suppl Figure 12 where we computed the \(\%\) increase in treatment related mutations when considering all (clonal + subclonal) treatment mutations as compared to the clonal treatment mutations (VAF \(>30\%\) ), which is equivalent to the fraction of missed 5- FU and platinum mutations to their subclonality. From the 25 oxaliplatin treated colon ASCs, 7 ASCs showed no increase in mutation load between all and clonal mutations (no missed subclonal treatment mutations), 16 showed an increase of \(\sim 50\%\) in platinum mutation load and 2 samples showed an increase of \(100\%\) or more in both platinum SBS and DBS mutations.  
+
+We have adapted the main text as follows: "Nevertheless, only 2 ASCs (from the 24- year- old donor) showed a remarkable subclonal platinum mutation contributions of \(100\%\) or more and the inclusion of subclonal platinum mutations had a low impact (less than \(50\%\) ) on the platinum mutational load for all other (n=25) ASCs (Supp. Fig. 12b,c)."  
+
+- "Our findings reveal that both CapOx chemotherapy and radiotherapy are mutagenic in colorectal ASCs and significantly increase the mutational burden in normal non-cancerous cells beyond typical age-related mutation accumulation." It would be great if the authors can expand on this observation by providing more insights into the type of mutations with respect to their functional consequences. This is indeed a very intriguing question. Although our approach is highly sensitive to detect the passenger mutational consequences of anticancer therapy, it is not optimal to identify potential drivers of secondary malignancies due to the low sample sizes and the relatively small proportion of coding sequence in the genome. Modelling functional impact and risks based on the limited number of coding mutations in our data set would result in highly unreliable risk estimates and thus incorrectly inform patients.  
+
+Minor comments:  
+
+- Overall, the way the experimental design has described is slightly confusing. It is unclear to me how many colonies were sequenced per individual. A simple summary table to include this information as well as, type of treatment, dosage, length of treatment and the time interval from the last treatment to sample collection would be very helpful for the readers.
+
+<--- Page Split --->
+
+Besides dosage, all this information is included in Suppl table 1 which we now have summarized in Panel A of main figure 1. We have also more comprehensively explained the experimental set up in the main text.  
+
+- Fig. 2 left and write panels are not labelled so I assumed the left panel is colon and right panel is liver.  
+
+Thank you for noticing. We have adapted the figure accordingly.  
+
+In Fig. 2f, the anti- correlation of indel with age even in healthy individuals is unexpected. In general, there is no strong correlation between INDEL load and age in the liver. For example, the healthy (and diseased) liver tissue study from Brunner et al 2019 shows a slightly higher SBS mutation load in the older healthy liver samples while the indel mutation load remains constant. Also in our current study, this correlation is weak/not significant and we therefore caution not to overinterpret the figure based on visual inspection.  
+
+## Reviewer #2 (Remarks to the Author): Expert in colorectal cancer organoids, therapy, and stem cells  
+
+The manuscript by Kuijk et al. identifies the mutational landscape in normal colonic and liver cells of CRC patients after being treated with several cycles of therapy (chemo and/or radiotherapy). This is an relevant study that addresses the key question of how normal tissue is damaged after treatment. The authors use whole genome sequencing to demonstrate that the mutational pattern in the liver does not change significantly after chemo but intestinal cells are affected (mainly by Oxaliplatin).  
+
+The work is original as the mutational landscape by NGS has not been well reported in normal intestinal and liver cells. It is an interesting study, with a good experimental set up. The group has an excellent background in NGS/cancer genetics and in identifying mutational landscapes of different cell types. The results are well interpreted, however a more exhaustive report would need to be addressed.  
+
+Major points:  
+
+1. My main concern is the low number of CRC samples analysed and the poor description in the text about the samples/crypts that were sequenced.  
+
+We have sequenced and included an additional set of 16 samples (13 from colon cohort and 3 from liver cohort) leading to a final set of 42 ASC samples. Overall, findings remain unchanged except that mutational signature analysis does not detect platinum related mutation in 2 additional samples of the 76- year old colon donor (n=3). This may be explained by the low number of CAPOX treatment cycles because this patient has undergone only a single treatment cycle. Therefore, the number of platinum- induced mutations is under the detection limit of mutation signature analysis. The information on the samples sequenced can be found in Suppl table 1 and as a table in Panel A of revised figure 1.
+
+<--- Page Split --->
+
+Are the results statistically significant? Can they do further analyses to be sure that the samples are representative? If they do power calculations/sample size, what would be the result? The liver data looks convincing but the authors would need to be sure that the rest shown is also representative.  
+
+In this study we relied on a sample size set similar to other studies that also used healthy tissue derived organoids to investigate mutation accumulation (Jager et al 2019, Drost et al 2017, Kuijk et al 2020, Nguyen et al, 2022). To further explore the sample size of the current study, we have conducted a power analysis as in Nguyen et al, 2022. For this, we first measured the Cohen's D to compute the standardized mean difference of the mutation load between platinum treated and untreated donors. Using the cohen.D function from effsize R package, we found a normalized effect size of 1.51. A similar normalized effect size (Cohen's D = 1.52) was observed when we assessed the excess in mutation load (observed mut load - predicted mut load using linear mixed model regression analysis) between platinum treated and untreated donors.  
+
+To assess the reliability of measured Cohen's D scores on mutation load given our number of samples, we performed a power analysis using the pwr.t2n.test() function of the pwr R package to calculate hypothetical detectable effect sizes using a significance level of 0.05 and a power of 0.8. It is generally accepted that power should be approximately \(80\%\) . Given that we have 5 untreated patients and 6 treated patients, we expect to be able to detect an effect size (Cohen's D) of 1.6 which is approximately equal to the computed Cohen's D values on the mutation load and excess mutation load. Thus, the sample sizes in our study would allow us to detect effect sizes as hypothetical detectable effect sizes with a power score of \(80\%\) .  
+
+Similarly, we applied this approach on the platinum and 5- FU mutation contributions. Using the SBS35 and SBS17 mutation contributions, we obtained Cohen's D scores of 2.7 (large) and 0.7 (medium) for platinum and 5- FU mutation contributions, respectively, between the treated and untreated donor group. The Cohen's D score on the platinum DBS (DBS5) contribution was 2.5, which is similar as the SBS35 mutation contribution. Thus, the normalized effect size for platinum mutation load is well beyond the hypothetical detectable effect size given the number of samples (1.6), but not for 5- FU mutation contribution. This is in line with our observations that 5- FU mutations were only detected in few samples of 2 (of 7) donors.  
+
+Lastly, we found that the Cohen's D score for radiation induced indels (ID8) between untreated and radiation treated donors was also higher (ie 1.8) than the hypothetical detectable effect size with higher power.  
+
+We have included this in Methods section and all details in a Suppl Note:  
+
+## Power Analysis  
+
+In this study we relied on a sample size set similar to other studies that also used healthy tissue derived organoids to investigate mutation accumulation. To further explore the sample size of the
+
+<--- Page Split --->
+
+current study, we have conducted a power analysis as in Nguyen et al, 2022. Details about the power analysis can be found in Suppl Note 1.  
+
+2. Why the authors discuss the 5FU influence as it has an impact, when 5 out of 7 show no effect and another individual only in one of the samples?  
+
+The specificity of the 5FU mutational signature allows us to very accurately determine the presence of 5FU- induced mutations. As we have demonstrated previously, this enables the detection of 5FU induced mutations in whole genome sequencing data of 5FU treated colorectal cancer patients, albeit not in every patient, the reason for which is unknown (Christensen et al). In our current study, most of the cells did not show 5FU induced mutations. However, based on the highly characteristic mutational footprint of 5FU, we are very confident that \(18\%\) of the colon cells harbor 5-FU induced mutations. We prefer not to trivialise this effect, because in some cells 5FU treatment equals \(\sim 10\) years of mutations caused by normal age- related processes. Therefore, we can conclude that 5FU can have a mutational impact on healthy cells and may cause secondary malignancies later in life. Fortunately for the patients, most healthy cells escape 5FU induced mutagenesis, which is also an important message. We have modified the abstract to address that in most of the colon ASCs we did not observe 5-FU induced mutations: "The effects of 5-FU treatment were more variable between cells and patients, ranging from a complete absence of 5-FU mutagenesis in \(>80\%\) of the individual colon ASCs to the accumulation of up to 500 mutations in the remaining individual colon ASCs."  
+
+3. Colorectal cancer histology/pathology is not shown/described however. which part of the colon the samples were resected from? it is important to know because may add more variability to the N analyzed (in addition to the age factor which is cover a big range). The text does not address it and this may have an impact. Did the authors see any difference in SBS and ID mutations by anatomical site in the colon and rectum? this may need to be discussed as a potential limitation.  
+
+Clinical info on tumor type and sublocation is included in Suppl table 1 and in panel A of figure 1. We have changed the column names accordingly in order to better interpret the clinical details.  
+
+The study from Lee- Six have shown that age- related mutations vary between anatomical location of the colon tissue, because of a slightly lower proliferation rate at the end of the colon ( \(\sim 12\) SBS1 mutations per year) compared to the right colon section ( \(\sim 18\) SBS1 mutations per year). Because all our samples are from the descending/sigmoid/rectal parts of the colon and not from the ascending/transverse sections, the effect of location on variation is minimal. Moreover, any putative location- dependent variation cannot explain the complete absence of 5-FU mutations in \(85\%\) of the colon ASCs. Because, our results demonstrate that chemotherapy mutation accumulation varies between tissues, this may also be true for the ascending and transverse sections of the colon, which were not examined in the present study. We have therefore included this possibility in the discussion section of our revised manuscript: "future work should aim to further elucidate the mutagenic effects of chemotherapies on other organ systems and organ anatomical sublocations"  
+
+4. In my opinion, the telomere section is roughly described and data are not included in the abstract/main figures. Makes sense to leave it? I would suggest to leave it for a better analysis.
+
+<--- Page Split --->
+
+We agree with the reviewer and we removed this section in the revised version of our manuscript.  
+
+5. The irradiation findings are not so novel as have already been described in similar approaches/cancers (e.g. 10.1038/ncomms12605). Authors refer to this in the introduction but do not discuss it later on.  
+
+The study from Behjati et al., is indeed the first study to assess radiation- induced mutagenesis in which they reported an increase in mutation contribution of an indel signature (sized between 1- 100 base pairs) and a balanced inversion signature. We have referred to this publication in radiation section as follows:  
+
+This signature was mostly similar to COSMIC ID- 8 (cos sim =0.66). ID- 8 mutations were previously found to be enriched in human cancers following radiation therapy \(^{2,3}\) . In addition to small deletions, we investigated SVs because radiation- treated tumors show radiotherapy- associated increases of inversions and large- scale deletions \(^{2,3}\) .  
+
+As with platinum and 5- FU mutational signatures, we do not claim that these radiotherapy- related signatures are novel, but we use these signatures as barcode to also quantify the mutagenic impact of radiotherapy in healthy rectum adult stem cells, which has not been described before.  
+
+6. The crypts were expanded from few cells, but as confirmed by the authors the results show that there could be more than one stem cell in each sample before expansion. Did they analyse this in depth? Using stem cell markers, flow cytometry and quantification? through- out the text is written continuously that they are working with single adult stem cells, is this always true? For example, figure legend states "single adult SCs solutions", I think this can take the reader to confusion.  
+
+Most of the treatment induced neutral passenger mutations are only present in one or a few cells. Therefore, sequencing and characterizing the genome- wide in vivo mutations of a single cell enables accurate quantification of the mutational contribution from late active mutation processes just before tissue sampling. Recently, we optimized and presented an alternative method for cataloging mutations in individual human ASCs without the necessity of using error- prone whole- genome amplification [5]. Here, tissue derived single ASCs are expanded in vitro into clonal organoid cultures to generate sufficient DNA for accurate WGS. The culture conditions have been optimized to promote the growth of intestinal stem cells over other cell types such as differentiated cells or stromal cells. This methods therefore particularly well suited to study and quantify treatment- induced mutations in ASCs in vivo. However, performing a clonal step is highly challenging with colon organoids that are not readily expanded from single cells, particularly from the relatively small biopsies used for the present study. Therefore, we adapted the protocol to maximize the chance of clonal expansion, by first establishing the culture followed by severe fragmentation and serial dilutions at the first split, where individual tiny fragments are most likely derived from individual cells at isolation or have a very recent common ancestor in vivo. As is evident from our observation that we can detect treatment induced mutations, this method is adequate for capturing very recent mutations. However, in rare cases, the small cell clump that gave rise to the clonal culture may
+
+<--- Page Split --->
+
+represent more than one stem cell of the crypt at the time of isolation. As experimental evidence of the success of the clonal step we examined the VAF peaks. The clonal step worked as expected in all of the lines as illustrated by a highly characteristic VAF- peak at \(50\%\) across all samples. Lower VAF peaks indicate that mutations are not shared by the single stem cell ancestor and are typically obtained by laser- capture microdissection of crypts (See for example Suppl Fig 1d,e in Lee- six et al., 2019).  
+
+The VAF- peaks at \(50\%\) , and the strong decrease of mutations at VAF at \(30\%\) , indicate that we have assessed somatic mutations of single adult stem cells (see also protocol guidelines in Jager et al., \(2017^{4}\) ). Moreover, the absence of subclonal treatment- induced mutations in \(\sim 85\%\) of colon ASC samples confirms the clonal outgrowth of a single ASCs. However, 2 out of 25 platinum treated colon samples showed subclonal platinum DBS and SBS mutations with similar contribution as clonal platinum DBS and SBS mutations. This indicates a fairly recent single stem cell ancestor at the beginning of the treatment but diverged into 2 (or more) stem cells during the course of the treatment. In these two clones we cannot exclude the possibility that these have been derived from two or more cells resulting in an underestimation of the real mutational impact of the anticancer therapy.  
+
+We have adapted the introduction and methods section of the manuscript to better explain the technical details of the protocol and included the experimental VAF analysis in the results.  
+
+7. A similar set up was done by Halazonetis group (Switzerland) where single crypts were expanded into organoids from mouse samples (AKP) before sequencing. A previous manuscript from the same group using APC min also revealed the mutation change when crypts were cultured more than 4 months. Authors should mention and compare results. The authors should also look at the AKP manuscript to do a scheme similar to show the number of crypts expanded/sequenced from each sample.  
+
+The mutational impact of culturing in human adult stem cells from colon and liver tissue have been well- documented in a previous study by our group<sup>5</sup>. In that study we have shown that in vitro culturing induces SBS- 18 mutations which are related to oxidative stress, and that low- oxygen culturing conditions decrease SBS- 18 mutation rate during culturing. We minimized the time in culture 7- 10 days before we performed the clonal step, which is far below the 4 month period of the study by the Halazonetis group. The potential small contribution of the derivation/culturing before the clonal step was applied is negligible (see also validation in our in vivo study<sup>6</sup>). Moreover, culturing induced mutations have a very different mutation context than the highly specific 5- FU and platinum- induced mutations (see also Christensen et al where we identified a clear 5FU signature apart from the in vitro signature by the in vitro treatment of intestinal organoids). Thus, even the few in vitro induced mutations that end up in the mutation matrices from clonal mutations (representing the in vivo induced mutations) will have no impact on the quantitative assessment of treatment induced mutations.  
+
+Finally, any in vitro induced mutations after the clonal step are subclonal after sequencing the bulk isogenic organoid culture and are discarded in downstream analysis.
+
+<--- Page Split --->
+
+Minor points:  
+
+1. Graphs in figure 3 and 4 are too small. Supplementary data should also be shown with a bigger size.  
+
+Main figure 3 and 4 as well as Suppl figures have been adapted.  
+
+2. The statistics section is poorly described. The normal and non-normal distribution should be better study than doing "assumptions" or "likely".  
+
+Thank you for pointing this. We have revised the statistical section.  
+
+3. It should be written in the abstract how many individuals for each sample type were used to do the study.  
+
+We have included the mean number of sequenced ASC samples per donor in the abstract. A complete overview on sample size per donor is now also provided in main figure 1 and in supplementary table 1.  
+
+4. Methods section: "Sorbitol and 0.5mM dithiothreitol (all purchased from Merck) until the supernatant was." (words are lacking here).  
+
+Thank you for pointing this out. We have completed this particular sentence ("...until the supernatant was clear.")  
+
+Reviewer #3 (Remarks to the Author): Expert in colorectal and hepatic organoids, stem cells  
+
+Kuijk et al. analyzed mutation patterns in chemotherapy- treated colon/liver tissue using organoid technology and estimated the number of mutations induced by chemotherapies. This is a fascinating study based on a valuable dataset from clinical specimens. The finding is novel in that chemotherapy- induced mutations in a tissue- specific manner. I would like to recommend its publication in Nature Communications, but several points are to be addressed as follows.  
+
+1. The "liver organoids" used in this study are actually derived from intrahepatic cholangiocytes and thus should be referred to as intrahepatic cholangiocyte organoids (ICO) (Marsee et al. Cell Stem Cell 2021).  
+
+Because colon stem cells and intrahepatic cholangiocyte organoids are both adult stem cells (ASCs) we use this latter terminology. To further specify the type of liver ASCs we use the suggested terminology and refer to the suggested paper at the end of the introduction: "We studied mutations in ASCs derived from the slowly renewing liver (cultured as intrahepatic cholangiocyte organoids)".  
+
+2. As described in the discussion, the weak point of this paper is a limited number of specimens
+
+<--- Page Split --->
+
+To overcome this limitation, we have sequenced and analyzed extra samples and performed a power analysis. See also our answers to the comments by reviewer 2.  
+
+3. and it is unclear why only a fraction of organoid clones showed the 5-FU signature. Are the two lines of organoids that showed the 5-FU signature associated with radiation therapy (or rectum origin)?  
+
+Of the 5 colon ASCs that show a contribution of the 5-FU mutational signature, 4 (of 4) samples came from the 24-year old donor and 1 (of 3) of the 66-year old donor. Both donors were diagnosed with rectum and thus treated with radiation, opening the possibility that 5-FU signature could be associated with these covariates. However, 5-FU related mutations have been documented in colon tissue in patients not being treated with radiation (https://doi.org/10.1101/2021.04.14.437578; and vice versa, no enrichment of 5-FU related mutations (SBS17a/b) have been described in radiation treated cancers \(^{2,3}\) . These results indicate that 5-FU and radiation mutation processes operate independently.  
+
+I think the authors should provide experimental evidence that ICOs were devoid of platinum signatures.  
+
+Platinum leads to interstrand crosslinks and thereby causes highly characteristic DBS mutations as has been experimentally demonstrated in platinum- treated isogenic pluripotent stem cells \(^{8}\) . Double base mutations are highly informative because they include two neighbouring SBS mutations located on the same strand. These platinum related mutation scars are thus highly robust to assess the mutational impact of platinum. In Suppl Figure 8, we show that all sequenced liver ASCs (ICOs) don't show any increase of CT>AT and CT>AA DBS mutations compared to untreated liver ASCs. In sharp contrast, nearly all colorectal ASCs from patients treated with platinum show these DBSs (Fig. 3d- e).  
+
+The authors provided an interesting explanation that liver ASCs have more effective inherent mechanism to protect against chemo- induced mutagenesis. However, there is an alternative explanation that all liver ASCs that gained platinum signature died out or underwent senescence in vivo (thus, they cannot be recovered as organoids).  
+
+This hypothesis was also included in the results section: "Damaged cells may have been effectively cleared from the liver because the time between treatment and collection was mostly longer for the liver than for the colon, even though also no mutations were observed in the 2 liver samples that were collected 1 month after treatment." See also comment 1 from reviewer 1.  
+
+In addition, liver ASCs are non- cycling, and non- cycling cells may gain fewer mutations by platinum drugs? To exclude these possibilities, the authors should treat ICO with platinum and determine whether ICO can gain platinum signature when proliferating in vitro.  
+
+We agree with the reviewer that more focused in vitro validation experiments must be conducted to dissect the mechanisms of chemotherapy induced mutation accumulation and the role of proliferation in this process. However, these mechanistic studies are beyond the scope of the
+
+<--- Page Split --->
+
+current manuscript that is focused on the in vivo mutational impact of anticancer therapies in different organ systems. We have therefore decided to discuss our results in the light of previous studies (Oriol et al., 2021 Nature communications; Bertrums et al., 2022 Cancer Discover). Together, these findings indicate that platinum mutation effect is independent of cell proliferation.  
+
+We have included the following sentence to the discussion part: "Similar observations in mutational effect between platinum and nucleobase analogue (5- FU, thiopurines) drugs has been reported in treated hematopoietic cells<sup>9,10</sup> and well- controlled in vitro validation experiments will dissect the mechanisms of chemotherapy- induced mutation accumulation and the role of proliferation in this process"  
+
+4. The authors excluded escaping from drugs in liver ASCs. However, oxaliplatin is a renal excretion drug. Why didn't they consider this possibility?  
+
+A highly efficient mechanism clearance of (oxali)platin based drugs from the vascular system (via renal excretion) could explain the absence of platinum mutations in liver ASCs, but it is not in line with the apparent and robust mutation effect in colon ASCs. Moreover, after uptake in the bloodstream, drugs and nutrients will travel through the portal vein into the liver, thereby exposing all liver cells to the platinum before it can be renally excreted.  
+
+## Minor points  
+
+1. Fig.1 is not very self-explanatory. I think it is better to label colons and cholangiocytes in the figure. The color codes for 5-FU+platinum+radiation and 5-FU+platinum are very similar, and I cannot distinguish the two colors. I cannot see which plots are significant in the figure.  
+
+We have changed Fig 1 accordingly.  
+
+1. Andrade, R. J. et al. Drug-induced liver injury. Nat Rev Dis Primers 5, 58 (2019).  
+
+2. Kocakavuk, E. et al. Radiotherapy is associated with a deletion signature that contributes to poor outcomes in patients with cancer. Nat. Genet. 53, 1088-1096 (2021).  
+
+3. Behjati, S. et al. Mutational signatures of ionizing radiation in second malignancies. Nat. Commun. 7, 12605 (2016).  
+
+4. Jager, M. et al. Measuring mutation accumulation in single human adult stem cells by whole-genome sequencing of organoid cultures. Nat. Protoc. 13, 59-78 (2018).  
+
+5. Kuijk, E. et al. The mutational impact of culturing human pluripotent and adult stem cells. Nat.
+
+<--- Page Split --->
+
+Commun. 11, 2493 (2020).  
+
+6. Blokzijl, F. et al. Tissue-specific mutation accumulation in human adult stem cells during life. Nature 538, 260–264 (2016).  
+
+7. Marsee, A. et al. Building consensus on definition and nomenclature of hepatic, pancreatic, and biliary organoids. Cell Stem Cell 28, 816–832 (2021).  
+
+8. Kucab, J. E. et al. A Compendium of Mutational Signatures of Environmental Agents. Cell 177, 821–836.e16 (2019).  
+
+9. Bertrums, E. J. M. et al. Elevated mutational age in blood of children treated for cancer contributes to therapy-related myeloid neoplasms. Cancer Discov. (2022) doi:10.1158/2159-8290.CD-22-0120.  
+
+10. Pich, O. et al. The evolution of hematopoietic cells under cancer therapy. Nat. Commun. 12, 4803 (2021).
+
+<--- Page Split --->
+
+## REVIEWERS' COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+The authors have address my questions and comments adequately. I have no further comments and I am happy with the revised version of the manuscript to go ahead for publication.  
+
+Reviewer #2 (Remarks to the Author):  
+
+I would like to thank the authors for their rebuttal letter. They have improved substantially the data and have discussed well their findings. They have also well discussed the limitations of their work.  
+
+Reviewer #3 (Remarks to the Author):  
+
+The labeling of "Liver ASC" is misleading for readers. Unfortunately, the authors were not responsive to my request and merely added a sentence in the introduction. I am afraid most readers would misunderstand that "hepatocytes" escape from mutations.
+
+<--- Page Split --->
+
+## REVIEWERS' COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+The authors have address my questions and comments adequately. I have no further comments and I am happy with the revised version of the manuscript to go ahead for publication.  
+
+We thank the reviewer for their appreciation of our responses and helpful comments.  
+
+Reviewer #2 (Remarks to the Author):  
+
+I would like to thank the authors for their rebuttal letter. They have improved substantially the data and have discussed well their findings. They have also well discussed the limitations of their work.  
+
+We thank the reviewer for their support and helpful comments throughout the review process.  
+
+Reviewer #3 (Remarks to the Author):  
+
+The labeling of "Liver ASC" is misleading for readers. Unfortunately, the authors were not responsive to my request and merely added a sentence in the introduction. I am afraid most readers would misunderstand that "hepatocytes" escape from mutations.  
+
+In the introduction we clearly state that we culture liver stem cells as intrahepatic cholangiocyte organoids. cholangiocytes are a distinct liver cell type than the hepatocytes in the liver and therefore we cannot conclude that "hepatocytes" escape from mutations.
+
+<--- Page Split --->

peer_reviews/df915f8909e53f3f2d709aa751c4ec3e7bb52f81abb1679add07da6037e2ba76/supplementary_0_Peer review file/supplementary_0_Peer review file_det.mmd

@@ -0,0 +1,514 @@
+<|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, 155, 891, 211]]<|/det|>
+Common anti- cancer therapies induce somatic mutations in stem cells of healthy tissue  
+
+<|ref|>image<|/ref|><|det|>[[57, 732, 240, 780]]<|/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|>[[115, 147, 848, 164]]<|/det|>
+Reviewer #1 (Remarks to the Author): Expert in somatic mutational signature analysis and genomics  
+
+<|ref|>text<|/ref|><|det|>[[115, 202, 881, 311]]<|/det|>
+Kuijk et al., have studied the effect of CapOx (5- FU and Oxaliplatin) and/or radiotherapy on normal stem cells of colon and liver tissues from individuals with either colorectal and or metastatic liver cancers. Their results highlight the tissue- specific respond to chemotherapy treatment across non- cancerous cells. They suggest colon is more vulnerable to chemo treatments, especially platinum- based treatments (e.g., SBS35) compared to liver which seem to be more protected. This vulnerability in turn can increase the risk of secondary cancer incident in colonic tissues.  
+
+<|ref|>text<|/ref|><|det|>[[115, 351, 865, 479]]<|/det|>
+Some recent studies of somatic mutations in normal cells using laser capture microdissection (e.g., Moore, et. al., 2021 and Lee- Six et al., 2019) or single cell derived colonies (e.g., Pich et al., 2021) have highlighted the footprints of chemotherapy on mutational landscape and clonal dynamics of normal cells. However, more work in this area is certainly required to systematically measure the effect of chemo on normal cells and to assess the risk of further cancer transformation. Therefore, this work is certainly relevant and can be interesting to the readers of this journal. However, I have listed some comments below to be clarified by the authors:  
+
+<|ref|>text<|/ref|><|det|>[[115, 517, 881, 662]]<|/det|>
+- What is the average number of treatment cycles for liver metastases? Biopsies from CRC patients were collected around \(\sim 1.5\) months after latest CapOx, while for liver metastases samples were biopsied \(\sim 15\) months post treatment. How variation to the time of biopsy may explain some of the variation in effect of chemo they have reported between these two groups? The authors indeed have suggested that damaged cells may have been removed from the liver due to longer time from the treatment to the biopsy collection. However, liver stem cell turnover is slower compared to colonic crypts stem cells. Hence, I wonder other factors might be at play such as variation in clonal dynamics between the two tissues?  
+
+<|ref|>text<|/ref|><|det|>[[115, 702, 870, 738]]<|/det|>
+- It is quite surprising that the 58 and 76 yrs-old CapOx-treated donors as well as the 5-FU-only treated donor did not show any elevation in SBS burden. How the authors explain this observation?  
+
+<|ref|>text<|/ref|><|det|>[[115, 777, 874, 812]]<|/det|>
+- Is there any corelation between SBS35 load and duration of the treatment and or different sections of colon that were biopsied?  
+
+<|ref|>text<|/ref|><|det|>[[115, 852, 867, 907]]<|/det|>
+- They have shown that the effect of 5-FU is more heterogenous across the individuals studied. Similar observation was also made in AML cases by Pich et al., 2021. Does the dose of 5-FU vary between individuals and if so, is there any correlation between the dose and SBS17 burden?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 117, 872, 190]]<|/det|>
+- The observation of sub-clonal platinum-induced mutations is quite unexpected. Although the authors suggested alternative possibility for this observation, yet it highlights that some of the chemo effect might be missed with their method. Can the authors give an estimate of the fraction of chemo effect that might have missed due to their sub-clonality nature?  
+
+<|ref|>text<|/ref|><|det|>[[115, 228, 879, 302]]<|/det|>
+- "Our findings reveal that both CapOx chemotherapy and radiotherapy are mutagenic in colorectal ASCs and significantly increase the mutational burden in normal non-cancerous cells beyond typical age-related mutation accumulation." It would be great if the authors can expand on this observation by providing more insights into the type of mutations with respect to their functional consequences.  
+
+<|ref|>text<|/ref|><|det|>[[115, 342, 248, 358]]<|/det|>
+Minor comments:  
+
+<|ref|>text<|/ref|><|det|>[[115, 397, 875, 470]]<|/det|>
+- Overall, the way the experimental design has described is slightly confusing. It is unclear to me how many colonies were sequenced per individual. A simple summary table to include this information as well as, type of treatment, dosage, length of treatment and the time interval from the last treatment to sample collection would be very helpful for the readers.  
+
+<|ref|>text<|/ref|><|det|>[[115, 508, 872, 545]]<|/det|>
+- Fig. 2 left and write panels are not labelled so I assumed the left panel is colon and right panel is liver. In Fig. 2f, the anti-correlation of indel with age even in healthy individuals is unexpected.  
+
+<|ref|>text<|/ref|><|det|>[[115, 670, 845, 688]]<|/det|>
+Reviewer #2 (Remarks to the Author): Expert in colorectal cancer organoids, therapy, and stem cells  
+
+<|ref|>text<|/ref|><|det|>[[115, 725, 880, 818]]<|/det|>
+The manuscript by Kuijk et al. identifies the mutational landscape in normal colonic and liver cells of CRC patients after being treated with several cycles of therapy (chemo and/or radiotherapy). This is an relevant study that addresses the key question of how normal tissue is damaged after treatment. The authors use whole genome sequencing to demonstrate that the mutational pattern in the liver does not change significantly after chemo but intestinal cells are affected (mainly by Oxaliplatin).  
+
+<|ref|>text<|/ref|><|det|>[[115, 856, 875, 892]]<|/det|>
+The work is original as the mutational landscape by NGS has not been well reported in normal intestinal and liver cells. It is an interesting study, with a good experimental set up. The group has an excellent
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 872, 125]]<|/det|>
+background in NGS/cancer genetics and in identifying mutational landscapes of different cell types. The results are well interpreted, however a more exhaustive report would need to be addressed.  
+
+<|ref|>text<|/ref|><|det|>[[115, 166, 215, 182]]<|/det|>
+Major points:  
+
+<|ref|>text<|/ref|><|det|>[[115, 220, 863, 312]]<|/det|>
+1. My main concern is the low number of CRC samples analysed and the poor description in the text about the samples/crypts that were sequenced. Are the results statistically significant? Can they do further analyses to be sure that the samples are representative? If they do power calculations/sample size, what would be the result? The liver data looks convincing but the authors would need to be sure that the rest shown is also representative.  
+
+<|ref|>text<|/ref|><|det|>[[115, 351, 857, 386]]<|/det|>
+2. Why the authors discuss the 5FU influence as it has an impact, when 5 out of 7 show no effect and another individual only in one of the samples?  
+
+<|ref|>text<|/ref|><|det|>[[115, 425, 866, 517]]<|/det|>
+3. Colorectal cancer histology/pathology is not shown/described however, which part of the colon the samples were resected from? It is important to know because may add more variability to the N analyzed (in addition to the age factor which is cover a big range). The text does not address it and this may have an impact. Did the authors see any difference in SBS and ID mutations by anatomical site in the colon and rectum? this may need to be discussed as a potential limitation.  
+
+<|ref|>text<|/ref|><|det|>[[115, 556, 808, 591]]<|/det|>
+4. In my opinion, the telomere section is roughly described and data are not included in the abstract/main figures. Makes sense to leave it? I would suggest to leave it for a better analysis.  
+
+<|ref|>text<|/ref|><|det|>[[115, 630, 880, 666]]<|/det|>
+5. The irradiation findings are not so novel as have already been described in similar approaches/cancers (e.g. 10.1038/ncomms12605). Authors refer to this in the introduction but do not discuss it later on.  
+
+<|ref|>text<|/ref|><|det|>[[115, 706, 883, 797]]<|/det|>
+6. The crypts were expanded from few cells, but as confirmed by the authors the results show that there could be more than one stem cell in each sample before expansion. Did they analyse this in depth? Using stem cell markers, flow cytometry and quantification? through-out the text is written continuously that they are working with single adult stem cells, is this always true? For example, figure legend states "single adult SCs solutions", I think this can take the reader to confusion.  
+
+<|ref|>text<|/ref|><|det|>[[115, 836, 875, 890]]<|/det|>
+7. A similar set up was done by Halazonetis group (Switzerland) where single crypts were expanded into organoids from mouse samples (AKP) before sequencing. A previous manuscript from the same group using APC min also revealed the mutation change when crypts were cultured more than 4 months.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 875, 126]]<|/det|>
+Authors should mention and compare results. The authors should also look at the AKP manuscript to do a scheme similar to show the number of crypts expanded/sequenced from each sample.  
+
+<|ref|>text<|/ref|><|det|>[[115, 195, 216, 210]]<|/det|>
+Minor points:  
+
+<|ref|>text<|/ref|><|det|>[[112, 221, 861, 238]]<|/det|>
+1. Graphs in figure 3 and 4 are too small. Supplementary data should also be shown with a bigger size.  
+
+<|ref|>text<|/ref|><|det|>[[115, 278, 860, 314]]<|/det|>
+2. The statistics section is poorly described. The normal and non-normal distribution should be better study than doing "assumptions" or "likely".  
+
+<|ref|>text<|/ref|><|det|>[[115, 353, 861, 389]]<|/det|>
+3. It should be written in the abstract how many individuals for each sample type were used to do the study.  
+
+<|ref|>text<|/ref|><|det|>[[115, 428, 881, 464]]<|/det|>
+4. Methods section: "Sorbitol and 0.5mM dithiothreitol (all purchased from Merck) until the supernatant was." (words are lacking here).  
+
+<|ref|>text<|/ref|><|det|>[[115, 588, 786, 605]]<|/det|>
+Reviewer #3 (Remarks to the Author): Expert in colorectal and hepatic organoids, stem cells  
+
+<|ref|>text<|/ref|><|det|>[[115, 644, 860, 736]]<|/det|>
+Kuijk et al. analyzed mutation patterns in chemotherapy- treated colon/liver tissue using organoid technology and estimated the number of mutations induced by chemotherapies. This is a fascinating study based on a valuable dataset from clinical specimens. The finding is novel in that chemotherapy- induced mutations in a tissue- specific manner. I would like to recommend its publication in Nature Communications, but several points are to be addressed as follows.  
+
+<|ref|>text<|/ref|><|det|>[[115, 775, 881, 811]]<|/det|>
+1. The "liver organoids" used in this study are actually derived from intrahepatic cholangiocytes and thus should be referred to as intrahepatic cholangiocyte organoids (ICO) (Marsee et al. Cell Stem Cell 2021).  
+
+<|ref|>text<|/ref|><|det|>[[115, 850, 880, 886]]<|/det|>
+2. As described in the discussion, the weak point of this paper is a limited number of specimens, and it is unclear why only a fraction of organoid clones showed the 5-FU signature. Are the two lines of
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 866, 125]]<|/det|>
+organoids that showed the 5- FU signature associated with radiation therapy (or rectum origin)? I think the authors should provide experimental evidence that  
+
+<|ref|>text<|/ref|><|det|>[[114, 165, 870, 292]]<|/det|>
+3. ICOs were devoid of platinum signatures. The authors provided an interesting explanation that liver ASCs have more effective inherent mechanism to protect against chemo-induced mutagenesis. However, there is an alternative explanation that all liver ASCs that gained platinum signature died out or underwent senescence in vivo (thus, they cannot be recovered as organoids). In addition, liver ASCs are non-cycling, and non-cycling cells may gain fewer mutations by platinum drugs? To exclude these possibilities, the authors should treat ICO with platinum and determine whether ICO can gain platinum signature when proliferating in vitro.  
+
+<|ref|>text<|/ref|><|det|>[[115, 331, 878, 366]]<|/det|>
+4. The authors excluded escaping from drugs in liver ASCs. However, oxaliplatin is a renal excretion drug. Why didn't they consider this possibility?  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 406, 212, 422]]<|/det|>
+## Minor points  
+
+<|ref|>text<|/ref|><|det|>[[115, 434, 857, 488]]<|/det|>
+1. Fig.1 is not very self-explanatory. I think it is better to label colons and cholangiocytes in the figure. The color codes for 5-FU+platinum+radiation and 5-FU+platinum are very similar, and I cannot distinguish the two colors. I cannot see which plots are significant in the figure.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[120, 86, 399, 101]]<|/det|>
+## Response to REVIEWER COMMENTS  
+
+<|ref|>sub_title<|/ref|><|det|>[[120, 122, 814, 156]]<|/det|>
+## Reviewer #1 (Remarks to the Author): Expert in somatic mutational signature analysis and genomics  
+
+<|ref|>text<|/ref|><|det|>[[120, 158, 864, 267]]<|/det|>
+Kuijk et al., have studied the effect of CapOx (5- FU and Oxaliplatin) and/or radiotherapy on normal stem cells of colon and liver tissues from individuals with either colorectal and or metastatic liver cancers. Their results highlight the tissue- specific respond to chemotherapy treatment across noncancerous cells. They suggest colon is more vulnerable to chemo treatments, especially platinum- based treatments (e.g., SBS35) compared to liver which seem to be more protected. This vulnerability in turn can increase the risk of secondary cancer incident in colonic tissues.  
+
+<|ref|>text<|/ref|><|det|>[[120, 286, 863, 395]]<|/det|>
+Some recent studies of somatic mutations in normal cells using laser capture microdissection (e.g., Moore, et. al., 2021 and Lee- Six et al., 2019) or single cell derived colonies (e.g., Pich et al., 2021) have highlighted the footprints of chemotherapy on mutational landscape and clonal dynamics of normal cells. However, more work in this area is certainly required to systematically measure the effect of chemo on normal cells and to assess the risk of further cancer transformation. Therefore, this work is certainly relevant and can be interesting to the readers of this journal.  
+
+<|ref|>text<|/ref|><|det|>[[120, 415, 693, 431]]<|/det|>
+However, I have listed some comments below to be clarified by the authors:  
+
+<|ref|>text<|/ref|><|det|>[[119, 451, 867, 597]]<|/det|>
+- What is the average number of treatment cycles for liver metastases? Biopsies from CRC patients were collected around \(\sim 1.5\) months after latest CapOx, while for liver metastases samples were biopsied \(\sim 15\) months post treatment. How variation to the time of biopsy may explain some of the variation in effect of chemo they have reported between these two groups? The authors indeed have suggested that damaged cells may have been removed from the liver due to longer time from the treatment to the biopsy collection. However, liver stem cell turnover is slower compared to colonic crypts stem cells. Hence, I wonder other factors might be at play such as variation in clonal dynamics between the two tissues?  
+
+<|ref|>text<|/ref|><|det|>[[119, 617, 878, 725]]<|/det|>
+Clinical details are described in Suppl table 1 including the number of treatment cycles. We have also included this information in figure 1 panel A. The liver patients included in this study have undergone more CAPOX treatment cyclic than the colon patients (CAPOX treatment cyclic of \(5.8 \pm 3.5\) for liver patients and \(3.2 \pm 1.7\) for colon patients; mean \(\pm\) standard deviation [SD]) and thus the lack of treatment- related mutations in liver ASCs cannot be explained by a lower number of treatment cycles. We have now included this sentence in the main text.  
+
+<|ref|>text<|/ref|><|det|>[[119, 744, 880, 890]]<|/det|>
+On average, there is indeed more time between the end of CAPOX treatment regime and surgical removal of tumor and surrounding healthy tissue (= time of biopsy) for liver patient than colon patients. However, healthy tissue of 2 liver donors (Donor 8, 10) was collected 1 month after the end of CAPOX treatment, which is below the average of 1.5 month for colon donors. All 5 liver ASCs of these 2 liver donors showed a complete absence of CAPOX induced mutations as well, indicating that time between treatment and biopsy is likely not the cause for the lack of CAPOX related mutations for liver ASCs. However, we cannot formally exclude the possibility of a highly efficient clearance of treatment damaged healthy liver ASCs. We consider this hypothesis unlikely because 5-
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[120, 85, 841, 120]]<|/det|>
+FU or platinum therapies rarely induce liver damage<sup>1</sup>. We have added this text to the discussion section of the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[119, 140, 880, 249]]<|/det|>
+Given the very slow liver stem cell turnover rate, we expect that healthy liver tissue harbour different clonal compositions than colon tissue. Sequencing a clonal group of nearby cells that have recently derived from a single (adult stem) cell (e.g. by micro laser biopsy sampling as conducted in Moore et al 2021) could shed more light on clonal dynamics of treatment- induced mutations. However, we used a single- cell based whole genome sequencing methodologies that integrate clonal expansion of a single cell and thus clonal dynamics cannot be studied with this approach.  
+
+<|ref|>text<|/ref|><|det|>[[119, 268, 873, 395]]<|/det|>
+Although it has been demonstrated that treatment changes the clonal architecture in cancer tissue due to selection of treatment driver resistance mechanisms (Martinez- Jimenez et al., 2022 BioRxiv), it may induce similar effects in healthy tissue. However, we have not found a different driver landscape between treated and untreated tissue in either liver or colon adult stem cells. Nevertheless, the number of samples we have sequenced here are not sufficient to be fully conclusive, but we also believe that studying clonal dynamics, although very interesting, would require different experimental approaches that are beyond the scope of this study.  
+
+<|ref|>text<|/ref|><|det|>[[120, 432, 870, 467]]<|/det|>
+- It is quite surprising that the 58 and 76 yrs-old CapOx-treated donors as well as the 5-FU-only treated donor did not show any elevation in SBS burden. How the authors explain this observation?  
+
+<|ref|>text<|/ref|><|det|>[[119, 487, 866, 594]]<|/det|>
+There is natural variation in the correlation between the age and the number of mutations, which increases with age. This means that the older the patients are, the more difficult it is to get significant effects of increases in mutational burden as illustrated by higher standard deviation at older age in Figure 2a and 2b. Thus, while we can clearly identify treatment- induced mutations in these donors by mutational signature analysis, the effect of the increase in mutations as a result of the treatment is not large enough to become apparent by mutational burden analysis alone.  
+
+<|ref|>text<|/ref|><|det|>[[119, 614, 878, 668]]<|/det|>
+We have clarified this by changing the results section as follows: "The SBS mutation burden of the 58 and 76 year old CapOx- treated donors, as well as the 5- FU- only treated donor, was within the natural variation in SBS mutation burden in the correlation with the age"  
+
+<|ref|>text<|/ref|><|det|>[[120, 688, 824, 723]]<|/det|>
+- Is there any correlation between SBS35 load and duration of the treatment and or different sections of colon that were biopsied?  
+
+<|ref|>text<|/ref|><|det|>[[120, 744, 866, 778]]<|/det|>
+There was no correlation between SBS35 mutation load and treatment duration. See comments on reviewer 2 related to colon anatomical site in the colon and rectum.  
+
+<|ref|>text<|/ref|><|det|>[[120, 798, 841, 851]]<|/det|>
+- They have shown that the effect of 5-FU is more heterogenous across the individuals studied. Similar observation was also made in AML cases by Pich et al., 2021. Does the dose of 5-FU vary between individuals and if so, is there any correlation between the dose and SBS17 burden?  
+
+<|ref|>text<|/ref|><|det|>[[120, 872, 853, 906]]<|/det|>
+Unfortunately, treatment dose information was not available for the current study. Nevertheless, only 5 of 28 colon ASCs showed 5-FU related mutations. Of these 5 colon ASCs, 4 (of 4) samples
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[119, 85, 880, 230]]<|/det|>
+came from the 24- year old donor and 1 (of 3) of the 66- year old donor. Thus, even if the 5- FU dose of each patient was shared, we would not observe any relationship between the given 5- FU dose and SBS- 17 mutation burden because \(\sim 85\%\) of the 5- FU exposed colon ASC show a complete absence of 5- FU related mutations. Moreover, internal 5- FU dose is also heavily influenced by pharmacogenetics as patients with (partial) DPD deficiency are less efficient in 5- FU clearance and typically display 5- FU toxicity during treatment (doi: 10.1016/j.ejca.2003.12.004). Finally, we observe some variability within the 24 year old patient, indicating that the variation in mutation accumulation is independent of 5- FU dosage.  
+
+<|ref|>text<|/ref|><|det|>[[120, 250, 880, 322]]<|/det|>
+- The observation of sub-clonal platinum-induced mutations is quite unexpected. Although the authors suggested alternative possibility for this observation, yet it highlights that some of the chemo effect might be missed with their method. Can the authors give an estimate of the fraction of chemo effect that might have missed due to their sub-clonality nature?  
+
+<|ref|>text<|/ref|><|det|>[[119, 342, 855, 469]]<|/det|>
+This information can be depicted from Suppl Figure 12 where we computed the \(\%\) increase in treatment related mutations when considering all (clonal + subclonal) treatment mutations as compared to the clonal treatment mutations (VAF \(>30\%\) ), which is equivalent to the fraction of missed 5- FU and platinum mutations to their subclonality. From the 25 oxaliplatin treated colon ASCs, 7 ASCs showed no increase in mutation load between all and clonal mutations (no missed subclonal treatment mutations), 16 showed an increase of \(\sim 50\%\) in platinum mutation load and 2 samples showed an increase of \(100\%\) or more in both platinum SBS and DBS mutations.  
+
+<|ref|>text<|/ref|><|det|>[[119, 489, 874, 576]]<|/det|>
+We have adapted the main text as follows: "Nevertheless, only 2 ASCs (from the 24- year- old donor) showed a remarkable subclonal platinum mutation contributions of \(100\%\) or more and the inclusion of subclonal platinum mutations had a low impact (less than \(50\%\) ) on the platinum mutational load for all other (n=25) ASCs (Supp. Fig. 12b,c)."  
+
+<|ref|>text<|/ref|><|det|>[[119, 596, 878, 779]]<|/det|>
+- "Our findings reveal that both CapOx chemotherapy and radiotherapy are mutagenic in colorectal ASCs and significantly increase the mutational burden in normal non-cancerous cells beyond typical age-related mutation accumulation." It would be great if the authors can expand on this observation by providing more insights into the type of mutations with respect to their functional consequences. This is indeed a very intriguing question. Although our approach is highly sensitive to detect the passenger mutational consequences of anticancer therapy, it is not optimal to identify potential drivers of secondary malignancies due to the low sample sizes and the relatively small proportion of coding sequence in the genome. Modelling functional impact and risks based on the limited number of coding mutations in our data set would result in highly unreliable risk estimates and thus incorrectly inform patients.  
+
+<|ref|>text<|/ref|><|det|>[[120, 799, 256, 813]]<|/det|>
+Minor comments:  
+
+<|ref|>text<|/ref|><|det|>[[120, 835, 853, 907]]<|/det|>
+- Overall, the way the experimental design has described is slightly confusing. It is unclear to me how many colonies were sequenced per individual. A simple summary table to include this information as well as, type of treatment, dosage, length of treatment and the time interval from the last treatment to sample collection would be very helpful for the readers.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[120, 85, 867, 138]]<|/det|>
+Besides dosage, all this information is included in Suppl table 1 which we now have summarized in Panel A of main figure 1. We have also more comprehensively explained the experimental set up in the main text.  
+
+<|ref|>text<|/ref|><|det|>[[120, 158, 860, 193]]<|/det|>
+- Fig. 2 left and write panels are not labelled so I assumed the left panel is colon and right panel is liver.  
+
+<|ref|>text<|/ref|><|det|>[[120, 196, 603, 212]]<|/det|>
+Thank you for noticing. We have adapted the figure accordingly.  
+
+<|ref|>text<|/ref|><|det|>[[119, 230, 866, 340]]<|/det|>
+In Fig. 2f, the anti- correlation of indel with age even in healthy individuals is unexpected. In general, there is no strong correlation between INDEL load and age in the liver. For example, the healthy (and diseased) liver tissue study from Brunner et al 2019 shows a slightly higher SBS mutation load in the older healthy liver samples while the indel mutation load remains constant. Also in our current study, this correlation is weak/not significant and we therefore caution not to overinterpret the figure based on visual inspection.  
+
+<|ref|>sub_title<|/ref|><|det|>[[120, 412, 850, 447]]<|/det|>
+## Reviewer #2 (Remarks to the Author): Expert in colorectal cancer organoids, therapy, and stem cells  
+
+<|ref|>text<|/ref|><|det|>[[119, 467, 875, 558]]<|/det|>
+The manuscript by Kuijk et al. identifies the mutational landscape in normal colonic and liver cells of CRC patients after being treated with several cycles of therapy (chemo and/or radiotherapy). This is an relevant study that addresses the key question of how normal tissue is damaged after treatment. The authors use whole genome sequencing to demonstrate that the mutational pattern in the liver does not change significantly after chemo but intestinal cells are affected (mainly by Oxaliplatin).  
+
+<|ref|>text<|/ref|><|det|>[[119, 578, 877, 668]]<|/det|>
+The work is original as the mutational landscape by NGS has not been well reported in normal intestinal and liver cells. It is an interesting study, with a good experimental set up. The group has an excellent background in NGS/cancer genetics and in identifying mutational landscapes of different cell types. The results are well interpreted, however a more exhaustive report would need to be addressed.  
+
+<|ref|>text<|/ref|><|det|>[[120, 689, 222, 704]]<|/det|>
+Major points:  
+
+<|ref|>text<|/ref|><|det|>[[120, 707, 872, 741]]<|/det|>
+1. My main concern is the low number of CRC samples analysed and the poor description in the text about the samples/crypts that were sequenced.  
+
+<|ref|>text<|/ref|><|det|>[[119, 761, 876, 907]]<|/det|>
+We have sequenced and included an additional set of 16 samples (13 from colon cohort and 3 from liver cohort) leading to a final set of 42 ASC samples. Overall, findings remain unchanged except that mutational signature analysis does not detect platinum related mutation in 2 additional samples of the 76- year old colon donor (n=3). This may be explained by the low number of CAPOX treatment cycles because this patient has undergone only a single treatment cycle. Therefore, the number of platinum- induced mutations is under the detection limit of mutation signature analysis. The information on the samples sequenced can be found in Suppl table 1 and as a table in Panel A of revised figure 1.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[120, 121, 870, 175]]<|/det|>
+Are the results statistically significant? Can they do further analyses to be sure that the samples are representative? If they do power calculations/sample size, what would be the result? The liver data looks convincing but the authors would need to be sure that the rest shown is also representative.  
+
+<|ref|>text<|/ref|><|det|>[[118, 194, 873, 359]]<|/det|>
+In this study we relied on a sample size set similar to other studies that also used healthy tissue derived organoids to investigate mutation accumulation (Jager et al 2019, Drost et al 2017, Kuijk et al 2020, Nguyen et al, 2022). To further explore the sample size of the current study, we have conducted a power analysis as in Nguyen et al, 2022. For this, we first measured the Cohen's D to compute the standardized mean difference of the mutation load between platinum treated and untreated donors. Using the cohen.D function from effsize R package, we found a normalized effect size of 1.51. A similar normalized effect size (Cohen's D = 1.52) was observed when we assessed the excess in mutation load (observed mut load - predicted mut load using linear mixed model regression analysis) between platinum treated and untreated donors.  
+
+<|ref|>text<|/ref|><|det|>[[118, 378, 870, 523]]<|/det|>
+To assess the reliability of measured Cohen's D scores on mutation load given our number of samples, we performed a power analysis using the pwr.t2n.test() function of the pwr R package to calculate hypothetical detectable effect sizes using a significance level of 0.05 and a power of 0.8. It is generally accepted that power should be approximately \(80\%\) . Given that we have 5 untreated patients and 6 treated patients, we expect to be able to detect an effect size (Cohen's D) of 1.6 which is approximately equal to the computed Cohen's D values on the mutation load and excess mutation load. Thus, the sample sizes in our study would allow us to detect effect sizes as hypothetical detectable effect sizes with a power score of \(80\%\) .  
+
+<|ref|>text<|/ref|><|det|>[[118, 543, 857, 689]]<|/det|>
+Similarly, we applied this approach on the platinum and 5- FU mutation contributions. Using the SBS35 and SBS17 mutation contributions, we obtained Cohen's D scores of 2.7 (large) and 0.7 (medium) for platinum and 5- FU mutation contributions, respectively, between the treated and untreated donor group. The Cohen's D score on the platinum DBS (DBS5) contribution was 2.5, which is similar as the SBS35 mutation contribution. Thus, the normalized effect size for platinum mutation load is well beyond the hypothetical detectable effect size given the number of samples (1.6), but not for 5- FU mutation contribution. This is in line with our observations that 5- FU mutations were only detected in few samples of 2 (of 7) donors.  
+
+<|ref|>text<|/ref|><|det|>[[119, 708, 870, 762]]<|/det|>
+Lastly, we found that the Cohen's D score for radiation induced indels (ID8) between untreated and radiation treated donors was also higher (ie 1.8) than the hypothetical detectable effect size with higher power.  
+
+<|ref|>text<|/ref|><|det|>[[120, 782, 667, 799]]<|/det|>
+We have included this in Methods section and all details in a Suppl Note:  
+
+<|ref|>sub_title<|/ref|><|det|>[[120, 840, 266, 857]]<|/det|>
+## Power Analysis  
+
+<|ref|>text<|/ref|><|det|>[[119, 866, 860, 901]]<|/det|>
+In this study we relied on a sample size set similar to other studies that also used healthy tissue derived organoids to investigate mutation accumulation. To further explore the sample size of the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[119, 85, 880, 120]]<|/det|>
+current study, we have conducted a power analysis as in Nguyen et al, 2022. Details about the power analysis can be found in Suppl Note 1.  
+
+<|ref|>text<|/ref|><|det|>[[119, 157, 880, 193]]<|/det|>
+2. Why the authors discuss the 5FU influence as it has an impact, when 5 out of 7 show no effect and another individual only in one of the samples?  
+
+<|ref|>text<|/ref|><|det|>[[118, 212, 878, 469]]<|/det|>
+The specificity of the 5FU mutational signature allows us to very accurately determine the presence of 5FU- induced mutations. As we have demonstrated previously, this enables the detection of 5FU induced mutations in whole genome sequencing data of 5FU treated colorectal cancer patients, albeit not in every patient, the reason for which is unknown (Christensen et al). In our current study, most of the cells did not show 5FU induced mutations. However, based on the highly characteristic mutational footprint of 5FU, we are very confident that \(18\%\) of the colon cells harbor 5-FU induced mutations. We prefer not to trivialise this effect, because in some cells 5FU treatment equals \(\sim 10\) years of mutations caused by normal age- related processes. Therefore, we can conclude that 5FU can have a mutational impact on healthy cells and may cause secondary malignancies later in life. Fortunately for the patients, most healthy cells escape 5FU induced mutagenesis, which is also an important message. We have modified the abstract to address that in most of the colon ASCs we did not observe 5-FU induced mutations: "The effects of 5-FU treatment were more variable between cells and patients, ranging from a complete absence of 5-FU mutagenesis in \(>80\%\) of the individual colon ASCs to the accumulation of up to 500 mutations in the remaining individual colon ASCs."  
+
+<|ref|>text<|/ref|><|det|>[[119, 488, 875, 578]]<|/det|>
+3. Colorectal cancer histology/pathology is not shown/described however. which part of the colon the samples were resected from? it is important to know because may add more variability to the N analyzed (in addition to the age factor which is cover a big range). The text does not address it and this may have an impact. Did the authors see any difference in SBS and ID mutations by anatomical site in the colon and rectum? this may need to be discussed as a potential limitation.  
+
+<|ref|>text<|/ref|><|det|>[[120, 598, 875, 633]]<|/det|>
+Clinical info on tumor type and sublocation is included in Suppl table 1 and in panel A of figure 1. We have changed the column names accordingly in order to better interpret the clinical details.  
+
+<|ref|>text<|/ref|><|det|>[[118, 653, 878, 854]]<|/det|>
+The study from Lee- Six have shown that age- related mutations vary between anatomical location of the colon tissue, because of a slightly lower proliferation rate at the end of the colon ( \(\sim 12\) SBS1 mutations per year) compared to the right colon section ( \(\sim 18\) SBS1 mutations per year). Because all our samples are from the descending/sigmoid/rectal parts of the colon and not from the ascending/transverse sections, the effect of location on variation is minimal. Moreover, any putative location- dependent variation cannot explain the complete absence of 5-FU mutations in \(85\%\) of the colon ASCs. Because, our results demonstrate that chemotherapy mutation accumulation varies between tissues, this may also be true for the ascending and transverse sections of the colon, which were not examined in the present study. We have therefore included this possibility in the discussion section of our revised manuscript: "future work should aim to further elucidate the mutagenic effects of chemotherapies on other organ systems and organ anatomical sublocations"  
+
+<|ref|>text<|/ref|><|det|>[[118, 873, 833, 908]]<|/det|>
+4. In my opinion, the telomere section is roughly described and data are not included in the abstract/main figures. Makes sense to leave it? I would suggest to leave it for a better analysis.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[120, 103, 860, 120]]<|/det|>
+We agree with the reviewer and we removed this section in the revised version of our manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[120, 140, 861, 193]]<|/det|>
+5. The irradiation findings are not so novel as have already been described in similar approaches/cancers (e.g. 10.1038/ncomms12605). Authors refer to this in the introduction but do not discuss it later on.  
+
+<|ref|>text<|/ref|><|det|>[[120, 213, 874, 285]]<|/det|>
+The study from Behjati et al., is indeed the first study to assess radiation- induced mutagenesis in which they reported an increase in mutation contribution of an indel signature (sized between 1- 100 base pairs) and a balanced inversion signature. We have referred to this publication in radiation section as follows:  
+
+<|ref|>text<|/ref|><|det|>[[120, 287, 856, 359]]<|/det|>
+This signature was mostly similar to COSMIC ID- 8 (cos sim =0.66). ID- 8 mutations were previously found to be enriched in human cancers following radiation therapy \(^{2,3}\) . In addition to small deletions, we investigated SVs because radiation- treated tumors show radiotherapy- associated increases of inversions and large- scale deletions \(^{2,3}\) .  
+
+<|ref|>text<|/ref|><|det|>[[120, 378, 868, 432]]<|/det|>
+As with platinum and 5- FU mutational signatures, we do not claim that these radiotherapy- related signatures are novel, but we use these signatures as barcode to also quantify the mutagenic impact of radiotherapy in healthy rectum adult stem cells, which has not been described before.  
+
+<|ref|>text<|/ref|><|det|>[[120, 469, 867, 560]]<|/det|>
+6. The crypts were expanded from few cells, but as confirmed by the authors the results show that there could be more than one stem cell in each sample before expansion. Did they analyse this in depth? Using stem cell markers, flow cytometry and quantification? through- out the text is written continuously that they are working with single adult stem cells, is this always true? For example, figure legend states "single adult SCs solutions", I think this can take the reader to confusion.  
+
+<|ref|>text<|/ref|><|det|>[[118, 598, 880, 909]]<|/det|>
+Most of the treatment induced neutral passenger mutations are only present in one or a few cells. Therefore, sequencing and characterizing the genome- wide in vivo mutations of a single cell enables accurate quantification of the mutational contribution from late active mutation processes just before tissue sampling. Recently, we optimized and presented an alternative method for cataloging mutations in individual human ASCs without the necessity of using error- prone whole- genome amplification [5]. Here, tissue derived single ASCs are expanded in vitro into clonal organoid cultures to generate sufficient DNA for accurate WGS. The culture conditions have been optimized to promote the growth of intestinal stem cells over other cell types such as differentiated cells or stromal cells. This methods therefore particularly well suited to study and quantify treatment- induced mutations in ASCs in vivo. However, performing a clonal step is highly challenging with colon organoids that are not readily expanded from single cells, particularly from the relatively small biopsies used for the present study. Therefore, we adapted the protocol to maximize the chance of clonal expansion, by first establishing the culture followed by severe fragmentation and serial dilutions at the first split, where individual tiny fragments are most likely derived from individual cells at isolation or have a very recent common ancestor in vivo. As is evident from our observation that we can detect treatment induced mutations, this method is adequate for capturing very recent mutations. However, in rare cases, the small cell clump that gave rise to the clonal culture may
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[119, 85, 880, 194]]<|/det|>
+represent more than one stem cell of the crypt at the time of isolation. As experimental evidence of the success of the clonal step we examined the VAF peaks. The clonal step worked as expected in all of the lines as illustrated by a highly characteristic VAF- peak at \(50\%\) across all samples. Lower VAF peaks indicate that mutations are not shared by the single stem cell ancestor and are typically obtained by laser- capture microdissection of crypts (See for example Suppl Fig 1d,e in Lee- six et al., 2019).  
+
+<|ref|>text<|/ref|><|det|>[[118, 213, 881, 396]]<|/det|>
+The VAF- peaks at \(50\%\) , and the strong decrease of mutations at VAF at \(30\%\) , indicate that we have assessed somatic mutations of single adult stem cells (see also protocol guidelines in Jager et al., \(2017^{4}\) ). Moreover, the absence of subclonal treatment- induced mutations in \(\sim 85\%\) of colon ASC samples confirms the clonal outgrowth of a single ASCs. However, 2 out of 25 platinum treated colon samples showed subclonal platinum DBS and SBS mutations with similar contribution as clonal platinum DBS and SBS mutations. This indicates a fairly recent single stem cell ancestor at the beginning of the treatment but diverged into 2 (or more) stem cells during the course of the treatment. In these two clones we cannot exclude the possibility that these have been derived from two or more cells resulting in an underestimation of the real mutational impact of the anticancer therapy.  
+
+<|ref|>text<|/ref|><|det|>[[120, 415, 840, 450]]<|/det|>
+We have adapted the introduction and methods section of the manuscript to better explain the technical details of the protocol and included the experimental VAF analysis in the results.  
+
+<|ref|>text<|/ref|><|det|>[[119, 505, 875, 615]]<|/det|>
+7. A similar set up was done by Halazonetis group (Switzerland) where single crypts were expanded into organoids from mouse samples (AKP) before sequencing. A previous manuscript from the same group using APC min also revealed the mutation change when crypts were cultured more than 4 months. Authors should mention and compare results. The authors should also look at the AKP manuscript to do a scheme similar to show the number of crypts expanded/sequenced from each sample.  
+
+<|ref|>text<|/ref|><|det|>[[118, 634, 880, 870]]<|/det|>
+The mutational impact of culturing in human adult stem cells from colon and liver tissue have been well- documented in a previous study by our group<sup>5</sup>. In that study we have shown that in vitro culturing induces SBS- 18 mutations which are related to oxidative stress, and that low- oxygen culturing conditions decrease SBS- 18 mutation rate during culturing. We minimized the time in culture 7- 10 days before we performed the clonal step, which is far below the 4 month period of the study by the Halazonetis group. The potential small contribution of the derivation/culturing before the clonal step was applied is negligible (see also validation in our in vivo study<sup>6</sup>). Moreover, culturing induced mutations have a very different mutation context than the highly specific 5- FU and platinum- induced mutations (see also Christensen et al where we identified a clear 5FU signature apart from the in vitro signature by the in vitro treatment of intestinal organoids). Thus, even the few in vitro induced mutations that end up in the mutation matrices from clonal mutations (representing the in vivo induced mutations) will have no impact on the quantitative assessment of treatment induced mutations.  
+
+<|ref|>text<|/ref|><|det|>[[119, 873, 866, 908]]<|/det|>
+Finally, any in vitro induced mutations after the clonal step are subclonal after sequencing the bulk isogenic organoid culture and are discarded in downstream analysis.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[120, 123, 224, 137]]<|/det|>
+Minor points:  
+
+<|ref|>text<|/ref|><|det|>[[120, 141, 852, 174]]<|/det|>
+1. Graphs in figure 3 and 4 are too small. Supplementary data should also be shown with a bigger size.  
+
+<|ref|>text<|/ref|><|det|>[[120, 196, 598, 211]]<|/det|>
+Main figure 3 and 4 as well as Suppl figures have been adapted.  
+
+<|ref|>text<|/ref|><|det|>[[120, 233, 833, 267]]<|/det|>
+2. The statistics section is poorly described. The normal and non-normal distribution should be better study than doing "assumptions" or "likely".  
+
+<|ref|>text<|/ref|><|det|>[[120, 288, 625, 303]]<|/det|>
+Thank you for pointing this. We have revised the statistical section.  
+
+<|ref|>text<|/ref|><|det|>[[120, 324, 860, 359]]<|/det|>
+3. It should be written in the abstract how many individuals for each sample type were used to do the study.  
+
+<|ref|>text<|/ref|><|det|>[[120, 379, 814, 431]]<|/det|>
+We have included the mean number of sequenced ASC samples per donor in the abstract. A complete overview on sample size per donor is now also provided in main figure 1 and in supplementary table 1.  
+
+<|ref|>text<|/ref|><|det|>[[119, 470, 815, 503]]<|/det|>
+4. Methods section: "Sorbitol and 0.5mM dithiothreitol (all purchased from Merck) until the supernatant was." (words are lacking here).  
+
+<|ref|>text<|/ref|><|det|>[[120, 506, 790, 540]]<|/det|>
+Thank you for pointing this out. We have completed this particular sentence ("...until the supernatant was clear.")  
+
+<|ref|>text<|/ref|><|det|>[[120, 561, 825, 578]]<|/det|>
+Reviewer #3 (Remarks to the Author): Expert in colorectal and hepatic organoids, stem cells  
+
+<|ref|>text<|/ref|><|det|>[[119, 598, 879, 688]]<|/det|>
+Kuijk et al. analyzed mutation patterns in chemotherapy- treated colon/liver tissue using organoid technology and estimated the number of mutations induced by chemotherapies. This is a fascinating study based on a valuable dataset from clinical specimens. The finding is novel in that chemotherapy- induced mutations in a tissue- specific manner. I would like to recommend its publication in Nature Communications, but several points are to be addressed as follows.  
+
+<|ref|>text<|/ref|><|det|>[[120, 708, 880, 761]]<|/det|>
+1. The "liver organoids" used in this study are actually derived from intrahepatic cholangiocytes and thus should be referred to as intrahepatic cholangiocyte organoids (ICO) (Marsee et al. Cell Stem Cell 2021).  
+
+<|ref|>text<|/ref|><|det|>[[120, 782, 880, 853]]<|/det|>
+Because colon stem cells and intrahepatic cholangiocyte organoids are both adult stem cells (ASCs) we use this latter terminology. To further specify the type of liver ASCs we use the suggested terminology and refer to the suggested paper at the end of the introduction: "We studied mutations in ASCs derived from the slowly renewing liver (cultured as intrahepatic cholangiocyte organoids)".  
+
+<|ref|>text<|/ref|><|det|>[[117, 872, 838, 889]]<|/det|>
+2. As described in the discussion, the weak point of this paper is a limited number of specimens
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 85, 880, 120]]<|/det|>
+To overcome this limitation, we have sequenced and analyzed extra samples and performed a power analysis. See also our answers to the comments by reviewer 2.  
+
+<|ref|>text<|/ref|><|det|>[[118, 157, 864, 211]]<|/det|>
+3. and it is unclear why only a fraction of organoid clones showed the 5-FU signature. Are the two lines of organoids that showed the 5-FU signature associated with radiation therapy (or rectum origin)?  
+
+<|ref|>text<|/ref|><|det|>[[118, 214, 880, 359]]<|/det|>
+Of the 5 colon ASCs that show a contribution of the 5-FU mutational signature, 4 (of 4) samples came from the 24-year old donor and 1 (of 3) of the 66-year old donor. Both donors were diagnosed with rectum and thus treated with radiation, opening the possibility that 5-FU signature could be associated with these covariates. However, 5-FU related mutations have been documented in colon tissue in patients not being treated with radiation (https://doi.org/10.1101/2021.04.14.437578; and vice versa, no enrichment of 5-FU related mutations (SBS17a/b) have been described in radiation treated cancers \(^{2,3}\) . These results indicate that 5-FU and radiation mutation processes operate independently.  
+
+<|ref|>text<|/ref|><|det|>[[120, 378, 819, 413]]<|/det|>
+I think the authors should provide experimental evidence that ICOs were devoid of platinum signatures.  
+
+<|ref|>text<|/ref|><|det|>[[118, 433, 880, 560]]<|/det|>
+Platinum leads to interstrand crosslinks and thereby causes highly characteristic DBS mutations as has been experimentally demonstrated in platinum- treated isogenic pluripotent stem cells \(^{8}\) . Double base mutations are highly informative because they include two neighbouring SBS mutations located on the same strand. These platinum related mutation scars are thus highly robust to assess the mutational impact of platinum. In Suppl Figure 8, we show that all sequenced liver ASCs (ICOs) don't show any increase of CT>AT and CT>AA DBS mutations compared to untreated liver ASCs. In sharp contrast, nearly all colorectal ASCs from patients treated with platinum show these DBSs (Fig. 3d- e).  
+
+<|ref|>text<|/ref|><|det|>[[119, 580, 866, 652]]<|/det|>
+The authors provided an interesting explanation that liver ASCs have more effective inherent mechanism to protect against chemo- induced mutagenesis. However, there is an alternative explanation that all liver ASCs that gained platinum signature died out or underwent senescence in vivo (thus, they cannot be recovered as organoids).  
+
+<|ref|>text<|/ref|><|det|>[[119, 672, 875, 744]]<|/det|>
+This hypothesis was also included in the results section: "Damaged cells may have been effectively cleared from the liver because the time between treatment and collection was mostly longer for the liver than for the colon, even though also no mutations were observed in the 2 liver samples that were collected 1 month after treatment." See also comment 1 from reviewer 1.  
+
+<|ref|>text<|/ref|><|det|>[[119, 764, 864, 817]]<|/det|>
+In addition, liver ASCs are non- cycling, and non- cycling cells may gain fewer mutations by platinum drugs? To exclude these possibilities, the authors should treat ICO with platinum and determine whether ICO can gain platinum signature when proliferating in vitro.  
+
+<|ref|>text<|/ref|><|det|>[[119, 855, 878, 909]]<|/det|>
+We agree with the reviewer that more focused in vitro validation experiments must be conducted to dissect the mechanisms of chemotherapy induced mutation accumulation and the role of proliferation in this process. However, these mechanistic studies are beyond the scope of the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[120, 85, 875, 157]]<|/det|>
+current manuscript that is focused on the in vivo mutational impact of anticancer therapies in different organ systems. We have therefore decided to discuss our results in the light of previous studies (Oriol et al., 2021 Nature communications; Bertrums et al., 2022 Cancer Discover). Together, these findings indicate that platinum mutation effect is independent of cell proliferation.  
+
+<|ref|>text<|/ref|><|det|>[[119, 177, 872, 267]]<|/det|>
+We have included the following sentence to the discussion part: "Similar observations in mutational effect between platinum and nucleobase analogue (5- FU, thiopurines) drugs has been reported in treated hematopoietic cells<sup>9,10</sup> and well- controlled in vitro validation experiments will dissect the mechanisms of chemotherapy- induced mutation accumulation and the role of proliferation in this process"  
+
+<|ref|>text<|/ref|><|det|>[[119, 287, 866, 322]]<|/det|>
+4. The authors excluded escaping from drugs in liver ASCs. However, oxaliplatin is a renal excretion drug. Why didn't they consider this possibility?  
+
+<|ref|>text<|/ref|><|det|>[[119, 342, 870, 432]]<|/det|>
+A highly efficient mechanism clearance of (oxali)platin based drugs from the vascular system (via renal excretion) could explain the absence of platinum mutations in liver ASCs, but it is not in line with the apparent and robust mutation effect in colon ASCs. Moreover, after uptake in the bloodstream, drugs and nutrients will travel through the portal vein into the liver, thereby exposing all liver cells to the platinum before it can be renally excreted.  
+
+<|ref|>sub_title<|/ref|><|det|>[[120, 490, 220, 504]]<|/det|>
+## Minor points  
+
+<|ref|>text<|/ref|><|det|>[[120, 507, 832, 560]]<|/det|>
+1. Fig.1 is not very self-explanatory. I think it is better to label colons and cholangiocytes in the figure. The color codes for 5-FU+platinum+radiation and 5-FU+platinum are very similar, and I cannot distinguish the two colors. I cannot see which plots are significant in the figure.  
+
+<|ref|>text<|/ref|><|det|>[[120, 581, 389, 597]]<|/det|>
+We have changed Fig 1 accordingly.  
+
+<|ref|>text<|/ref|><|det|>[[118, 650, 750, 667]]<|/det|>
+1. Andrade, R. J. et al. Drug-induced liver injury. Nat Rev Dis Primers 5, 58 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[118, 681, 847, 730]]<|/det|>
+2. Kocakavuk, E. et al. Radiotherapy is associated with a deletion signature that contributes to poor outcomes in patients with cancer. Nat. Genet. 53, 1088-1096 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[118, 744, 816, 793]]<|/det|>
+3. Behjati, S. et al. Mutational signatures of ionizing radiation in second malignancies. Nat. Commun. 7, 12605 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[118, 808, 857, 858]]<|/det|>
+4. Jager, M. et al. Measuring mutation accumulation in single human adult stem cells by whole-genome sequencing of organoid cultures. Nat. Protoc. 13, 59-78 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[115, 871, 866, 888]]<|/det|>
+5. Kuijk, E. et al. The mutational impact of culturing human pluripotent and adult stem cells. Nat.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[156, 85, 359, 101]]<|/det|>
+Commun. 11, 2493 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[118, 116, 850, 165]]<|/det|>
+6. Blokzijl, F. et al. Tissue-specific mutation accumulation in human adult stem cells during life. Nature 538, 260–264 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[118, 180, 876, 231]]<|/det|>
+7. Marsee, A. et al. Building consensus on definition and nomenclature of hepatic, pancreatic, and biliary organoids. Cell Stem Cell 28, 816–832 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[118, 244, 860, 294]]<|/det|>
+8. Kucab, J. E. et al. A Compendium of Mutational Signatures of Environmental Agents. Cell 177, 821–836.e16 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[118, 308, 855, 388]]<|/det|>
+9. Bertrums, E. J. M. et al. Elevated mutational age in blood of children treated for cancer contributes to therapy-related myeloid neoplasms. Cancer Discov. (2022) doi:10.1158/2159-8290.CD-22-0120.  
+
+<|ref|>text<|/ref|><|det|>[[118, 403, 848, 453]]<|/det|>
+10. Pich, O. et al. The evolution of hematopoietic cells under cancer therapy. Nat. Commun. 12, 4803 (2021).
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 90, 300, 106]]<|/det|>
+## REVIEWERS' COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[115, 146, 393, 163]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 202, 860, 238]]<|/det|>
+The authors have address my questions and comments adequately. I have no further comments and I am happy with the revised version of the manuscript to go ahead for publication.  
+
+<|ref|>text<|/ref|><|det|>[[115, 308, 393, 324]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 364, 876, 399]]<|/det|>
+I would like to thank the authors for their rebuttal letter. They have improved substantially the data and have discussed well their findings. They have also well discussed the limitations of their work.  
+
+<|ref|>text<|/ref|><|det|>[[115, 525, 393, 541]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 580, 872, 634]]<|/det|>
+The labeling of "Liver ASC" is misleading for readers. Unfortunately, the authors were not responsive to my request and merely added a sentence in the introduction. I am afraid most readers would misunderstand that "hepatocytes" escape from mutations.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[151, 85, 344, 101]]<|/det|>
+## REVIEWERS' COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[151, 119, 437, 136]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[151, 154, 793, 205]]<|/det|>
+The authors have address my questions and comments adequately. I have no further comments and I am happy with the revised version of the manuscript to go ahead for publication.  
+
+<|ref|>text<|/ref|><|det|>[[150, 223, 794, 240]]<|/det|>
+We thank the reviewer for their appreciation of our responses and helpful comments.  
+
+<|ref|>text<|/ref|><|det|>[[151, 275, 437, 292]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[151, 310, 847, 362]]<|/det|>
+I would like to thank the authors for their rebuttal letter. They have improved substantially the data and have discussed well their findings. They have also well discussed the limitations of their work.  
+
+<|ref|>text<|/ref|><|det|>[[150, 397, 800, 431]]<|/det|>
+We thank the reviewer for their support and helpful comments throughout the review process.  
+
+<|ref|>text<|/ref|><|det|>[[151, 450, 437, 466]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[151, 484, 845, 536]]<|/det|>
+The labeling of "Liver ASC" is misleading for readers. Unfortunately, the authors were not responsive to my request and merely added a sentence in the introduction. I am afraid most readers would misunderstand that "hepatocytes" escape from mutations.  
+
+<|ref|>text<|/ref|><|det|>[[151, 571, 844, 623]]<|/det|>
+In the introduction we clearly state that we culture liver stem cells as intrahepatic cholangiocyte organoids. cholangiocytes are a distinct liver cell type than the hepatocytes in the liver and therefore we cannot conclude that "hepatocytes" escape from mutations.
+
+<--- Page Split --->

peer_reviews/e170d279c61e529e3456c6e4a695fc43f21f6570df2098e8ddcd42b11ae2af23/supplementary_1_Transparent Peer Review file/images_list.json

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

peer_reviews/e170d279c61e529e3456c6e4a695fc43f21f6570df2098e8ddcd42b11ae2af23/supplementary_1_Transparent Peer Review file/supplementary_1_Transparent Peer Review file.mmd

@@ -0,0 +1,454 @@
+
+# nature portfolio  
+
+# Peer Review File  
+
+# N-cadherin crosstalk with integrin weakens the molecular clutch in response to surface viscosity  
+
+Corresponding Author: Professor Manuel Salmeron- Sanchez  
+
+Attachments originally included by the reviewers as part of their assessment can be found at the end of this file.  
+
+Version 1:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+1. What are the noteworthy results?  
+
+This is a nice work showing how N- cadherin crosstalks with integrin to affect sensation of surface viscosity. Both the model and experiments are well designed and conducted, which implicate that cell- cell and cell- ECM adhesions compete with each other for actin cytoskeleton under the viscous environment.  
+
+2. Will the work be of significance to the field and related fields? How does it compare to the established literature? If the work is not original, please provide relevant references.  
+
+The significance of the work in relation to the field and related fields lies in its contribution to our understanding of how adherens junctions influence mechanosensation of viscosity. While previous research has established the crucial role of integrin- mediated focal adhesion in mechanosensation of viscoelasticity (Nature Materials, 2021, 20 (9): 1290- 1299; Proceedings of the National Academy of Sciences, 2018, 115 (12): E2686- E2695), the present work expands upon this by investigating the impact of cadherins on integrin- mediated mechanosensation of viscosity. But, what are the differences in the antagonistic effects of the two receptors on the elastic and viscous environment, and what is the mechanical mechanism behind this difference?  
+
+To effectively situate the current work within the established literature, it is important for the authors to cite and compare the previous research on integrin- mediated focal adhesion and highlight the novel insights and observations provided by their study. By doing so, the authors can demonstrate the unique contributions of their work and provide a comprehensive overview of the existing knowledge in the field. This approach not only strengthens the credibility of the current study but also facilitates a clearer understanding of how it advances the current state of knowledge.  
+
+3. Does the work support the conclusions and claims, or is additional evidence needed? Yes, the evidence is sufficient.  
+
+4. Are there any flaws in the data analysis, interpretation and conclusions? Do these prohibit publication or require revision? There are several flaws as list following:  
+
+(1) Quantitative Difference in Viscosity and Comparison between Experiments and Simulations: It's crucial for the authors to provide a clear quantitative comparison of the viscosity of different supported lipid bilayers (SLBs) such as DOPC and DPPC, as well as glass. This comparison should include specific measurements and values to elucidate the differences in viscosity. Additionally, direct comparisons between the results of experiments and simulations, particularly in Fig. 3e, f, Fig. 4c, and d, would enhance the robustness of the findings and strengthen the conclusions. Providing a side-by-side comparison of experimental and simulated data can help validate the simulation models and their relevance to the experimental observations.  
+
+experimental and simulated data can help validate the simulation models and their relevance to the experimental observations.  
+
+(2) Physiological Relevance of Integrins and Cadherins in Sensing Viscosity:  
+
+The authors should explicitly discuss the potential in vivo scenarios where integrins and cadherins cooperate to sense the viscosity of the extracellular matrix (ECM). Exploring the physiological relevance of their findings will help establish the broader implications of the research in a biological context.
+
+<--- Page Split --->
+
+(3) Surface Density of Peptides, Functionalization Efficiency, and Theoretical Distance Among RGD Ligands: It is imperative for the authors to provide detailed information about the surface density of peptides on functionalized SLBs and glass, along with the efficiency of functionalization processes. This information is critical for understanding the experimental setup and ensuring the reproducibility of the results.  
+
+Furthermore, the method and basis for obtaining the "theoretical distance among RGD ligands of \(44 \text{nm}\) " should be clearly explained and supported by relevant references or experimental validation.  
+
+5. Is the methodology sound? Does the work meet the expected standards in your field? Yes  
+
+6. Is there enough detail provided in the methods for the work to be reproduced? Yes.  
+
+## Reviewer #2  
+
+(Remarks to the Author)  
+
+This report explores the mechanism of cadherin - integrin crosstalk in response to mobility of their respective ligands. As reviewed properly in the manuscript, this cross- talk has been the subject of previous studies, including the specific ligands used in this study; what is new is a cytoskeleton- based competition mechanism to explain this phenomenon. The fundamental demonstration that engagement of cadherins reduces cell response to integrins, in a mobility- dependent manner, is overall strong. The subsequent examination of the underlying mechanisms is interesting, but incomplete. In summary, two issues - one technical and the other conceptual - limit my overall enthusiasm for this high- potential study. If addressed, this report has potential to advance the field by demonstrating a new mechanism of crosstalk between two signaling pathways.  
+
+1) Supported lipid bilayer stability. The cellular responses largely look at a 24 hour timepoint, which represents the integration of several complex molecular functions. Underlying much of this is the requirement that the substrate remain stable and intact over the experiment, which is not assured for supported lipid bilayers. Cells can displace or update lipid structures from such a surface. The extent of this type of interaction varies between cells and culture conditions. Some experiment showing that the lipid bilayer remains intact, with few holes through which proteins and cells can reach the underlying surface, is needed to address this issue. It should be certainly explored for the two lipid formulations with both ligands, but need not be an extensive experiment.  
+
+2) Mechanism of cross-talk. The proposed mechanism of competition between integrins and cadherins for actin fibers is interesting. However, and as reviewed in the manuscript, cadherin engagement modulates cytoskeletal dynamics through signaling pathways including Rac1 which can alter actin polymerization and receptor cluster formation, leading to complex and somewhat counterintuitive impacts on actin flow. The talin and Jasplakinolide experiments are good steps, but don't address changes in these upstream pathways. Some measure or modulation of actin polymerization activity could help address this issue. Alternatively, engagement of cadherins at locations distinct from integrin interactions might rule out signaling-based effects as there would not be local competition for actin fibers.  
+
+Of minor note, there is a moderate level of typographical errors in the manuscript, including the Methods section. These do not dramatically impact the study, but should be addressed if this moves forward to publication.  
+
+## Reviewer #3  
+
+(Remarks to the Author)  
+
+In this article, Barcelona- Estaje et al describe the effect of substrate viscosity on the crosstalk between N- cadherin and integrin adhesion sites. The article contains an impressive amount of data and some interesting results. However, in some cases, the conclusions drawn are not fully supported by the data and some experimental details, especially concerning the data analysis, are missing. As a general comment, I wonder if the 3 types of surfaces used to evaluate the effect of viscosity, these being DPPC, DOPC and glass, were the best possible choice to draw the conclusions that are present in the article. Importantly, the glass surface was functionalized with a concentration of neutravidin 5 times higher than the one used for lipid bilayers) and biotin- PEG was added, while in the lipid bilayers, only the biotinylated adhesion ligands were added. In most of the results shown, the difference between lipid bilayers and glass is larger than between DPPC and DOPC layers, and the glass control is often used to 'fix' the trend. My concern is that this surface is too different to be used to set the trend or make a conclusion regarding the increasing viscosity. It would have been more interesting to use a different lipid bilayer, for instance, DMPC, POPC or sphingomyelin layers. It would have also been interesting to see the results of having layers with HAVDI only.  
+
+In addition, I don't really see what the big conclusion of this study is. Is it that the viscosity influences the crosstalk between N- cadherin and integrin? How is this relevant for stem cells, or other cells? Are the values of viscosity used here related to biologically relevant values? Are the effects of the viscosity relevant in a biological setting, where many other cues are present?  
+
+More detailed comments on the manuscript can be found in the attached PDF.  
+
+Version 2:
+
+<--- Page Split --->
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+The reviewer is satisfied with the revised manuscript and thus recommend publication of the manuscript in Nature Communications.  
+
+Reviewer #2  
+
+(Remarks to the Author)  
+
+The additional studies and narrative in this revised manuscript are much appreciated. However, they do not address the two concerns raised in my initial review. I cannot recommend this report for publication, based on the comments below.  
+
+1) Supported lipid bilayer stability.  
+
+The images of Supplementary Fig. 2i illustrate the stability problem. Cells are appearing in the fluorescence channel, indicating uptake of the BODIPY- functionalized lipids. The impact of removal of materials at some point will be production of holes, potentially below the limit of optical resolution, that will allow proteins from the media to attach to the surface. The representative image of Day 1 on DPPC in fact shows local depletion of lipids around the four adherent cells in the upper left quadrant of the image. A screenshot of this area with arrows indicating such regions is attached. The contrast has been increased to better highlight these issues.  
+
+If there is some other explanation for the cells appearing green in these images and the local depletion, it should be discussed in the narrative. Otherwise, reanalysis of key experiments where the analysis focuses on cells not exhibiting local disruption, is needed.  
+
+2) Mechanism of cross-talk.  
+
+The anti- N- cadherin experiments are interesting and much appreciated. However, Supplementary Fig. 13a shows that the anti- N- cadherin application does increase actin flow compared to RGD alone, particularly for DPPC. This could be due to the fact that the antibody could also be reaching the ventral side of the cells, as it is applied in solution. Regardless, addressing this change in actin flow is needed, whether through more extensive discussion on this mechanism on the overall impact of this study or additional experiments that more completely separate the signals. These experiments could include exposure of cells to beads coated with the anti- N- cadherin antibodies or using micropatterned surfaces with small regions of anti- N- cadherin interspersed into the lipid bilayer.  
+
+Reviewer #3  
+
+(Remarks to the Author)  
+
+I would like to extend my congratulations to the authors for their heroic effort in addressing all of my comments and remarks. All of my questions have been successfully addressed, and I believe this work is now fit for publication in its current state.  
+
+Version 3:  
+
+Reviewer comments:  
+
+Reviewer #2  
+
+(Remarks to the Author)  
+
+The revisions regarding lipid bilayer structure and impact on conclusions is appreciated. As noted, the stability issue is seen for many cell types, and remedying it would require a change in biological model or substrate design. However, the term "reorganization" fails to capture that the cells are likely exposing part of the underlying substrate. I would recommend being more specific with "disruption" or "removal".  
+
+The revisions on cross- talk mechanism are also much appreciated, and have addressed my concern.
+
+<--- Page Split --->
+
+Open Access This Peer Review 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 cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+The images or other third party material in this Peer Review 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 https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+
+## Response to reviewers  
+
+## Reviewer #1:  
+
+1. What are the noteworthy results?  
+
+This is a nice work showing how N- cadherin crosstalks with integrin to affect sensation of surface viscosity. Both the model and experiments are well designed and conducted, which implicate that cell- cell and cell- ECM adhesions compete with each other for actin cytoskeleton under the viscous environment.  
+
+We thank the reviewer for their comments and feedback on our study, which have helped in improving our manuscript; please find our responses to their specific comments below.  
+
+2. Will the work be of significance to the field and related fields? How does it compare to the established literature? If the work is not original, please provide relevant references. The significance of the work in relation to the field and related fields lies in its contribution to our understanding of how adherens junctions influence mechanosensation of viscosity. While previous research has established the crucial role of integrin-mediated focal adhesion in mechanosensation of viscoelasticity (Nature Materials, 2021, 20 (9): 1290-1299; Proceedings of the National Academy of Sciences, 2018, 115 (12): E2686-E2695), the present work expands upon this by investigating the impact of cadherins on integrin-mediated mechanosensation of viscosity. But, what are the differences in the antagonistic effects of the two receptors on the elastic and viscous environment, and what is the mechanical mechanism behind this difference?  
+
+To effectively situate the current work within the established literature, it is important for the authors to cite and compare the previous research on integrin- mediated focal adhesion and highlight the novel insights and observations provided by their study. By doing so, the authors can demonstrate the unique contributions of their work and provide a comprehensive overview of the existing knowledge in the field. This approach not only strengthens the credibility of the current study but also facilitates a clearer understanding of how it advances the current state of knowledge.  
+
+The reviewer is correct that the novelty of the work lies in putting together integrins and cadherins in response to surface viscosity. Viscosity is an important component of viscoelasticity, a key property of the ECM (Nature 2020, 584: 535- 546; https://doi.org/10.1038/s41586- 020- 2612- 2). Yet, while significant amount of work has been done to understand cell response to elasticity, including the role of integrins and cadherins (Nature Materials 2016, 15(12): 1297- 1306; https://doi.org/10.1038/nmat4725), the role of viscosity is not understood yet. We have expanded the introduction to include this.  
+
+3. Does the work support the conclusions and claims, or is additional evidence needed? Yes, the evidence is sufficient.  
+
+We thank the reviewer for their support.
+
+<--- Page Split --->
+
+4. Are there any flaws in the data analysis, interpretation and conclusions? Do these prohibit publication or require revision?  
+
+There are several flaws as list following:  
+
+(1) Quantitative Difference in Viscosity and Comparison between Experiments and Simulations:  
+
+It's crucial for the authors to provide a clear quantitative comparison of the viscosity of different supported lipid bilayers (SLBs) such as DOPC and DPPC, as well as glass. This comparison should include specific measurements and values to elucidate the differences in viscosity. Additionally, direct comparisons between the results of experiments and simulations, particularly in Fig. 3e, f, Fig. 4c, and d, would enhance the robustness of the findings and strengthen the conclusions. Providing a side-by-side comparison of experimental and simulated data can help validate the simulation models and their relevance to the experimental observations.  
+
+We thank the reviewer for raising this point; we have previously quantified the viscosity of the different bilayers (Bennett et al., PNAS 2018; https://doi.org/10.1073/pnas.1710653115) to be \(1 \times 10^{- 6} \mathrm{~Pa} \cdot \mathrm{s} \cdot \mathrm{m}\) for DOPC and \(1 \times 10^{- 4} \mathrm{~Pa} \cdot \mathrm{s} \cdot \mathrm{m}\) for DPPC; the glass control, being not laterally mobile and essentially not deformable by the cell contractile machinery, is considered infinitely viscous in terms of viscosity. We have added this information in the revised version of the paper. We have modified the figures to facilitate a comparison between the trends predicted in the models and observed in the experimental data (Figures 3f and 4d). Specifically, we now show model data starting at a viscosity of DOPC, \(1 \times 10^{- 6} \mathrm{~Pa} \cdot \mathrm{s} \cdot \mathrm{m}\) , and we represent the model predictions at three distinct viscosity values, representing DOPC, DPPC and glass (highest modelled viscosity); predictions for the full ranges of viscosity are now shown in Supplementary Figure 15. We note that an exact match between experiments and model is not expected, as the model is a highly simplified system. Instead, the model is meant to show that the fundamental interactions of a molecular clutch system, combined with competition between integrins and cadherins for actin, can predict the observed trends in actin flows and adhesions as a function of changes in RGD and HAVDI concentrations. Further, the predicted trends take place in a viscosity range that matches the order of measured DOPC and DPPC viscosities.  
+
+(2) Physiological Relevance of Integrins and Cadherins in Sensing Viscosity: The authors should explicitly discuss the potential in vivo scenarios where integrins and cadherins cooperate to sense the viscosity of the extracellular matrix (ECM). Exploring the physiological relevance of their findings will help establish the broader implications of the research in a biological context.  
+
+The ECM is viscoelastic in nature and there is increasing interest in understanding cell response to it. Examples such as collective cell migration – that underpins development, cancer progression and regeneration – are explained through integrin and cadherin crosstalks in response to matrix stiffness. Yet, the ECM is viscoelastic and apart from a few seminal papers in the field, some excellent reviews have been published in the last few years (see e.g. Nature 2020, 584: 535–546; https://doi.org/10.1038/s41586-020-2612-2). Our work lies in the context of understanding the role of the viscous part of viscoelasticity and how this impacts integrins and cadherins. Integrins are key to interact with the ECM – of viscoelastic nature – whereas cadherins link cells and their role is modulated by the viscosity of the cell membrane (Adv
+
+<--- Page Split --->
+
+Healthc Mater. 2020;9(8):e1901259; https://doi.org/10.1002/adhm.201901259). This has been discussed in the revised version of the paper.  
+
+(3) Surface Density of Peptides, Functionalization Efficiency, and Theoretical Distance Among RGD Ligands:  
+
+It is imperative for the authors to provide detailed information about the surface density of peptides on functionalized SLBs and glass, along with the efficiency of functionalization processes. This information is critical for understanding the experimental setup and ensuring the reproducibility of the results.  
+
+Furthermore, the method and basis for obtaining the "theoretical distance among RGD ligands of \(44~\mathrm{nm}\) " should be clearly explained and supported by relevant references or experimental validation.  
+
+Peptide density is regulated via the amount of biotinylated lipid added in the lipid mixture; we have previously confirmed peptide densities via quantitative fluorescence microscopy (Bennett et al., PNAS 2018; https://doi.org/10.1073/pnas.1710653115). The theoretical distance between lipids is calculated as \(1 / \sqrt{n}\) , where \(n\) is the particle density, as now explained in the manuscript.  
+
+5. Is the methodology sound? Does the work meet the expected standards in your field? Yes.  
+
+6. Is there enough detail provided in the methods for the work to be reproduced? Yes.  
+
+We thank the reviewer for their support.
+
+<--- Page Split --->
+
+## Reviewer #2:  
+
+This report explores the mechanism of cadherin - integrin crosstalk in response to mobility of their respective ligands. As reviewed properly in the manuscript, this cross- talk has been the subject of previous studies, including the specific ligands used in this study; what is new is a cytoskeleton- based competition mechanism to explain this phenomenon. The fundamental demonstration that engagement of cadherins reduces cell response to integrins, in a mobility- dependent manner, is overall strong. The subsequent examination of the underlying mechanisms is interesting, but incomplete. In summary, two issues - one technical and the other conceptual - limit my overall enthusiasm for this high- potential study. If addressed, this report has potential to advance the field by demonstrating a new mechanism of crosstalk between two signaling pathways.  
+
+1) Supported lipid bilayer stability. The cellular responses largely look at a 24 hour timepoint, which represents the integration of several complex molecular functions. Underlying much of this is the requirement that the substrate remain stable and intact over the experiment, which is not assured for supported lipid bilayers. Cells can displace or update lipid structures from such a surface. The extent of this type of interaction varies between cells and culture conditions. Some experiment showing that the lipid bilayer remains intact, with few holes through which proteins and cells can reach the underlying surface, is needed to address this issue. It should be certainly explored for the two lipid formulations with both ligands, but need not be an extensive experiment.  
+
+We agree with the reviewer about the importance of demonstrating the bilayer's stability during culture. We have added a new Supplementary Figure 2i, which shows bilayer stability after 1 and 5 days in the presence of cells; this is done through observation of bilayers containing 0.1 mol% BODIPY-functionalized lipids.  
+
+2) Mechanism of cross-talk. The proposed mechanism of competition between integrins and cadherins for actin fibers is interesting. However, and as reviewed in the manuscript, cadherin engagement modulates cytoskeletal dynamics through signaling pathways including Rac1 which can alter actin polymerization and receptor cluster formation, leading to complex and somewhat counterintuitive impacts on actin flow. The talin and Jasplakinolide experiments are good steps, but don't address changes in these upstream pathways. Some measure or modulation of actin polymerization activity could help address this issue. Alternatively, engagement of cadherins at locations distinct from integrin interactions might rule out signaling-based effects as there would not be local competition for actin fibers.  
+
+We thank the reviewer for raising this interesting point. As suggested, we have engaged cadherins at locations distinct from integrins using an anti- N- cadherin antibody to activate dorsal N- cadherins after cells have adhered on the bilayers. We observed that dorsal cadherin engagement (demonstrated through changes in \(\beta\) - catenin signalling) does not lead to a decrease in FA formation on DPPC or glass, as HAVDI- functionalised bilayers instead do. This is shown in new Supplementary Figure 11. Similarly, actin flow is not increased to the same levels as on HAVDI- functionalised bilayers (new Supplementary Figure 13). These observations support the talin overexpression and the jasplakinolide experiments in pointing to the competition between integrins and cadherins for actin fibres as the main mechanism for crosstalk.
+
+<--- Page Split --->
+
+Of minor note, there is a moderate level of typographical errors in the manuscript, including the Methods section. These do not dramatically impact the study, but should be addressed if this moves forward to publication.  
+
+We apologise for these errors, which we have now corrected.
+
+<--- Page Split --->
+
+## Reviewer #3:  
+
+In this article, Barcelona- Estaje et al describe the effect of substrate viscosity on the crosstalk between N- cadherin and integrin adhesion sites. The article contains an impressive amount of data and some interesting results. However, in some cases, the conclusions drawn are not fully supported by the data and some experimental details, especially concerning the data analysis, are missing. As a general comment, I wonder if the 3 types of surfaces used to evaluate the effect of viscosity, these being DPPC, DOPC and glass, were the best possible choice to draw the conclusions that are present in the article. Importantly, the glass surface was functionalized with a concentration of neutravidin 5 times higher than the one used for lipid bilayers and biotin- PEG was added, while in the lipid bilayers, only the biotinylated adhesion ligands were added. In most of the results shown, the difference between lipid bilayers and glass is larger than between DPPC and DOPC layers, and the glass control is often used to 'fix' the trend. My concern is that this surface is too different to be used to set the trend or make a conclusion regarding the increasing viscosity. It would have been more interesting to use a different lipid bilayer, for instance, DMPC, POPC or sphingomyelin layers. It would have also been interesting to see the results of having layers with HAVDI only.  
+
+We thank the reviewer for raising these issues as it has allowed us to clarify details regarding the material platform used in our study. In terms of viscosity and bilayers choice, we have used DOPC and DPPC because we have previously demonstrated that they provide a range of viscosity able to modulate the molecular clutch engagement (Bennett et al., PNAS 2018; https://doi.org/10.1073/pnas.1710653115). As in the previous study, we consider glass as an infinitely viscous substrate, with the ligands being not laterally mobile and essentially not deformable by the cell contractile machinery. On glass, a higher amount of neutravidin is used compared to the bilayers because the entire surface is coated with neutravidin, which is then bound to defined mixtures of biotin- PEG/biotin- ligand which determine the number of ligands available. On the bilayers the addition of biotin- PEG is instead not necessary because ligands' amounts are controlled by the amount of biotinylated lipid in the bilayer. Using these strategies, the density of ligands is similarly controlled on both the glass surfaces and the bilayers. With regards to bilayers functionalised only with HAVDI, these are not able to support cell adhesion and for this reason they are not included.  
+
+In addition, I don't really see what the big conclusion of this study is. Is it that the viscosity influences the crosstalk between N- cadherin and integrin? How is this relevant for stem cells, or other cells? Are the values of viscosity used here related to biologically relevant values? Are the effects of the viscosity relevant in a biological setting, where many other cues are present?  
+
+The reviewer is correct that we demonstrate that viscosity influences the crosstalk between N- cadherins and integrins. Further, we present a mechanism by which N- cadherin binding to ligands modulates integrin mediated- adhesion through competition for acting fibres that in turn reduce the force loading rate via the molecular clutch model. The study is relevant as viscoelasticity is a key property of the ECM and, e.g., stem cells differentiate to osteoblasts on soft hydrogels that maintain elasticity and increased viscosity (i.e. stress relaxation in Nature Materials 2016;15:326- 34; https://doi.org/10.1038/nmat4489). Here, we decouple elasticity from viscosity to understand its role in building up integrin adhesion in an environment where cadherins are present. Of course, we fully agree with the reviewer that in a biological setting in vivo many other factors apart from viscosity would be present. However, isolating its
+
+<--- Page Split --->
+
+contribution, as done here, is useful to understand its potential role in any context where a viscous component is present.  
+
+More detailed comments on the manuscript can be found in the attached PDF.  
+
+1. I have some questions regarding the concentration of peptides used, and what they mean. In the methods section the authors wrote that biotinylated lipids were used in different quantities to achieve 0.02, 0.2, 0.22, 2 or \(2.2\%\) of functionalization (correct?). Is this the limiting factor on the functionalization of the bilayers? For instance, for a bilayer with \(0.2\%\) RGD \(+2\%\) HAVDI, a lipid mixture containing \(2.2\%\) of biotinylated lipids is used and a solution of 1:10 of RGD:HAVDI is added to the solution? Can you please clarify this. Since the biotinylated lipids are always DPPC, will there be an effect on the viscosity between the layers with 0.02 and the \(2.2\%\) (for the DOPC samples)?  
+
+The reviewer is correct. The amount of biotinylated lipids controls the amount of functionalisation, and ligands are then added in the appropriate ratio. We apologise if this was not clear in our methodological section, and we have now addressed this. Varying the amount of biotinylated lipid within the explored ranges does affect bilayer mobility, as measured via fluorescence correlation spectroscopy and shown in the new Supplementary Figure 2j.  
+
+2. The authors state that they use \(0.2\%\) and \(2\%\) HAVDI concentration as the 'low' and 'high' HAVDI. In supporting figure 4, \(10\%\) HAVDI was used. Can the authors provide some rationale to why the 0.2 and \(2\%\) HAVDI conditions were chosen? And how was the layer with \(10\%\) HAVDI prepared?  
+
+\(0.02\%\) and \(2\%\) were chosen after an initial screening of concentrations ranging from \(0.02\%\) to \(10\%\) . The layer with \(10\%\) HAVDI was obtained by adding up to \(12\%\) biotinylated lipids in the lipid mixture (depending on the required RGD concentration).  
+
+3. For the quantification of the peptides in supplementary figure 3 I have a couple of comments. First, it would be useful to have a control with the intensities of the bilayers containing only streptavidin. Then we could also confirm the presence of RGD and HAVDI at the lower concentration. Secondly, the image presented on panel seems brighter, while the quantification shows that there is no difference. Is this because of the LUT of the image or was this not a very representative image? Also, the amount of HAVDI is \(10x\) higher but the fluorescence intensity detected is only \(2 - 2.5x\) higher – any explanations? And, finally, since there are a lot of experiments presented in the paper with varying concentrations of RGD, I would also quantify the amount of RGD only in the layers, to be sure that there is an increase.  
+
+We thank the reviewer for raising these comments. The images with neutravidin only are not shown because there was no signal present. In terms of concentrations, the images for lower HAVDI concentration are shown in panels b) and f), and their intensity is similar. We have previously confirmed peptide densities using this lipid bilayer platform via quantitative fluorescence microscopy of fluorescent neutravidin (Bennett et al., PNAS 2018; https://doi.org/10.1073/pnas.1710653115). Here, we did not adopt a quantitative fluorescence microscopy technique and instead focussed on comparing single ligands with mixtures to ensure adequate and consistent functionalisation when ligand mixtures were used.
+
+<--- Page Split --->
+
+4. What do the author mean by 'a higher intensity of N-cadherin expression' (end page 5). The supplementary figure 3 shows the quantification of the antibody in panel a, but the raw images are not shown. How did the authors quantify this? Is this per cell, an average? Just a cross-section or the whole volume? Depending on how the images are analyzed, the difference can be related to the different cellular localization instead of an expression level.  
+
+We thank the reviewer for this comment. Images were analyzed by creating a mask of the cell and then using it to define the region of interest for the stained N- Cadherins. Once this was done, the intensity of the N- Cadherin staining was calculated per cell and normalized by the area. Images of the N- Cadherin staining have been added to the supplementary material (Supplementary Figure 3b).  
+
+5. "when lower concentrations (0.02 % mol) of RGD were tested (i.e. higher RGD spacing), changes in the size of hMSC area due to HAVDI were observed only for DPPC and not for DOPC" – it is not clear why are there differences between DOPC and DPPC in this case; please add some explanation.  
+
+At very low RGD concentrations and high intermolecular distances, cell attachment and cell spreading are minimal on DOPC. In this condition, the disruption introduced by the addition of HAVDI cannot have any further effect in reducing cell area. This is not the case on DPPC substrates, where cells can still spread despite the low amount of RGD ligands.  
+
+6. The data presented in figure 1e,f,g and in supplementary figure 4b is partially referent to the same experimental conditions but the trends shown are different. For instance, DPPC with \(0.2\% \mathrm{RGD}\) , the cell area in supplementary figure 4 increases when going from 0.02 to \(0.2\%\) HAVDI, which is not in agreement with the trend discussed in the main text.  
+
+We thank the reviewer for this observation. In Figures 1e,f,g, we show how HAVDI, which corresponds to \(0.02\%\) , and high HAVDI, which corresponds to \(2\%\) ; \(0.2\%\) HAVDI mentioned by the reviewer is not reported in this figure. In Supplementary Figure 4, we show varying HAVDI concentrations, from 0.02, to 0.2, 2 and \(10\%\) ; the change from 0.02 to \(0.2\%\) HAVDI in Supplementary Figure 4b is not statistically significant.  
+
+7. Regarding the colocalization of MIIA with the focal adhesions, I think that the data does not support the conclusions in the manuscript. The authors stated "adding HAVDI decreases MIIA levels in eth focal adhesions". By looking at the images, it seems that the presence of HAVDI leads to a much higher expression of MIIA (brighter cells). It is difficult to see if there is MIIA present at the focal adhesions because the level in the cytosol is much higher. The authors show intensity profiles, but it is difficult to interpret these without knowing their location of the image. To retrieve some quantitative analysis, the authors could do some image correlation analysis, rather than showing one intensity profile for each condition.  
+
+We thank the reviewer for raising this point. Even though we agree that image correlation analysis would be ideal for this statement, it cannot be done with the current images presented
+
+<--- Page Split --->
+
+in this manuscript. In order to try to clarify this, we have added lines where the intensity profiles have been taken so that a clear correlation between figures and profiles can be established.  
+
+8. On figure 2, the data shown in panel b, and on panels c and d is different, while it is acquired for the experimental conditions (e.g. compare DPPC RGD only in panel b and panel d). Can the authors explain this? This is especially important since in panel d there is no decrease in the nuc/cyt ratio of YAP between RGD only and low HAVDI, while he authors claim a difference in panel b (with \(\mathrm{p}< 0.0001\) )...  
+
+The decision not to include the comparisons between RGD and low and high HAVDI conditions in every panel was made to avoid overcrowding the graph and to ensure clarity of interpretation. As the data in Figure 2 panel b and panels c- e are the same experimental conditions for RGD and HAVDI, they have now been merged; merged data is shown in panel b. In panels c- e, only the conditions of interest for these panels (HAVDI or scrambled HAVDI) and the corresponding differences (HAVDI vs. scrambled), if present, are now indicated.  
+
+9. For the differentiation markers (figure 1g, h, i), to be able to draw conclusion on the effect of viscosity, it is necessary to calculate/show the statistical differences between the same condition in different substrates. For a first loo, it seems that the data is too spread to be able to draw any conclusions (differences not statistically significant). This is valid for all the expression markers analysed.  
+
+We understand the reviewer's concern regarding the need to present statistical differences between the same conditions on different substrates. However, we would like to clarify that while we acknowledge the perceived spread in the data, we have indeed compared the data between conditions on different substrates. We would like to note that all statistical differences among conditions have been now included in Supporting information as Tables 2, 3 and 4.  
+
+10. "Interestingly, at increasing concentrations of HAVDI, SOX9 expression increased independently of viscosity." Except for the glass, where there seems to be a drop. Can you comment on this?  
+
+While HAVDI is known to promote chondrogenesis, there are reports of osteogenesis being induced in its presence (Zhu et al. 2016, Biomaterials 77, 44- 52; https://doi.org/10.1016/j.biomaterials.2015.10.072). It can be argued that this is the case on glass, where an increase in pRUNX2 is observed. We have included this observation in the manuscript.  
+
+11. The results regarding the intensity and length of the focal adhesion are difficult to interpret due to the lack of information regarding the analysis. How is the intensity calculated? Is there any image segmentation performed? How?  
+
+We apologize for not including this in the text. In response to this query, we have incorporated detailed information regarding the calculation of intensity, as well as any image segmentation procedures, in the Methods section of our manuscript.
+
+<--- Page Split --->
+
+12. On p14 the authors claim "only for more viscous substrates, smaller FAs when adding HAVDI". The differences between RGD, low HADVI, high HAVDI should be quantified in all viscosity conditions (also graph 3e).  
+
+We apologise as we don't understand what the reviewer means here. Quantification of the length of FAs are shown in Figure 3e for all the conditions in the paper (i.e. RGD, low HAVDI and high HAVDI).  
+
+13. For Vinculin, only the quantification is shown (Figure 3e). Please add the fluorescence images as well. Add also the images, for FAK and vinculin, for low HAVDI (even if just in SI).  
+
+We thank the reviewer for this comment. Representative images for all the conditions have been added as Supplementary Figure 8b.  
+
+14. The authors claim that "by overexpressing talin, FAs increase." They should calculate statistical significances between each condition for non-transfected and transfected cells.  
+
+We have considered the reviewer's suggestion and statistical significances between each condition for non-transfected and transfected cells have been calculated and added to the relevant sections of the revised manuscript. We also note that all statistical differences for transfected and not transfected cells have been included in the supporting information as Tables 5, 6 and 7.  
+
+15. The authors claim that "as expected and in contrast to what we observed in control cells on DPPC, the length of FAs in transfected Y201 cells was not affected by HAVDI" and "as expected and in contrast to what we observed in control cells (= no transfection I assume) on DPPC AND GLASS (?), the length of FAs in transfected Y201 cells was not affected by HAVDI". The authors should show statistical differences between RGD-low-high in transfected cells in all conditions.  
+
+We thank the reviewer for this comment. Following their suggestion, we have now included the result of the statistical analyses comparing RGD-low-high conditions in transfected cells across all experimental conditions. We note that all statistical differences have been added in the supporting information as Tables 5, 6, and 7.  
+
+16. "In this model, talin-vinculin and \(\alpha\) -catenin-vinculin would compete with each other to bind to actin, given the limited availability of actin filaments". Could this also be related to the availability of vinculin?  
+
+The reviewer has raised here a very important point that in essence suggests an alternative hypothesis to the one we included in our manuscript, i.e. whether the competition between integrins and N-cadherins could be for vinculin instead of actin filaments. To demonstrate this, we performed additional experiments measuring the actin flow on DPPC and glass with RGD and RGD+HAVDI using wild type cells and also cells that have been transfected to overexpress
+
+<--- Page Split --->
+
+vinculin. Results are shown in Figure 4f- g and Supplementary Figure 14 and demonstrate that, even in the transfected cells, the actin flow increases for cells on RGD+HAVDI, supporting that the competition is actually for actin filaments. We thank the reviewer for this insightful comment that made us perform additional experiments that validate our hypothesis and strengthen the manuscript.  
+
+17. "When low-viscosity bilayers (DOPC) are functionalized with HAVDI, this functionalization does not affect FA formation". Can the authors provide some hypothesis as to why this is happening?  
+
+The overarching message of this manuscript is that when the molecular clutch is engaged, then the presence of HAVDI leads to a weakening in cell adhesion as reflected by a decrease in the size of focal adhesions and increase in the actin flow. However, the molecular clutch is not engaged on DOPC when the substrate is functionalised with RGD (as already demonstrated in Bennett et al., PNAS 2018; https://doi.org/10.1073/pnas.1710653115). Therefore, the effect of adding HAVDI on DOPC substrates does not alter focal adhesion formation, which is already minimal before HAVDI addition.  
+
+18. "hTERT Y201 MSCs behave in the same way as primary hMSCs, being mechanosensitive and with a similar differentiation potential (Supplementary Figure 9)". I disagree with this statement. If the authors want to claim this, they should put both cell lines next to each other in the same graphs. Supplementary figure 9 is only about Y201 cells. For example SOX9 levels seem quite different in both cell lines: primary cells have SOX9 0.05-0.15 (fig 2h) while Y201 have 2-4 (fig S9h). Also, the trend between the RGD-low-high for DPPC and glass are different between the 2 cell lines. And the cyt/nuc ratio for YAP is very different, suggesting that the cells ARE responding differently.  
+
+We are sorry for the misrepresentation of the Y201 cells in this statement. We meant to say that hTERT Y201 MSCs behave in a similar way as primary hMSCs, maintaining mechanosensitivity and differentiation potential. We have made this clear in the revised version of the manuscript.  
+
+19. "We observed that increased talin expression led to more vinculin being recruited to the site of FAs". So maybe vinculin is drawn away from N-cadherin adhesions, and this is the rate-limiting step (not actin)? This claim could be support by imaging N-cadherin with talin overexpression to check if there was less cadherin present.  
+
+This comment is related to number 16 before, where the reviewer explores the idea of whether the limiting factor is vinculin instead of actin. Further to the data already mentioned in comment 16 (i.e., actin flow experiments with cells transfected to overexpress vinculin), we have also performed additional experiments to show N- cadherin staining in cells transfected to overexpress talin. In agreement with our hypothesis, high levels of HAVDI still leads to higher expression of N- cadherin for the transfected cells. This is now shown in Supplementary Figure 10a.
+
+<--- Page Split --->
+
+20. Add more information on the quantification and statistical analysis. For instance, in many cases, the total number of observations is not mentioned, nor the number of biological replicates.  
+
+We thank the reviewer for this comment. We have added the number of observations in the figures. The number of biological replicates for all figures in the revised version of the manuscript has been added in the methods section.  
+
+21. % mol, mol %, % are used interchangeably, make more uniform throughout the Manuscript  
+
+We apologize about the change of nomenclature. It has been homogenised along the text in the revised version of the manuscript.  
+
+22. Please add also the lower error bars in the bar graphs (in both the manuscript and the SI)  
+
+Lower error bars in the graph have been added in the revised version of the manuscript (including Supporting Information).  
+
+23. The letters and text describing the viscosity values in figure 1a are too small to read. Can you also mention the source of these values or how were they obtained?  
+
+We apologise for this. The size of the text in figure 1a has been increased and their source has also been included in the revised version of the manuscript.  
+
+24. Figure 1c, there seems to be no data for the non-functionalized surface (no error bar visible)  
+
+On non-functionalized DOPC, there were no cells on any sample, hence the cell density value was 0. This has been made clear in the revised version of the manuscript.  
+
+25. For figure 1h, it would be better to have an overview image (similar to that shown in sup. Fig. 3b). If you opt for showing the individual cells, please center the images...  
+
+We appreciate the reviewer's observation and overview images have been included for figure 1h in the revised version of the manuscript.  
+
+26. Scale bars: either always in the figures or in the caption (eg. fig 1b versus 1h). Also, please use μm instead of um.  
+
+This has now been corrected.  
+
+27. Mention supplementary figure 5 in the main text.
+
+<--- Page Split --->
+
+Supplementary Figure 5 is now mentioned in the "N- cadherin ligation affects hMSC mechanosensing and differentiation" section.  
+
+28. Add the time scales for the kymographs in figure 4.  
+
+We appreciate this comment and the time scales have been added in the kymographs in Figure 4.  
+
+29. "at a density of 10000 cells/cm² for fixed cell experiments and 20000 cells/cm² for in-cell westerns assays." Why the different densities? Does this have an impact on the results?  
+
+We used higher cell densities for In-Cell Western assays because of their low sensitivity, requiring higher numbers of cells to get a measurable signal.  
+
+30. Please review the methods section of typos (there are a LOT).  
+
+We apologise for the typos, which have been corrected throughout the manuscript.
+
+<--- Page Split --->
+
+## Response to reviewers  
+
+## Reviewer #1:  
+
+The reviewer is satisfied with the revised manuscript and thus recommend publication of the manuscript in Nature Communications.  
+
+We are pleased that the reviewer is satisfied and recommends publication of this manuscript.  
+
+## Reviewer #2:  
+
+The additional studies and narrative in this revised manuscript are much appreciated. However, they do not address the two concerns raised in my initial review. I cannot recommend this report for publication, based on the comments below.  
+
+1) Supported lipid bilayer stability.  
+
+The images of Supplementary Fig. 2i illustrate the stability problem. Cells are appearing in the fluorescence channel, indicating uptake of the BODIPY- functionalized lipids. The impact of removal of materials at some point will be production of holes, potentially below the limit of optical resolution, that will allow proteins from the media to attach to the surface. The representative image of Day 1 on DPPC in fact shows local depletion of lipids around the four adherent cells in the upper left quadrant of the image. A screenshot of this area with arrows indicating such regions is attached. The contrast has been increased to better highlight these issues.  
+
+If there is some other explanation for the cells appearing green in these images and the local depletion, it should be discussed in the narrative. Otherwise, reanalysis of key experiments where the analysis focuses on cells not exhibiting local disruption, is needed.  
+
+The reviewer questions the stability of the substrates and has pointed out the local depletion of lipids around cells. This is a point that we did not note in the previous version of the manuscript and so we need to thank the reviewer again for bringing it up. It is known that cells remodel proteins at the cell- material interface and that when proteins are loosely attached to the substrate they can be 'removed' from it and eventually internalised, e.g. fibronectin on glass (see the pioneering work of Grinnell and Geiger: Cell, 25, 121- 132 (1981) https://doi.org/10.1016/0092- 8674(81)90236- 1 and J Cell Biol 103, 2697 (1986) https://doi.org/10.1083/jcb.103.6.2697; and the work of others such as Altankov J Biomed Mater Res 30, 385 (1996) https://doi.org/10.1002/(SICI)1097- 4636(199603)30:3<385::AID- JBM13>3.0. CO;2- J). This mechanical remodelling at the interface determines the compatibility of substrates. Insomuch that when mechanical remodelling does not happen because, e.g., proteins are strongly attached to the underlying material then cells increase protease secretion and the bioactivity of the interface is compromised (see e.g. our work in Acta Biomater 77, 74 (2018) https://doi.org/10.1016/j.actbio.2018.07.016).  
+
+This phenomenon highlights the importance of the initial cell- material interactions and how this determines cell response in the mid and long term. The reviewer made the comment of whether these holes in the substrate can be afterwards occupied by, e.g., proteins coming from the media. We would point out that cells very quickly start producing their own ECM (see e.g. Nat Mater 18, 883, 2019 https://doi.org/10.1038/s41563- 019- 0307- 6) and this does not prevent the initial effect from the substrate. We therefore argue that, while in the mid/long terms some reorganization of the bilayers occurs, cellular responses are still governed by the initial
+
+<--- Page Split --->
+
+interactions and hence by the varying degree of viscosity of the substrates. Indeed, if cell response was driven by the effect of these defects, we would then expect a similar response to all surfaces, corresponding to the phenotype of cells seeded directly on glass. Instead, we do observe clear differences in cell behaviour depending on the substrate, here and in our previous work (Bennett et al., PNAS 2018; https://doi.org/10.1073/pnas.1710653115).  
+
+Following the reviewer's advice we included additional text in the revised version R2 of the manuscript to address this comment.  
+
+2) Mechanism of cross-talk.  
+
+The anti- N- cadherin experiments are interesting and much appreciated. However, Supplementary Fig. 13a shows that the anti- N- cadherin application does increase actin flow compared to RGD alone, particularly for DPPC. This could be due to the fact that the antibody could also be reaching the ventral side of the cells, as it is applied in solution. Regardless, addressing this change in actin flow is needed, whether through more extensive discussion on this mechanism on the overall impact of this study or additional experiments that more completely separate the signals. These experiments could include exposure of cells to beads coated with the anti- N- cadherin antibodies or using micropatterned surfaces with small regions of anti- N- cadherin interspersed into the lipid bilayer.  
+
+We thank the reviewer for pointing out that we did not discuss the increase in actin flow following treatment with anti- N- cadherin on RGD- functionalised DPPC. We acknowledge that a small effect is observed, whereby the actin flow increases after N- cadherin application on RGD- DPPC as seen in Supplementary Figure 13a. However, this effect is smaller than the one observed when ventral HAVDI ligation occurs, and, importantly, is not accompanied by a significant change in focal adhesion formation, as seen in Supplementary Figure 11b. We therefore argue that the local competition for actin fibres between integrins and cadherins remains the key mechanism to regulate cross- talk following ventral HAVDI engagement. As the reviewer suggests, the effect that we see on the actin flow may be due to some of the antibody reaching the ventral side of the cells and eliciting this subtle change. We have improved our discussion of these results in the revised version R2 of the manuscript to point out these changes.  
+
+## Reviewer #3:  
+
+I would like to extend my congratulations to the authors for their heroic effort in addressing all of my comments and remarks. All of my questions have been successfully addressed, and I believe this work is now fit for publication in its current state.  
+
+We are delighted that the reviewer appreciates the significant amount of work that we put into R1 to address their concerns.
+
+<--- Page Split --->
+
+1. I have some questions regarding the concentration of peptides used, and what they mean. In the methods section the authors wrote that biotinylated lipids were used in different quantities to achieve 0.02, 0.2, 0.22, 2 or 2.2 % of functionalization (correct?). Is this the limiting factor on the functionalization of the bilayers? For instance, for a bilayer with \(0.2\% \text{RGD} + 2\% \text{HAVDI}\) , a lipid mixture containing \(2.2\%\) of biotinylated lipids is used and a solution of 1:10 of RGD:HAVDI is added to the solution? Can you please clarify this. Since the biotinylated lipids are always DPPC, will there be an effect on the viscosity between the layers with 0.02 and the \(2.2\%\) (for the DOPC samples)?  
+
+2. The authors state that they use \(0.2\%\) and \(2\%\) HAVDI concentration as the 'low' and 'high' HAVDI. In supporting figure 4, \(10\%\) HAVDI was used. Can the authors provide some rationale to why the 0.2 and \(2\%\) HAVDI conditions were chosen? And how was the layer with \(10\%\) HAVDI prepared?  
+
+3. For the quantification of the peptides in supplementary figure 3 I have a couple of comments. First, it would be useful to have a control with the intensities of the bilayers containing only streptavidin. Then we could also confirm the presence of RGD and HAVDI at the lower concentration. Secondly, the image presented on panel seems brighter, while the quantification shows that there is no difference. Is this because of the LUT of the image or was this not a very representative image? Also, the amount of HAVDI is \(10x\) higher but the fluorescence intensity detected is only \(2 - 2.5x\) higher – any explanations? And, finally, since there are a lot of experiments presented in the paper with varying concentrations of RGD, I would also quantify the amount of RGD only in the layers, to be sure that there is an increase.  
+
+4. What do the author mean by 'a higher intensity of N-cadherin expression' (end page 5). The supplementary figure 3 shows the quantification of the antibody in panel a, but the raw images are not shown. How did the authors quantify this? Is this per cell, an average? Just a cross-section or the whole volume? Depending on how the images are analyzed, the difference can be related to the different cellular localization instead of an expression level.  
+
+5. "when lower concentrations (0.02 % mol) of RGD were tested (i.e. higher RGD spacing), changes in the size of hMSC area due to HAVDI were observed only for DPPC and not for DOPC" – it is not clear why are there differences between DOPC and DPPC in this case; please add some explanation.  
+
+6. The data presented in figure 1e,f,g and in supplementary figure 4b is partially referent to the same experimental conditions but the trends shown are different. For instance, DPPC with \(0.2\% \text{RGD}\) , the cell area in supplementary figure 4 increases when going from 0.02 to \(0.2\%\) HAVDI, which is not in agreement with the trend discussed in the main text.  
+
+7. Regarding the colocalization of MIIA with the focal adhesions, I think that the data does not support the conclusions in the manuscript. The authors stated "adding HAVDI decreases MIIA levels in eth focal adhesions". By looking at the images, it seems that the presence of HAVDI leads to a much higher expression of MIIA (brighter cells). It is difficult to see if there is MIIA present at the focal adhesions because the level in the cytosol is much higher. The authors show intensity profiles, but it is difficult to interpret these without knowing their location of the image. To retrieve some quantitative analysis, the authors could do some image correlation analysis, rather than showing one intensity profile for each condition.  
+
+8. On figure 2, the data shown in panel b, and on panels c and d is different, while it is acquired for the experimental conditions (e.g. compare DPPC RGD only in panel b and
+
+<--- Page Split --->
+
+panel d). Can the authors explain this? This is especially important since in panel d there is no decrease in the nuc/cyt ratio of YAP between RGD only and low HAVDI, while he authors claim a difference in panel b (with \(p< 0.0001\) )...  
+
+9. For the differentiation markers (figure 1g, h, i), to be able to draw conclusion on the effect of viscosity, it is necessary to calculate/show the statistical differences between the same condition in different substrates. For a first loo, it seems that the data is too spread to be able to draw any conclusions (differences not statistically significant). This is valid for all the expression markers analysed.  
+
+10. "Interestingly, at increasing concentrations of HAVDI, SOX9 expression increased independently of viscosity." Except for the glass, where there seems to be a drop. Can you comment on this?  
+
+11. The results regarding the intensity and length of the focal adhesion are difficult to interpret due to the lack of information regarding the analysis. How is the intensity calculated? Is there any image segmentation performed? How?  
+
+12. On p14 the authors claim "only for more viscous substrates, smaller FAs when adding HAVDI". The differences between RGD, low HAVDI, high HAVDI should be quantified in all viscosity conditions (also graph 3e).  
+
+13. For Vinculin, only the quantification is shown (Figure 3e). Please add the fluorescence images as well. Add also the images, for FAK and vinculin, for low HAVDI (even if just in SI).  
+
+14. The authors claim that "by overexpressing talin, FAs increase." They should calculate statistical significances between each condition for non-transfected and transfected cells.  
+
+15. The authors claim that "as expected and in contrast to what we observed in control cells on DPPC, the length of FAs in transfected Y201 cells was not affected by HAVDI" and "as expected and in contrast to what we observed in control cells (= no transfection I assume) on DPPC AND GLASS (?), the length of FAs in transfected Y201 cells was not affected by HAVDI". The authors should show statistical differences between RGD-low-high in transfected cells in all conditions.  
+
+16. "In this model, talin-vinculin and \(\alpha\) -catenin-vinculin would compete with each other to bind to actin, given the limited availability of actin filaments". Could this also be related to the availability of vinculin?  
+
+17. "When low-viscosity bilayers (DOPC) are functionalized with HAVDI, this functionalization does not affect FA formation". Can the authors provide some hypothesis as to why this is happening?  
+
+18. "hTERT Y201 MSCs behave in the same way as primary hMSCs, being mechanosensitive and with a similar differentiation potential (Supplementary Figure 9)". I disagree with this statement. If the authors want to claim this, they should put both cell lines next to each other in the same graphs. Supplementary figure 9 is only about Y201 cells. For example SOX9 levels seem quite different in both cell lines: primary cells have SOX9 0.05-0.15 (fig 2h) while Y201 have 2-4 (fig S9h). Also, the trend between the RGD-low-high for DPPC and glass are different between the 2 cell lines. And the cyt/nuc ratio for YAP is very different, suggesting that the cells ARE responding differently.  
+
+19. "We observed that increased talin expression led to more vinculin being recruited to the site of FAs". So maybe vinculin is drawn away from N-cadherin adhesions, and this is the rate-limiting step (not actin)? This claim could be support by imaging N-cadherin with talin overexpression to check if there was less cadherin present.
+
+<--- Page Split --->
+
+20. Add more information on the quantification and statistical analysis. For instance, in many cases, the total number of observations is not mentioned, nor the number of biological replicates.21. % mol, mol %, % are used interchangeably, make more uniform throughout the manuscript22. Please add also the lower error bars in the bar graphs (in both the manuscript and the SI)23. The letters and text describing the viscosity values in figure 1a are too small to read. Can you also mention the source of these values or how were they obtained?24. Figure 1c, there seems to be no data for the non-functionalized surface (no error bar visible)25. For figure 1h, it would be better to have an overview image (similar to that shown in sup. Fig. 3b). If you opt for showing the individual cells, please center the images...26. Scale bars: either always in the figures or in the caption (eg. fig 1b versus 1h). Also, please use μm instead of um.27. Mention supplementary figure 5 in the main text.28. Add the time scales for the kymographs in figure 4.29. "at a density of 10000 cells/cm2 for fixed cell experiments and 20000 cells/cm2 for in-cell westerns assays." Why the different densities? Does this have an impact on the results?30. Please review the methods section of typos (there are a LOT).
+
+<--- Page Split --->
+![PLACEHOLDER_22_0]
+
+
+<--- Page Split --->

peer_reviews/e170d279c61e529e3456c6e4a695fc43f21f6570df2098e8ddcd42b11ae2af23/supplementary_1_Transparent Peer Review file/supplementary_1_Transparent Peer Review file_det.mmd

@@ -0,0 +1,634 @@
+<|ref|>title<|/ref|><|det|>[[72, 50, 295, 80]]<|/det|>
+# nature portfolio  
+
+<|ref|>title<|/ref|><|det|>[[74, 96, 296, 120]]<|/det|>
+# Peer Review File  
+
+<|ref|>title<|/ref|><|det|>[[73, 161, 869, 211]]<|/det|>
+# N-cadherin crosstalk with integrin weakens the molecular clutch in response to surface viscosity  
+
+<|ref|>text<|/ref|><|det|>[[73, 224, 597, 241]]<|/det|>
+Corresponding Author: Professor Manuel Salmeron- Sanchez  
+
+<|ref|>text<|/ref|><|det|>[[73, 274, 890, 289]]<|/det|>
+Attachments originally included by the reviewers as part of their assessment can be found at the end of this file.  
+
+<|ref|>text<|/ref|><|det|>[[73, 325, 144, 339]]<|/det|>
+Version 1:  
+
+<|ref|>text<|/ref|><|det|>[[73, 352, 220, 366]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[73, 378, 160, 392]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[73, 404, 238, 417]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[73, 418, 316, 431]]<|/det|>
+1. What are the noteworthy results?  
+
+<|ref|>text<|/ref|><|det|>[[72, 431, 911, 471]]<|/det|>
+This is a nice work showing how N- cadherin crosstalks with integrin to affect sensation of surface viscosity. Both the model and experiments are well designed and conducted, which implicate that cell- cell and cell- ECM adhesions compete with each other for actin cytoskeleton under the viscous environment.  
+
+<|ref|>text<|/ref|><|det|>[[72, 483, 896, 510]]<|/det|>
+2. Will the work be of significance to the field and related fields? How does it compare to the established literature? If the work is not original, please provide relevant references.  
+
+<|ref|>text<|/ref|><|det|>[[72, 510, 918, 600]]<|/det|>
+The significance of the work in relation to the field and related fields lies in its contribution to our understanding of how adherens junctions influence mechanosensation of viscosity. While previous research has established the crucial role of integrin- mediated focal adhesion in mechanosensation of viscoelasticity (Nature Materials, 2021, 20 (9): 1290- 1299; Proceedings of the National Academy of Sciences, 2018, 115 (12): E2686- E2695), the present work expands upon this by investigating the impact of cadherins on integrin- mediated mechanosensation of viscosity. But, what are the differences in the antagonistic effects of the two receptors on the elastic and viscous environment, and what is the mechanical mechanism behind this difference?  
+
+<|ref|>text<|/ref|><|det|>[[72, 612, 911, 678]]<|/det|>
+To effectively situate the current work within the established literature, it is important for the authors to cite and compare the previous research on integrin- mediated focal adhesion and highlight the novel insights and observations provided by their study. By doing so, the authors can demonstrate the unique contributions of their work and provide a comprehensive overview of the existing knowledge in the field. This approach not only strengthens the credibility of the current study but also facilitates a clearer understanding of how it advances the current state of knowledge.  
+
+<|ref|>text<|/ref|><|det|>[[72, 690, 675, 717]]<|/det|>
+3. Does the work support the conclusions and claims, or is additional evidence needed? Yes, the evidence is sufficient.  
+
+<|ref|>text<|/ref|><|det|>[[72, 729, 920, 757]]<|/det|>
+4. Are there any flaws in the data analysis, interpretation and conclusions? Do these prohibit publication or require revision? There are several flaws as list following:  
+
+<|ref|>text<|/ref|><|det|>[[72, 757, 920, 833]]<|/det|>
+(1) Quantitative Difference in Viscosity and Comparison between Experiments and Simulations: It's crucial for the authors to provide a clear quantitative comparison of the viscosity of different supported lipid bilayers (SLBs) such as DOPC and DPPC, as well as glass. This comparison should include specific measurements and values to elucidate the differences in viscosity. Additionally, direct comparisons between the results of experiments and simulations, particularly in Fig. 3e, f, Fig. 4c, and d, would enhance the robustness of the findings and strengthen the conclusions. Providing a side-by-side comparison of experimental and simulated data can help validate the simulation models and their relevance to the experimental observations.  
+
+<|ref|>text<|/ref|><|det|>[[72, 836, 911, 860]]<|/det|>
+experimental and simulated data can help validate the simulation models and their relevance to the experimental observations.  
+
+<|ref|>text<|/ref|><|det|>[[72, 873, 602, 887]]<|/det|>
+(2) Physiological Relevance of Integrins and Cadherins in Sensing Viscosity:  
+
+<|ref|>text<|/ref|><|det|>[[72, 887, 907, 926]]<|/det|>
+The authors should explicitly discuss the potential in vivo scenarios where integrins and cadherins cooperate to sense the viscosity of the extracellular matrix (ECM). Exploring the physiological relevance of their findings will help establish the broader implications of the research in a biological context.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 46, 905, 99]]<|/det|>
+(3) Surface Density of Peptides, Functionalization Efficiency, and Theoretical Distance Among RGD Ligands: It is imperative for the authors to provide detailed information about the surface density of peptides on functionalized SLBs and glass, along with the efficiency of functionalization processes. This information is critical for understanding the experimental setup and ensuring the reproducibility of the results.  
+
+<|ref|>text<|/ref|><|det|>[[72, 98, 905, 127]]<|/det|>
+Furthermore, the method and basis for obtaining the "theoretical distance among RGD ligands of \(44 \text{nm}\) " should be clearly explained and supported by relevant references or experimental validation.  
+
+<|ref|>text<|/ref|><|det|>[[72, 139, 671, 166]]<|/det|>
+5. Is the methodology sound? Does the work meet the expected standards in your field? Yes  
+
+<|ref|>text<|/ref|><|det|>[[72, 177, 620, 204]]<|/det|>
+6. Is there enough detail provided in the methods for the work to be reproduced? Yes.  
+
+<|ref|>sub_title<|/ref|><|det|>[[73, 217, 162, 230]]<|/det|>
+## Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[73, 243, 238, 256]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 256, 905, 360]]<|/det|>
+This report explores the mechanism of cadherin - integrin crosstalk in response to mobility of their respective ligands. As reviewed properly in the manuscript, this cross- talk has been the subject of previous studies, including the specific ligands used in this study; what is new is a cytoskeleton- based competition mechanism to explain this phenomenon. The fundamental demonstration that engagement of cadherins reduces cell response to integrins, in a mobility- dependent manner, is overall strong. The subsequent examination of the underlying mechanisms is interesting, but incomplete. In summary, two issues - one technical and the other conceptual - limit my overall enthusiasm for this high- potential study. If addressed, this report has potential to advance the field by demonstrating a new mechanism of crosstalk between two signaling pathways.  
+
+<|ref|>text<|/ref|><|det|>[[72, 371, 899, 464]]<|/det|>
+1) Supported lipid bilayer stability. The cellular responses largely look at a 24 hour timepoint, which represents the integration of several complex molecular functions. Underlying much of this is the requirement that the substrate remain stable and intact over the experiment, which is not assured for supported lipid bilayers. Cells can displace or update lipid structures from such a surface. The extent of this type of interaction varies between cells and culture conditions. Some experiment showing that the lipid bilayer remains intact, with few holes through which proteins and cells can reach the underlying surface, is needed to address this issue. It should be certainly explored for the two lipid formulations with both ligands, but need not be an extensive experiment.  
+
+<|ref|>text<|/ref|><|det|>[[72, 475, 910, 569]]<|/det|>
+2) Mechanism of cross-talk. The proposed mechanism of competition between integrins and cadherins for actin fibers is interesting. However, and as reviewed in the manuscript, cadherin engagement modulates cytoskeletal dynamics through signaling pathways including Rac1 which can alter actin polymerization and receptor cluster formation, leading to complex and somewhat counterintuitive impacts on actin flow. The talin and Jasplakinolide experiments are good steps, but don't address changes in these upstream pathways. Some measure or modulation of actin polymerization activity could help address this issue. Alternatively, engagement of cadherins at locations distinct from integrin interactions might rule out signaling-based effects as there would not be local competition for actin fibers.  
+
+<|ref|>text<|/ref|><|det|>[[70, 580, 905, 608]]<|/det|>
+Of minor note, there is a moderate level of typographical errors in the manuscript, including the Methods section. These do not dramatically impact the study, but should be addressed if this moves forward to publication.  
+
+<|ref|>sub_title<|/ref|><|det|>[[73, 620, 162, 633]]<|/det|>
+## Reviewer #3  
+
+<|ref|>text<|/ref|><|det|>[[73, 647, 237, 660]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 660, 916, 817]]<|/det|>
+In this article, Barcelona- Estaje et al describe the effect of substrate viscosity on the crosstalk between N- cadherin and integrin adhesion sites. The article contains an impressive amount of data and some interesting results. However, in some cases, the conclusions drawn are not fully supported by the data and some experimental details, especially concerning the data analysis, are missing. As a general comment, I wonder if the 3 types of surfaces used to evaluate the effect of viscosity, these being DPPC, DOPC and glass, were the best possible choice to draw the conclusions that are present in the article. Importantly, the glass surface was functionalized with a concentration of neutravidin 5 times higher than the one used for lipid bilayers) and biotin- PEG was added, while in the lipid bilayers, only the biotinylated adhesion ligands were added. In most of the results shown, the difference between lipid bilayers and glass is larger than between DPPC and DOPC layers, and the glass control is often used to 'fix' the trend. My concern is that this surface is too different to be used to set the trend or make a conclusion regarding the increasing viscosity. It would have been more interesting to use a different lipid bilayer, for instance, DMPC, POPC or sphingomyelin layers. It would have also been interesting to see the results of having layers with HAVDI only.  
+
+<|ref|>text<|/ref|><|det|>[[72, 816, 909, 867]]<|/det|>
+In addition, I don't really see what the big conclusion of this study is. Is it that the viscosity influences the crosstalk between N- cadherin and integrin? How is this relevant for stem cells, or other cells? Are the values of viscosity used here related to biologically relevant values? Are the effects of the viscosity relevant in a biological setting, where many other cues are present?  
+
+<|ref|>text<|/ref|><|det|>[[72, 879, 606, 894]]<|/det|>
+More detailed comments on the manuscript can be found in the attached PDF.  
+
+<|ref|>text<|/ref|><|det|>[[72, 919, 144, 932]]<|/det|>
+Version 2:
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[73, 47, 219, 60]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[73, 73, 160, 86]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[73, 100, 238, 112]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[73, 113, 850, 139]]<|/det|>
+The reviewer is satisfied with the revised manuscript and thus recommend publication of the manuscript in Nature Communications.  
+
+<|ref|>text<|/ref|><|det|>[[73, 152, 161, 165]]<|/det|>
+Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[73, 178, 238, 191]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[73, 191, 920, 218]]<|/det|>
+The additional studies and narrative in this revised manuscript are much appreciated. However, they do not address the two concerns raised in my initial review. I cannot recommend this report for publication, based on the comments below.  
+
+<|ref|>text<|/ref|><|det|>[[73, 231, 306, 244]]<|/det|>
+1) Supported lipid bilayer stability.  
+
+<|ref|>text<|/ref|><|det|>[[72, 244, 924, 320]]<|/det|>
+The images of Supplementary Fig. 2i illustrate the stability problem. Cells are appearing in the fluorescence channel, indicating uptake of the BODIPY- functionalized lipids. The impact of removal of materials at some point will be production of holes, potentially below the limit of optical resolution, that will allow proteins from the media to attach to the surface. The representative image of Day 1 on DPPC in fact shows local depletion of lipids around the four adherent cells in the upper left quadrant of the image. A screenshot of this area with arrows indicating such regions is attached. The contrast has been increased to better highlight these issues.  
+
+<|ref|>text<|/ref|><|det|>[[72, 320, 920, 360]]<|/det|>
+If there is some other explanation for the cells appearing green in these images and the local depletion, it should be discussed in the narrative. Otherwise, reanalysis of key experiments where the analysis focuses on cells not exhibiting local disruption, is needed.  
+
+<|ref|>text<|/ref|><|det|>[[73, 373, 260, 386]]<|/det|>
+2) Mechanism of cross-talk.  
+
+<|ref|>text<|/ref|><|det|>[[72, 386, 905, 477]]<|/det|>
+The anti- N- cadherin experiments are interesting and much appreciated. However, Supplementary Fig. 13a shows that the anti- N- cadherin application does increase actin flow compared to RGD alone, particularly for DPPC. This could be due to the fact that the antibody could also be reaching the ventral side of the cells, as it is applied in solution. Regardless, addressing this change in actin flow is needed, whether through more extensive discussion on this mechanism on the overall impact of this study or additional experiments that more completely separate the signals. These experiments could include exposure of cells to beads coated with the anti- N- cadherin antibodies or using micropatterned surfaces with small regions of anti- N- cadherin interspersed into the lipid bilayer.  
+
+<|ref|>text<|/ref|><|det|>[[73, 490, 161, 503]]<|/det|>
+Reviewer #3  
+
+<|ref|>text<|/ref|><|det|>[[73, 516, 238, 529]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 530, 914, 556]]<|/det|>
+I would like to extend my congratulations to the authors for their heroic effort in addressing all of my comments and remarks. All of my questions have been successfully addressed, and I believe this work is now fit for publication in its current state.  
+
+<|ref|>text<|/ref|><|det|>[[73, 569, 142, 581]]<|/det|>
+Version 3:  
+
+<|ref|>text<|/ref|><|det|>[[73, 595, 219, 608]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[73, 621, 161, 633]]<|/det|>
+Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[73, 647, 238, 660]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 660, 920, 714]]<|/det|>
+The revisions regarding lipid bilayer structure and impact on conclusions is appreciated. As noted, the stability issue is seen for many cell types, and remedying it would require a change in biological model or substrate design. However, the term "reorganization" fails to capture that the cells are likely exposing part of the underlying substrate. I would recommend being more specific with "disruption" or "removal".  
+
+<|ref|>text<|/ref|><|det|>[[72, 725, 763, 739]]<|/det|>
+The revisions on cross- talk mechanism are also much appreciated, and have addressed my concern.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 45, 916, 99]]<|/det|>
+Open Access This Peer Review 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.  
+
+<|ref|>text<|/ref|><|det|>[[72, 99, 796, 113]]<|/det|>
+In cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+<|ref|>text<|/ref|><|det|>[[72, 112, 910, 165]]<|/det|>
+The images or other third party material in this Peer Review 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.  
+
+<|ref|>text<|/ref|><|det|>[[72, 165, 618, 179]]<|/det|>
+To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[119, 85, 310, 101]]<|/det|>
+## Response to reviewers  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 118, 227, 134]]<|/det|>
+## Reviewer #1:  
+
+<|ref|>text<|/ref|><|det|>[[120, 150, 407, 166]]<|/det|>
+1. What are the noteworthy results?  
+
+<|ref|>text<|/ref|><|det|>[[119, 167, 880, 232]]<|/det|>
+This is a nice work showing how N- cadherin crosstalks with integrin to affect sensation of surface viscosity. Both the model and experiments are well designed and conducted, which implicate that cell- cell and cell- ECM adhesions compete with each other for actin cytoskeleton under the viscous environment.  
+
+<|ref|>text<|/ref|><|det|>[[119, 248, 880, 282]]<|/det|>
+We thank the reviewer for their comments and feedback on our study, which have helped in improving our manuscript; please find our responses to their specific comments below.  
+
+<|ref|>text<|/ref|><|det|>[[118, 313, 880, 495]]<|/det|>
+2. Will the work be of significance to the field and related fields? How does it compare to the established literature? If the work is not original, please provide relevant references. The significance of the work in relation to the field and related fields lies in its contribution to our understanding of how adherens junctions influence mechanosensation of viscosity. While previous research has established the crucial role of integrin-mediated focal adhesion in mechanosensation of viscoelasticity (Nature Materials, 2021, 20 (9): 1290-1299; Proceedings of the National Academy of Sciences, 2018, 115 (12): E2686-E2695), the present work expands upon this by investigating the impact of cadherins on integrin-mediated mechanosensation of viscosity. But, what are the differences in the antagonistic effects of the two receptors on the elastic and viscous environment, and what is the mechanical mechanism behind this difference?  
+
+<|ref|>text<|/ref|><|det|>[[118, 510, 880, 625]]<|/det|>
+To effectively situate the current work within the established literature, it is important for the authors to cite and compare the previous research on integrin- mediated focal adhesion and highlight the novel insights and observations provided by their study. By doing so, the authors can demonstrate the unique contributions of their work and provide a comprehensive overview of the existing knowledge in the field. This approach not only strengthens the credibility of the current study but also facilitates a clearer understanding of how it advances the current state of knowledge.  
+
+<|ref|>text<|/ref|><|det|>[[118, 641, 880, 756]]<|/det|>
+The reviewer is correct that the novelty of the work lies in putting together integrins and cadherins in response to surface viscosity. Viscosity is an important component of viscoelasticity, a key property of the ECM (Nature 2020, 584: 535- 546; https://doi.org/10.1038/s41586- 020- 2612- 2). Yet, while significant amount of work has been done to understand cell response to elasticity, including the role of integrins and cadherins (Nature Materials 2016, 15(12): 1297- 1306; https://doi.org/10.1038/nmat4725), the role of viscosity is not understood yet. We have expanded the introduction to include this.  
+
+<|ref|>text<|/ref|><|det|>[[118, 789, 825, 821]]<|/det|>
+3. Does the work support the conclusions and claims, or is additional evidence needed? Yes, the evidence is sufficient.  
+
+<|ref|>text<|/ref|><|det|>[[119, 839, 444, 855]]<|/det|>
+We thank the reviewer for their support.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 84, 866, 117]]<|/det|>
+4. Are there any flaws in the data analysis, interpretation and conclusions? Do these prohibit publication or require revision?  
+
+<|ref|>text<|/ref|><|det|>[[119, 133, 448, 150]]<|/det|>
+There are several flaws as list following:  
+
+<|ref|>text<|/ref|><|det|>[[117, 166, 878, 199]]<|/det|>
+(1) Quantitative Difference in Viscosity and Comparison between Experiments and Simulations:  
+
+<|ref|>text<|/ref|><|det|>[[118, 199, 880, 331]]<|/det|>
+It's crucial for the authors to provide a clear quantitative comparison of the viscosity of different supported lipid bilayers (SLBs) such as DOPC and DPPC, as well as glass. This comparison should include specific measurements and values to elucidate the differences in viscosity. Additionally, direct comparisons between the results of experiments and simulations, particularly in Fig. 3e, f, Fig. 4c, and d, would enhance the robustness of the findings and strengthen the conclusions. Providing a side-by-side comparison of experimental and simulated data can help validate the simulation models and their relevance to the experimental observations.  
+
+<|ref|>text<|/ref|><|det|>[[118, 346, 880, 609]]<|/det|>
+We thank the reviewer for raising this point; we have previously quantified the viscosity of the different bilayers (Bennett et al., PNAS 2018; https://doi.org/10.1073/pnas.1710653115) to be \(1 \times 10^{- 6} \mathrm{~Pa} \cdot \mathrm{s} \cdot \mathrm{m}\) for DOPC and \(1 \times 10^{- 4} \mathrm{~Pa} \cdot \mathrm{s} \cdot \mathrm{m}\) for DPPC; the glass control, being not laterally mobile and essentially not deformable by the cell contractile machinery, is considered infinitely viscous in terms of viscosity. We have added this information in the revised version of the paper. We have modified the figures to facilitate a comparison between the trends predicted in the models and observed in the experimental data (Figures 3f and 4d). Specifically, we now show model data starting at a viscosity of DOPC, \(1 \times 10^{- 6} \mathrm{~Pa} \cdot \mathrm{s} \cdot \mathrm{m}\) , and we represent the model predictions at three distinct viscosity values, representing DOPC, DPPC and glass (highest modelled viscosity); predictions for the full ranges of viscosity are now shown in Supplementary Figure 15. We note that an exact match between experiments and model is not expected, as the model is a highly simplified system. Instead, the model is meant to show that the fundamental interactions of a molecular clutch system, combined with competition between integrins and cadherins for actin, can predict the observed trends in actin flows and adhesions as a function of changes in RGD and HAVDI concentrations. Further, the predicted trends take place in a viscosity range that matches the order of measured DOPC and DPPC viscosities.  
+
+<|ref|>text<|/ref|><|det|>[[118, 640, 879, 724]]<|/det|>
+(2) Physiological Relevance of Integrins and Cadherins in Sensing Viscosity: The authors should explicitly discuss the potential in vivo scenarios where integrins and cadherins cooperate to sense the viscosity of the extracellular matrix (ECM). Exploring the physiological relevance of their findings will help establish the broader implications of the research in a biological context.  
+
+<|ref|>text<|/ref|><|det|>[[118, 740, 880, 887]]<|/det|>
+The ECM is viscoelastic in nature and there is increasing interest in understanding cell response to it. Examples such as collective cell migration – that underpins development, cancer progression and regeneration – are explained through integrin and cadherin crosstalks in response to matrix stiffness. Yet, the ECM is viscoelastic and apart from a few seminal papers in the field, some excellent reviews have been published in the last few years (see e.g. Nature 2020, 584: 535–546; https://doi.org/10.1038/s41586-020-2612-2). Our work lies in the context of understanding the role of the viscous part of viscoelasticity and how this impacts integrins and cadherins. Integrins are key to interact with the ECM – of viscoelastic nature – whereas cadherins link cells and their role is modulated by the viscosity of the cell membrane (Adv
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 878, 117]]<|/det|>
+Healthc Mater. 2020;9(8):e1901259; https://doi.org/10.1002/adhm.201901259). This has been discussed in the revised version of the paper.  
+
+<|ref|>text<|/ref|><|det|>[[118, 149, 880, 183]]<|/det|>
+(3) Surface Density of Peptides, Functionalization Efficiency, and Theoretical Distance Among RGD Ligands:  
+
+<|ref|>text<|/ref|><|det|>[[118, 183, 880, 249]]<|/det|>
+It is imperative for the authors to provide detailed information about the surface density of peptides on functionalized SLBs and glass, along with the efficiency of functionalization processes. This information is critical for understanding the experimental setup and ensuring the reproducibility of the results.  
+
+<|ref|>text<|/ref|><|det|>[[118, 249, 879, 297]]<|/det|>
+Furthermore, the method and basis for obtaining the "theoretical distance among RGD ligands of \(44~\mathrm{nm}\) " should be clearly explained and supported by relevant references or experimental validation.  
+
+<|ref|>text<|/ref|><|det|>[[118, 314, 880, 395]]<|/det|>
+Peptide density is regulated via the amount of biotinylated lipid added in the lipid mixture; we have previously confirmed peptide densities via quantitative fluorescence microscopy (Bennett et al., PNAS 2018; https://doi.org/10.1073/pnas.1710653115). The theoretical distance between lipids is calculated as \(1 / \sqrt{n}\) , where \(n\) is the particle density, as now explained in the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[118, 427, 880, 460]]<|/det|>
+5. Is the methodology sound? Does the work meet the expected standards in your field? Yes.  
+
+<|ref|>text<|/ref|><|det|>[[118, 494, 880, 526]]<|/det|>
+6. Is there enough detail provided in the methods for the work to be reproduced? Yes.  
+
+<|ref|>text<|/ref|><|det|>[[119, 543, 444, 559]]<|/det|>
+We thank the reviewer for their support.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 85, 226, 100]]<|/det|>
+## Reviewer #2:  
+
+<|ref|>text<|/ref|><|det|>[[118, 116, 880, 281]]<|/det|>
+This report explores the mechanism of cadherin - integrin crosstalk in response to mobility of their respective ligands. As reviewed properly in the manuscript, this cross- talk has been the subject of previous studies, including the specific ligands used in this study; what is new is a cytoskeleton- based competition mechanism to explain this phenomenon. The fundamental demonstration that engagement of cadherins reduces cell response to integrins, in a mobility- dependent manner, is overall strong. The subsequent examination of the underlying mechanisms is interesting, but incomplete. In summary, two issues - one technical and the other conceptual - limit my overall enthusiasm for this high- potential study. If addressed, this report has potential to advance the field by demonstrating a new mechanism of crosstalk between two signaling pathways.  
+
+<|ref|>text<|/ref|><|det|>[[118, 296, 880, 445]]<|/det|>
+1) Supported lipid bilayer stability. The cellular responses largely look at a 24 hour timepoint, which represents the integration of several complex molecular functions. Underlying much of this is the requirement that the substrate remain stable and intact over the experiment, which is not assured for supported lipid bilayers. Cells can displace or update lipid structures from such a surface. The extent of this type of interaction varies between cells and culture conditions. Some experiment showing that the lipid bilayer remains intact, with few holes through which proteins and cells can reach the underlying surface, is needed to address this issue. It should be certainly explored for the two lipid formulations with both ligands, but need not be an extensive experiment.  
+
+<|ref|>text<|/ref|><|det|>[[118, 460, 879, 527]]<|/det|>
+We agree with the reviewer about the importance of demonstrating the bilayer's stability during culture. We have added a new Supplementary Figure 2i, which shows bilayer stability after 1 and 5 days in the presence of cells; this is done through observation of bilayers containing 0.1 mol% BODIPY-functionalized lipids.  
+
+<|ref|>text<|/ref|><|det|>[[118, 558, 880, 707]]<|/det|>
+2) Mechanism of cross-talk. The proposed mechanism of competition between integrins and cadherins for actin fibers is interesting. However, and as reviewed in the manuscript, cadherin engagement modulates cytoskeletal dynamics through signaling pathways including Rac1 which can alter actin polymerization and receptor cluster formation, leading to complex and somewhat counterintuitive impacts on actin flow. The talin and Jasplakinolide experiments are good steps, but don't address changes in these upstream pathways. Some measure or modulation of actin polymerization activity could help address this issue. Alternatively, engagement of cadherins at locations distinct from integrin interactions might rule out signaling-based effects as there would not be local competition for actin fibers.  
+
+<|ref|>text<|/ref|><|det|>[[118, 722, 880, 886]]<|/det|>
+We thank the reviewer for raising this interesting point. As suggested, we have engaged cadherins at locations distinct from integrins using an anti- N- cadherin antibody to activate dorsal N- cadherins after cells have adhered on the bilayers. We observed that dorsal cadherin engagement (demonstrated through changes in \(\beta\) - catenin signalling) does not lead to a decrease in FA formation on DPPC or glass, as HAVDI- functionalised bilayers instead do. This is shown in new Supplementary Figure 11. Similarly, actin flow is not increased to the same levels as on HAVDI- functionalised bilayers (new Supplementary Figure 13). These observations support the talin overexpression and the jasplakinolide experiments in pointing to the competition between integrins and cadherins for actin fibres as the main mechanism for crosstalk.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 100, 880, 150]]<|/det|>
+Of minor note, there is a moderate level of typographical errors in the manuscript, including the Methods section. These do not dramatically impact the study, but should be addressed if this moves forward to publication.  
+
+<|ref|>text<|/ref|><|det|>[[118, 166, 612, 183]]<|/det|>
+We apologise for these errors, which we have now corrected.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 85, 226, 100]]<|/det|>
+## Reviewer #3:  
+
+<|ref|>text<|/ref|><|det|>[[118, 116, 880, 364]]<|/det|>
+In this article, Barcelona- Estaje et al describe the effect of substrate viscosity on the crosstalk between N- cadherin and integrin adhesion sites. The article contains an impressive amount of data and some interesting results. However, in some cases, the conclusions drawn are not fully supported by the data and some experimental details, especially concerning the data analysis, are missing. As a general comment, I wonder if the 3 types of surfaces used to evaluate the effect of viscosity, these being DPPC, DOPC and glass, were the best possible choice to draw the conclusions that are present in the article. Importantly, the glass surface was functionalized with a concentration of neutravidin 5 times higher than the one used for lipid bilayers and biotin- PEG was added, while in the lipid bilayers, only the biotinylated adhesion ligands were added. In most of the results shown, the difference between lipid bilayers and glass is larger than between DPPC and DOPC layers, and the glass control is often used to 'fix' the trend. My concern is that this surface is too different to be used to set the trend or make a conclusion regarding the increasing viscosity. It would have been more interesting to use a different lipid bilayer, for instance, DMPC, POPC or sphingomyelin layers. It would have also been interesting to see the results of having layers with HAVDI only.  
+
+<|ref|>text<|/ref|><|det|>[[118, 379, 880, 609]]<|/det|>
+We thank the reviewer for raising these issues as it has allowed us to clarify details regarding the material platform used in our study. In terms of viscosity and bilayers choice, we have used DOPC and DPPC because we have previously demonstrated that they provide a range of viscosity able to modulate the molecular clutch engagement (Bennett et al., PNAS 2018; https://doi.org/10.1073/pnas.1710653115). As in the previous study, we consider glass as an infinitely viscous substrate, with the ligands being not laterally mobile and essentially not deformable by the cell contractile machinery. On glass, a higher amount of neutravidin is used compared to the bilayers because the entire surface is coated with neutravidin, which is then bound to defined mixtures of biotin- PEG/biotin- ligand which determine the number of ligands available. On the bilayers the addition of biotin- PEG is instead not necessary because ligands' amounts are controlled by the amount of biotinylated lipid in the bilayer. Using these strategies, the density of ligands is similarly controlled on both the glass surfaces and the bilayers. With regards to bilayers functionalised only with HAVDI, these are not able to support cell adhesion and for this reason they are not included.  
+
+<|ref|>text<|/ref|><|det|>[[118, 640, 879, 707]]<|/det|>
+In addition, I don't really see what the big conclusion of this study is. Is it that the viscosity influences the crosstalk between N- cadherin and integrin? How is this relevant for stem cells, or other cells? Are the values of viscosity used here related to biologically relevant values? Are the effects of the viscosity relevant in a biological setting, where many other cues are present?  
+
+<|ref|>text<|/ref|><|det|>[[118, 723, 879, 887]]<|/det|>
+The reviewer is correct that we demonstrate that viscosity influences the crosstalk between N- cadherins and integrins. Further, we present a mechanism by which N- cadherin binding to ligands modulates integrin mediated- adhesion through competition for acting fibres that in turn reduce the force loading rate via the molecular clutch model. The study is relevant as viscoelasticity is a key property of the ECM and, e.g., stem cells differentiate to osteoblasts on soft hydrogels that maintain elasticity and increased viscosity (i.e. stress relaxation in Nature Materials 2016;15:326- 34; https://doi.org/10.1038/nmat4489). Here, we decouple elasticity from viscosity to understand its role in building up integrin adhesion in an environment where cadherins are present. Of course, we fully agree with the reviewer that in a biological setting in vivo many other factors apart from viscosity would be present. However, isolating its
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 878, 117]]<|/det|>
+contribution, as done here, is useful to understand its potential role in any context where a viscous component is present.  
+
+<|ref|>text<|/ref|><|det|>[[118, 149, 876, 166]]<|/det|>
+More detailed comments on the manuscript can be found in the attached PDF.  
+
+<|ref|>text<|/ref|><|det|>[[118, 183, 880, 315]]<|/det|>
+1. I have some questions regarding the concentration of peptides used, and what they mean. In the methods section the authors wrote that biotinylated lipids were used in different quantities to achieve 0.02, 0.2, 0.22, 2 or \(2.2\%\) of functionalization (correct?). Is this the limiting factor on the functionalization of the bilayers? For instance, for a bilayer with \(0.2\%\) RGD \(+2\%\) HAVDI, a lipid mixture containing \(2.2\%\) of biotinylated lipids is used and a solution of 1:10 of RGD:HAVDI is added to the solution? Can you please clarify this. Since the biotinylated lipids are always DPPC, will there be an effect on the viscosity between the layers with 0.02 and the \(2.2\%\) (for the DOPC samples)?  
+
+<|ref|>text<|/ref|><|det|>[[118, 330, 880, 414]]<|/det|>
+The reviewer is correct. The amount of biotinylated lipids controls the amount of functionalisation, and ligands are then added in the appropriate ratio. We apologise if this was not clear in our methodological section, and we have now addressed this. Varying the amount of biotinylated lipid within the explored ranges does affect bilayer mobility, as measured via fluorescence correlation spectroscopy and shown in the new Supplementary Figure 2j.  
+
+<|ref|>text<|/ref|><|det|>[[118, 444, 879, 511]]<|/det|>
+2. The authors state that they use \(0.2\%\) and \(2\%\) HAVDI concentration as the 'low' and 'high' HAVDI. In supporting figure 4, \(10\%\) HAVDI was used. Can the authors provide some rationale to why the 0.2 and \(2\%\) HAVDI conditions were chosen? And how was the layer with \(10\%\) HAVDI prepared?  
+
+<|ref|>text<|/ref|><|det|>[[118, 527, 879, 577]]<|/det|>
+\(0.02\%\) and \(2\%\) were chosen after an initial screening of concentrations ranging from \(0.02\%\) to \(10\%\) . The layer with \(10\%\) HAVDI was obtained by adding up to \(12\%\) biotinylated lipids in the lipid mixture (depending on the required RGD concentration).  
+
+<|ref|>text<|/ref|><|det|>[[118, 608, 880, 757]]<|/det|>
+3. For the quantification of the peptides in supplementary figure 3 I have a couple of comments. First, it would be useful to have a control with the intensities of the bilayers containing only streptavidin. Then we could also confirm the presence of RGD and HAVDI at the lower concentration. Secondly, the image presented on panel seems brighter, while the quantification shows that there is no difference. Is this because of the LUT of the image or was this not a very representative image? Also, the amount of HAVDI is \(10x\) higher but the fluorescence intensity detected is only \(2 - 2.5x\) higher – any explanations? And, finally, since there are a lot of experiments presented in the paper with varying concentrations of RGD, I would also quantify the amount of RGD only in the layers, to be sure that there is an increase.  
+
+<|ref|>text<|/ref|><|det|>[[118, 772, 880, 903]]<|/det|>
+We thank the reviewer for raising these comments. The images with neutravidin only are not shown because there was no signal present. In terms of concentrations, the images for lower HAVDI concentration are shown in panels b) and f), and their intensity is similar. We have previously confirmed peptide densities using this lipid bilayer platform via quantitative fluorescence microscopy of fluorescent neutravidin (Bennett et al., PNAS 2018; https://doi.org/10.1073/pnas.1710653115). Here, we did not adopt a quantitative fluorescence microscopy technique and instead focussed on comparing single ligands with mixtures to ensure adequate and consistent functionalisation when ligand mixtures were used.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 116, 879, 199]]<|/det|>
+4. What do the author mean by 'a higher intensity of N-cadherin expression' (end page 5). The supplementary figure 3 shows the quantification of the antibody in panel a, but the raw images are not shown. How did the authors quantify this? Is this per cell, an average? Just a cross-section or the whole volume? Depending on how the images are analyzed, the difference can be related to the different cellular localization instead of an expression level.  
+
+<|ref|>text<|/ref|><|det|>[[118, 215, 879, 298]]<|/det|>
+We thank the reviewer for this comment. Images were analyzed by creating a mask of the cell and then using it to define the region of interest for the stained N- Cadherins. Once this was done, the intensity of the N- Cadherin staining was calculated per cell and normalized by the area. Images of the N- Cadherin staining have been added to the supplementary material (Supplementary Figure 3b).  
+
+<|ref|>text<|/ref|><|det|>[[118, 330, 879, 396]]<|/det|>
+5. "when lower concentrations (0.02 % mol) of RGD were tested (i.e. higher RGD spacing), changes in the size of hMSC area due to HAVDI were observed only for DPPC and not for DOPC" – it is not clear why are there differences between DOPC and DPPC in this case; please add some explanation.  
+
+<|ref|>text<|/ref|><|det|>[[118, 411, 879, 477]]<|/det|>
+At very low RGD concentrations and high intermolecular distances, cell attachment and cell spreading are minimal on DOPC. In this condition, the disruption introduced by the addition of HAVDI cannot have any further effect in reducing cell area. This is not the case on DPPC substrates, where cells can still spread despite the low amount of RGD ligands.  
+
+<|ref|>text<|/ref|><|det|>[[118, 509, 879, 576]]<|/det|>
+6. The data presented in figure 1e,f,g and in supplementary figure 4b is partially referent to the same experimental conditions but the trends shown are different. For instance, DPPC with \(0.2\% \mathrm{RGD}\) , the cell area in supplementary figure 4 increases when going from 0.02 to \(0.2\%\) HAVDI, which is not in agreement with the trend discussed in the main text.  
+
+<|ref|>text<|/ref|><|det|>[[118, 592, 879, 674]]<|/det|>
+We thank the reviewer for this observation. In Figures 1e,f,g, we show how HAVDI, which corresponds to \(0.02\%\) , and high HAVDI, which corresponds to \(2\%\) ; \(0.2\%\) HAVDI mentioned by the reviewer is not reported in this figure. In Supplementary Figure 4, we show varying HAVDI concentrations, from 0.02, to 0.2, 2 and \(10\%\) ; the change from 0.02 to \(0.2\%\) HAVDI in Supplementary Figure 4b is not statistically significant.  
+
+<|ref|>text<|/ref|><|det|>[[118, 707, 879, 839]]<|/det|>
+7. Regarding the colocalization of MIIA with the focal adhesions, I think that the data does not support the conclusions in the manuscript. The authors stated "adding HAVDI decreases MIIA levels in eth focal adhesions". By looking at the images, it seems that the presence of HAVDI leads to a much higher expression of MIIA (brighter cells). It is difficult to see if there is MIIA present at the focal adhesions because the level in the cytosol is much higher. The authors show intensity profiles, but it is difficult to interpret these without knowing their location of the image. To retrieve some quantitative analysis, the authors could do some image correlation analysis, rather than showing one intensity profile for each condition.  
+
+<|ref|>text<|/ref|><|det|>[[118, 855, 879, 888]]<|/det|>
+We thank the reviewer for raising this point. Even though we agree that image correlation analysis would be ideal for this statement, it cannot be done with the current images presented
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 878, 117]]<|/det|>
+in this manuscript. In order to try to clarify this, we have added lines where the intensity profiles have been taken so that a clear correlation between figures and profiles can be established.  
+
+<|ref|>text<|/ref|><|det|>[[118, 148, 880, 232]]<|/det|>
+8. On figure 2, the data shown in panel b, and on panels c and d is different, while it is acquired for the experimental conditions (e.g. compare DPPC RGD only in panel b and panel d). Can the authors explain this? This is especially important since in panel d there is no decrease in the nuc/cyt ratio of YAP between RGD only and low HAVDI, while he authors claim a difference in panel b (with \(\mathrm{p}< 0.0001\) )...  
+
+<|ref|>text<|/ref|><|det|>[[118, 248, 880, 347]]<|/det|>
+The decision not to include the comparisons between RGD and low and high HAVDI conditions in every panel was made to avoid overcrowding the graph and to ensure clarity of interpretation. As the data in Figure 2 panel b and panels c- e are the same experimental conditions for RGD and HAVDI, they have now been merged; merged data is shown in panel b. In panels c- e, only the conditions of interest for these panels (HAVDI or scrambled HAVDI) and the corresponding differences (HAVDI vs. scrambled), if present, are now indicated.  
+
+<|ref|>text<|/ref|><|det|>[[118, 379, 880, 461]]<|/det|>
+9. For the differentiation markers (figure 1g, h, i), to be able to draw conclusion on the effect of viscosity, it is necessary to calculate/show the statistical differences between the same condition in different substrates. For a first loo, it seems that the data is too spread to be able to draw any conclusions (differences not statistically significant). This is valid for all the expression markers analysed.  
+
+<|ref|>text<|/ref|><|det|>[[118, 477, 880, 560]]<|/det|>
+We understand the reviewer's concern regarding the need to present statistical differences between the same conditions on different substrates. However, we would like to clarify that while we acknowledge the perceived spread in the data, we have indeed compared the data between conditions on different substrates. We would like to note that all statistical differences among conditions have been now included in Supporting information as Tables 2, 3 and 4.  
+
+<|ref|>text<|/ref|><|det|>[[118, 592, 880, 641]]<|/det|>
+10. "Interestingly, at increasing concentrations of HAVDI, SOX9 expression increased independently of viscosity." Except for the glass, where there seems to be a drop. Can you comment on this?  
+
+<|ref|>text<|/ref|><|det|>[[118, 658, 880, 740]]<|/det|>
+While HAVDI is known to promote chondrogenesis, there are reports of osteogenesis being induced in its presence (Zhu et al. 2016, Biomaterials 77, 44- 52; https://doi.org/10.1016/j.biomaterials.2015.10.072). It can be argued that this is the case on glass, where an increase in pRUNX2 is observed. We have included this observation in the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[118, 773, 880, 822]]<|/det|>
+11. The results regarding the intensity and length of the focal adhesion are difficult to interpret due to the lack of information regarding the analysis. How is the intensity calculated? Is there any image segmentation performed? How?  
+
+<|ref|>text<|/ref|><|det|>[[118, 839, 880, 887]]<|/det|>
+We apologize for not including this in the text. In response to this query, we have incorporated detailed information regarding the calculation of intensity, as well as any image segmentation procedures, in the Methods section of our manuscript.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 100, 878, 151]]<|/det|>
+12. On p14 the authors claim "only for more viscous substrates, smaller FAs when adding HAVDI". The differences between RGD, low HADVI, high HAVDI should be quantified in all viscosity conditions (also graph 3e).  
+
+<|ref|>text<|/ref|><|det|>[[118, 166, 878, 216]]<|/det|>
+We apologise as we don't understand what the reviewer means here. Quantification of the length of FAs are shown in Figure 3e for all the conditions in the paper (i.e. RGD, low HAVDI and high HAVDI).  
+
+<|ref|>text<|/ref|><|det|>[[118, 247, 878, 298]]<|/det|>
+13. For Vinculin, only the quantification is shown (Figure 3e). Please add the fluorescence images as well. Add also the images, for FAK and vinculin, for low HAVDI (even if just in SI).  
+
+<|ref|>text<|/ref|><|det|>[[118, 314, 878, 347]]<|/det|>
+We thank the reviewer for this comment. Representative images for all the conditions have been added as Supplementary Figure 8b.  
+
+<|ref|>text<|/ref|><|det|>[[118, 378, 878, 412]]<|/det|>
+14. The authors claim that "by overexpressing talin, FAs increase." They should calculate statistical significances between each condition for non-transfected and transfected cells.  
+
+<|ref|>text<|/ref|><|det|>[[118, 427, 879, 510]]<|/det|>
+We have considered the reviewer's suggestion and statistical significances between each condition for non-transfected and transfected cells have been calculated and added to the relevant sections of the revised manuscript. We also note that all statistical differences for transfected and not transfected cells have been included in the supporting information as Tables 5, 6 and 7.  
+
+<|ref|>text<|/ref|><|det|>[[118, 542, 879, 641]]<|/det|>
+15. The authors claim that "as expected and in contrast to what we observed in control cells on DPPC, the length of FAs in transfected Y201 cells was not affected by HAVDI" and "as expected and in contrast to what we observed in control cells (= no transfection I assume) on DPPC AND GLASS (?), the length of FAs in transfected Y201 cells was not affected by HAVDI". The authors should show statistical differences between RGD-low-high in transfected cells in all conditions.  
+
+<|ref|>text<|/ref|><|det|>[[118, 657, 879, 723]]<|/det|>
+We thank the reviewer for this comment. Following their suggestion, we have now included the result of the statistical analyses comparing RGD-low-high conditions in transfected cells across all experimental conditions. We note that all statistical differences have been added in the supporting information as Tables 5, 6, and 7.  
+
+<|ref|>text<|/ref|><|det|>[[118, 755, 879, 805]]<|/det|>
+16. "In this model, talin-vinculin and \(\alpha\) -catenin-vinculin would compete with each other to bind to actin, given the limited availability of actin filaments". Could this also be related to the availability of vinculin?  
+
+<|ref|>text<|/ref|><|det|>[[118, 821, 879, 903]]<|/det|>
+The reviewer has raised here a very important point that in essence suggests an alternative hypothesis to the one we included in our manuscript, i.e. whether the competition between integrins and N-cadherins could be for vinculin instead of actin filaments. To demonstrate this, we performed additional experiments measuring the actin flow on DPPC and glass with RGD and RGD+HAVDI using wild type cells and also cells that have been transfected to overexpress
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 879, 168]]<|/det|>
+vinculin. Results are shown in Figure 4f- g and Supplementary Figure 14 and demonstrate that, even in the transfected cells, the actin flow increases for cells on RGD+HAVDI, supporting that the competition is actually for actin filaments. We thank the reviewer for this insightful comment that made us perform additional experiments that validate our hypothesis and strengthen the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[118, 199, 879, 250]]<|/det|>
+17. "When low-viscosity bilayers (DOPC) are functionalized with HAVDI, this functionalization does not affect FA formation". Can the authors provide some hypothesis as to why this is happening?  
+
+<|ref|>text<|/ref|><|det|>[[118, 265, 880, 379]]<|/det|>
+The overarching message of this manuscript is that when the molecular clutch is engaged, then the presence of HAVDI leads to a weakening in cell adhesion as reflected by a decrease in the size of focal adhesions and increase in the actin flow. However, the molecular clutch is not engaged on DOPC when the substrate is functionalised with RGD (as already demonstrated in Bennett et al., PNAS 2018; https://doi.org/10.1073/pnas.1710653115). Therefore, the effect of adding HAVDI on DOPC substrates does not alter focal adhesion formation, which is already minimal before HAVDI addition.  
+
+<|ref|>text<|/ref|><|det|>[[118, 411, 880, 544]]<|/det|>
+18. "hTERT Y201 MSCs behave in the same way as primary hMSCs, being mechanosensitive and with a similar differentiation potential (Supplementary Figure 9)". I disagree with this statement. If the authors want to claim this, they should put both cell lines next to each other in the same graphs. Supplementary figure 9 is only about Y201 cells. For example SOX9 levels seem quite different in both cell lines: primary cells have SOX9 0.05-0.15 (fig 2h) while Y201 have 2-4 (fig S9h). Also, the trend between the RGD-low-high for DPPC and glass are different between the 2 cell lines. And the cyt/nuc ratio for YAP is very different, suggesting that the cells ARE responding differently.  
+
+<|ref|>text<|/ref|><|det|>[[118, 559, 879, 625]]<|/det|>
+We are sorry for the misrepresentation of the Y201 cells in this statement. We meant to say that hTERT Y201 MSCs behave in a similar way as primary hMSCs, maintaining mechanosensitivity and differentiation potential. We have made this clear in the revised version of the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[118, 657, 879, 724]]<|/det|>
+19. "We observed that increased talin expression led to more vinculin being recruited to the site of FAs". So maybe vinculin is drawn away from N-cadherin adhesions, and this is the rate-limiting step (not actin)? This claim could be support by imaging N-cadherin with talin overexpression to check if there was less cadherin present.  
+
+<|ref|>text<|/ref|><|det|>[[118, 740, 879, 855]]<|/det|>
+This comment is related to number 16 before, where the reviewer explores the idea of whether the limiting factor is vinculin instead of actin. Further to the data already mentioned in comment 16 (i.e., actin flow experiments with cells transfected to overexpress vinculin), we have also performed additional experiments to show N- cadherin staining in cells transfected to overexpress talin. In agreement with our hypothesis, high levels of HAVDI still leads to higher expression of N- cadherin for the transfected cells. This is now shown in Supplementary Figure 10a.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 880, 135]]<|/det|>
+20. Add more information on the quantification and statistical analysis. For instance, in many cases, the total number of observations is not mentioned, nor the number of biological replicates.  
+
+<|ref|>text<|/ref|><|det|>[[118, 150, 880, 202]]<|/det|>
+We thank the reviewer for this comment. We have added the number of observations in the figures. The number of biological replicates for all figures in the revised version of the manuscript has been added in the methods section.  
+
+<|ref|>text<|/ref|><|det|>[[115, 236, 880, 255]]<|/det|>
+21. % mol, mol %, % are used interchangeably, make more uniform throughout the Manuscript  
+
+<|ref|>text<|/ref|><|det|>[[118, 270, 880, 304]]<|/det|>
+We apologize about the change of nomenclature. It has been homogenised along the text in the revised version of the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[118, 335, 864, 353]]<|/det|>
+22. Please add also the lower error bars in the bar graphs (in both the manuscript and the SI)  
+
+<|ref|>text<|/ref|><|det|>[[118, 368, 880, 402]]<|/det|>
+Lower error bars in the graph have been added in the revised version of the manuscript (including Supporting Information).  
+
+<|ref|>text<|/ref|><|det|>[[118, 433, 880, 467]]<|/det|>
+23. The letters and text describing the viscosity values in figure 1a are too small to read. Can you also mention the source of these values or how were they obtained?  
+
+<|ref|>text<|/ref|><|det|>[[118, 482, 880, 516]]<|/det|>
+We apologise for this. The size of the text in figure 1a has been increased and their source has also been included in the revised version of the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[118, 548, 880, 565]]<|/det|>
+24. Figure 1c, there seems to be no data for the non-functionalized surface (no error bar visible)  
+
+<|ref|>text<|/ref|><|det|>[[118, 580, 880, 614]]<|/det|>
+On non-functionalized DOPC, there were no cells on any sample, hence the cell density value was 0. This has been made clear in the revised version of the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[118, 646, 880, 680]]<|/det|>
+25. For figure 1h, it would be better to have an overview image (similar to that shown in sup. Fig. 3b). If you opt for showing the individual cells, please center the images...  
+
+<|ref|>text<|/ref|><|det|>[[118, 695, 880, 728]]<|/det|>
+We appreciate the reviewer's observation and overview images have been included for figure 1h in the revised version of the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[118, 761, 880, 794]]<|/det|>
+26. Scale bars: either always in the figures or in the caption (eg. fig 1b versus 1h). Also, please use μm instead of um.  
+
+<|ref|>text<|/ref|><|det|>[[120, 810, 355, 826]]<|/det|>
+This has now been corrected.  
+
+<|ref|>text<|/ref|><|det|>[[118, 860, 547, 877]]<|/det|>
+27. Mention supplementary figure 5 in the main text.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 83, 880, 118]]<|/det|>
+Supplementary Figure 5 is now mentioned in the "N- cadherin ligation affects hMSC mechanosensing and differentiation" section.  
+
+<|ref|>text<|/ref|><|det|>[[118, 148, 565, 167]]<|/det|>
+28. Add the time scales for the kymographs in figure 4.  
+
+<|ref|>text<|/ref|><|det|>[[118, 183, 880, 216]]<|/det|>
+We appreciate this comment and the time scales have been added in the kymographs in Figure 4.  
+
+<|ref|>text<|/ref|><|det|>[[118, 247, 880, 282]]<|/det|>
+29. "at a density of 10000 cells/cm² for fixed cell experiments and 20000 cells/cm² for in-cell westerns assays." Why the different densities? Does this have an impact on the results?  
+
+<|ref|>text<|/ref|><|det|>[[118, 297, 880, 331]]<|/det|>
+We used higher cell densities for In-Cell Western assays because of their low sensitivity, requiring higher numbers of cells to get a measurable signal.  
+
+<|ref|>text<|/ref|><|det|>[[118, 346, 650, 364]]<|/det|>
+30. Please review the methods section of typos (there are a LOT).  
+
+<|ref|>text<|/ref|><|det|>[[118, 379, 783, 397]]<|/det|>
+We apologise for the typos, which have been corrected throughout the manuscript.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[119, 85, 311, 101]]<|/det|>
+## Response to reviewers  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 118, 227, 134]]<|/det|>
+## Reviewer #1:  
+
+<|ref|>text<|/ref|><|det|>[[118, 150, 878, 183]]<|/det|>
+The reviewer is satisfied with the revised manuscript and thus recommend publication of the manuscript in Nature Communications.  
+
+<|ref|>text<|/ref|><|det|>[[118, 199, 867, 215]]<|/det|>
+We are pleased that the reviewer is satisfied and recommends publication of this manuscript.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 232, 227, 248]]<|/det|>
+## Reviewer #2:  
+
+<|ref|>text<|/ref|><|det|>[[118, 264, 878, 313]]<|/det|>
+The additional studies and narrative in this revised manuscript are much appreciated. However, they do not address the two concerns raised in my initial review. I cannot recommend this report for publication, based on the comments below.  
+
+<|ref|>text<|/ref|><|det|>[[120, 330, 401, 346]]<|/det|>
+1) Supported lipid bilayer stability.  
+
+<|ref|>text<|/ref|><|det|>[[118, 346, 879, 476]]<|/det|>
+The images of Supplementary Fig. 2i illustrate the stability problem. Cells are appearing in the fluorescence channel, indicating uptake of the BODIPY- functionalized lipids. The impact of removal of materials at some point will be production of holes, potentially below the limit of optical resolution, that will allow proteins from the media to attach to the surface. The representative image of Day 1 on DPPC in fact shows local depletion of lipids around the four adherent cells in the upper left quadrant of the image. A screenshot of this area with arrows indicating such regions is attached. The contrast has been increased to better highlight these issues.  
+
+<|ref|>text<|/ref|><|det|>[[118, 477, 878, 527]]<|/det|>
+If there is some other explanation for the cells appearing green in these images and the local depletion, it should be discussed in the narrative. Otherwise, reanalysis of key experiments where the analysis focuses on cells not exhibiting local disruption, is needed.  
+
+<|ref|>text<|/ref|><|det|>[[117, 543, 879, 774]]<|/det|>
+The reviewer questions the stability of the substrates and has pointed out the local depletion of lipids around cells. This is a point that we did not note in the previous version of the manuscript and so we need to thank the reviewer again for bringing it up. It is known that cells remodel proteins at the cell- material interface and that when proteins are loosely attached to the substrate they can be 'removed' from it and eventually internalised, e.g. fibronectin on glass (see the pioneering work of Grinnell and Geiger: Cell, 25, 121- 132 (1981) https://doi.org/10.1016/0092- 8674(81)90236- 1 and J Cell Biol 103, 2697 (1986) https://doi.org/10.1083/jcb.103.6.2697; and the work of others such as Altankov J Biomed Mater Res 30, 385 (1996) https://doi.org/10.1002/(SICI)1097- 4636(199603)30:3<385::AID- JBM13>3.0. CO;2- J). This mechanical remodelling at the interface determines the compatibility of substrates. Insomuch that when mechanical remodelling does not happen because, e.g., proteins are strongly attached to the underlying material then cells increase protease secretion and the bioactivity of the interface is compromised (see e.g. our work in Acta Biomater 77, 74 (2018) https://doi.org/10.1016/j.actbio.2018.07.016).  
+
+<|ref|>text<|/ref|><|det|>[[118, 789, 879, 903]]<|/det|>
+This phenomenon highlights the importance of the initial cell- material interactions and how this determines cell response in the mid and long term. The reviewer made the comment of whether these holes in the substrate can be afterwards occupied by, e.g., proteins coming from the media. We would point out that cells very quickly start producing their own ECM (see e.g. Nat Mater 18, 883, 2019 https://doi.org/10.1038/s41563- 019- 0307- 6) and this does not prevent the initial effect from the substrate. We therefore argue that, while in the mid/long terms some reorganization of the bilayers occurs, cellular responses are still governed by the initial
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 879, 168]]<|/det|>
+interactions and hence by the varying degree of viscosity of the substrates. Indeed, if cell response was driven by the effect of these defects, we would then expect a similar response to all surfaces, corresponding to the phenotype of cells seeded directly on glass. Instead, we do observe clear differences in cell behaviour depending on the substrate, here and in our previous work (Bennett et al., PNAS 2018; https://doi.org/10.1073/pnas.1710653115).  
+
+<|ref|>text<|/ref|><|det|>[[118, 183, 878, 216]]<|/det|>
+Following the reviewer's advice we included additional text in the revised version R2 of the manuscript to address this comment.  
+
+<|ref|>text<|/ref|><|det|>[[118, 265, 346, 281]]<|/det|>
+2) Mechanism of cross-talk.  
+
+<|ref|>text<|/ref|><|det|>[[118, 281, 879, 430]]<|/det|>
+The anti- N- cadherin experiments are interesting and much appreciated. However, Supplementary Fig. 13a shows that the anti- N- cadherin application does increase actin flow compared to RGD alone, particularly for DPPC. This could be due to the fact that the antibody could also be reaching the ventral side of the cells, as it is applied in solution. Regardless, addressing this change in actin flow is needed, whether through more extensive discussion on this mechanism on the overall impact of this study or additional experiments that more completely separate the signals. These experiments could include exposure of cells to beads coated with the anti- N- cadherin antibodies or using micropatterned surfaces with small regions of anti- N- cadherin interspersed into the lipid bilayer.  
+
+<|ref|>text<|/ref|><|det|>[[118, 444, 879, 643]]<|/det|>
+We thank the reviewer for pointing out that we did not discuss the increase in actin flow following treatment with anti- N- cadherin on RGD- functionalised DPPC. We acknowledge that a small effect is observed, whereby the actin flow increases after N- cadherin application on RGD- DPPC as seen in Supplementary Figure 13a. However, this effect is smaller than the one observed when ventral HAVDI ligation occurs, and, importantly, is not accompanied by a significant change in focal adhesion formation, as seen in Supplementary Figure 11b. We therefore argue that the local competition for actin fibres between integrins and cadherins remains the key mechanism to regulate cross- talk following ventral HAVDI engagement. As the reviewer suggests, the effect that we see on the actin flow may be due to some of the antibody reaching the ventral side of the cells and eliciting this subtle change. We have improved our discussion of these results in the revised version R2 of the manuscript to point out these changes.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 675, 227, 691]]<|/det|>
+## Reviewer #3:  
+
+<|ref|>text<|/ref|><|det|>[[118, 707, 879, 756]]<|/det|>
+I would like to extend my congratulations to the authors for their heroic effort in addressing all of my comments and remarks. All of my questions have been successfully addressed, and I believe this work is now fit for publication in its current state.  
+
+<|ref|>text<|/ref|><|det|>[[118, 772, 878, 805]]<|/det|>
+We are delighted that the reviewer appreciates the significant amount of work that we put into R1 to address their concerns.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 85, 880, 223]]<|/det|>
+1. I have some questions regarding the concentration of peptides used, and what they mean. In the methods section the authors wrote that biotinylated lipids were used in different quantities to achieve 0.02, 0.2, 0.22, 2 or 2.2 % of functionalization (correct?). Is this the limiting factor on the functionalization of the bilayers? For instance, for a bilayer with \(0.2\% \text{RGD} + 2\% \text{HAVDI}\) , a lipid mixture containing \(2.2\%\) of biotinylated lipids is used and a solution of 1:10 of RGD:HAVDI is added to the solution? Can you please clarify this. Since the biotinylated lipids are always DPPC, will there be an effect on the viscosity between the layers with 0.02 and the \(2.2\%\) (for the DOPC samples)?  
+
+<|ref|>text<|/ref|><|det|>[[118, 225, 879, 292]]<|/det|>
+2. The authors state that they use \(0.2\%\) and \(2\%\) HAVDI concentration as the 'low' and 'high' HAVDI. In supporting figure 4, \(10\%\) HAVDI was used. Can the authors provide some rationale to why the 0.2 and \(2\%\) HAVDI conditions were chosen? And how was the layer with \(10\%\) HAVDI prepared?  
+
+<|ref|>text<|/ref|><|det|>[[118, 294, 880, 466]]<|/det|>
+3. For the quantification of the peptides in supplementary figure 3 I have a couple of comments. First, it would be useful to have a control with the intensities of the bilayers containing only streptavidin. Then we could also confirm the presence of RGD and HAVDI at the lower concentration. Secondly, the image presented on panel seems brighter, while the quantification shows that there is no difference. Is this because of the LUT of the image or was this not a very representative image? Also, the amount of HAVDI is \(10x\) higher but the fluorescence intensity detected is only \(2 - 2.5x\) higher – any explanations? And, finally, since there are a lot of experiments presented in the paper with varying concentrations of RGD, I would also quantify the amount of RGD only in the layers, to be sure that there is an increase.  
+
+<|ref|>text<|/ref|><|det|>[[118, 468, 880, 570]]<|/det|>
+4. What do the author mean by 'a higher intensity of N-cadherin expression' (end page 5). The supplementary figure 3 shows the quantification of the antibody in panel a, but the raw images are not shown. How did the authors quantify this? Is this per cell, an average? Just a cross-section or the whole volume? Depending on how the images are analyzed, the difference can be related to the different cellular localization instead of an expression level.  
+
+<|ref|>text<|/ref|><|det|>[[118, 572, 879, 640]]<|/det|>
+5. "when lower concentrations (0.02 % mol) of RGD were tested (i.e. higher RGD spacing), changes in the size of hMSC area due to HAVDI were observed only for DPPC and not for DOPC" – it is not clear why are there differences between DOPC and DPPC in this case; please add some explanation.  
+
+<|ref|>text<|/ref|><|det|>[[118, 642, 879, 710]]<|/det|>
+6. The data presented in figure 1e,f,g and in supplementary figure 4b is partially referent to the same experimental conditions but the trends shown are different. For instance, DPPC with \(0.2\% \text{RGD}\) , the cell area in supplementary figure 4 increases when going from 0.02 to \(0.2\%\) HAVDI, which is not in agreement with the trend discussed in the main text.  
+
+<|ref|>text<|/ref|><|det|>[[118, 712, 880, 866]]<|/det|>
+7. Regarding the colocalization of MIIA with the focal adhesions, I think that the data does not support the conclusions in the manuscript. The authors stated "adding HAVDI decreases MIIA levels in eth focal adhesions". By looking at the images, it seems that the presence of HAVDI leads to a much higher expression of MIIA (brighter cells). It is difficult to see if there is MIIA present at the focal adhesions because the level in the cytosol is much higher. The authors show intensity profiles, but it is difficult to interpret these without knowing their location of the image. To retrieve some quantitative analysis, the authors could do some image correlation analysis, rather than showing one intensity profile for each condition.  
+
+<|ref|>text<|/ref|><|det|>[[118, 869, 879, 902]]<|/det|>
+8. On figure 2, the data shown in panel b, and on panels c and d is different, while it is acquired for the experimental conditions (e.g. compare DPPC RGD only in panel b and
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[150, 84, 880, 137]]<|/det|>
+panel d). Can the authors explain this? This is especially important since in panel d there is no decrease in the nuc/cyt ratio of YAP between RGD only and low HAVDI, while he authors claim a difference in panel b (with \(p< 0.0001\) )...  
+
+<|ref|>text<|/ref|><|det|>[[118, 138, 880, 223]]<|/det|>
+9. For the differentiation markers (figure 1g, h, i), to be able to draw conclusion on the effect of viscosity, it is necessary to calculate/show the statistical differences between the same condition in different substrates. For a first loo, it seems that the data is too spread to be able to draw any conclusions (differences not statistically significant). This is valid for all the expression markers analysed.  
+
+<|ref|>text<|/ref|><|det|>[[118, 225, 880, 276]]<|/det|>
+10. "Interestingly, at increasing concentrations of HAVDI, SOX9 expression increased independently of viscosity." Except for the glass, where there seems to be a drop. Can you comment on this?  
+
+<|ref|>text<|/ref|><|det|>[[118, 277, 880, 328]]<|/det|>
+11. The results regarding the intensity and length of the focal adhesion are difficult to interpret due to the lack of information regarding the analysis. How is the intensity calculated? Is there any image segmentation performed? How?  
+
+<|ref|>text<|/ref|><|det|>[[118, 330, 880, 380]]<|/det|>
+12. On p14 the authors claim "only for more viscous substrates, smaller FAs when adding HAVDI". The differences between RGD, low HAVDI, high HAVDI should be quantified in all viscosity conditions (also graph 3e).  
+
+<|ref|>text<|/ref|><|det|>[[118, 382, 880, 432]]<|/det|>
+13. For Vinculin, only the quantification is shown (Figure 3e). Please add the fluorescence images as well. Add also the images, for FAK and vinculin, for low HAVDI (even if just in SI).  
+
+<|ref|>text<|/ref|><|det|>[[118, 434, 880, 468]]<|/det|>
+14. The authors claim that "by overexpressing talin, FAs increase." They should calculate statistical significances between each condition for non-transfected and transfected cells.  
+
+<|ref|>text<|/ref|><|det|>[[118, 469, 880, 570]]<|/det|>
+15. The authors claim that "as expected and in contrast to what we observed in control cells on DPPC, the length of FAs in transfected Y201 cells was not affected by HAVDI" and "as expected and in contrast to what we observed in control cells (= no transfection I assume) on DPPC AND GLASS (?), the length of FAs in transfected Y201 cells was not affected by HAVDI". The authors should show statistical differences between RGD-low-high in transfected cells in all conditions.  
+
+<|ref|>text<|/ref|><|det|>[[118, 572, 880, 623]]<|/det|>
+16. "In this model, talin-vinculin and \(\alpha\) -catenin-vinculin would compete with each other to bind to actin, given the limited availability of actin filaments". Could this also be related to the availability of vinculin?  
+
+<|ref|>text<|/ref|><|det|>[[118, 625, 880, 676]]<|/det|>
+17. "When low-viscosity bilayers (DOPC) are functionalized with HAVDI, this functionalization does not affect FA formation". Can the authors provide some hypothesis as to why this is happening?  
+
+<|ref|>text<|/ref|><|det|>[[118, 678, 880, 815]]<|/det|>
+18. "hTERT Y201 MSCs behave in the same way as primary hMSCs, being mechanosensitive and with a similar differentiation potential (Supplementary Figure 9)". I disagree with this statement. If the authors want to claim this, they should put both cell lines next to each other in the same graphs. Supplementary figure 9 is only about Y201 cells. For example SOX9 levels seem quite different in both cell lines: primary cells have SOX9 0.05-0.15 (fig 2h) while Y201 have 2-4 (fig S9h). Also, the trend between the RGD-low-high for DPPC and glass are different between the 2 cell lines. And the cyt/nuc ratio for YAP is very different, suggesting that the cells ARE responding differently.  
+
+<|ref|>text<|/ref|><|det|>[[118, 817, 880, 885]]<|/det|>
+19. "We observed that increased talin expression led to more vinculin being recruited to the site of FAs". So maybe vinculin is drawn away from N-cadherin adhesions, and this is the rate-limiting step (not actin)? This claim could be support by imaging N-cadherin with talin overexpression to check if there was less cadherin present.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 84, 881, 418]]<|/det|>
+20. Add more information on the quantification and statistical analysis. For instance, in many cases, the total number of observations is not mentioned, nor the number of biological replicates.21. % mol, mol %, % are used interchangeably, make more uniform throughout the manuscript22. Please add also the lower error bars in the bar graphs (in both the manuscript and the SI)23. The letters and text describing the viscosity values in figure 1a are too small to read. Can you also mention the source of these values or how were they obtained?24. Figure 1c, there seems to be no data for the non-functionalized surface (no error bar visible)25. For figure 1h, it would be better to have an overview image (similar to that shown in sup. Fig. 3b). If you opt for showing the individual cells, please center the images...26. Scale bars: either always in the figures or in the caption (eg. fig 1b versus 1h). Also, please use μm instead of um.27. Mention supplementary figure 5 in the main text.28. Add the time scales for the kymographs in figure 4.29. "at a density of 10000 cells/cm2 for fixed cell experiments and 20000 cells/cm2 for in-cell westerns assays." Why the different densities? Does this have an impact on the results?30. Please review the methods section of typos (there are a LOT).
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[0, 0, 999, 1005]]<|/det|>
+
+<--- Page Split --->

peer_reviews/e313ef03312a9cd7d704bf32c944399605e13b93b88ea466757c97a2091ebf07/supplementary_0_Peer review File/images_list.json

@@ -0,0 +1,880 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Fig. S4. SEM morphology observation of q2D-cMOF. (a, b) SEM image of q2D-cZ67. (c, d) SEM image of q2D-U66. (e, f) SEM image of q2D-NiMOF. The crystal size of q2D-cZ67 and q2D-cU66 are hexahedrons and octahedrons, measuring 150 nm and 200 nm, respectively. q2D-cNiMOF displays a lamellar microstructure with a diameter of 300 nm",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 0
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_18f.jpg",
+    "caption": "Fig. S18. C1s XPS spectra of q2D-cMOF and q2D-FcMOF. (f) q2D-cU66 Note, the data is added as Supplementary Fig. 18f.",
+    "footnote": [],
+    "bbox": [
+      [
+        310,
+        108,
+        720,
+        301
+      ]
+    ],
+    "page_idx": 7
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_1.jpg",
+    "caption": "Fig. S20. N1s XPS spectra of q2D-cMOF and q2D-FcMOF. (f) q2D-FcNiMOF.",
+    "footnote": [],
+    "bbox": [
+      [
+        310,
+        370,
+        720,
+        562
+      ]
+    ],
+    "page_idx": 8
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3b.jpg",
+    "caption": "Fig. 3b Voltage profiles of 50th Li plating/stripping in Fig. 3a.",
+    "footnote": [],
+    "bbox": [
+      [
+        291,
+        106,
+        720,
+        339
+      ]
+    ],
+    "page_idx": 8
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_2.jpg",
+    "caption": "Fig. S33. Polarization analysis of \\(\\mathrm{Li}||\\mathrm{Cu}\\) . Li nucleation overpotentials of asymmetric \\(\\mathrm{CFx@Li}||\\mathrm{Cu}\\) , \\(\\mathrm{q2D - FcU66@Li}||\\mathrm{Cu}\\) , \\(\\mathrm{q2D - FcNiMOF@Li}||\\mathrm{Cu}\\) , cell in ether electrolytes with a capacity of \\(1\\mathrm{mAhcm^{-2}}\\) at \\(1\\mathrm{mAcm^{-2}}\\)",
+    "footnote": [],
+    "bbox": [
+      [
+        293,
+        402,
+        722,
+        633
+      ]
+    ],
+    "page_idx": 9
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_3.jpg",
+    "caption": "Fig. S15. XPS survey spectra of (a) q2D-cMOF, (b) q2D-FcMOF.",
+    "footnote": [],
+    "bbox": [
+      [
+        275,
+        108,
+        737,
+        500
+      ]
+    ],
+    "page_idx": 9
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3a.jpg",
+    "caption": "Fig. S34. Voltage profiles of Li plating/stripping at different cycle. (a) q2D-FcZ8@Li|Cu, (b) Li|Cu, (c) q2D-FcZ67@Li|Cu, (d) q2D-FcU66@Li|Cu, (e) q2D-FcNiMOF@Li|Cu, (f) CFx@Li|Cu in Fig. 3a",
+    "footnote": [],
+    "bbox": [
+      [
+        176,
+        564,
+        844,
+        821
+      ]
+    ],
+    "page_idx": 10
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3c.jpg",
+    "caption": "Fig. S37. Voltage profiles of Li plating/stripping at different cycle. (a) q2D-FcZ8@Li|Cu, (b) Li|Cu plating/stripping at different cycle number in Fig. 3c. (c) q2D-FcZ8@Li|Cu, (d) Li|Cu plating/stripping at different cycle number in Fig. S36",
+    "footnote": [],
+    "bbox": [
+      [
+        252,
+        110,
+        769,
+        384
+      ]
+    ],
+    "page_idx": 10
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_4.jpg",
+    "caption": "Fig. S40. Tafel plots of symmetric cells with q2D-FcZ8@Li, q2D-FcZ67@Li, bare Li in exchange current density test",
+    "footnote": [],
+    "bbox": [
+      [
+        290,
+        480,
+        707,
+        705
+      ]
+    ],
+    "page_idx": 11
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_5.jpg",
+    "caption": "Fig. S46. Voltage profiles of charge/discharge. (a) q2D-FcZ8@Li||LFP, q2D-FcZ67@Li||LFP, Li||LFP, (c) q2D-FcU66@Li||LFP, q2D-FcNiMOF@Li||LFP, CFx@Li||LFP at 50th cycle. (b) q2D-FcZ8@Li||LFP, q2D-FcZ67@Li||LFP, Li||LFP, (d) q2D-FcU66@Li||LFP, q2D-FcNiMOF@Li||LFP, CFx@Li||LFP at 200th cycle",
+    "footnote": [],
+    "bbox": [
+      [
+        234,
+        110,
+        787,
+        410
+      ]
+    ],
+    "page_idx": 11
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_6.jpg",
+    "caption": "Fig. S47. Voltage profiles of charge/discharge. (a) q2D-FcZ8@Li||LFP, (b) q2D-FcZ67@Li||LFP, (c) Li||LFP, (d) q2D-FcU66@Li||LFP, (e) q2D-FcNiMOF@Li||LFP, (f) CFx@Li||LFP at different cycles",
+    "footnote": [],
+    "bbox": [
+      [
+        168,
+        559,
+        860,
+        813
+      ]
+    ],
+    "page_idx": 12
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_7.jpg",
+    "caption": "Fig. S49. Voltage profiles of charge/discharge. (a) q2D-FcZ8@Li||LFP, (b) q2D-FcZ67@Li||LFP, (c) Li||LFP, (d) q2D-FcU66@Li||LFP, (e) q2D-FcNiMOF@Li||LFP, (f) CFx@Li||LFP at different current density",
+    "footnote": [],
+    "bbox": [
+      [
+        168,
+        113,
+        860,
+        358
+      ]
+    ],
+    "page_idx": 12
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_8.jpg",
+    "caption": "Fig. S17. Metal XPS spectra of q2D-cMOF and q2D-FcMOF. Zn2p XPS spectra of (a) q2D-cZ8, (c) q2D-FcZ8. Co2p XPS spectra of (b) q2D-cZ67, (f) q2D-FcZ67. Zr3d XPS spectra of (c) q2D-cU66, (g) q2D-FcU66. Ni2p XPS spectra of (d) q2D-cNiMOF, (h) q2D-FcNiMOF",
+    "footnote": [],
+    "bbox": [
+      [
+        245,
+        111,
+        775,
+        575
+      ]
+    ],
+    "page_idx": 13
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_9.jpg",
+    "caption": "Fig. S20. N1s XPS spectra of q2D-cMOF and q2D-FcMOF. (a) q2D-cZ8 and (b) q2D-FcZ8, (c) q2D-cZ67 and (d) q2D-FcZ67, (e) q2D-cNiMOF and (f) q2D-FcNiMOF.",
+    "footnote": [],
+    "bbox": [
+      [
+        240,
+        112,
+        785,
+        472
+      ]
+    ],
+    "page_idx": 14
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_10.jpg",
+    "caption": "Fig. S46. Voltage profiles of charge/discharge. (a) q2D-FcZ8@Li||LFP, q2D-FcZ67@Li||LFP, Li||LFP, (c) q2D-FcU66@Li||LFP, q2D-FcNiMOF@Li||LFP, CFx@Li||LFP at 50th cycle. (b) q2D-FcZ8@Li||LFP, q2D-FcZ67@Li||LFP, Li||LFP, (d) q2D-FcU66@Li||LFP, q2D-FcNiMOF@Li||LFP, CFx@Li||LFP at 200th cycle",
+    "footnote": [],
+    "bbox": [
+      [
+        238,
+        110,
+        784,
+        407
+      ]
+    ],
+    "page_idx": 15
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_11.jpg",
+    "caption": "Fig. S28 FT-IR spectra of LiTFSI mixed with q2D-FcMOF",
+    "footnote": [],
+    "bbox": [
+      [
+        257,
+        106,
+        744,
+        384
+      ]
+    ],
+    "page_idx": 16
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_12.jpg",
+    "caption": "Fig. S29 \\(^{7}\\mathrm{Li}\\) NMR of LiTFSI mixed with q2D-FcMOF.",
+    "footnote": [],
+    "bbox": [
+      [
+        232,
+        475,
+        777,
+        732
+      ]
+    ],
+    "page_idx": 19
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_13.jpg",
+    "caption": "Fig. S30. Fittings of signal intensity in pulsed field gradient nuclear magnetic resonance (PFG-NMR) as a function of gradient strength for the (a-d) Li+, (e-h) TFSI- in electrolyte with different q2DFcMOF.",
+    "footnote": [],
+    "bbox": [
+      [
+        250,
+        113,
+        770,
+        703
+      ]
+    ],
+    "page_idx": 19
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_31.jpg",
+    "caption": "Fig. S31. Self-diffusivities of ionic species in electrolyte with different q2D-FcMOF. Note, the data is added as Supplementary Fig. 31.",
+    "footnote": [],
+    "bbox": [
+      [
+        234,
+        118,
+        770,
+        325
+      ]
+    ],
+    "page_idx": 20
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_14.jpg",
+    "caption": "Fig. S32. Current time curves and EIS Nyquist spectra of symmetric cells before, after polarization test using (a) q2D-FcZ8@Li, (b) q2D-FcZ67@Li, (c) q2D-FcU66@Li, (d) q2D-FcNiMOF@Li, (e) bare Li",
+    "footnote": [],
+    "bbox": [
+      [
+        217,
+        110,
+        810,
+        790
+      ]
+    ],
+    "page_idx": 21
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5a.jpg",
+    "caption": "Fig. 5a. The charge density differences and maximum adsorption energy of q2D-FcMOF adsorbed Li (The yellow and blue represent the charge accumulate and loss region).",
+    "footnote": [],
+    "bbox": [
+      [
+        110,
+        494,
+        891,
+        655
+      ]
+    ],
+    "page_idx": 22
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_15.jpg",
+    "caption": "Fig. S26. XRD spectra the double-layer ASEI on the Li deposition: before cycling and after 50th cycling.",
+    "footnote": [],
+    "bbox": [
+      [
+        228,
+        177,
+        770,
+        469
+      ]
+    ],
+    "page_idx": 23
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_16.jpg",
+    "caption": "Li-Zn phase diagram (Copyright 1991, Springer)",
+    "footnote": [],
+    "bbox": [
+      [
+        214,
+        548,
+        780,
+        835
+      ]
+    ],
+    "page_idx": 24
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_17.jpg",
+    "caption": "Fig. S21. Surface morphology of the (a, b) q2D-FcZ8@Li",
+    "footnote": [],
+    "bbox": [
+      [
+        174,
+        108,
+        838,
+        290
+      ]
+    ],
+    "page_idx": 24
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_18.jpg",
+    "caption": "Fig. S27. Raman spectra the double-layer ASEI on the Li deposition: before cycling and after 50th cycling.",
+    "footnote": [],
+    "bbox": [
+      [
+        268,
+        380,
+        737,
+        633
+      ]
+    ],
+    "page_idx": 26
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2b.jpg",
+    "caption": "Fig 2b. The strength of each group of TOF-SIMS of the double-layer ASEI on the q2D-FcZ8@Li before cycling under \\(\\mathrm{Cs}^+\\) sputtering.",
+    "footnote": [],
+    "bbox": [
+      [
+        193,
+        131,
+        767,
+        358
+      ]
+    ],
+    "page_idx": 26
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_19.jpg",
+    "caption": "Fig. S2. The different nucleated LiF crystal phases are not fully fused with tiny defects that contribute to the transport of \\(\\mathrm{Li}^{+}\\)",
+    "footnote": [],
+    "bbox": [
+      [
+        156,
+        280,
+        861,
+        420
+      ]
+    ],
+    "page_idx": 27
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3a.jpg",
+    "caption": "Fig. S35. Nyquist plots at 10th, 50th, 100th, 200th cycle in Fig. 3a. (a) q2D-FcZ8@Li|Cu, (b) Li|Cu, (c) q2D-FcZ67@Li|Cu, (d) q2D-FcNiMOF@Li|Cu, (e) q2D-FcU66@Li|Cu, (f) CFx@Li|Cu",
+    "footnote": [],
+    "bbox": [
+      [
+        141,
+        282,
+        880,
+        565
+      ]
+    ],
+    "page_idx": 28
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_20.jpg",
+    "caption": "Fig. S43. Voltage profiles of charge/discharge. (a) q2D-FcZ8||Li in the voltage range of 0.5 to 3.0 V, (b) q2D-FcZ8||Li in the voltage range of 0.5 t 4.5 V.",
+    "footnote": [],
+    "bbox": [
+      [
+        163,
+        115,
+        827,
+        306
+      ]
+    ],
+    "page_idx": 29
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_21.jpg",
+    "caption": "Fig. S44. Cyclic Voltammetry results from the first five laps of \\(\\mathrm{Li}\\| \\mathrm{q2D - FcZ8@Li}\\)",
+    "footnote": [],
+    "bbox": [
+      [
+        210,
+        230,
+        732,
+        513
+      ]
+    ],
+    "page_idx": 31
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2a.jpg",
+    "caption": "Fig. 2a 3D rendering of the TOF-SIMS of the double-layer ASEI on the q2D-FcZ8@Li before cycling under \\(\\mathrm{Cs}^+\\) sputtering.",
+    "footnote": [],
+    "bbox": [
+      [
+        280,
+        106,
+        728,
+        328
+      ]
+    ],
+    "page_idx": 32
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_22.jpg",
+    "caption": "Fig. S22. 3D rendering of the TOF-SIMS of the double-layer ASEI on the q2D-FcZ8@Li (a) before and (b) after 50th cycling under \\(\\mathrm{Cs}^+\\) sputtering.",
+    "footnote": [],
+    "bbox": [
+      [
+        295,
+        420,
+        728,
+        841
+      ]
+    ],
+    "page_idx": 35
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2b.jpg",
+    "caption": "Fig. 2b. The strength of each group of TOF-SIMS of the double-layer ASEI on the q2D-FcZ8@Li before cycling under \\(\\mathrm{Cs}^+\\) sputtering.",
+    "footnote": [],
+    "bbox": [
+      [
+        210,
+        180,
+        750,
+        393
+      ]
+    ],
+    "page_idx": 35
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_23.jpg",
+    "caption": "Fig. S23. The strength of each group of TOF-SIMS of the double-layer ASEI on the q2D-FcZ8@Li after 50th cycling under \\(\\mathrm{Cs}^+\\) sputtering.",
+    "footnote": [],
+    "bbox": [
+      [
+        210,
+        499,
+        750,
+        711
+      ]
+    ],
+    "page_idx": 36
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Fig. 2 XPS spectra at various depths of the SEI on the Li deposition: Zn2p, C1s, F1s, and Li1s spectra of q2D-FcZ8@Li (c) before cycling, (d) after 50th cycling.",
+    "footnote": [],
+    "bbox": [
+      [
+        110,
+        106,
+        904,
+        300
+      ]
+    ],
+    "page_idx": 36
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_24.jpg",
+    "caption": "Fig. S26. XRD spectra the double-layer ASEI on the Li deposition: before cycling and after 50th cycling.",
+    "footnote": [],
+    "bbox": [
+      [
+        240,
+        410,
+        757,
+        692
+      ]
+    ],
+    "page_idx": 37
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_25.jpg",
+    "caption": "Fig. S27. Raman spectra the double-layer ASEI on the Li deposition: before cycling and after 50th cycling.",
+    "footnote": [],
+    "bbox": [
+      [
+        274,
+        135,
+        732,
+        380
+      ]
+    ],
+    "page_idx": 37
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2e.jpg",
+    "caption": "Fig. 2e SEM image of surface morphology for bare Li and q2D-FcZ8 at different deposition rates. Note, the data is added as Fig. 2e.",
+    "footnote": [],
+    "bbox": [
+      [
+        193,
+        295,
+        817,
+        465
+      ]
+    ],
+    "page_idx": 38
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_26.jpg",
+    "caption": "Fig. S53. Surface morphology of the (a, b) q2D-FcZ8@Li and (c, d) bare Li electrodes paired with LFP and NCM811 after the different cycle",
+    "footnote": [],
+    "bbox": [
+      [
+        205,
+        530,
+        813,
+        845
+      ]
+    ],
+    "page_idx": 39
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_27.jpg",
+    "caption": "Fig. S4. SEM morphology observation of q2D-cMOF. (a, b) SEM image of q2D-cZ67. (c, d) SEM image of q2D-U66. (e, f) SEM image of q2D-NiMOF. The crystal size of q2D-cZ67 and q2D-cU66 are hexahedrons and octahedrons, measuring \\(150\\mathrm{nm}\\) and \\(200\\mathrm{nm}\\) , respectively. q2D-cNiMOF displays a lamellar microstructure with a diameter of \\(300\\mathrm{nm}\\)",
+    "footnote": [],
+    "bbox": [
+      [
+        147,
+        280,
+        840,
+        530
+      ]
+    ],
+    "page_idx": 39
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_28.jpg",
+    "caption": "Fig. S3. SEM morphology observation of q2D-cZ8. (a) SEM image of q2D-cZ8 under the scale of 100 nm and 50 nm",
+    "footnote": [],
+    "bbox": [
+      [
+        270,
+        110,
+        748,
+        370
+      ]
+    ],
+    "page_idx": 43
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_1e.jpg",
+    "caption": "Fig. 1e. STEM image with corresponding elemental mapping of q2D-FcZ8",
+    "footnote": [],
+    "bbox": [
+      [
+        219,
+        697,
+        790,
+        858
+      ]
+    ],
+    "page_idx": 44
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_29.jpg",
+    "caption": "Fig. S15b. XPS survey spectra of q2D-FcMOF",
+    "footnote": [],
+    "bbox": [
+      [
+        188,
+        130,
+        828,
+        400
+      ]
+    ],
+    "page_idx": 44
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Fig. 2 XPS spectra at various depths of the SEI on the Li deposition: Zn2p, C1s, F1s, and Li1s spectra of q2D-FcZ8@Li (b) before cycling, (c) after 50th cycling.",
+    "footnote": [],
+    "bbox": [
+      [
+        130,
+        105,
+        872,
+        290
+      ]
+    ],
+    "page_idx": 45
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2c.jpg",
+    "caption": "Fig. 2c and 2d after modification",
+    "footnote": [],
+    "bbox": [
+      [
+        110,
+        378,
+        907,
+        574
+      ]
+    ],
+    "page_idx": 46
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_30.jpg",
+    "caption": "Fig. S25a. XPS spectra at various depths of the SEI on the Li deposition: Zn2p, C1s, F1s, and Li1s spectra after 50th cycling.",
+    "footnote": [],
+    "bbox": [
+      [
+        174,
+        118,
+        840,
+        391
+      ]
+    ],
+    "page_idx": 46
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_31.jpg",
+    "caption": "Fig. S18. C1s XPS spectra of q2D-cMOF and q2D-FcMOF. (a) q2D-cZ8, (b) q2D-cZ67, (c) q2D-cU66, (d) q2D-cNiMOF, (e) q2D-FcZ67, (f) q2D-cU66, (g) q2D-FcNiMOF",
+    "footnote": [],
+    "bbox": [
+      [
+        175,
+        115,
+        855,
+        715
+      ]
+    ],
+    "page_idx": 47
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_32.jpg",
+    "caption": "Fig. S20. N1s XPS spectra of q2D-cMOF and q2D-FcMOF. (a) q2D-cZ8 and (b) q2D-FcZ8, (c) q2D-cZ67 and (d) q2D-FcZ67, (e) q2D-cNiMOF and (f) q2D-FcNiMOF",
+    "footnote": [],
+    "bbox": [
+      [
+        170,
+        115,
+        857,
+        568
+      ]
+    ],
+    "page_idx": 48
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_55.jpg",
+    "caption": "Fig. S55. Li1s XPS spectra of air-exposed bare Li for different time periods Note, the data is added as Supplementary Fig. 55.",
+    "footnote": [],
+    "bbox": [
+      [
+        204,
+        108,
+        816,
+        446
+      ]
+    ],
+    "page_idx": 49
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_56.jpg",
+    "caption": "Fig. S56. F1s XPS spectra of air-exposed Li. (a) bare Li (b) q2D-FcZ8@Li for different time periods Note, the data is added as Supplementary Fig. 56.",
+    "footnote": [],
+    "bbox": [
+      [
+        179,
+        115,
+        860,
+        565
+      ]
+    ],
+    "page_idx": 50
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_58.jpg",
+    "caption": "Fig. S58. C1s XPS spectra of air-exposed Li. (a) bare Li (b) q2D-FcZ8@Li for different time periods Note, the data is added as Supplementary Fig. 58.",
+    "footnote": [],
+    "bbox": [
+      [
+        171,
+        115,
+        858,
+        551
+      ]
+    ],
+    "page_idx": 51
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2a.jpg",
+    "caption": "Fig. 2a 3D rendering of the TOF-SIMS of the double-layer ASEI on the q2D-FcZ8@Li after cycling under \\(\\mathrm{Cs}^+\\) sputtering.",
+    "footnote": [],
+    "bbox": [
+      [
+        266,
+        370,
+        746,
+        608
+      ]
+    ],
+    "page_idx": 52
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_33.jpg",
+    "caption": "Fig. S22a. 3D rendering of the TOF-SIMS of the double-layer ASEI on the q2D-FcZ8@Li after cycling under \\(\\mathrm{Cs}^+\\) sputtering.",
+    "footnote": [],
+    "bbox": [
+      [
+        264,
+        688,
+        747,
+        813
+      ]
+    ],
+    "page_idx": 53
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2b.jpg",
+    "caption": "Fig. 2b. The strength of each group of TOF-SIMS of the double-layer ASEI on the q2D-FcZ8@Li after cycling under \\(\\mathrm{Cs}^+\\) sputtering",
+    "footnote": [],
+    "bbox": [
+      [
+        210,
+        130,
+        750,
+        344
+      ]
+    ],
+    "page_idx": 53
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2e.jpg",
+    "caption": "Fig. 2e SEM image of surface morphology for bare Li and q2D-FcZ8 at different deposition rates (current density in Fig. 2e4 is \\(1\\mathrm{mAcm}^{-2}\\) ).",
+    "footnote": [],
+    "bbox": [
+      [
+        155,
+        103,
+        855,
+        295
+      ]
+    ],
+    "page_idx": 54
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_34.jpg",
+    "caption": "Fig. S53. Surface morphology of the (a, b) q2D-FcZ8@Li and (c, d) bare Li electrodes paired with LFP and NCM811 after the different cycle",
+    "footnote": [],
+    "bbox": [
+      [
+        163,
+        111,
+        857,
+        472
+      ]
+    ],
+    "page_idx": 55
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_35.jpg",
+    "caption": "Fig. S59. XRD patterns evolution of pristine \\(\\mathrm{q2D - FcZ8}\\) (a) and bare Li (b) after exposure to air for different times",
+    "footnote": [],
+    "bbox": [
+      [
+        170,
+        335,
+        840,
+        560
+      ]
+    ],
+    "page_idx": 56
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4i.jpg",
+    "caption": "Fig. 4i. The specific capacity of air-exposed \\(\\mathrm{q2D - FcZ8@Li}\\) and bare Li for different time periods Note, the data is added as Fig. 4i.",
+    "footnote": [],
+    "bbox": [
+      [
+        308,
+        662,
+        704,
+        861
+      ]
+    ],
+    "page_idx": 57
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3i.jpg",
+    "caption": "Fig. 3i. Comparison of current density and the corresponding cycle life of the q2D-FcZ8@Li with the representative protected Li metal anode (Data is obtained from Table S1 in Supporting Information).",
+    "footnote": [],
+    "bbox": [
+      [
+        200,
+        297,
+        800,
+        476
+      ]
+    ],
+    "page_idx": 57
+  }
+]

peer_reviews/e9d0f5a2deb03447593c3ad30539546fd4d0753326f2e48a3f07a7dfc4c10876/supplementary_0_Peer Review File NEW/images_list.json

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

peer_reviews/e9d0f5a2deb03447593c3ad30539546fd4d0753326f2e48a3f07a7dfc4c10876/supplementary_0_Peer Review File NEW/supplementary_0_Peer Review File NEW.mmd

@@ -0,0 +1,89 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+# Physiological Basis for Atmospheric Methane Oxidation and Methanotrophic Growth on Air  
+
+![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.
+
+<--- Page Split --->
+
+Reviewer #1 (Remarks to the Author):  
+
+I read the latest version of the manuscript and the accompanying rebuttal with keen interest. My primary concern from the previous revision pertains to the difficulty in computing the cell- specific energy requirements of trace gas- oxidizing microorganisms. Specifically, I raised questions about the assessment of the number of living cells on the membranes used to measure trace gas (TG) oxidation rates.  
+
+The challenge lies in accurately measuring this variable, and the authors addressed it by the monitoring of N2 incorporation in biomass by NanoSIMS to justify their assumption that all cells were alive and fixed N2 during the assay. While their argument is plausible, the estimates remain somewhat delicate due to the inherent difficulty in gauging the number of cells on the membrane (potential loss of cells during the washing steps and cell sorting), the indirect evidence of living cells supported by labeled N incorporated in cell biomass (TG oxidation rate measurements were decoupled from NanoSIMS experiments), and the potential utilization of other energy sources in the air.  
+
+I believe it is important to acknowledge these complexities in the manuscript section that compares the energy yield with previous estimates. That being said, I believe the work represents a significant progress towards a better understanding of atmospheric chemosynthesis.  
+
+Reviewer #2 (Remarks to the Author):  
+
+The revised manuscript from Schmider et al. investigates the ability of a number of methanotrophs to grow on air by utilising the trace gases CH4, H2, and CO as a source of energy, as well as nitrogen sources (N2 and otherwise).  
+
+The authors have paid careful attention to the queries and concerns of myself and the other reviewers, and their revision of the manuscript and explanations in the response to reviewers have done much to strengthen the manuscript. In it's current form the manuscript significantly expands on our understanding of bacterial trace gas oxidation and makes a strong case that methanotrophs are able to utilise air as their sole source of energy and nitrogen.  
+
+I especially appreciate the authors inclusion of a control experiment showing that M. gorgona MG08 is unable to grow when supplied with air which lacks trace quantities of CH4, CO, H2. In response to my comments the reviewers point out the difficulty of creating a contaminant free environment, that doesn't contain substrates in addition of atmospheric CH4, CO, H2, that may contribute to growth. I am sympathetic of the difficulty of these experiments. However, if the authors wish to state that these methanotrophs are growing on air, it is important that they demonstrate that the reduced trace gases in air are required as at least the main sources of energy for growth. Extraordinary claims require extraordinary evidence. In this case, I believe it is sufficient to show this for one of the 4 species they identify as able to live on air. But in future work, I would suggest that these controls should be included as standard.  
+
+Minor comments:  
+
+Pg. 6 line 127 - 'performed' not 'performed'  
+
+Reviewer #3 (Remarks to the Author):  
+
+I have reviewed the author's responses to the reviewers, including the comments and questions I submitted (Reviewer 3). In all cases, I think they have provided appropriate answers, or have made the necessary changes.
+
+<--- Page Split --->
+
+I thank the authors for this attention to detail.
+
+<--- Page Split --->
+
+## REVIEWERS' COMMENTS  
+
+We would like to thank the reviewers for their constructive comments and suggestions throughout the review process that were of great help in improving the overall quality of the study.  
+
+Reviewer #1 (Remarks to the Author):  
+
+I read the latest version of the manuscript and the accompanying rebuttal with keen interest. My primary concern from the previous revision pertains to the difficulty in computing the cell- specific energy requirements of trace gas- oxidizing microorganisms. Specifically, I raised questions about the assessment of the number of living cells on the membranes used to measure trace gas (TG) oxidation rates.  
+
+The challenge lies in accurately measuring this variable, and the authors addressed it by the monitoring of N2 incorporation in biomass by NanoSIMS to justify their assumption that all cells were alive and fixed N2 during the assay. While their argument is plausible, the estimates remain somewhat delicate due to the inherent difficulty in gauging the number of cells on the membrane (potential loss of cells during the washing steps and cell sorting), the indirect evidence of living cells supported by labeled N incorporated in cell biomass (TG oxidation rate measurements were decoupled from NanoSIMS experiments), and the potential utilization of other energy sources in the air.  
+
+I believe it is important to acknowledge these complexities in the manuscript section that compares the energy yield with previous estimates. That being said, I believe the work represents a significant progress towards a better understanding of atmospheric chemosynthesis.  
+
+We have included the following sentence at the end of the relevant section to make readers aware of the complexities:  
+
+Line 204 - 206: "However, we acknowledge that differences in activity between individual cells and potential inaccuracies in the quantification of cells contributing to observed activities might have introduced a minor error to our estimations."  
+
+Reviewer #2 (Remarks to the Author):  
+
+The revised manuscript from Schmider et al. investigates the ability of a number of methanotrophs to grow on air by utilising the trace gases CH4, H2, and CO as a source of energy, as well as nitrogen sources (N2 and otherwise).  
+
+The authors have paid careful attention to the queries and concerns of myself and the other reviewers, and their revision of the manuscript and explanations in the response to reviewers have done much to strengthen the manuscript. In it's current form the manuscript significantly expands on our understanding of bacterial trace gas oxidation and makes a strong case that methanotrophs are able to
+
+<--- Page Split --->
+
+utilise air as their sole source of energy and nitrogen.  
+
+I especially appreciate the authors inclusion of a control experiment showing that M. gorgona MG08 is unable to grow when supplied with air which lacks trace quantities of CH4, CO, H2. In response to my comments the reviewers point out the difficulty of creating a contaminant free environment, that doesn't contain substrates in addition of atmospheric CH4, CO, H2, that may contribute to growth. I am sympathetic of the difficulty of these experiments. However, if the authors wish to state that these methanotrophs are growing on air, it is important that they demonstrate that the reduced trace gases in air are required as at least the main sources of energy for growth. Extraordinary claims require extraordinary evidence. In this case, I believe it is sufficient to show this for one of the 4 species they identify as able to live on air. But in future work, I would suggest that these controls should be included as standard.  
+
+Minor comments:  
+
+Pg. 6 line 127 - 'performed' not 'performed'  
+
+The word has been changed accordingly.  
+
+Reviewer #3 (Remarks to the Author):  
+
+I have reviewed the author's responses to the reviewers, including the comments and questions I submitted (Reviewer 3). In all cases, I think they have provided appropriate answers, or have made the necessary changes.  
+
+I thank the authors for this attention to detail.
+
+<--- Page Split --->

peer_reviews/e9d0f5a2deb03447593c3ad30539546fd4d0753326f2e48a3f07a7dfc4c10876/supplementary_0_Peer Review File NEW/supplementary_0_Peer Review File NEW_det.mmd

@@ -0,0 +1,119 @@
+<|ref|>title<|/ref|><|det|>[[61, 40, 508, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[67, 110, 362, 139]]<|/det|>
+Peer Review File  
+
+<|ref|>title<|/ref|><|det|>[[90, 154, 907, 211]]<|/det|>
+# Physiological Basis for Atmospheric Methane Oxidation and Methanotrophic Growth on Air  
+
+<|ref|>image<|/ref|><|det|>[[56, 732, 240, 783]]<|/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, 878, 133]]<|/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.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[120, 84, 414, 97]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 111, 864, 182]]<|/det|>
+I read the latest version of the manuscript and the accompanying rebuttal with keen interest. My primary concern from the previous revision pertains to the difficulty in computing the cell- specific energy requirements of trace gas- oxidizing microorganisms. Specifically, I raised questions about the assessment of the number of living cells on the membranes used to measure trace gas (TG) oxidation rates.  
+
+<|ref|>text<|/ref|><|det|>[[118, 195, 864, 308]]<|/det|>
+The challenge lies in accurately measuring this variable, and the authors addressed it by the monitoring of N2 incorporation in biomass by NanoSIMS to justify their assumption that all cells were alive and fixed N2 during the assay. While their argument is plausible, the estimates remain somewhat delicate due to the inherent difficulty in gauging the number of cells on the membrane (potential loss of cells during the washing steps and cell sorting), the indirect evidence of living cells supported by labeled N incorporated in cell biomass (TG oxidation rate measurements were decoupled from NanoSIMS experiments), and the potential utilization of other energy sources in the air.  
+
+<|ref|>text<|/ref|><|det|>[[119, 321, 870, 364]]<|/det|>
+I believe it is important to acknowledge these complexities in the manuscript section that compares the energy yield with previous estimates. That being said, I believe the work represents a significant progress towards a better understanding of atmospheric chemosynthesis.  
+
+<|ref|>text<|/ref|><|det|>[[120, 405, 414, 419]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[119, 433, 875, 476]]<|/det|>
+The revised manuscript from Schmider et al. investigates the ability of a number of methanotrophs to grow on air by utilising the trace gases CH4, H2, and CO as a source of energy, as well as nitrogen sources (N2 and otherwise).  
+
+<|ref|>text<|/ref|><|det|>[[118, 489, 876, 560]]<|/det|>
+The authors have paid careful attention to the queries and concerns of myself and the other reviewers, and their revision of the manuscript and explanations in the response to reviewers have done much to strengthen the manuscript. In it's current form the manuscript significantly expands on our understanding of bacterial trace gas oxidation and makes a strong case that methanotrophs are able to utilise air as their sole source of energy and nitrogen.  
+
+<|ref|>text<|/ref|><|det|>[[118, 573, 874, 714]]<|/det|>
+I especially appreciate the authors inclusion of a control experiment showing that M. gorgona MG08 is unable to grow when supplied with air which lacks trace quantities of CH4, CO, H2. In response to my comments the reviewers point out the difficulty of creating a contaminant free environment, that doesn't contain substrates in addition of atmospheric CH4, CO, H2, that may contribute to growth. I am sympathetic of the difficulty of these experiments. However, if the authors wish to state that these methanotrophs are growing on air, it is important that they demonstrate that the reduced trace gases in air are required as at least the main sources of energy for growth. Extraordinary claims require extraordinary evidence. In this case, I believe it is sufficient to show this for one of the 4 species they identify as able to live on air. But in future work, I would suggest that these controls should be included as standard.  
+
+<|ref|>text<|/ref|><|det|>[[119, 756, 252, 770]]<|/det|>
+Minor comments:  
+
+<|ref|>text<|/ref|><|det|>[[119, 783, 453, 798]]<|/det|>
+Pg. 6 line 127 - 'performed' not 'performed'  
+
+<|ref|>text<|/ref|><|det|>[[119, 826, 414, 840]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[119, 854, 866, 897]]<|/det|>
+I have reviewed the author's responses to the reviewers, including the comments and questions I submitted (Reviewer 3). In all cases, I think they have provided appropriate answers, or have made the necessary changes.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 97, 473, 111]]<|/det|>
+I thank the authors for this attention to detail.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 125, 303, 140]]<|/det|>
+## REVIEWERS' COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[115, 167, 857, 200]]<|/det|>
+We would like to thank the reviewers for their constructive comments and suggestions throughout the review process that were of great help in improving the overall quality of the study.  
+
+<|ref|>text<|/ref|><|det|>[[115, 226, 392, 241]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 260, 856, 343]]<|/det|>
+I read the latest version of the manuscript and the accompanying rebuttal with keen interest. My primary concern from the previous revision pertains to the difficulty in computing the cell- specific energy requirements of trace gas- oxidizing microorganisms. Specifically, I raised questions about the assessment of the number of living cells on the membranes used to measure trace gas (TG) oxidation rates.  
+
+<|ref|>text<|/ref|><|det|>[[114, 360, 880, 476]]<|/det|>
+The challenge lies in accurately measuring this variable, and the authors addressed it by the monitoring of N2 incorporation in biomass by NanoSIMS to justify their assumption that all cells were alive and fixed N2 during the assay. While their argument is plausible, the estimates remain somewhat delicate due to the inherent difficulty in gauging the number of cells on the membrane (potential loss of cells during the washing steps and cell sorting), the indirect evidence of living cells supported by labeled N incorporated in cell biomass (TG oxidation rate measurements were decoupled from NanoSIMS experiments), and the potential utilization of other energy sources in the air.  
+
+<|ref|>text<|/ref|><|det|>[[115, 494, 880, 543]]<|/det|>
+I believe it is important to acknowledge these complexities in the manuscript section that compares the energy yield with previous estimates. That being said, I believe the work represents a significant progress towards a better understanding of atmospheric chemosynthesis.  
+
+<|ref|>text<|/ref|><|det|>[[115, 579, 875, 612]]<|/det|>
+We have included the following sentence at the end of the relevant section to make readers aware of the complexities:  
+
+<|ref|>text<|/ref|><|det|>[[115, 622, 828, 672]]<|/det|>
+Line 204 - 206: "However, we acknowledge that differences in activity between individual cells and potential inaccuracies in the quantification of cells contributing to observed activities might have introduced a minor error to our estimations."  
+
+<|ref|>text<|/ref|><|det|>[[115, 699, 392, 714]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 732, 880, 781]]<|/det|>
+The revised manuscript from Schmider et al. investigates the ability of a number of methanotrophs to grow on air by utilising the trace gases CH4, H2, and CO as a source of energy, as well as nitrogen sources (N2 and otherwise).  
+
+<|ref|>text<|/ref|><|det|>[[115, 799, 867, 865]]<|/det|>
+The authors have paid careful attention to the queries and concerns of myself and the other reviewers, and their revision of the manuscript and explanations in the response to reviewers have done much to strengthen the manuscript. In it's current form the manuscript significantly expands on our understanding of bacterial trace gas oxidation and makes a strong case that methanotrophs are able to
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 124, 503, 139]]<|/det|>
+utilise air as their sole source of energy and nitrogen.  
+
+<|ref|>text<|/ref|><|det|>[[115, 157, 875, 325]]<|/det|>
+I especially appreciate the authors inclusion of a control experiment showing that M. gorgona MG08 is unable to grow when supplied with air which lacks trace quantities of CH4, CO, H2. In response to my comments the reviewers point out the difficulty of creating a contaminant free environment, that doesn't contain substrates in addition of atmospheric CH4, CO, H2, that may contribute to growth. I am sympathetic of the difficulty of these experiments. However, if the authors wish to state that these methanotrophs are growing on air, it is important that they demonstrate that the reduced trace gases in air are required as at least the main sources of energy for growth. Extraordinary claims require extraordinary evidence. In this case, I believe it is sufficient to show this for one of the 4 species they identify as able to live on air. But in future work, I would suggest that these controls should be included as standard.  
+
+<|ref|>text<|/ref|><|det|>[[115, 336, 247, 350]]<|/det|>
+Minor comments:  
+
+<|ref|>text<|/ref|><|det|>[[115, 368, 434, 384]]<|/det|>
+Pg. 6 line 127 - 'performed' not 'performed'  
+
+<|ref|>text<|/ref|><|det|>[[117, 410, 411, 425]]<|/det|>
+The word has been changed accordingly.  
+
+<|ref|>text<|/ref|><|det|>[[115, 470, 392, 485]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 503, 864, 553]]<|/det|>
+I have reviewed the author's responses to the reviewers, including the comments and questions I submitted (Reviewer 3). In all cases, I think they have provided appropriate answers, or have made the necessary changes.  
+
+<|ref|>text<|/ref|><|det|>[[115, 571, 451, 586]]<|/det|>
+I thank the authors for this attention to detail.
+
+<--- Page Split --->

peer_reviews/eb68748872af85199990c7aa3967e103ea56513dea1570d6c6e723d67ca9c019/supplementary_1_Peer Review File./images_list.json

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

peer_reviews/eb68748872af85199990c7aa3967e103ea56513dea1570d6c6e723d67ca9c019/supplementary_1_Peer Review File./supplementary_1_Peer Review File..mmd

@@ -0,0 +1,546 @@
+
+# nature portfolio  
+
+# Peer Review Information  
+
+Journal: Nature Microbiology  Manuscript Title: Pseudomonas putida mediates bacterial killing, biofilm invasion and biocontrol with a type IVB secretion system  Corresponding author name(s): Leo Eberl
+
+<--- Page Split --->
+
+# nature portfolio  
+
+## Reviewer Comments & Decisions:  
+
+## Decision Letter, initial version:  
+
+Dear Professor Eberl,  
+
+Thank you for your patience while your manuscript "A type IVB secretion system adapted for bacterial killing, biofilm invasion and biocontrol" was under peer- review at Nature Microbiology. It has now been seen by 3 referees, whose expertise and comments you will find at the end of this email. Although they find your work of some potential interest, they have raised a number of concerns that will need to be addressed before we can consider publication of the work in Nature Microbiology.  
+
+In particular, you will see that referee #1 raise concerns regarding the plant experiment and asks you to use a more biologically relevant approach, and suggests you to tone down some statements and improve the data presentation and description. Referee #2 raises issues regarding the lack of data availability. Referee #3 mainly comments on the need for more mechanistic insight to strengthen the manuscript.  
+
+Should further experimental data allow you to address these criticisms, we would be happy to look at a revised manuscript.  
+
+We are committed to providing a fair and constructive peer- review process. Please do not hesitate to contact us if there are specific requests from the reviewers that you believe are technically impossible or unlikely to yield a meaningful outcome.  
+
+We strongly support public availability of data. Please place the data used in your paper into a public data repository, if one exists, or alternatively, present the data as Source Data or Supplementary Information. If data can only be shared on request, please explain why in your Data Availability Statement, and also in the correspondence with your editor. For some data types, deposition in a public repository is mandatory - more information on our data deposition policies and available repositories can be found at https://www.nature.com/nature- research/editorial- policies/reporting- standards#availability- of- data.  
+
+Please include a data availability statement as a separate section after Methods but before references, under the heading "Data Availability". This section should inform readers about the availability of the data used to support the conclusions of your study. This information includes accession codes to public repositories (data banks for protein, DNA or RNA sequences, microarray, proteomics data etc...), references to source data published alongside the paper, unique identifiers such as URLs to data repository entries, or data set DOIs, and any other statement about data availability. At a minimum, you should include the following statement: "The data that support the findings of this study are
+
+<--- Page Split --->
+
+# nature portfolio  
+
+available from the corresponding author upon request", mentioning any restrictions on availability. If DOIs are provided, we also strongly encourage including these in the Reference list (authors, title, publisher (repository name), identifier, year). For more guidance on how to write this section please see:  
+
+http://www.nature.com/authors/policies/data/data- availability- statements- data- citations.pdf  
+
+If revising your manuscript:  
+
+\* Include a "Response to referees" document detailing, point- by- point, how you addressed each referee comment. If no action was taken to address a point, you must provide a compelling argument. This response will be sent back to the referees along with the revised manuscript.  
+
+\* If you have not done so already we suggest that you begin to revise your manuscript so that it conforms to our Article format instructions at http://www.nature.com/nmicrobiol/info/final- submission. Refer also to any guidelines provided in this letter.  
+
+\* Include a revised version of any required reporting checklist. It will be available to referees (and, potentially, statisticians) to aid in their evaluation if the manuscript goes back for peer review. A revised checklist is essential for re- review of the paper.  
+
+When submitting the revised version of your manuscript, please pay close attention to our href="https://www.nature.com/nature- research/editorial- policies/image- integrity">Digital Image Integrity Guidelines.</a> and to the following points below:  
+
+- that unprocessed scans are clearly labelled and match the gels and western blots presented in figures. 
+- that control panels for gels and western blots are appropriately described as loading on sample processing controls 
+- all images in the paper are checked for duplication of panels and for splicing of gel lanes.  
+
+Finally, please ensure that you retain unprocessed data and metadata files after publication, ideally archiving data in perpetuity, as these may be requested during the peer review and production process or after publication if any issues arise.  
+
+Please use the link below to submit a revised paper:  
+
+{redacted}  
+
+<strong>Note: </strong> This url links to your confidential homepage and associated information about manuscripts you may have submitted or be reviewing for us. If you wish to forward this e- mail to co- authors, please delete this link to your homepage first.
+
+<--- Page Split --->
+
+# nature portfolio  
+
+Nature Microbiology is committed to improving transparency in authorship. As part of our efforts in this direction, we are now requesting that all authors identified as 'corresponding author' on published papers create and link their Open Researcher and Contributor Identifier (ORCID) with their account on the Manuscript Tracking System (MTS), prior to acceptance. This applies to primary research papers only. ORCID helps the scientific community achieve unambiguous attribution of all scholarly contributions. You can create and link your ORCID from the home page of the MTS by clicking on 'Modify my Springer Nature account'. For more information please visit please visit <a href="http://www.springernature.com/orcid">www.springernature.com/orcid</a>.  
+
+If you wish to submit a suitably revised manuscript we would hope to receive it within 6 months. If you cannot send it within this time, please let us know. We will be happy to consider your revision, even if a similar study has been accepted for publication at Nature Microbiology or published elsewhere (up to a maximum of 6 months).  
+
+In the meantime we hope that you find our referees' comments helpful.  
+
+Yours sincerely,  
+
+{redacted}  
+
+\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*  
+
+Reviewer Expertise:  
+
+Referee #1: microbial ecology of plant bacteria/Pseudomonas Referee #2: Type IV secretion systems Referee #3: plant microbe signaling/QS  
+
+Reviewer Comments:  
+
+Reviewer #1 (Remarks to the Author):  
+
+This very interesting study has done a good job of describing a novel form of contact- dependent killing of bacteria. Contact- dependent killing schemes have become quite the rage in microbial ecology, and yet most have been associated with variants of type VI secretion systems. This study has shown a very novel method of delivering what appear to be toxic effectors using a variant of the type IVB secretion system. The study is quite solid, and the authors have employed a variety of powerful tools including Tn sequencing and various microscopy techniques to both demonstrate the contact dependence of the killing of various other bacteria by the Pseudomonas putida strain, and to identify the genetic loci associated with both the toxic effector as well as the immunity function. It is quite remarkable that this type IVB secretion system seems to be quite promiscuous in its ability to kill other bacteria, as they did not seem to find any taxa that were immune to its effect. They might want to comment further on the relative breadth of this killing compared to many other type VI systems that have been studied. As such, there are indeed important translational aspects of the study, as exemplified here in their demonstration that the biological control of bacterial wilt caused by Ralstonia solanacearum could be achieved using this strain capable of contact killing. The manuscript was quite
+
+<--- Page Split --->
+
+# nature portfolio  
+
+well written, and the powerful and logical experimental design leading to the discovery of the system was well considered. I have only a few relatively minor comments about both the way certain findings should be described, a few conclusions that may be a bit overstated, as well as suggestions for how the demonstration of plant disease control could have been better assessed and described.  
+
+## Specifics:  
+
+Lines 67- 69. We are told here that a Xanthomonas strain had exhibited a type IV dependent killing of other bacterial strains - while it is only in a couple of sentences later that the authors note that there is more than one kind of type IV secretion system, and that there had been no prior demonstration of the type IVB system being used in antimicrobial activities. However, there was never a mention until much later in the discussion that the Xanthomonas example, was in fact, a type IVA secretion system. I feel that this point should have been made earlier, because I was bothered by the concern that this study might not have been as novel as they were suggesting, because they had not noted that this earlier example was not a type IVB system.  
+
+Line 90 and elsewhere. Here and in many locations throughout the manuscript, the authors have used the term "competition" to describe the interaction of IsoF with various other bacteria. At least in my mind, the term "competition" implies some sort of growth reduction that would be a result of the need to share certain nutrient resources with a neighbor, or perhaps an effect of the chemical environment by a neighbor etc. Given that they are demonstrating here that "competition" is actually the direct killing of their neighbor, I wonder if it would be cleaner to refer to "interaction" or "killing" or some other term that does not seem to have this nutrient competition connotation.  
+
+## Line 97 misspelling: iodide  
+
+Line 112. I had to take a step back and try to figure out why the authors had decided to use Pseudomonas aureofaciens in the study discussed in this section because there had not been a previous mention that this particular species was in fact susceptible to the killing effect of strain IsoF. One had to dig into the results of Figure 1, and specifically recognize that Pseudomonas aureofaciens was one of the strains that was in fact susceptible to the killing by IsoF. I think it would benefit from a short note to discuss why there had been a change in the killing assay away from KT2442 to this new bacterial target.  
+
+Line 126: Again, in this sentence they use the term "outcompete" when it's clearly was killing the P. aureofaciens. I think it would be better to note killing rather than competition in such examples.  
+
+Line 210: it is a bit disconcerting to hear that they were not able to restore killing by complimentation. I do not recall there being a discussion of this negative finding.  
+
+Line 236 through 238. Upon initially reading the sentence, I found it surprising and unexpected, since it did not seem like the simple process of killing a bacterium would cause it to disappear from the biofilm. It was only later when the authors noted that this replacement of the dead cells by the IsoF strain only occurred in flow chambers where turbulence etc could in fact have dislodged the dead cells. That said, after reading the entire manuscript I was left with a feeling that there is a bit of an
+
+<--- Page Split --->
+
+# nature portfolio  
+
+overstatement in the authors conclusions that the IsoF strain could invade biofilms, as this seemed to have been restricted to the flow cells where there is some ability to remove cells from the biofilm, and was not observed in pre- established biofilms that had formed on agar plates etc. This made me wonder whether the invasion of biofilms that might have formed on plants and other habitats where they feel the type IV secretion system could be useful in killing target organisms was as likely to occur as they suggest. I feel a bit of further clarification of the situations in which biofilm invasion/replacement is likely to occur is warranted.  
+
+There were several aspects of the plant protection experiment that I found a bit awkward or simplistic. The experiment, using small seedlings growing on agar plates is a somewhat unnatural setting, and one that might have facilitated both microbial survival on, and multiplication on the damaged plant tissue. The experiment was done in a way that would have maximized the potential effect of the presence of IsoF together with the pathogen, since they were mixed together and then immediately applied to the wounded plant. There thus would have been maximal opportunity for IsoF to have been in close proximity to the pathogen and thus for it to have killed it before the pathogen could have multiplied and caused disease. I was disappointed that the authors did not attempt to do a more biologically relevant experiment in which plants grown in some sort of soil matrix or even sand would have had the soil flooded simultaneously with a mixture of the bacteria, or even better, flooded first with the pathogen and then shortly thereafter with the beneficial strain. This would have been a more powerful test of the ability of IsoF to kill developing biofilms on the roots, thereby alleviating the likelihood of disease. These would have been quite simple studies to perform and would better reflect the normal process of infection. I was surprised at the way the authors presented the information on the effects of IsoF on the disease process, relating root weight, leaf area, chlorophyll content etc in a PCO plot in figure 5C. I had never seen such results presented in this way, and found it very difficult to interpret. I feel it would have been much cleaner to have simply prepared a small table showing the effect of the three or four treatments on each of these dependent variables.  
+
+Line 372 to 375: this important statement was not given the weight that I feel it deserved, as it does point to perhaps a limited scenario in which type IVB killing will prove to be biologically important.  
+
+Line 388 - 389: this sentence seems to be a bit of an overstatement given what they have just noted on lines 372 to 375.  
+
+I found the discussion a very interesting and pleasant read, although I had the strong impression that it reiterated almost word for word statements that had been made in the results. Some condensation seems possible.  
+
+Reviewer #2 (Remarks to the Author):  
+
+The manuscript by Purtschert- Montenegro et al describes the phenomenon of contact- dependent killing of other bacterial species by the soil bacterium Pseudomonas putida IsoF, shown to be mediated by a Type IV secretion system coded by the newly named kib cluster. This phenomenon has
+
+<--- Page Split --->
+
+# natureportfolio  
+
+been shown before in Xanthomonas and Stenotrophomonas bacteria. The difference here is that the P. putida isoF T4SS belongs to the larger Type IVB class while that of X. citri and S. maltophilia belong to the Type IVA systems. The authors use fluorescently- labelled knockout strains to show that the killing is dependent on structural components of the T4SS apparatus as well as on a putative effector coded by the kib locus. The authors go on to perform some elegant experiments that show that the kib locus, and specifically the effector Piso_02333, allow the IsoF strain to invade biofilms formed by the P. putida K2442 strain (that does not carry a T4SS, but rather a bacteria- killing T6SS). They then show that the IsoF strain can prevent the phytopathogenic bacterium Ralstonia solanacearum from infecting tomato seedlings. The manuscript is very well written, the results are clearly explained and presented and the experiments appear to have been carefully carried out. I do, however have a few comments and suggestions that I believe should be addressed.  
+
+Perhaps my major criticism (easily addressed) is that the reader is not provided with the accession numbers of the P. putida IsoF genome nor the accession numbers of the genes coded by the kib locus. Therefore the reader cannot confer whether the genes described in the paper do in fact correspond to the specified T4SSB components. For example, Lines 139- 145: Here the kib cluster coding 61 genes is described with reference to Figure 2b and Extended Data Table 1. However, in neither this figure or this table, nor anywhere else in the manuscript or in the Supplementary Information are the accession numbers provided for the Pseudomonas putida IsoF strain genome or the genes in the kib cluster. I searched for these sequences in the NCBI database and could not find any of them. Also, searching for the gene names Piso_02333 and Piso_02332 in these databases did not turn up any hits. Finally, no Data Availability Statement was provided, which should contain these accession codes. Is it possible case that the genome of this organism has not yet been made publicly available? I think that it should definitely be deposited and released before being sent off for review. Especially since this strain has been the subject of study by the Eberle group for at least two decades (for example: Steidle et al, 2001. doi: 10.1128/AEM.67.12.5761- 5770.2001. ). The authors should provide a table with all accession numbers of the 61 genes genes in the kib locus. If this is not possible, then the full nucleotide and translated protein sequences each open reading frame in the kib locus should be provided in fasta format as supplementary information.  
+
+Another point, in a way related to that above, is that the nature of the Piso_02332/Piso_02333 pair is not explored to any significant extent. How many homologs of these two proteins are found in the public databases and what can we say about their phylogenetic distribution (individually or as a pair)? Is the Piso_02332 immunity protein expected to be localized in the cytosol of the periplasmid (does it have a signal peptide?). This could provide clues regarding the site of action of its cognate effector. Regarding the Piso_02333 effector: Does it have any putative motifs that could be used as a recognition signal for secretion by the T4SSB apparatus?  
+
+Other points.  
+
+Lines 67- 69. A bacterial killing T4SS has been characterized in X. citri and Stenotrophomonas maltophilia (Souza et al, 2015; Bayer- Santos et al, 2019). And homologous systems have been identified in over a hundred other bacterial species (Sgro et al, 2019).  
+
+Lines 71- 73: The authors imply that DNA transfer is mediated only by class A systems while class B
+
+<--- Page Split --->
+
+# nature portfolio  
+
+systems have until now been restricted to the role of transferring effectors into eukaryotic cells. This is not quite true. For example, Class B T4SSs are responsible for the horizontal transfer (conjugation) of IncI plasmids.  
+
+Lines 113- 116: Eight out of sixteen killing defective mutants localize to four genes in the kib cluster. What about the other eight insertion mutants? What were their insertion sites?  
+
+Lines 103,106- 107 and 296: in several places in the manuscript, the term "host range" is used to represent competitor bacterial species. This is not really appropriate. A more suitable term would be "range of target species" or "range of target organisms".  
+
+Line 162: the references 22, 32, 55 and 66 refer to studies on only T6SS effector- immunity protein pairs. However, a few thousand bacteria- killing T4SSA effector- immunity protein pairs have been identified in the genomes of over a hundred species.  
+
+Lines 202- 204. It is reported that the delta32- 33 strain grows more slowly than the wild- type strain or the delta33 strain. This could suggest that Piso_02332 may be neutralizing more than one effector.  
+
+Lines 208- 211: Why were you not able to restore killing of P. aureofaciens and KT2442 by complementing the mutant delta32- 33 strain with a plasmid pBBR::32- 33 coding for the effector- immunity pair. Did the authors confirm that this plasmid in fact expresses the two proteins?  
+
+Lines 226- 229: It is written that after 3 days of incubation, isoF had formed a mature biofilm by invading and replacing the KT2442 biofilm. This would correspond to time point 120 hours (48 hrs of KT2442 growth on its own plus 72 after addition of IsoF). No Figure is mentioned to show this. Figures 4a and 4b only show growth up to two days after isoF addition to KT2442 biofilms (total of 96 hours).  
+
+Lines 281- 283: In Figure 5d, why are the deltaT4B CFUs the same as the IsoF CFUs? I would expect the IsoF numbers to be greater than the deltaT4B numbers in these experiments.  
+
+Lines 298- 302: This is only partially correct. All of the Xanthomonadaceae- like bacterial killing T4SSs have one or more effector/immunity pairs at the same locus containing all of the structural components of the secretion system PLUS other effector/immunity pairs found in other chromosomal locations (Sgro et al, 2019). Here, since we have no access the genome sequence of the IsoF strain under study, we have no way to investigate whether other possible effector/immunity pairs are found in the genome.  
+
+Reviewer #3 (Remarks to the Author):  
+
+This study reports that a P. putida strain contains a locus encoding a typeIVB secretion system which is involved in bacterial killing via cell- cell contact. This system possesses the novel feature since typeIV SS have been reported thus far to only infect eukaryotic cells. Other aspects of this work include that the locus is most likely found in a genomic island which has been recently acquired and
+
+<--- Page Split --->
+
+# nature portfolio  
+
+that this locus is not widespread since it is only found thus far in a bunch of Pseudomonas isolates. This work reports on an interesting novel locus found in a plant associated bacterium with a possible role in competition as well as biocontrol. This work is well performed, described and discussed in the context of microbial ecology. The weakness is the lack of mechanistic data and insight on the mechanism(s) of this system and these initial interesting results raise several questions which need more attention in order to make this work more tangible and less 'preliminary'.  
+
+Apparently, the method of killing is not via lysis however no insight on possible targets or mechanism is provided; any data on this aspect would considerably increase the impact of this article. The effector is thought to be the 33 locus/protein, has any further experimentation been performed to confirm unequivocally that this is the effector and on its possible target and mechanism?  
+
+It is surprising that mutants in this locus significantly affect bacterial growth since these are thought to be accessory present in a recently acquired genomic island. This aspect is is unclear and needs more attention/explanation.  
+
+No reference is made on whether this system is also able to infect eukaryotic cells or should it be considered specific for bacteria- bacteria interactions? Has this aspect been tested?  
+
+The immunity aspect of this system is not entirely clear since two loci appear to be involved; one major locusPiso_02332 encodes for an immunity protein, however no mechanistic insight is provided and in addition it is not tested whether this locus alone can provide immunity if transferred to other bacteria.  
+
+## Author Rebuttal to Initial comments  
+
+## Reviewer #1 (Remarks to the Author):  
+
+This very interesting study has done a good job of describing a novel form of contact- dependent killing of bacteria. Contact- dependent killing schemes have become quite the rage in microbial ecology, and yet most have been associated with variants of type VI secretion systems. This study has shown a very novel method of delivering what appear to be toxic effectors using a variant of the type IVB secretion system. The study is quite solid, and the authors have employed a variety of powerful tools including Tn sequencing and various microscopy techniques to both demonstrate the contact dependence of the killing of various other bacteria by the Pseudomonas putida strain, and to identify the genetic loci associated with both the toxic effector as well as the immunity function. It is quite remarkable that this type IVB secretion system seems to be quite promiscuous in its ability to kill other bacteria, as they did not seem to find any taxa that were immune to its effect. They might  
+
+want to comment further on the relative breadth of this killing compared to many other type VI systems that have been studied. As such, there are indeed important translational aspects of the study, as exemplified here in their demonstration that the biological control of bacterial wilt caused by Ralstonia solanacearum could be achieved using this strain capable of contact killing. The manuscript was quite
+
+<--- Page Split --->
+
+# nature portfolio  
+
+well written, and the powerful and logical experimental design leading to the discovery of the system was well considered. I have only a few relatively minor comments about both the way certain findings should be described, a few conclusions that may be a bit overstated, as well as suggestions for how the demonstration of plant disease control could have been better assessed and described.  
+
+We are very grateful for the positive evaluation of our study and the helpful comments to improve our manuscript.  
+
+Specifics:  
+
+Lines 67- 69. We are told here that a Xanthomonas strain had exhibited a type IV dependent killing of other bacterial strains - while it is only in a couple of sentences later that the authors note that there is more than one kind of type IV secretion system, and that there had been no prior demonstration of the type IVB system being used in antimicrobial activities. However, there was never a mention until much later in the discussion that the Xanthomonas example, was in fact, a type IVA secretion system. I feel that this point should have been made earlier, because I was bothered by the concern that this study might not have been as novel as they were suggesting, because they had not noted that this earlier example was not a type IVB system.  
+
+The criticism was noticed and we now specify that previous work in Xanthomonas identified a T4ASS in contrast to the T4BSS we identified in P. putida IsoF.  
+
+Line 90 and elsewhere. Here and in many locations throughout the manuscript, the authors have used the term "competition" to describe the interaction of IsoF with various other bacteria. At least in my mind, the term "competition" implies some sort of growth reduction that would be a result of the need to share certain nutrient resources with a neighbor, or perhaps an effect of the chemical environment by a neighbor etc. Given that they are demonstrating here that "competition" is actually the direct killing of their neighbor, I wonder if it would be cleaner to refer to "interaction" or "killing" or some other term that does not seem to have this nutrient competition connotation.  
+
+We are thankful for this valuable comment. Accordingly, we have changed the term competition to 'interaction' or 'killing' depending on the context throughout the text. We also changed 'contact- dependent competition (CDC)' to 'contact- dependent killing (CDK)'.  
+
+Line 97 misspelling: iodide  
+
+## Corrected  
+
+Line 112. I had to take a step back and try to figure out why the authors had decided to use
+
+<--- Page Split --->
+
+# nature portfolio  
+
+Pseudomonas aureofaciens in the study discussed in this section because there had not been a previous mention that this particular species was in fact susceptible to the killing effect of strain IsoF. One had to dig into the results of Figure 1, and specifically recognize that Pseudomonas aureofaciens was one of the strains that was in fact susceptible to the killing by IsoF. I think it would benefit from a short note to discuss why there had been a change in the killing assay away from KT2442 to this new bacterial target.  
+
+Thank you for making us aware of this inconsistency. We have chosen P. aureofaciens in these assays because this strain was more sensitive to killing by IsoF than KT2442. We have added this information in the revised version of the manuscript.  
+
+Line 126: Again, in this sentence they use the term "outcompete" when it's clearly was killing the P. aureofaciens. I think it would be better to note killing rather than competition in such examples.  
+
+## Changed to kill as suggested.  
+
+Line 210: it is a bit disconcerting to hear that they were not able to restore killing by complementation. I do not recall there being a discussion of this negative finding.  
+
+To address the concern of the reviewer, we have analyzed the strains by SDS- PAGE in order to investigate ectopic expression of PisoF_02332 and PisoF_02333. In the complemented strain, we observed a band corresponding to the of the immunity protein but could not detect the effector, indicating that the immunity protein is produced in excess over the toxin. We therefore hypothesize that killing was not restored as a consequence of an unphysiological overexpression of the immunity protein in the complemented strain that effectively neutralized all effector molecules. We have added this information to the Results section and show the SDS- PAGE in the new Fig. 8b of the Extended Data.  
+
+Line 236 through 238. Upon initially reading the sentence, I found it surprising and unexpected, since it did not seem like the simple process of killing a bacterium would cause it to disappear from the biofilm. It was only later when the authors noted that this replacement of the dead cells by the IsoF strain only occurred in flow chambers where turbulence etc could in fact have dislodged the dead cells. That said, after reading the entire manuscript I was left with a feeling that there is a bit of an overstatement in the authors conclusions that the IsoF strain could invade biofilms, as this seemed to have been restricted to the flow cells where there is some ability to remove cells from the biofilm, and was not observed in preestablished biofilms that had formed on agar plates etc. This made me wonder whether the invasion of biofilms that might have formed on plants and other habitats where they feel the type IV secretion system could be useful in killing target organisms was
+
+<--- Page Split --->
+
+as likely to occur as they suggest. I feel a bit of further clarification of the situations in which biofilm invasion/replacement is likely to occur is warranted.  
+
+This is a very valuable and interesting comment. To investigate whether IsoF could also invade a biofilm grown on an agar plate we followed the fate of the strain in CDK assays against KT2442 over 72 hours. While growth of IsoF was restricted to the initial inoculation area after 24 hours, we observed that IsoF invaded the space occupied by the target strain and formed satellite colonies after 72 hours. We hypothesize that killed cells eventually lyse and no longer form a barrier that prevents invasion. Another important factor for invasion competence is that IsoF produces the very powerful biosurfactant putisolin, which was not only shown to disperse pre- established biofilms but also allows the strain to translocate over semisolid surfaces by means of swarming motility. We have added this information in the Results section, as new Fig. 12 in the Extended Data, and amended the discussion to clarify this issue.  
+
+There were several aspects of the plant protection experiment that I found a bit awkward or simplistic. The experiment, using small seedlings growing on agar plates is a somewhat unnatural setting, and one that might have facilitated both microbial survival on, and multiplication on the damaged plant tissue. The experiment was done in a way that would have maximized the potential effect of the presence of IsoF together with the pathogen, since they were mixed together and then immediately applied to the wounded plant. There thus would have been maximal opportunity for IsoF to have been in close proximity to the pathogen and thus for it to have killed it before the pathogen could have multiplied and caused disease. I was disappointed that the authors did not attempt to do a more biologically relevant experiment in which plants grown in some sort of soil matrix or even sand would have had the soil flooded simultaneously with a mixture of the bacteria, or even better, flooded first with the pathogen and then shortly thereafter with the beneficial strain. This would have been a more powerful test of the ability of IsoF to kill developing biofilms on the roots, thereby alleviating the likelihood of disease. These would have been quite simple studies to perform and would better reflect the normal process of infection. I was surprised at the way the authors presented the information on the effects of IsoF on the disease process, relating root weight, leaf area, chlorophyll content etc in a PCO plot in figure 5C. I had never seen such results presented in this way, and found it very difficult to interpret. I feel it would have been much cleaner to have simply prepared a small table showing the effect of the three or four treatments on each of these dependent variables.  
+
+We agree that the plant protection experiment is somewhat simplistic and artificial. To follow the advice of the reviewer, we established a soil- based infection model as previously described by Medina and López- Baena, 2018 and used it to investigate the effect of the kib cluster on biocontrol activity
+
+<--- Page Split --->
+
+# nature portfolio  
+
+under more natural conditions. In this setup non- sterile soil is drenched simultaneously with a mixture of IsoF or AT4B and Ralstonia. The results, which are shown in the modified Fig. 5 as well as in the new Fig. 20 of the Extended Data, demonstrated that IsoF successfully prevents disease and that this effect is dependent on the kib gene cluster. The results of these experiments were included in the main text. We feel that the PCA graph is the best way to present our data. However, the criticism was noticed and to satisfy the concerns of the reviewer we also added the independent graphs for each parameter measured in Fig. 18 of the Extended Data.  
+
+Line 372 to 375: this important statement was not given the weight that I feel it deserved, as it does point to perhaps a limited scenario in which type IVB killing will prove to be biologically important.  
+
+As mentioned before, we have added additional data showing that IsoF in fact is capable of invading a pre- established biofilm on an agar plate, it just requires more time (72 h versus 24 h, which we used in our routine assays). We have included this information in Fig. 4c and amended the discussion accordingly. Moreover, we have rephrased our statement on the importance of putisolvin to replace cells in biofilms, as previous work has demonstrated that this biosurfactant can efficiently remove pre- established biofilms (Kuiper et al., 2004).  
+
+Line 388 - 389: this sentence seems to be a bit of an overstatement given what they have just noted on lines 372 to 375.  
+
+As mentioned above, IsoF not only invaded a pre- established biofilm in flow- cells but was also capable of invading a pre- established biofilm on an agar plate.  
+
+I found the discussion a very interesting and pleasant read, although I had the strong impression that it reiterated almost word for word statements that had been made in the results. Some condensation seems possible.  
+
+We have tightened up the text as suggested.  
+
+## Reviewer #2 (Remarks to the Author):  
+
+The manuscript by Purtschert- Montenegro et al describes the phenomenon of contact- dependent killing of other bacterial species by the soil bacterium Pseudomonas putida IsoF, shown to be mediated by a Type IV secretion system coded by the newly named kib cluster. This phenomenon has been shown
+
+<--- Page Split --->
+
+# nature portfolio  
+
+before in Xanthomonas and Stenotrophomonas bacteria. The difference here is that the P. putida isoF T4SS belongs to the larger Type IVB class while that of X. citri and S. maltophilia belong to the Type IVA systems. The authors use fluorescently- labelled knockout strains to show that the killing is dependent on structural components of the T4SS apparatus as well as on a putative effector coded by the kib locus. The authors go on to perform some elegant experiments that show that the kib locus, and specifically the effector Piso_02333, allow the IsoF strain to invade biofilms formed by the P. putida K2442 strain (that does not carry a T4SS, but rather a bacteria- killing T6SS). They then show that the IsoF strain can prevent the phytopathogenic bacterium Ralstonia solanacearum from infecting tomato seedlings. The manuscript is very well written, the results are clearly explained and presented and the experiments appear to have been carefully carried out. I do, however have a few comments and suggestions that I believe should be addressed.  
+
+Perhaps my major criticism (easily addressed) is that the reader is not provided with the accession numbers of the P. putida IsoF genome nor the accession numbers of the genes coded by the kib locus. Therefore the reader cannot confer whether the genes described in the paper do in fact correspond to the specified T4SSB components. For example, Lines 139- 145: Here the kib cluster coding 61 genes is described with reference to Figure 2b and Extended Data Table 1. However, in neither this figure or this table, nor anywhere else in the manuscript or in the Supplementary Information are the accession numbers provided for the Pseudomonas putida IsoF strain genome or the genes in the kib cluster. I searched for these sequences in the NCBI database and could not find any of them. Also, searching for the gene names Piso_02333 and Piso_02332 in these databases did not turn up any hits. Finally, no Data Availability Statement was provided, which should contain these accession codes. Is it possible case that the genome of this organism has not yet been made publicly available? I think that it should definitely be deposited and released before being sent off for review. Especially since this strain has been the subject of study by the Eberle group for at least two decades (for example: Steidle et al, 2001. doi: 10.1128/AEM.67.12.5761- 5770.2001. ). The authors should provide a table with all accession numbers of the 61 genes genes in the kib locus. If this is not possible, then the full nucleotide and translated protein sequences each open reading frame in the kib locus should be provided in fasta format as supplementary information.  
+
+We are very sorry for this mistake. The genome sequence of IsoF is now available at NCBI under the accession number CP072013. We have added this in a Data Availability Statement as requested. In addition, we have added the accession numbers of the 61 genes of the kib locus in the new Table 4 in the Extended Dataset.
+
+<--- Page Split --->
+
+# nature portfolio  
+
+Another point, in a way related to that above, is that the nature of the Piso_02332/Piso_02333 pair is not explored to any significant extent.  
+
+How many homologs of these two proteins are found in the public databases and what can we say about their phylogenetic distribution (individually or as a pair)?  
+
+We are thankful for this comment. We searched the NCBI database for homologs of PisoF_02332 and PisoF_02333 and found that the operon structure is fully conserved, supporting the idea that they represent an E- I pair. Moreover, the genes were exclusively found within homologs of the kib locus in a few Pseudomonas strains, we unable to identify orphan homologs. The comparison of the phylogenetic trees of the PisoF_02332- 33 genes, all orthologs of the kib cluster and eight housekeeping genes of the strains carrying the kib locus revealed that the tree topology is congruent, suggesting that strains carrying the kib cluster form a defined lineage that originated from a common ancestor. We have added this information as a new section in the results and in the new Fig. 9 of the Extended Data.  
+
+Is the Piso_02332 immunity protein expected to be localized in the cytosol of the periplasmid (does it have a signal peptide?). This could provide clues regarding the site of action of its cognate effector.  
+
+Using LocTREE and PSORTb the subcellular localization of PisoF_02332 and PisoF_02333 was predicted to be to be cytoplasmic. Additionally, using SignalP- 6.0 web tool we found that neither protein had a signal peptide. We have added this information in the results section.  
+
+Regarding the Piso_02333 effector: Does it have any putative motifs that could be used as a recognition signal for secretion by the T4SSB apparatus?  
+
+Although the T4BSS of Legionella translocates more than 330 effectors only few recognition sequences have been described. For some effectors the following C terminal motifs have been described: hydrophobic residues (Nagai et al., 2005; Voth et al., 2012), an EExxE domain (Huang et al., 2011) and a FxxxLxxxK domain (Kim et al., 2020). However, none of these motifs could be identified in the C- terminal region of the effector PisoF_02333. Interestingly, a FxxxLxxxK domain was found to be present in the C- terminal region of the immunity protein PisoF_02332, suggesting that this protein may be transferred together with its cognate effector toxin. Moreover, we noticed that PisoF_02333 has an unusual glutamine- rich domain in the C- terminal region of the protein (9 Q of 14 aa; Extended Data Fig. 8a) and speculate that these may play a role in effector recognition. Interestingly, a conserved glutamine- rich domain was also identified in the C- terminal regions of VirD4 coupling proteins of phylogenetically diverse T4ASSs (Das, 2020). This region was shown to be required for
+
+<--- Page Split --->
+
+# nature portfolio  
+
+recognition of T- strand DNA but not of the second transferred substrate, the single- stranded DNA- binding protein VirE2. This information has been added.  
+
+Other points.  
+
+Lines 67- 69. A bacterial killing T4SS has been characterized in X. citri and Stenotrophomonas maltophilia (Souza et al, 2015; Bayer- Santos et al, 2019). And homologous systems have been identified in over a hundred other bacterial species (Sgro et al, 2019).  
+
+We have added this information and references as requested.  
+
+Lines 71- 73: The authors imply that DNA transfer is mediated only by class A systems while class B systems have until now been restricted to the role of transferring effectors into eukaryotic cells. This is not quite true. For example, Class B T4SSs are responsible for the horizontal transfer (conjugation) of Incl plasmids.  
+
+We are grateful for this comment and have re- phrased the sentence to indicate that transferring effectors into eukaryotic cells is not the only function of T4BSS.  
+
+Lines 113- 116: Eight out of sixteen killing defective mutants localize to four genes in the kib cluster. What about the other eight insertion mutants? What were their insertion sites?  
+
+We were not able to amplify the regions of the Tn5 insertions in the other 8 mutants and, given that all successfully sequenced mutants were within the kib gene cluster, we did not further analyze these mutants.  
+
+Lines 103,106- 107 and 296: in several places in the manuscript, the term "host range" is used to represent competitor bacterial species. This is not really appropriate. A more suitable term would be "range of target species" or "range of target organisms".  
+
+This is a good point and we have changed the wording to 'range of target species' throughout the text.  
+
+Line 162: the references 22, 32, 55 and 66 refer to studies on only T6SS effector- immunity protein pairs. However, a few thousand bacteria- killing T4SSA effector- immunity protein pairs have been identified in the genomes of over a hundred species.
+
+<--- Page Split --->
+
+# nature portfolio  
+
+This is true, thank you for pointing this out. We have added the following publications in the text: Bayer- Santos et al., 2019; Sgro et al., 2019; Souza et al., 2015.  
+
+Lines 202- 204. It is reported that the delta32- 33 strain grows more slowly than the wild- type strain or the delta33 strain. This could suggest that Piso_02332 may be neutralizing more than one effector.  
+
+In fact, we speculated that Piso_02351 and PisoF_02352 may encode an additional E- I pair. However, recent work in our lab did not support the idea of another E- I pair within the kib gene cluster and as a consequence we have removed this statement. Our most recent results rather suggest that that the observed retarded growth of the mutant is a consequence of a defective kib nanomachinery, which appears to impact viability. In this context, it is interesting to note that inactivation of dotL in certain Legionella pneumophila strains is lethal (Buscher et al., 2005). However, lethality of a dotL mutation is suppressed by mutation of other components of the T4BSS, indicating that the interactions between the different protein components is finely tuned and that a disturbance may lead to self- toxicity.  
+
+Lines 208- 211: Why were you not able to restore killing of P. aureofaciens and KT2442 by complementing the mutant delta32- 33 strain with a plasmid pBBR::32- 33 coding for the effector- immunity pair. Did the authors confirm that this plasmid in fact expresses the two proteins?  
+
+As already mentioned in our response #1, we have analyzed the strains by SDS- PAGE in order to investigate ectopic expression of PisoF_02332 and PisoF_02333. In the complemented strain, we observed a band corresponding to the of the immunity protein but could not detect the toxin, indicating that the immunity protein is produced in excess over the toxin. We therefore hypothesize that killing was not restored because of an unphysiological overexpression of the immunity protein in the complemented strain that effectively neutralized the effector. We have added this information to the discussion and show the SDS- PAGE in the new Fig. 8b of the Extended Data.  
+
+Lines 226- 229: It is written that after 3 days of incubation, isoF had formed a mature biofilm by invading and replacing the KT2442 biofilm. This would correspond to time point 120 hours (48 hrs of KT2442 growth on its own plus 72 after addition of IsoF). No Figure is mentioned to show this. Figures 4a and 4b only show growth up to two days after isoF addition to KT2442 biofilms (total of 96 hours).  
+
+The reviewer is right that the data shown only cover 72 hours of co- cultivation. The text was corrected accordingly. Thank you for pointing this out.
+
+<--- Page Split --->
+
+# nature portfolio  
+
+Lines 281- 283: In Figure 5d, why are the deltaT4B CFUs the same as the IsoF CFUs? I would expect the IsoF numbers to be greater than the deltaT4B numbers in these experiments.  
+
+IsoF was shown to be an excellent tomato root colonizer that forms biofilms on the root surface (Steidle et al., 2001). We speculate that that the constant CFU numbers reflect that the loading capacity of the root for IsoF has been reached and that this is not affected by the kib locus. In this context, it is also important to keep in mind that Ralstonia only transiently colonizes the root surface before the bacteria gain access to host root systems through natural wounds caused by the emergence of lateral roots or through wounds acquired as roots grow through the soil (Xue et al., 2020).  
+
+Lines 298- 302: This is only partially correct. All of the Xanthomonadaceae- like bacterial killing T4SSs have one or more effector/immunity pairs at the same locus containing all of the structural components of the secretion system PLUS other effector/immunity pairs found in other chromosomal locations (Sgro et al, 2019). Here, since we have no access the genome sequence of the IsoF strain under study, we have no way to investigate whether other possible effector/immunity pairs are found in the genome.  
+
+We are sorry that the genome sequence was not available at the time of submission. We have made it available now. We have reworded the statement and shortened the text.  
+
+## Reviewer #3 (Remarks to the Author):  
+
+This study reports that a P. putida strain contains a locus encoding a type IVB secretion system which is involved in bacterial killing via cell- cell contact. This system possesses the novel feature since type IV SS have been reported thus far to only infect eukaryotic cells. Other aspects of this work include that the locus is most likely found in a genomic island which has been recently acquired and that this locus is not widespread since it is only found thus far in a bunch of Pseudomonas isolates.  
+
+This work reports on an interesting novel locus found in a plant associated bacterium with a possible role in competition as well as biocontrol. This work is well performed, described and discussed in the context of microbial ecology. The weakness is the lack of mechanistic data and insight on the mechanism(s) of this system and these initial interesting results raise several questions which need more attention in order to make this work more tangible and less 'preliminary'.  
+
+We are thankful for the supportive comments. We agree with the reviewer that it would be nice to have more insights into the underlying molecular mechanisms of killing by this nanomachinery. The main difficulty in this respect is the absolute novelty of the system that shares no homology with any other killing machinery. However, the criticism was noticed and we have performed additional
+
+<--- Page Split --->
+
+# nature portfolio  
+
+experiments to shed more light on the mode of killing. However, the main focus of our study was indeed the identification and characterization of this novel killing machinery and the evaluation of its biocontrol potential.  
+
+Apparently, the method of killing is not via lysis however no insight on possible targets or mechanism is provided; any data on this aspect would considerably increase the impact of this article. The effector is thought to be the 33 locus/protein, has any further experimentation been performed to confirm unequivocally that this is the effector and on its possible target and mechanism?  
+
+Bioinformatic analysis of PisoF_02332 and PisoF_02333 revealed that both proteins are located in the cytoplasm. We have added this information to the text and in the new Fig. 8 of the Extended Data. As mentioned in our response to reviewer #2, we also noticed that PisoF_02333 has an unusual glutamine- rich domain in the C- terminal region of the protein (9 Q of 14 aa; Extended Data Fig. 8a) and speculate that it may play a role in effector recognition. Interestingly, a conserved glutamine- rich domain was also identified in the C- terminal regions of VirD4 coupling proteins of phylogenetically diverse T4ASSs. This region was shown to be required for recognition of T- strand DNA but not of the second transferred substrate, the single- stranded DNA- binding protein VirE2 (Das, 2020). This information has been added to the Discussion.  
+
+Moreover, work that is currently ongoing in the lab aims at isolating resistant mutants of susceptible bacteria with the aim to identify the molecular target(s) of kib- encoded effectors. If successful, these investigations may allow us to gain insights into the mode of killing. However, we feel that this will be a story on its own.  
+
+It is surprising that mutants in this locus significantly affect bacterial growth since these are thought to be accessory present in a recently acquired genomic island. This aspect is is unclear and needs more attention/explanation.  
+
+As discussed before, we hypothesize that the observed retarded growth of the mutant is a consequence of a defective kib nanomachinery, which may impact viability. It appears that the interactions between the different protein components of kib nanomachinery is finely tuned and that a disturbance by deleting the E- I pair may lead to self- toxicity. We have added this information to the Discussion to improve clarity.  
+
+No reference is made on whether this system is also able to infect eukaryotic cells or should it be considered specific for bacteria- bacteria interactions? Has this aspect been tested?
+
+<--- Page Split --->
+
+# nature portfolio  
+
+We have tested IsoF in a C. elegans infection model and found the strain to be avirulent. Likewise, we could not observe antifungal or anti- oomycete activities. It is also noteworthy that IsoF does not grow at \(37^{\circ}C\) and thus will be unable to infect mammalian cells. Furthermore, in our tomato root colonization assays we could not observe an effect of the kib locus on plant growth. Collectively, these data suggest that this killing machinery is specific for bacterial cells interactions.  
+
+The immunity aspect of this system is not entirely clear since two loci appear to be involved; one major locus Piso_02332 encodes for an immunity protein, however no mechanistic insight is provided and in addition it is not tested whether this locus alone can provide immunity if transferred to other bacteria.  
+
+We showed that complementation of the \(\Delta 32 - 33\) mutant by providing PisoF_02332 in trans on a plasmid (pBBR::32) rendered the strain resistant to killing by the wildtype. However, we were unable to complement the \(\Delta T4B\) mutant, which lacks the entire kib gene cluster, with pBBR::32. Analysis of the strains by SDS- PAGE revealed that a band corresponding to the PisoF_02332 protein is visible in the complemented \(\Delta 32 - 33\) mutant but not the complemented \(\Delta T4B\) mutant. The lack of PisoF_02332 expression in the \(\Delta T4B\) mutant background explains its sensitivity to kib- mediated killing and suggests that the kib gene cluster encodes functions required for the expression of PisoF_02332 or affects its stability. We have added this information to the main text and have removed the speculation that another E- I pair is encoded by the cluster, as work currently under way in our laboratory does not support this idea.  
+
+## Decision Letter, first revision:  
+
+Dear Dr. Eberl,  
+
+Thank you for submitting your revised manuscript "A type IVB secretion system adapted for bacterial killing, biofilm invasion and biocontrol" (NMICROBIOL- 21092434A). It has now been seen by the original referees and their comments are below. The reviewers find that the paper has improved in revision, and therefore we'll be happy in principle to publish it in Nature Microbiology, pending minor revisions to satisfy the referees' final requests and to comply with our editorial and formatting guidelines.  
+
+If the current version of your manuscript is in a PDF format, please email us a copy of the file in an editable format (Microsoft Word or LaTex)- - we can not proceed with PDFs at this stage.  
+
+We are now performing detailed checks on your paper and will send you a checklist detailing our editorial and formatting requirements in about a week. Please do not upload the final materials and make any revisions until you receive this additional information from us.
+
+<--- Page Split --->
+
+# nature portfolio  
+
+Thank you again for your interest in Nature Microbiology Please do not hesitate to contact me if you have any questions.  
+
+Sincerely,  
+
+{redacted}  
+
+Reviewer #1 (Remarks to the Author):  
+
+see below  
+
+Reviewer #3 (Remarks to the Author):  
+
+This revised version is an improvement as authors have carefully revised most of the points raised by the three reviewers.  
+
+Most additional experimenting and data was also provided:  
+
+- SDS PAGE analysis, the immunity protein was believed to be under-expressed compared to the toxin hence there was a lack of complementation.- The authors have performed a new plant experiment based on a soil infection model in order to test biocontrol activity under conditions which resemble more the wild. The results confirmed what was previously observed in more controlled conditions.- An additional biofilm experiment was also performed involving a longer time frame further evidencing the effects of bacterial invasion.- The genome sequence was properly deposited in data banks.- Performed a more complete phylogenetic search of the two ORFs (2332 and 2333)- More in silico protein information of the 2332 and 2333 ORFs- Have tested the killing model on C. elegans indicating that it does not appear to infect eukaryotic cells  
+
+In the revised text, these new additional results and figures are well integrated in the document and have no comments.  
+
+Minor comment:  
+
+Not clear the reasoning why 8 Tn5 mutants could not be mapped - nowadays with NGS it is ultimately possible to map any mutation in the genome. I suggest to either map these mutants or just remove this information and simply state that a subset was mapped.
+
+<--- Page Split --->
+
+# nature portfolio  
+
+## Decision Letter, final checks  
+
+Dear Dr. Eberl,  
+
+Thank you for your patience as we've prepared the guidelines for final submission of your Nature Microbiology manuscript, "A type IVB secretion system adapted for bacterial killing, biofilm invasion and biocontrol" (NMICROBIOL- 21092434A). Please carefully follow the step- by- step instructions provided in the attached file, and add a response in each row of the table to indicate the changes that you have made. Please also check and comment on any additional marked- up edits we have proposed within the text. Ensuring that each point is addressed will help to ensure that your revised manuscript can be swiftly handed over to our production team.  
+
+We would like to start working on your revised paper, with all of the requested files and forms, as soon as possible (preferably within two weeks). Please get in contact with us if you anticipate delays.  
+
+When you upload your final materials, please include a point- by- point response to any remaining reviewer comments.  
+
+If you have not done so already, please alert us to any related manuscripts from your group that are under consideration or in press at other journals, or are being written up for submission to other journals (see: https://www.nature.com/nature- research/editorial- policies/plagiarism#policy- on- duplicate- publication for details).  
+
+In recognition of the time and expertise our reviewers provide to Nature Microbiology's editorial process, we would like to formally acknowledge their contribution to the external peer review of your manuscript entitled "A type IVB secretion system adapted for bacterial killing, biofilm invasion and biocontrol". For those reviewers who give their assent, we will be publishing their names alongside the published article.  
+
+Nature Microbiology offers a Transparent Peer Review option for new original research manuscripts submitted after December 1st, 2019. As part of this initiative, we encourage our authors to support increased transparency into the peer review process by agreeing to have the reviewer comments, author rebuttal letters, and editorial decision letters published as a Supplementary item. When you submit your final files please clearly state in your cover letter whether or not you would like to participate in this initiative. Please note that failure to state your preference will result in delays in accepting your manuscript for publication.  
+
+## Cover suggestions  
+
+As you prepare your final files we encourage you to consider whether you have any images or illustrations that may be appropriate for use on the cover of Nature Microbiology.
+
+<--- Page Split --->
+
+# nature portfolio  
+
+Covers should be both aesthetically appealing and scientifically relevant, and should be supplied at the best quality available. Due to the prominence of these images, we do not generally select images featuring faces, children, text, graphs, schematic drawings, or collages on our covers.  
+
+We accept TIFF, JPEG, PNG or PSD file formats (a layered PSD file would be ideal), and the image should be at least 300ppi resolution (preferably 600- 1200 ppi), in CMYK colour mode.  
+
+If your image is selected, we may also use it on the journal website as a banner image, and may need to make artistic alterations to fit our journal style.  
+
+Please submit your suggestions, clearly labeled, along with your final files. We'll be in touch if more information is needed.  
+
+Nature Microbiology has now transitioned to a unified Rights Collection system which will allow our Author Services team to quickly and easily collect the rights and permissions required to publish your work. Approximately 10 days after your paper is formally accepted, you will receive an email in providing you with a link to complete the grant of rights. If your paper is eligible for Open Access, our Author Services team will also be in touch regarding any additional information that may be required to arrange payment for your article.  
+
+Please note that you will not receive your proofs until the publishing agreement has been received through our system.  
+
+Please note that <i>Nature Microbiology</i> is a Transformative Journal (TJ). Authors may publish their research with us through the traditional subscription access route or make their paper immediately open access through payment of an article- processing charge (APC). Authors will not be required to make a final decision about access to their article until it has been accepted. <a href="https://www.springernature.com/gp/open- research/transformative- journals"> Find out more about Transformative Journals</a>  
+
+## Authors may need to take specific actions to achieve <a  
+
+href="https://www.springernature.com/gp/open- research/funding/policy- compliance- faqs"> compliance</a> with funder and institutional open access mandates. If your research is supported by a funder that requires immediate open access (e.g. according to <a href="https://www.springernature.com/gp/open- research/plan- s- compliance">Plan S principles</a>) then you should select the gold OA route, and we will direct you to the compliant route where possible. For authors selecting the subscription publication route, the journal's standard licensing terms will need to be accepted, including <a href="https://www.nature.com/nature- portfolio/editorial- policies/self- archiving- and- license- to- publish">self- archiving policies</a>. Those licensing terms will supersede any other terms that the author or any third party may assert apply to any version of the manuscript.  
+
+For information regarding our different publishing models please see our <a
+
+<--- Page Split --->
+
+# nature portfolio  
+
+href="https://www.springernature.com/gp/open- research/transformative- journals"> Transformative Journals </a> page. If you have any questions about costs, Open Access requirements, or our legal forms, please contact ASJournals@springernature.com.  
+
+Please use the following link for uploading these materials: {redacted}  
+
+If you have any further questions, please feel free to contact me.  
+
+Best regards,  
+
+{redacted}  
+
+Reviewer #1: Remarks to the Author: see below  
+
+Reviewer #3:  
+
+Remarks to the Author:  
+
+This revised version is an improvement as authors have carefully revised most of the points raised by the three reviewers.  
+
+Most additional experimenting and data was also provided:  
+
+- SDS PAGE analysis, the immunity protein was believed to be under-expressed compared to the toxin hence there was a lack of complementation.- The authors have performed a new plant experiment based on a soil infection model in order to test biocontrol activity under conditions which resemble more the wild. The results confirmed what was previously observed in more controlled conditions.- An additional biofilm experiment was also performed involving a longer time frame further evidencing the effects of bacterial invasion.- The genome sequence was properly deposited in data banks.- Performed a more complete phylogenetic search of the two ORFs (2332 and 2333)- More in silico protein information of the 2332 and 2333 ORFs- Have tested the killing model on C. elegans indicating that it does not appear to infect eukaryotic cells  
+
+In the revised text, these new additional results and figures are well integrated in the document and have no comments.
+
+<--- Page Split --->
+
+# nature portfolio  
+
+Minor comment:  
+
+Not clear the reasoning why 8 Tn5 mutants could not be mapped - nowadays with NGS it is ultimately possible to map any mutation in the genome. I suggest to either map these mutants or just remove this information and simply state that a subset was mapped.  
+
+## Final Decision Letter:  
+
+Dear Leo,  
+
+I am pleased to accept your Article "<i>Pseudomonas putida</i> mediates bacterial killing, biofilm invasion and biocontrol with a type IVB secretion system" for publication in Nature Microbiology. Thank you for having chosen to submit your work to us and many congratulations.  
+
+Over the next few weeks, your paper will be copyedited to ensure that it conforms to Nature Microbiology style. We look particularly carefully at the titles of all papers to ensure that they are relatively brief and understandable.  
+
+Once your paper is typeset, you will receive an email with a link to choose the appropriate publishing options for your paper and our Author Services team will be in touch regarding any additional information that may be required. Once your paper has been scheduled for online publication, the Nature press office will be in touch to confirm the details.  
+
+After the grant of rights is completed, you will receive a link to your electronic proof via email with a request to make any corrections within 48 hours. If, when you receive your proof, you cannot meet this deadline, please inform us at rjsproduction@springernature.com immediately. You will not receive your proofs until the publishing agreement has been received through our system  
+
+Due to the importance of these deadlines, we ask you please us know now whether you will be difficult to contact over the next month. If this is the case, we ask you provide us with the contact information (email, phone and fax) of someone who will be able to check the proofs on your behalf, and who will be available to address any last- minute problems.  
+
+Acceptance of your manuscript is conditional on all authors' agreement with our publication policies (see https://www.nature.com/nmicrobiol/editorial- policies). In particular your manuscript must not be published elsewhere and there must be no announcement of the work to any media outlet until the publication date (the day on which it is uploaded onto our website).  
+
+Please note that <i>Nature Microbiology</i> is a Transformative Journal (TJ). Authors may publish their research with us through the traditional subscription access route or make their paper immediately open access through payment of an article- processing charge (APC). Authors will not be required to make a final decision about access to their article until it has been accepted. <a href="https://www.springernature.com/gp/open- research/transformative- journals"> Find out more about Transformative Journals</a>
+
+<--- Page Split --->
+
+# nature portfolio  
+
+Authors may need to take specific actions to achieve <a href="https://www.springernature.com/gp/open- research/funding/policy- compliance- faqs"> compliance</a> with funder and institutional open access mandates. If your research is supported by a funder that requires immediate open access (e.g. according to <a href="https://www.springernature.com/gp/open- research/plan- s- compliance">Plan S principles</a>) then you should select the gold OA route, and we will direct you to the compliant route where possible. For authors selecting the subscription publication route, the journal's standard licensing terms will need to be accepted, including <a href="https://www.nature.com/nature- portfolio/editorial- policies/self- archiving- and- license- to- publish">self- archiving policies</a>. Those licensing terms will supersede any other terms that the author or any third party may assert apply to any version of the manuscript.  
+
+If you have any questions about our publishing options, costs, Open Access requirements, or our legal forms, please contact ASJournals@springernature.com  
+
+An online order form for reprints of your paper is available at <a  
+
+href="https://www.nature.com/reprints/author-  
+
+reprints.html">https://www.nature.com/reprints/author- reprints.html</a>. All co- authors, authors' institutions and authors' funding agencies can order reprints using the form appropriate to their geographical region.  
+
+We welcome the submission of potential cover material (including a short caption of around 40 words) related to your manuscript; suggestions should be sent to Nature Microbiology as electronic files (the image should be 300 dpi at \(210 \times 297 \text{mm}\) in either TIFF or JPEG format). Please note that such pictures should be selected more for their aesthetic appeal than for their scientific content, and that colour images work better than black and white or grayscale images. Please do not try to design a cover with the Nature Microbiology logo etc., and please do not submit composites of images related to your work. I am sure you will understand that we cannot make any promise as to whether any of your suggestions might be selected for the cover of the journal.  
+
+You can now use a single sign- on for all your accounts, view the status of all your manuscript submissions and reviews, access usage statistics for your published articles and download a record of your refereeing activity for the Nature journals.  
+
+To assist our authors in disseminating their research to the broader community, our Shared initiative provides you with a unique shareable link that will allow anyone (with or without a subscription) to read the published article. Recipients of the link with a subscription will also be able to download and print the PDF.  
+
+As soon as your article is published, you will receive an automated email with your shareable link.
+
+<--- Page Split --->

peer_reviews/eb68748872af85199990c7aa3967e103ea56513dea1570d6c6e723d67ca9c019/supplementary_1_Peer Review File./supplementary_1_Peer Review File._det.mmd

@@ -0,0 +1,766 @@
+<|ref|>title<|/ref|><|det|>[[595, 44, 970, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>title<|/ref|><|det|>[[115, 161, 571, 194]]<|/det|>
+# Peer Review Information  
+
+<|ref|>text<|/ref|><|det|>[[114, 220, 857, 293]]<|/det|>
+Journal: Nature Microbiology  Manuscript Title: Pseudomonas putida mediates bacterial killing, biofilm invasion and biocontrol with a type IVB secretion system  Corresponding author name(s): Leo Eberl
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[595, 44, 970, 88]]<|/det|>
+# nature portfolio  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 148, 568, 173]]<|/det|>
+## Reviewer Comments & Decisions:  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 207, 343, 222]]<|/det|>
+## Decision Letter, initial version:  
+
+<|ref|>text<|/ref|><|det|>[[116, 242, 273, 257]]<|/det|>
+Dear Professor Eberl,  
+
+<|ref|>text<|/ref|><|det|>[[115, 271, 880, 348]]<|/det|>
+Thank you for your patience while your manuscript "A type IVB secretion system adapted for bacterial killing, biofilm invasion and biocontrol" was under peer- review at Nature Microbiology. It has now been seen by 3 referees, whose expertise and comments you will find at the end of this email. Although they find your work of some potential interest, they have raised a number of concerns that will need to be addressed before we can consider publication of the work in Nature Microbiology.  
+
+<|ref|>text<|/ref|><|det|>[[115, 361, 875, 437]]<|/det|>
+In particular, you will see that referee #1 raise concerns regarding the plant experiment and asks you to use a more biologically relevant approach, and suggests you to tone down some statements and improve the data presentation and description. Referee #2 raises issues regarding the lack of data availability. Referee #3 mainly comments on the need for more mechanistic insight to strengthen the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 451, 870, 482]]<|/det|>
+Should further experimental data allow you to address these criticisms, we would be happy to look at a revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 495, 874, 541]]<|/det|>
+We are committed to providing a fair and constructive peer- review process. Please do not hesitate to contact us if there are specific requests from the reviewers that you believe are technically impossible or unlikely to yield a meaningful outcome.  
+
+<|ref|>text<|/ref|><|det|>[[115, 554, 866, 660]]<|/det|>
+We strongly support public availability of data. Please place the data used in your paper into a public data repository, if one exists, or alternatively, present the data as Source Data or Supplementary Information. If data can only be shared on request, please explain why in your Data Availability Statement, and also in the correspondence with your editor. For some data types, deposition in a public repository is mandatory - more information on our data deposition policies and available repositories can be found at https://www.nature.com/nature- research/editorial- policies/reporting- standards#availability- of- data.  
+
+<|ref|>text<|/ref|><|det|>[[115, 674, 880, 780]]<|/det|>
+Please include a data availability statement as a separate section after Methods but before references, under the heading "Data Availability". This section should inform readers about the availability of the data used to support the conclusions of your study. This information includes accession codes to public repositories (data banks for protein, DNA or RNA sequences, microarray, proteomics data etc...), references to source data published alongside the paper, unique identifiers such as URLs to data repository entries, or data set DOIs, and any other statement about data availability. At a minimum, you should include the following statement: "The data that support the findings of this study are
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[595, 44, 969, 88]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[115, 116, 870, 179]]<|/det|>
+available from the corresponding author upon request", mentioning any restrictions on availability. If DOIs are provided, we also strongly encourage including these in the Reference list (authors, title, publisher (repository name), identifier, year). For more guidance on how to write this section please see:  
+
+<|ref|>text<|/ref|><|det|>[[115, 179, 808, 194]]<|/det|>
+http://www.nature.com/authors/policies/data/data- availability- statements- data- citations.pdf  
+
+<|ref|>text<|/ref|><|det|>[[116, 222, 323, 237]]<|/det|>
+If revising your manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[115, 252, 878, 297]]<|/det|>
+\* Include a "Response to referees" document detailing, point- by- point, how you addressed each referee comment. If no action was taken to address a point, you must provide a compelling argument. This response will be sent back to the referees along with the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 311, 840, 356]]<|/det|>
+\* If you have not done so already we suggest that you begin to revise your manuscript so that it conforms to our Article format instructions at http://www.nature.com/nmicrobiol/info/final- submission. Refer also to any guidelines provided in this letter.  
+
+<|ref|>text<|/ref|><|det|>[[115, 370, 852, 415]]<|/det|>
+\* Include a revised version of any required reporting checklist. It will be available to referees (and, potentially, statisticians) to aid in their evaluation if the manuscript goes back for peer review. A revised checklist is essential for re- review of the paper.  
+
+<|ref|>text<|/ref|><|det|>[[115, 444, 843, 490]]<|/det|>
+When submitting the revised version of your manuscript, please pay close attention to our href="https://www.nature.com/nature- research/editorial- policies/image- integrity">Digital Image Integrity Guidelines.</a> and to the following points below:  
+
+<|ref|>text<|/ref|><|det|>[[115, 504, 840, 580]]<|/det|>
+- that unprocessed scans are clearly labelled and match the gels and western blots presented in figures. 
+- that control panels for gels and western blots are appropriately described as loading on sample processing controls 
+- all images in the paper are checked for duplication of panels and for splicing of gel lanes.  
+
+<|ref|>text<|/ref|><|det|>[[115, 594, 856, 639]]<|/det|>
+Finally, please ensure that you retain unprocessed data and metadata files after publication, ideally archiving data in perpetuity, as these may be requested during the peer review and production process or after publication if any issues arise.  
+
+<|ref|>text<|/ref|><|det|>[[115, 668, 503, 683]]<|/det|>
+Please use the link below to submit a revised paper:  
+
+<|ref|>text<|/ref|><|det|>[[116, 698, 199, 713]]<|/det|>
+{redacted}  
+
+<|ref|>text<|/ref|><|det|>[[115, 728, 870, 773]]<|/det|>
+<strong>Note: </strong> This url links to your confidential homepage and associated information about manuscripts you may have submitted or be reviewing for us. If you wish to forward this e- mail to co- authors, please delete this link to your homepage first.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[596, 44, 969, 88]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[115, 116, 879, 238]]<|/det|>
+Nature Microbiology is committed to improving transparency in authorship. As part of our efforts in this direction, we are now requesting that all authors identified as 'corresponding author' on published papers create and link their Open Researcher and Contributor Identifier (ORCID) with their account on the Manuscript Tracking System (MTS), prior to acceptance. This applies to primary research papers only. ORCID helps the scientific community achieve unambiguous attribution of all scholarly contributions. You can create and link your ORCID from the home page of the MTS by clicking on 'Modify my Springer Nature account'. For more information please visit please visit <a href="http://www.springernature.com/orcid">www.springernature.com/orcid</a>.  
+
+<|ref|>text<|/ref|><|det|>[[115, 251, 856, 312]]<|/det|>
+If you wish to submit a suitably revised manuscript we would hope to receive it within 6 months. If you cannot send it within this time, please let us know. We will be happy to consider your revision, even if a similar study has been accepted for publication at Nature Microbiology or published elsewhere (up to a maximum of 6 months).  
+
+<|ref|>text<|/ref|><|det|>[[115, 326, 649, 342]]<|/det|>
+In the meantime we hope that you find our referees' comments helpful.  
+
+<|ref|>text<|/ref|><|det|>[[115, 356, 232, 371]]<|/det|>
+Yours sincerely,  
+
+<|ref|>text<|/ref|><|det|>[[115, 385, 198, 401]]<|/det|>
+{redacted}  
+
+<|ref|>text<|/ref|><|det|>[[115, 415, 610, 427]]<|/det|>
+\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*  
+
+<|ref|>text<|/ref|><|det|>[[115, 431, 264, 445]]<|/det|>
+Reviewer Expertise:  
+
+<|ref|>text<|/ref|><|det|>[[115, 460, 576, 506]]<|/det|>
+Referee #1: microbial ecology of plant bacteria/Pseudomonas Referee #2: Type IV secretion systems Referee #3: plant microbe signaling/QS  
+
+<|ref|>text<|/ref|><|det|>[[115, 521, 274, 536]]<|/det|>
+Reviewer Comments:  
+
+<|ref|>text<|/ref|><|det|>[[115, 550, 404, 565]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 579, 880, 789]]<|/det|>
+This very interesting study has done a good job of describing a novel form of contact- dependent killing of bacteria. Contact- dependent killing schemes have become quite the rage in microbial ecology, and yet most have been associated with variants of type VI secretion systems. This study has shown a very novel method of delivering what appear to be toxic effectors using a variant of the type IVB secretion system. The study is quite solid, and the authors have employed a variety of powerful tools including Tn sequencing and various microscopy techniques to both demonstrate the contact dependence of the killing of various other bacteria by the Pseudomonas putida strain, and to identify the genetic loci associated with both the toxic effector as well as the immunity function. It is quite remarkable that this type IVB secretion system seems to be quite promiscuous in its ability to kill other bacteria, as they did not seem to find any taxa that were immune to its effect. They might want to comment further on the relative breadth of this killing compared to many other type VI systems that have been studied. As such, there are indeed important translational aspects of the study, as exemplified here in their demonstration that the biological control of bacterial wilt caused by Ralstonia solanacearum could be achieved using this strain capable of contact killing. The manuscript was quite
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[595, 44, 969, 87]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[115, 117, 875, 179]]<|/det|>
+well written, and the powerful and logical experimental design leading to the discovery of the system was well considered. I have only a few relatively minor comments about both the way certain findings should be described, a few conclusions that may be a bit overstated, as well as suggestions for how the demonstration of plant disease control could have been better assessed and described.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 193, 187, 208]]<|/det|>
+## Specifics:  
+
+<|ref|>text<|/ref|><|det|>[[115, 222, 875, 342]]<|/det|>
+Lines 67- 69. We are told here that a Xanthomonas strain had exhibited a type IV dependent killing of other bacterial strains - while it is only in a couple of sentences later that the authors note that there is more than one kind of type IV secretion system, and that there had been no prior demonstration of the type IVB system being used in antimicrobial activities. However, there was never a mention until much later in the discussion that the Xanthomonas example, was in fact, a type IVA secretion system. I feel that this point should have been made earlier, because I was bothered by the concern that this study might not have been as novel as they were suggesting, because they had not noted that this earlier example was not a type IVB system.  
+
+<|ref|>text<|/ref|><|det|>[[115, 356, 877, 460]]<|/det|>
+Line 90 and elsewhere. Here and in many locations throughout the manuscript, the authors have used the term "competition" to describe the interaction of IsoF with various other bacteria. At least in my mind, the term "competition" implies some sort of growth reduction that would be a result of the need to share certain nutrient resources with a neighbor, or perhaps an effect of the chemical environment by a neighbor etc. Given that they are demonstrating here that "competition" is actually the direct killing of their neighbor, I wonder if it would be cleaner to refer to "interaction" or "killing" or some other term that does not seem to have this nutrient competition connotation.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 475, 315, 490]]<|/det|>
+## Line 97 misspelling: iodide  
+
+<|ref|>text<|/ref|><|det|>[[115, 504, 877, 609]]<|/det|>
+Line 112. I had to take a step back and try to figure out why the authors had decided to use Pseudomonas aureofaciens in the study discussed in this section because there had not been a previous mention that this particular species was in fact susceptible to the killing effect of strain IsoF. One had to dig into the results of Figure 1, and specifically recognize that Pseudomonas aureofaciens was one of the strains that was in fact susceptible to the killing by IsoF. I think it would benefit from a short note to discuss why there had been a change in the killing assay away from KT2442 to this new bacterial target.  
+
+<|ref|>text<|/ref|><|det|>[[115, 623, 863, 653]]<|/det|>
+Line 126: Again, in this sentence they use the term "outcompete" when it's clearly was killing the P. aureofaciens. I think it would be better to note killing rather than competition in such examples.  
+
+<|ref|>text<|/ref|><|det|>[[115, 668, 877, 698]]<|/det|>
+Line 210: it is a bit disconcerting to hear that they were not able to restore killing by complimentation. I do not recall there being a discussion of this negative finding.  
+
+<|ref|>text<|/ref|><|det|>[[115, 712, 879, 788]]<|/det|>
+Line 236 through 238. Upon initially reading the sentence, I found it surprising and unexpected, since it did not seem like the simple process of killing a bacterium would cause it to disappear from the biofilm. It was only later when the authors noted that this replacement of the dead cells by the IsoF strain only occurred in flow chambers where turbulence etc could in fact have dislodged the dead cells. That said, after reading the entire manuscript I was left with a feeling that there is a bit of an
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[596, 45, 970, 88]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[115, 116, 880, 222]]<|/det|>
+overstatement in the authors conclusions that the IsoF strain could invade biofilms, as this seemed to have been restricted to the flow cells where there is some ability to remove cells from the biofilm, and was not observed in pre- established biofilms that had formed on agar plates etc. This made me wonder whether the invasion of biofilms that might have formed on plants and other habitats where they feel the type IV secretion system could be useful in killing target organisms was as likely to occur as they suggest. I feel a bit of further clarification of the situations in which biofilm invasion/replacement is likely to occur is warranted.  
+
+<|ref|>text<|/ref|><|det|>[[115, 236, 880, 505]]<|/det|>
+There were several aspects of the plant protection experiment that I found a bit awkward or simplistic. The experiment, using small seedlings growing on agar plates is a somewhat unnatural setting, and one that might have facilitated both microbial survival on, and multiplication on the damaged plant tissue. The experiment was done in a way that would have maximized the potential effect of the presence of IsoF together with the pathogen, since they were mixed together and then immediately applied to the wounded plant. There thus would have been maximal opportunity for IsoF to have been in close proximity to the pathogen and thus for it to have killed it before the pathogen could have multiplied and caused disease. I was disappointed that the authors did not attempt to do a more biologically relevant experiment in which plants grown in some sort of soil matrix or even sand would have had the soil flooded simultaneously with a mixture of the bacteria, or even better, flooded first with the pathogen and then shortly thereafter with the beneficial strain. This would have been a more powerful test of the ability of IsoF to kill developing biofilms on the roots, thereby alleviating the likelihood of disease. These would have been quite simple studies to perform and would better reflect the normal process of infection. I was surprised at the way the authors presented the information on the effects of IsoF on the disease process, relating root weight, leaf area, chlorophyll content etc in a PCO plot in figure 5C. I had never seen such results presented in this way, and found it very difficult to interpret. I feel it would have been much cleaner to have simply prepared a small table showing the effect of the three or four treatments on each of these dependent variables.  
+
+<|ref|>text<|/ref|><|det|>[[115, 519, 870, 550]]<|/det|>
+Line 372 to 375: this important statement was not given the weight that I feel it deserved, as it does point to perhaps a limited scenario in which type IVB killing will prove to be biologically important.  
+
+<|ref|>text<|/ref|><|det|>[[115, 564, 870, 594]]<|/det|>
+Line 388 - 389: this sentence seems to be a bit of an overstatement given what they have just noted on lines 372 to 375.  
+
+<|ref|>text<|/ref|><|det|>[[115, 609, 875, 654]]<|/det|>
+I found the discussion a very interesting and pleasant read, although I had the strong impression that it reiterated almost word for word statements that had been made in the results. Some condensation seems possible.  
+
+<|ref|>text<|/ref|><|det|>[[116, 712, 404, 727]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 742, 880, 808]]<|/det|>
+The manuscript by Purtschert- Montenegro et al describes the phenomenon of contact- dependent killing of other bacterial species by the soil bacterium Pseudomonas putida IsoF, shown to be mediated by a Type IV secretion system coded by the newly named kib cluster. This phenomenon has
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[595, 45, 970, 88]]<|/det|>
+# natureportfolio  
+
+<|ref|>text<|/ref|><|det|>[[115, 115, 881, 283]]<|/det|>
+been shown before in Xanthomonas and Stenotrophomonas bacteria. The difference here is that the P. putida isoF T4SS belongs to the larger Type IVB class while that of X. citri and S. maltophilia belong to the Type IVA systems. The authors use fluorescently- labelled knockout strains to show that the killing is dependent on structural components of the T4SS apparatus as well as on a putative effector coded by the kib locus. The authors go on to perform some elegant experiments that show that the kib locus, and specifically the effector Piso_02333, allow the IsoF strain to invade biofilms formed by the P. putida K2442 strain (that does not carry a T4SS, but rather a bacteria- killing T6SS). They then show that the IsoF strain can prevent the phytopathogenic bacterium Ralstonia solanacearum from infecting tomato seedlings. The manuscript is very well written, the results are clearly explained and presented and the experiments appear to have been carefully carried out. I do, however have a few comments and suggestions that I believe should be addressed.  
+
+<|ref|>text<|/ref|><|det|>[[115, 295, 881, 551]]<|/det|>
+Perhaps my major criticism (easily addressed) is that the reader is not provided with the accession numbers of the P. putida IsoF genome nor the accession numbers of the genes coded by the kib locus. Therefore the reader cannot confer whether the genes described in the paper do in fact correspond to the specified T4SSB components. For example, Lines 139- 145: Here the kib cluster coding 61 genes is described with reference to Figure 2b and Extended Data Table 1. However, in neither this figure or this table, nor anywhere else in the manuscript or in the Supplementary Information are the accession numbers provided for the Pseudomonas putida IsoF strain genome or the genes in the kib cluster. I searched for these sequences in the NCBI database and could not find any of them. Also, searching for the gene names Piso_02333 and Piso_02332 in these databases did not turn up any hits. Finally, no Data Availability Statement was provided, which should contain these accession codes. Is it possible case that the genome of this organism has not yet been made publicly available? I think that it should definitely be deposited and released before being sent off for review. Especially since this strain has been the subject of study by the Eberle group for at least two decades (for example: Steidle et al, 2001. doi: 10.1128/AEM.67.12.5761- 5770.2001. ). The authors should provide a table with all accession numbers of the 61 genes genes in the kib locus. If this is not possible, then the full nucleotide and translated protein sequences each open reading frame in the kib locus should be provided in fasta format as supplementary information.  
+
+<|ref|>text<|/ref|><|det|>[[115, 563, 876, 669]]<|/det|>
+Another point, in a way related to that above, is that the nature of the Piso_02332/Piso_02333 pair is not explored to any significant extent. How many homologs of these two proteins are found in the public databases and what can we say about their phylogenetic distribution (individually or as a pair)? Is the Piso_02332 immunity protein expected to be localized in the cytosol of the periplasmid (does it have a signal peptide?). This could provide clues regarding the site of action of its cognate effector. Regarding the Piso_02333 effector: Does it have any putative motifs that could be used as a recognition signal for secretion by the T4SSB apparatus?  
+
+<|ref|>text<|/ref|><|det|>[[116, 683, 212, 697]]<|/det|>
+Other points.  
+
+<|ref|>text<|/ref|><|det|>[[116, 712, 827, 758]]<|/det|>
+Lines 67- 69. A bacterial killing T4SS has been characterized in X. citri and Stenotrophomonas maltophilia (Souza et al, 2015; Bayer- Santos et al, 2019). And homologous systems have been identified in over a hundred other bacterial species (Sgro et al, 2019).  
+
+<|ref|>text<|/ref|><|det|>[[115, 772, 860, 788]]<|/det|>
+Lines 71- 73: The authors imply that DNA transfer is mediated only by class A systems while class B
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[596, 45, 970, 88]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[115, 117, 881, 163]]<|/det|>
+systems have until now been restricted to the role of transferring effectors into eukaryotic cells. This is not quite true. For example, Class B T4SSs are responsible for the horizontal transfer (conjugation) of IncI plasmids.  
+
+<|ref|>text<|/ref|><|det|>[[115, 177, 865, 208]]<|/det|>
+Lines 113- 116: Eight out of sixteen killing defective mutants localize to four genes in the kib cluster. What about the other eight insertion mutants? What were their insertion sites?  
+
+<|ref|>text<|/ref|><|det|>[[115, 221, 864, 268]]<|/det|>
+Lines 103,106- 107 and 296: in several places in the manuscript, the term "host range" is used to represent competitor bacterial species. This is not really appropriate. A more suitable term would be "range of target species" or "range of target organisms".  
+
+<|ref|>text<|/ref|><|det|>[[115, 281, 860, 327]]<|/det|>
+Line 162: the references 22, 32, 55 and 66 refer to studies on only T6SS effector- immunity protein pairs. However, a few thousand bacteria- killing T4SSA effector- immunity protein pairs have been identified in the genomes of over a hundred species.  
+
+<|ref|>text<|/ref|><|det|>[[115, 340, 880, 372]]<|/det|>
+Lines 202- 204. It is reported that the delta32- 33 strain grows more slowly than the wild- type strain or the delta33 strain. This could suggest that Piso_02332 may be neutralizing more than one effector.  
+
+<|ref|>text<|/ref|><|det|>[[115, 385, 844, 431]]<|/det|>
+Lines 208- 211: Why were you not able to restore killing of P. aureofaciens and KT2442 by complementing the mutant delta32- 33 strain with a plasmid pBBR::32- 33 coding for the effector- immunity pair. Did the authors confirm that this plasmid in fact expresses the two proteins?  
+
+<|ref|>text<|/ref|><|det|>[[115, 444, 874, 505]]<|/det|>
+Lines 226- 229: It is written that after 3 days of incubation, isoF had formed a mature biofilm by invading and replacing the KT2442 biofilm. This would correspond to time point 120 hours (48 hrs of KT2442 growth on its own plus 72 after addition of IsoF). No Figure is mentioned to show this. Figures 4a and 4b only show growth up to two days after isoF addition to KT2442 biofilms (total of 96 hours).  
+
+<|ref|>text<|/ref|><|det|>[[115, 519, 864, 550]]<|/det|>
+Lines 281- 283: In Figure 5d, why are the deltaT4B CFUs the same as the IsoF CFUs? I would expect the IsoF numbers to be greater than the deltaT4B numbers in these experiments.  
+
+<|ref|>text<|/ref|><|det|>[[115, 563, 872, 655]]<|/det|>
+Lines 298- 302: This is only partially correct. All of the Xanthomonadaceae- like bacterial killing T4SSs have one or more effector/immunity pairs at the same locus containing all of the structural components of the secretion system PLUS other effector/immunity pairs found in other chromosomal locations (Sgro et al, 2019). Here, since we have no access the genome sequence of the IsoF strain under study, we have no way to investigate whether other possible effector/immunity pairs are found in the genome.  
+
+<|ref|>text<|/ref|><|det|>[[116, 699, 404, 714]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 728, 870, 789]]<|/det|>
+This study reports that a P. putida strain contains a locus encoding a typeIVB secretion system which is involved in bacterial killing via cell- cell contact. This system possesses the novel feature since typeIV SS have been reported thus far to only infect eukaryotic cells. Other aspects of this work include that the locus is most likely found in a genomic island which has been recently acquired and
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[595, 45, 970, 88]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[115, 116, 866, 208]]<|/det|>
+that this locus is not widespread since it is only found thus far in a bunch of Pseudomonas isolates. This work reports on an interesting novel locus found in a plant associated bacterium with a possible role in competition as well as biocontrol. This work is well performed, described and discussed in the context of microbial ecology. The weakness is the lack of mechanistic data and insight on the mechanism(s) of this system and these initial interesting results raise several questions which need more attention in order to make this work more tangible and less 'preliminary'.  
+
+<|ref|>text<|/ref|><|det|>[[115, 221, 880, 283]]<|/det|>
+Apparently, the method of killing is not via lysis however no insight on possible targets or mechanism is provided; any data on this aspect would considerably increase the impact of this article. The effector is thought to be the 33 locus/protein, has any further experimentation been performed to confirm unequivocally that this is the effector and on its possible target and mechanism?  
+
+<|ref|>text<|/ref|><|det|>[[115, 295, 866, 342]]<|/det|>
+It is surprising that mutants in this locus significantly affect bacterial growth since these are thought to be accessory present in a recently acquired genomic island. This aspect is is unclear and needs more attention/explanation.  
+
+<|ref|>text<|/ref|><|det|>[[115, 355, 840, 386]]<|/det|>
+No reference is made on whether this system is also able to infect eukaryotic cells or should it be considered specific for bacteria- bacteria interactions? Has this aspect been tested?  
+
+<|ref|>text<|/ref|><|det|>[[115, 400, 867, 461]]<|/det|>
+The immunity aspect of this system is not entirely clear since two loci appear to be involved; one major locusPiso_02332 encodes for an immunity protein, however no mechanistic insight is provided and in addition it is not tested whether this locus alone can provide immunity if transferred to other bacteria.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 480, 386, 496]]<|/det|>
+## Author Rebuttal to Initial comments  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 515, 399, 532]]<|/det|>
+## Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 548, 875, 718]]<|/det|>
+This very interesting study has done a good job of describing a novel form of contact- dependent killing of bacteria. Contact- dependent killing schemes have become quite the rage in microbial ecology, and yet most have been associated with variants of type VI secretion systems. This study has shown a very novel method of delivering what appear to be toxic effectors using a variant of the type IVB secretion system. The study is quite solid, and the authors have employed a variety of powerful tools including Tn sequencing and various microscopy techniques to both demonstrate the contact dependence of the killing of various other bacteria by the Pseudomonas putida strain, and to identify the genetic loci associated with both the toxic effector as well as the immunity function. It is quite remarkable that this type IVB secretion system seems to be quite promiscuous in its ability to kill other bacteria, as they did not seem to find any taxa that were immune to its effect. They might  
+
+<|ref|>text<|/ref|><|det|>[[115, 719, 877, 787]]<|/det|>
+want to comment further on the relative breadth of this killing compared to many other type VI systems that have been studied. As such, there are indeed important translational aspects of the study, as exemplified here in their demonstration that the biological control of bacterial wilt caused by Ralstonia solanacearum could be achieved using this strain capable of contact killing. The manuscript was quite
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[596, 45, 969, 88]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[115, 118, 870, 185]]<|/det|>
+well written, and the powerful and logical experimental design leading to the discovery of the system was well considered. I have only a few relatively minor comments about both the way certain findings should be described, a few conclusions that may be a bit overstated, as well as suggestions for how the demonstration of plant disease control could have been better assessed and described.  
+
+<|ref|>text<|/ref|><|det|>[[115, 203, 855, 237]]<|/det|>
+We are very grateful for the positive evaluation of our study and the helpful comments to improve our manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 254, 184, 270]]<|/det|>
+Specifics:  
+
+<|ref|>text<|/ref|><|det|>[[115, 287, 874, 422]]<|/det|>
+Lines 67- 69. We are told here that a Xanthomonas strain had exhibited a type IV dependent killing of other bacterial strains - while it is only in a couple of sentences later that the authors note that there is more than one kind of type IV secretion system, and that there had been no prior demonstration of the type IVB system being used in antimicrobial activities. However, there was never a mention until much later in the discussion that the Xanthomonas example, was in fact, a type IVA secretion system. I feel that this point should have been made earlier, because I was bothered by the concern that this study might not have been as novel as they were suggesting, because they had not noted that this earlier example was not a type IVB system.  
+
+<|ref|>text<|/ref|><|det|>[[115, 440, 880, 474]]<|/det|>
+The criticism was noticed and we now specify that previous work in Xanthomonas identified a T4ASS in contrast to the T4BSS we identified in P. putida IsoF.  
+
+<|ref|>text<|/ref|><|det|>[[115, 491, 874, 608]]<|/det|>
+Line 90 and elsewhere. Here and in many locations throughout the manuscript, the authors have used the term "competition" to describe the interaction of IsoF with various other bacteria. At least in my mind, the term "competition" implies some sort of growth reduction that would be a result of the need to share certain nutrient resources with a neighbor, or perhaps an effect of the chemical environment by a neighbor etc. Given that they are demonstrating here that "competition" is actually the direct killing of their neighbor, I wonder if it would be cleaner to refer to "interaction" or "killing" or some other term that does not seem to have this nutrient competition connotation.  
+
+<|ref|>text<|/ref|><|det|>[[115, 626, 883, 678]]<|/det|>
+We are thankful for this valuable comment. Accordingly, we have changed the term competition to 'interaction' or 'killing' depending on the context throughout the text. We also changed 'contact- dependent competition (CDC)' to 'contact- dependent killing (CDK)'.  
+
+<|ref|>text<|/ref|><|det|>[[116, 695, 310, 710]]<|/det|>
+Line 97 misspelling: iodide  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 729, 191, 744]]<|/det|>
+## Corrected  
+
+<|ref|>text<|/ref|><|det|>[[115, 762, 781, 780]]<|/det|>
+Line 112. I had to take a step back and try to figure out why the authors had decided to use
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[596, 45, 970, 88]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[115, 118, 881, 208]]<|/det|>
+Pseudomonas aureofaciens in the study discussed in this section because there had not been a previous mention that this particular species was in fact susceptible to the killing effect of strain IsoF. One had to dig into the results of Figure 1, and specifically recognize that Pseudomonas aureofaciens was one of the strains that was in fact susceptible to the killing by IsoF. I think it would benefit from a short note to discuss why there had been a change in the killing assay away from KT2442 to this new bacterial target.  
+
+<|ref|>text<|/ref|><|det|>[[115, 220, 881, 273]]<|/det|>
+Thank you for making us aware of this inconsistency. We have chosen P. aureofaciens in these assays because this strain was more sensitive to killing by IsoF than KT2442. We have added this information in the revised version of the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 301, 842, 335]]<|/det|>
+Line 126: Again, in this sentence they use the term "outcompete" when it's clearly was killing the P. aureofaciens. I think it would be better to note killing rather than competition in such examples.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 353, 330, 369]]<|/det|>
+## Changed to kill as suggested.  
+
+<|ref|>text<|/ref|><|det|>[[115, 386, 876, 420]]<|/det|>
+Line 210: it is a bit disconcerting to hear that they were not able to restore killing by complementation. I do not recall there being a discussion of this negative finding.  
+
+<|ref|>text<|/ref|><|det|>[[114, 436, 882, 573]]<|/det|>
+To address the concern of the reviewer, we have analyzed the strains by SDS- PAGE in order to investigate ectopic expression of PisoF_02332 and PisoF_02333. In the complemented strain, we observed a band corresponding to the of the immunity protein but could not detect the effector, indicating that the immunity protein is produced in excess over the toxin. We therefore hypothesize that killing was not restored as a consequence of an unphysiological overexpression of the immunity protein in the complemented strain that effectively neutralized all effector molecules. We have added this information to the Results section and show the SDS- PAGE in the new Fig. 8b of the Extended Data.  
+
+<|ref|>text<|/ref|><|det|>[[114, 589, 877, 758]]<|/det|>
+Line 236 through 238. Upon initially reading the sentence, I found it surprising and unexpected, since it did not seem like the simple process of killing a bacterium would cause it to disappear from the biofilm. It was only later when the authors noted that this replacement of the dead cells by the IsoF strain only occurred in flow chambers where turbulence etc could in fact have dislodged the dead cells. That said, after reading the entire manuscript I was left with a feeling that there is a bit of an overstatement in the authors conclusions that the IsoF strain could invade biofilms, as this seemed to have been restricted to the flow cells where there is some ability to remove cells from the biofilm, and was not observed in preestablished biofilms that had formed on agar plates etc. This made me wonder whether the invasion of biofilms that might have formed on plants and other habitats where they feel the type IV secretion system could be useful in killing target organisms was
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 118, 844, 152]]<|/det|>
+as likely to occur as they suggest. I feel a bit of further clarification of the situations in which biofilm invasion/replacement is likely to occur is warranted.  
+
+<|ref|>text<|/ref|><|det|>[[114, 180, 883, 362]]<|/det|>
+This is a very valuable and interesting comment. To investigate whether IsoF could also invade a biofilm grown on an agar plate we followed the fate of the strain in CDK assays against KT2442 over 72 hours. While growth of IsoF was restricted to the initial inoculation area after 24 hours, we observed that IsoF invaded the space occupied by the target strain and formed satellite colonies after 72 hours. We hypothesize that killed cells eventually lyse and no longer form a barrier that prevents invasion. Another important factor for invasion competence is that IsoF produces the very powerful biosurfactant putisolin, which was not only shown to disperse pre- established biofilms but also allows the strain to translocate over semisolid surfaces by means of swarming motility. We have added this information in the Results section, as new Fig. 12 in the Extended Data, and amended the discussion to clarify this issue.  
+
+<|ref|>text<|/ref|><|det|>[[114, 390, 880, 719]]<|/det|>
+There were several aspects of the plant protection experiment that I found a bit awkward or simplistic. The experiment, using small seedlings growing on agar plates is a somewhat unnatural setting, and one that might have facilitated both microbial survival on, and multiplication on the damaged plant tissue. The experiment was done in a way that would have maximized the potential effect of the presence of IsoF together with the pathogen, since they were mixed together and then immediately applied to the wounded plant. There thus would have been maximal opportunity for IsoF to have been in close proximity to the pathogen and thus for it to have killed it before the pathogen could have multiplied and caused disease. I was disappointed that the authors did not attempt to do a more biologically relevant experiment in which plants grown in some sort of soil matrix or even sand would have had the soil flooded simultaneously with a mixture of the bacteria, or even better, flooded first with the pathogen and then shortly thereafter with the beneficial strain. This would have been a more powerful test of the ability of IsoF to kill developing biofilms on the roots, thereby alleviating the likelihood of disease. These would have been quite simple studies to perform and would better reflect the normal process of infection. I was surprised at the way the authors presented the information on the effects of IsoF on the disease process, relating root weight, leaf area, chlorophyll content etc in a PCO plot in figure 5C. I had never seen such results presented in this way, and found it very difficult to interpret. I feel it would have been much cleaner to have simply prepared a small table showing the effect of the three or four treatments on each of these dependent variables.  
+
+<|ref|>text<|/ref|><|det|>[[115, 730, 882, 784]]<|/det|>
+We agree that the plant protection experiment is somewhat simplistic and artificial. To follow the advice of the reviewer, we established a soil- based infection model as previously described by Medina and López- Baena, 2018 and used it to investigate the effect of the kib cluster on biocontrol activity
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[595, 45, 970, 88]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[114, 118, 883, 246]]<|/det|>
+under more natural conditions. In this setup non- sterile soil is drenched simultaneously with a mixture of IsoF or AT4B and Ralstonia. The results, which are shown in the modified Fig. 5 as well as in the new Fig. 20 of the Extended Data, demonstrated that IsoF successfully prevents disease and that this effect is dependent on the kib gene cluster. The results of these experiments were included in the main text. We feel that the PCA graph is the best way to present our data. However, the criticism was noticed and to satisfy the concerns of the reviewer we also added the independent graphs for each parameter measured in Fig. 18 of the Extended Data.  
+
+<|ref|>text<|/ref|><|det|>[[114, 274, 848, 310]]<|/det|>
+Line 372 to 375: this important statement was not given the weight that I feel it deserved, as it does point to perhaps a limited scenario in which type IVB killing will prove to be biologically important.  
+
+<|ref|>text<|/ref|><|det|>[[114, 339, 883, 449]]<|/det|>
+As mentioned before, we have added additional data showing that IsoF in fact is capable of invading a pre- established biofilm on an agar plate, it just requires more time (72 h versus 24 h, which we used in our routine assays). We have included this information in Fig. 4c and amended the discussion accordingly. Moreover, we have rephrased our statement on the importance of putisolvin to replace cells in biofilms, as previous work has demonstrated that this biosurfactant can efficiently remove pre- established biofilms (Kuiper et al., 2004).  
+
+<|ref|>text<|/ref|><|det|>[[114, 460, 870, 494]]<|/det|>
+Line 388 - 389: this sentence seems to be a bit of an overstatement given what they have just noted on lines 372 to 375.  
+
+<|ref|>text<|/ref|><|det|>[[114, 506, 881, 542]]<|/det|>
+As mentioned above, IsoF not only invaded a pre- established biofilm in flow- cells but was also capable of invading a pre- established biofilm on an agar plate.  
+
+<|ref|>text<|/ref|><|det|>[[115, 570, 870, 623]]<|/det|>
+I found the discussion a very interesting and pleasant read, although I had the strong impression that it reiterated almost word for word statements that had been made in the results. Some condensation seems possible.  
+
+<|ref|>text<|/ref|><|det|>[[115, 636, 448, 653]]<|/det|>
+We have tightened up the text as suggested.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 692, 399, 708]]<|/det|>
+## Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 728, 880, 781]]<|/det|>
+The manuscript by Purtschert- Montenegro et al describes the phenomenon of contact- dependent killing of other bacterial species by the soil bacterium Pseudomonas putida IsoF, shown to be mediated by a Type IV secretion system coded by the newly named kib cluster. This phenomenon has been shown
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[596, 44, 970, 88]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[114, 117, 876, 318]]<|/det|>
+before in Xanthomonas and Stenotrophomonas bacteria. The difference here is that the P. putida isoF T4SS belongs to the larger Type IVB class while that of X. citri and S. maltophilia belong to the Type IVA systems. The authors use fluorescently- labelled knockout strains to show that the killing is dependent on structural components of the T4SS apparatus as well as on a putative effector coded by the kib locus. The authors go on to perform some elegant experiments that show that the kib locus, and specifically the effector Piso_02333, allow the IsoF strain to invade biofilms formed by the P. putida K2442 strain (that does not carry a T4SS, but rather a bacteria- killing T6SS). They then show that the IsoF strain can prevent the phytopathogenic bacterium Ralstonia solanacearum from infecting tomato seedlings. The manuscript is very well written, the results are clearly explained and presented and the experiments appear to have been carefully carried out. I do, however have a few comments and suggestions that I believe should be addressed.  
+
+<|ref|>text<|/ref|><|det|>[[114, 337, 881, 648]]<|/det|>
+Perhaps my major criticism (easily addressed) is that the reader is not provided with the accession numbers of the P. putida IsoF genome nor the accession numbers of the genes coded by the kib locus. Therefore the reader cannot confer whether the genes described in the paper do in fact correspond to the specified T4SSB components. For example, Lines 139- 145: Here the kib cluster coding 61 genes is described with reference to Figure 2b and Extended Data Table 1. However, in neither this figure or this table, nor anywhere else in the manuscript or in the Supplementary Information are the accession numbers provided for the Pseudomonas putida IsoF strain genome or the genes in the kib cluster. I searched for these sequences in the NCBI database and could not find any of them. Also, searching for the gene names Piso_02333 and Piso_02332 in these databases did not turn up any hits. Finally, no Data Availability Statement was provided, which should contain these accession codes. Is it possible case that the genome of this organism has not yet been made publicly available? I think that it should definitely be deposited and released before being sent off for review. Especially since this strain has been the subject of study by the Eberle group for at least two decades (for example: Steidle et al, 2001. doi: 10.1128/AEM.67.12.5761- 5770.2001. ). The authors should provide a table with all accession numbers of the 61 genes genes in the kib locus. If this is not possible, then the full nucleotide and translated protein sequences each open reading frame in the kib locus should be provided in fasta format as supplementary information.  
+
+<|ref|>text<|/ref|><|det|>[[115, 676, 883, 749]]<|/det|>
+We are very sorry for this mistake. The genome sequence of IsoF is now available at NCBI under the accession number CP072013. We have added this in a Data Availability Statement as requested. In addition, we have added the accession numbers of the 61 genes of the kib locus in the new Table 4 in the Extended Dataset.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[595, 45, 970, 88]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[115, 118, 860, 153]]<|/det|>
+Another point, in a way related to that above, is that the nature of the Piso_02332/Piso_02333 pair is not explored to any significant extent.  
+
+<|ref|>text<|/ref|><|det|>[[115, 165, 839, 200]]<|/det|>
+How many homologs of these two proteins are found in the public databases and what can we say about their phylogenetic distribution (individually or as a pair)?  
+
+<|ref|>text<|/ref|><|det|>[[114, 211, 883, 375]]<|/det|>
+We are thankful for this comment. We searched the NCBI database for homologs of PisoF_02332 and PisoF_02333 and found that the operon structure is fully conserved, supporting the idea that they represent an E- I pair. Moreover, the genes were exclusively found within homologs of the kib locus in a few Pseudomonas strains, we unable to identify orphan homologs. The comparison of the phylogenetic trees of the PisoF_02332- 33 genes, all orthologs of the kib cluster and eight housekeeping genes of the strains carrying the kib locus revealed that the tree topology is congruent, suggesting that strains carrying the kib cluster form a defined lineage that originated from a common ancestor. We have added this information as a new section in the results and in the new Fig. 9 of the Extended Data.  
+
+<|ref|>text<|/ref|><|det|>[[115, 386, 844, 421]]<|/det|>
+Is the Piso_02332 immunity protein expected to be localized in the cytosol of the periplasmid (does it have a signal peptide?). This could provide clues regarding the site of action of its cognate effector.  
+
+<|ref|>text<|/ref|><|det|>[[115, 432, 882, 484]]<|/det|>
+Using LocTREE and PSORTb the subcellular localization of PisoF_02332 and PisoF_02333 was predicted to be to be cytoplasmic. Additionally, using SignalP- 6.0 web tool we found that neither protein had a signal peptide. We have added this information in the results section.  
+
+<|ref|>text<|/ref|><|det|>[[115, 511, 875, 546]]<|/det|>
+Regarding the Piso_02333 effector: Does it have any putative motifs that could be used as a recognition signal for secretion by the T4SSB apparatus?  
+
+<|ref|>text<|/ref|><|det|>[[114, 557, 883, 758]]<|/det|>
+Although the T4BSS of Legionella translocates more than 330 effectors only few recognition sequences have been described. For some effectors the following C terminal motifs have been described: hydrophobic residues (Nagai et al., 2005; Voth et al., 2012), an EExxE domain (Huang et al., 2011) and a FxxxLxxxK domain (Kim et al., 2020). However, none of these motifs could be identified in the C- terminal region of the effector PisoF_02333. Interestingly, a FxxxLxxxK domain was found to be present in the C- terminal region of the immunity protein PisoF_02332, suggesting that this protein may be transferred together with its cognate effector toxin. Moreover, we noticed that PisoF_02333 has an unusual glutamine- rich domain in the C- terminal region of the protein (9 Q of 14 aa; Extended Data Fig. 8a) and speculate that these may play a role in effector recognition. Interestingly, a conserved glutamine- rich domain was also identified in the C- terminal regions of VirD4 coupling proteins of phylogenetically diverse T4ASSs (Das, 2020). This region was shown to be required for
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[595, 44, 970, 88]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[114, 118, 883, 154]]<|/det|>
+recognition of T- strand DNA but not of the second transferred substrate, the single- stranded DNA- binding protein VirE2. This information has been added.  
+
+<|ref|>text<|/ref|><|det|>[[115, 166, 213, 181]]<|/det|>
+Other points.  
+
+<|ref|>text<|/ref|><|det|>[[115, 201, 877, 255]]<|/det|>
+Lines 67- 69. A bacterial killing T4SS has been characterized in X. citri and Stenotrophomonas maltophilia (Souza et al, 2015; Bayer- Santos et al, 2019). And homologous systems have been identified in over a hundred other bacterial species (Sgro et al, 2019).  
+
+<|ref|>text<|/ref|><|det|>[[115, 267, 577, 283]]<|/det|>
+We have added this information and references as requested.  
+
+<|ref|>text<|/ref|><|det|>[[115, 313, 870, 385]]<|/det|>
+Lines 71- 73: The authors imply that DNA transfer is mediated only by class A systems while class B systems have until now been restricted to the role of transferring effectors into eukaryotic cells. This is not quite true. For example, Class B T4SSs are responsible for the horizontal transfer (conjugation) of Incl plasmids.  
+
+<|ref|>text<|/ref|><|det|>[[115, 414, 883, 450]]<|/det|>
+We are grateful for this comment and have re- phrased the sentence to indicate that transferring effectors into eukaryotic cells is not the only function of T4BSS.  
+
+<|ref|>text<|/ref|><|det|>[[115, 489, 840, 524]]<|/det|>
+Lines 113- 116: Eight out of sixteen killing defective mutants localize to four genes in the kib cluster. What about the other eight insertion mutants? What were their insertion sites?  
+
+<|ref|>text<|/ref|><|det|>[[115, 536, 883, 589]]<|/det|>
+We were not able to amplify the regions of the Tn5 insertions in the other 8 mutants and, given that all successfully sequenced mutants were within the kib gene cluster, we did not further analyze these mutants.  
+
+<|ref|>text<|/ref|><|det|>[[115, 600, 855, 654]]<|/det|>
+Lines 103,106- 107 and 296: in several places in the manuscript, the term "host range" is used to represent competitor bacterial species. This is not really appropriate. A more suitable term would be "range of target species" or "range of target organisms".  
+
+<|ref|>text<|/ref|><|det|>[[115, 665, 880, 683]]<|/det|>
+This is a good point and we have changed the wording to 'range of target species' throughout the text.  
+
+<|ref|>text<|/ref|><|det|>[[115, 702, 881, 755]]<|/det|>
+Line 162: the references 22, 32, 55 and 66 refer to studies on only T6SS effector- immunity protein pairs. However, a few thousand bacteria- killing T4SSA effector- immunity protein pairs have been identified in the genomes of over a hundred species.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[595, 45, 969, 88]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[115, 118, 883, 154]]<|/det|>
+This is true, thank you for pointing this out. We have added the following publications in the text: Bayer- Santos et al., 2019; Sgro et al., 2019; Souza et al., 2015.  
+
+<|ref|>text<|/ref|><|det|>[[115, 184, 883, 219]]<|/det|>
+Lines 202- 204. It is reported that the delta32- 33 strain grows more slowly than the wild- type strain or the delta33 strain. This could suggest that Piso_02332 may be neutralizing more than one effector.  
+
+<|ref|>text<|/ref|><|det|>[[114, 230, 883, 375]]<|/det|>
+In fact, we speculated that Piso_02351 and PisoF_02352 may encode an additional E- I pair. However, recent work in our lab did not support the idea of another E- I pair within the kib gene cluster and as a consequence we have removed this statement. Our most recent results rather suggest that that the observed retarded growth of the mutant is a consequence of a defective kib nanomachinery, which appears to impact viability. In this context, it is interesting to note that inactivation of dotL in certain Legionella pneumophila strains is lethal (Buscher et al., 2005). However, lethality of a dotL mutation is suppressed by mutation of other components of the T4BSS, indicating that the interactions between the different protein components is finely tuned and that a disturbance may lead to self- toxicity.  
+
+<|ref|>text<|/ref|><|det|>[[115, 414, 882, 468]]<|/det|>
+Lines 208- 211: Why were you not able to restore killing of P. aureofaciens and KT2442 by complementing the mutant delta32- 33 strain with a plasmid pBBR::32- 33 coding for the effector- immunity pair. Did the authors confirm that this plasmid in fact expresses the two proteins?  
+
+<|ref|>text<|/ref|><|det|>[[114, 496, 882, 616]]<|/det|>
+As already mentioned in our response #1, we have analyzed the strains by SDS- PAGE in order to investigate ectopic expression of PisoF_02332 and PisoF_02333. In the complemented strain, we observed a band corresponding to the of the immunity protein but could not detect the toxin, indicating that the immunity protein is produced in excess over the toxin. We therefore hypothesize that killing was not restored because of an unphysiological overexpression of the immunity protein in the complemented strain that effectively neutralized the effector. We have added this information to the discussion and show the SDS- PAGE in the new Fig. 8b of the Extended Data.  
+
+<|ref|>text<|/ref|><|det|>[[115, 633, 879, 705]]<|/det|>
+Lines 226- 229: It is written that after 3 days of incubation, isoF had formed a mature biofilm by invading and replacing the KT2442 biofilm. This would correspond to time point 120 hours (48 hrs of KT2442 growth on its own plus 72 after addition of IsoF). No Figure is mentioned to show this. Figures 4a and 4b only show growth up to two days after isoF addition to KT2442 biofilms (total of 96 hours).  
+
+<|ref|>text<|/ref|><|det|>[[115, 716, 882, 751]]<|/det|>
+The reviewer is right that the data shown only cover 72 hours of co- cultivation. The text was corrected accordingly. Thank you for pointing this out.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[595, 45, 969, 88]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[115, 118, 861, 154]]<|/det|>
+Lines 281- 283: In Figure 5d, why are the deltaT4B CFUs the same as the IsoF CFUs? I would expect the IsoF numbers to be greater than the deltaT4B numbers in these experiments.  
+
+<|ref|>text<|/ref|><|det|>[[114, 165, 883, 291]]<|/det|>
+IsoF was shown to be an excellent tomato root colonizer that forms biofilms on the root surface (Steidle et al., 2001). We speculate that that the constant CFU numbers reflect that the loading capacity of the root for IsoF has been reached and that this is not affected by the kib locus. In this context, it is also important to keep in mind that Ralstonia only transiently colonizes the root surface before the bacteria gain access to host root systems through natural wounds caused by the emergence of lateral roots or through wounds acquired as roots grow through the soil (Xue et al., 2020).  
+
+<|ref|>text<|/ref|><|det|>[[114, 302, 882, 393]]<|/det|>
+Lines 298- 302: This is only partially correct. All of the Xanthomonadaceae- like bacterial killing T4SSs have one or more effector/immunity pairs at the same locus containing all of the structural components of the secretion system PLUS other effector/immunity pairs found in other chromosomal locations (Sgro et al, 2019). Here, since we have no access the genome sequence of the IsoF strain under study, we have no way to investigate whether other possible effector/immunity pairs are found in the genome.  
+
+<|ref|>text<|/ref|><|det|>[[114, 404, 883, 440]]<|/det|>
+We are sorry that the genome sequence was not available at the time of submission. We have made it available now. We have reworded the statement and shortened the text.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 480, 400, 496]]<|/det|>
+## Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[114, 498, 877, 585]]<|/det|>
+This study reports that a P. putida strain contains a locus encoding a type IVB secretion system which is involved in bacterial killing via cell- cell contact. This system possesses the novel feature since type IV SS have been reported thus far to only infect eukaryotic cells. Other aspects of this work include that the locus is most likely found in a genomic island which has been recently acquired and that this locus is not widespread since it is only found thus far in a bunch of Pseudomonas isolates.  
+
+<|ref|>text<|/ref|><|det|>[[114, 587, 856, 678]]<|/det|>
+This work reports on an interesting novel locus found in a plant associated bacterium with a possible role in competition as well as biocontrol. This work is well performed, described and discussed in the context of microbial ecology. The weakness is the lack of mechanistic data and insight on the mechanism(s) of this system and these initial interesting results raise several questions which need more attention in order to make this work more tangible and less 'preliminary'.  
+
+<|ref|>text<|/ref|><|det|>[[114, 708, 883, 781]]<|/det|>
+We are thankful for the supportive comments. We agree with the reviewer that it would be nice to have more insights into the underlying molecular mechanisms of killing by this nanomachinery. The main difficulty in this respect is the absolute novelty of the system that shares no homology with any other killing machinery. However, the criticism was noticed and we have performed additional
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[595, 45, 970, 88]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[115, 118, 883, 172]]<|/det|>
+experiments to shed more light on the mode of killing. However, the main focus of our study was indeed the identification and characterization of this novel killing machinery and the evaluation of its biocontrol potential.  
+
+<|ref|>text<|/ref|><|det|>[[115, 202, 873, 272]]<|/det|>
+Apparently, the method of killing is not via lysis however no insight on possible targets or mechanism is provided; any data on this aspect would considerably increase the impact of this article. The effector is thought to be the 33 locus/protein, has any further experimentation been performed to confirm unequivocally that this is the effector and on its possible target and mechanism?  
+
+<|ref|>text<|/ref|><|det|>[[114, 284, 883, 448]]<|/det|>
+Bioinformatic analysis of PisoF_02332 and PisoF_02333 revealed that both proteins are located in the cytoplasm. We have added this information to the text and in the new Fig. 8 of the Extended Data. As mentioned in our response to reviewer #2, we also noticed that PisoF_02333 has an unusual glutamine- rich domain in the C- terminal region of the protein (9 Q of 14 aa; Extended Data Fig. 8a) and speculate that it may play a role in effector recognition. Interestingly, a conserved glutamine- rich domain was also identified in the C- terminal regions of VirD4 coupling proteins of phylogenetically diverse T4ASSs. This region was shown to be required for recognition of T- strand DNA but not of the second transferred substrate, the single- stranded DNA- binding protein VirE2 (Das, 2020). This information has been added to the Discussion.  
+
+<|ref|>text<|/ref|><|det|>[[115, 460, 882, 531]]<|/det|>
+Moreover, work that is currently ongoing in the lab aims at isolating resistant mutants of susceptible bacteria with the aim to identify the molecular target(s) of kib- encoded effectors. If successful, these investigations may allow us to gain insights into the mode of killing. However, we feel that this will be a story on its own.  
+
+<|ref|>text<|/ref|><|det|>[[115, 560, 867, 613]]<|/det|>
+It is surprising that mutants in this locus significantly affect bacterial growth since these are thought to be accessory present in a recently acquired genomic island. This aspect is is unclear and needs more attention/explanation.  
+
+<|ref|>text<|/ref|><|det|>[[115, 625, 883, 715]]<|/det|>
+As discussed before, we hypothesize that the observed retarded growth of the mutant is a consequence of a defective kib nanomachinery, which may impact viability. It appears that the interactions between the different protein components of kib nanomachinery is finely tuned and that a disturbance by deleting the E- I pair may lead to self- toxicity. We have added this information to the Discussion to improve clarity.  
+
+<|ref|>text<|/ref|><|det|>[[115, 727, 825, 761]]<|/det|>
+No reference is made on whether this system is also able to infect eukaryotic cells or should it be considered specific for bacteria- bacteria interactions? Has this aspect been tested?
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[595, 45, 970, 88]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[114, 118, 883, 208]]<|/det|>
+We have tested IsoF in a C. elegans infection model and found the strain to be avirulent. Likewise, we could not observe antifungal or anti- oomycete activities. It is also noteworthy that IsoF does not grow at \(37^{\circ}C\) and thus will be unable to infect mammalian cells. Furthermore, in our tomato root colonization assays we could not observe an effect of the kib locus on plant growth. Collectively, these data suggest that this killing machinery is specific for bacterial cells interactions.  
+
+<|ref|>text<|/ref|><|det|>[[115, 220, 872, 273]]<|/det|>
+The immunity aspect of this system is not entirely clear since two loci appear to be involved; one major locus Piso_02332 encodes for an immunity protein, however no mechanistic insight is provided and in addition it is not tested whether this locus alone can provide immunity if transferred to other bacteria.  
+
+<|ref|>text<|/ref|><|det|>[[114, 285, 883, 467]]<|/det|>
+We showed that complementation of the \(\Delta 32 - 33\) mutant by providing PisoF_02332 in trans on a plasmid (pBBR::32) rendered the strain resistant to killing by the wildtype. However, we were unable to complement the \(\Delta T4B\) mutant, which lacks the entire kib gene cluster, with pBBR::32. Analysis of the strains by SDS- PAGE revealed that a band corresponding to the PisoF_02332 protein is visible in the complemented \(\Delta 32 - 33\) mutant but not the complemented \(\Delta T4B\) mutant. The lack of PisoF_02332 expression in the \(\Delta T4B\) mutant background explains its sensitivity to kib- mediated killing and suggests that the kib gene cluster encodes functions required for the expression of PisoF_02332 or affects its stability. We have added this information to the main text and have removed the speculation that another E- I pair is encoded by the cluster, as work currently under way in our laboratory does not support this idea.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 515, 335, 530]]<|/det|>
+## Decision Letter, first revision:  
+
+<|ref|>text<|/ref|><|det|>[[119, 550, 230, 565]]<|/det|>
+Dear Dr. Eberl,  
+
+<|ref|>text<|/ref|><|det|>[[115, 580, 870, 669]]<|/det|>
+Thank you for submitting your revised manuscript "A type IVB secretion system adapted for bacterial killing, biofilm invasion and biocontrol" (NMICROBIOL- 21092434A). It has now been seen by the original referees and their comments are below. The reviewers find that the paper has improved in revision, and therefore we'll be happy in principle to publish it in Nature Microbiology, pending minor revisions to satisfy the referees' final requests and to comply with our editorial and formatting guidelines.  
+
+<|ref|>text<|/ref|><|det|>[[115, 684, 857, 715]]<|/det|>
+If the current version of your manuscript is in a PDF format, please email us a copy of the file in an editable format (Microsoft Word or LaTex)- - we can not proceed with PDFs at this stage.  
+
+<|ref|>text<|/ref|><|det|>[[115, 729, 857, 775]]<|/det|>
+We are now performing detailed checks on your paper and will send you a checklist detailing our editorial and formatting requirements in about a week. Please do not upload the final materials and make any revisions until you receive this additional information from us.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[596, 45, 970, 88]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[115, 117, 861, 149]]<|/det|>
+Thank you again for your interest in Nature Microbiology Please do not hesitate to contact me if you have any questions.  
+
+<|ref|>text<|/ref|><|det|>[[115, 162, 191, 178]]<|/det|>
+Sincerely,  
+
+<|ref|>text<|/ref|><|det|>[[115, 192, 198, 220]]<|/det|>
+{redacted}  
+
+<|ref|>text<|/ref|><|det|>[[115, 237, 404, 252]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 267, 191, 281]]<|/det|>
+see below  
+
+<|ref|>text<|/ref|><|det|>[[115, 310, 404, 326]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 340, 872, 371]]<|/det|>
+This revised version is an improvement as authors have carefully revised most of the points raised by the three reviewers.  
+
+<|ref|>text<|/ref|><|det|>[[115, 384, 555, 400]]<|/det|>
+Most additional experimenting and data was also provided:  
+
+<|ref|>text<|/ref|><|det|>[[112, 413, 877, 595]]<|/det|>
+- SDS PAGE analysis, the immunity protein was believed to be under-expressed compared to the toxin hence there was a lack of complementation.- The authors have performed a new plant experiment based on a soil infection model in order to test biocontrol activity under conditions which resemble more the wild. The results confirmed what was previously observed in more controlled conditions.- An additional biofilm experiment was also performed involving a longer time frame further evidencing the effects of bacterial invasion.- The genome sequence was properly deposited in data banks.- Performed a more complete phylogenetic search of the two ORFs (2332 and 2333)- More in silico protein information of the 2332 and 2333 ORFs- Have tested the killing model on C. elegans indicating that it does not appear to infect eukaryotic cells  
+
+<|ref|>text<|/ref|><|det|>[[115, 608, 864, 639]]<|/det|>
+In the revised text, these new additional results and figures are well integrated in the document and have no comments.  
+
+<|ref|>text<|/ref|><|det|>[[115, 653, 237, 668]]<|/det|>
+Minor comment:  
+
+<|ref|>text<|/ref|><|det|>[[115, 682, 878, 728]]<|/det|>
+Not clear the reasoning why 8 Tn5 mutants could not be mapped - nowadays with NGS it is ultimately possible to map any mutation in the genome. I suggest to either map these mutants or just remove this information and simply state that a subset was mapped.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[595, 44, 970, 88]]<|/det|>
+# nature portfolio  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 138, 325, 153]]<|/det|>
+## Decision Letter, final checks  
+
+<|ref|>text<|/ref|><|det|>[[115, 191, 228, 205]]<|/det|>
+Dear Dr. Eberl,  
+
+<|ref|>text<|/ref|><|det|>[[115, 220, 877, 325]]<|/det|>
+Thank you for your patience as we've prepared the guidelines for final submission of your Nature Microbiology manuscript, "A type IVB secretion system adapted for bacterial killing, biofilm invasion and biocontrol" (NMICROBIOL- 21092434A). Please carefully follow the step- by- step instructions provided in the attached file, and add a response in each row of the table to indicate the changes that you have made. Please also check and comment on any additional marked- up edits we have proposed within the text. Ensuring that each point is addressed will help to ensure that your revised manuscript can be swiftly handed over to our production team.  
+
+<|ref|>text<|/ref|><|det|>[[115, 339, 870, 370]]<|/det|>
+We would like to start working on your revised paper, with all of the requested files and forms, as soon as possible (preferably within two weeks). Please get in contact with us if you anticipate delays.  
+
+<|ref|>text<|/ref|><|det|>[[115, 384, 840, 414]]<|/det|>
+When you upload your final materials, please include a point- by- point response to any remaining reviewer comments.  
+
+<|ref|>text<|/ref|><|det|>[[115, 429, 869, 488]]<|/det|>
+If you have not done so already, please alert us to any related manuscripts from your group that are under consideration or in press at other journals, or are being written up for submission to other journals (see: https://www.nature.com/nature- research/editorial- policies/plagiarism#policy- on- duplicate- publication for details).  
+
+<|ref|>text<|/ref|><|det|>[[115, 503, 878, 577]]<|/det|>
+In recognition of the time and expertise our reviewers provide to Nature Microbiology's editorial process, we would like to formally acknowledge their contribution to the external peer review of your manuscript entitled "A type IVB secretion system adapted for bacterial killing, biofilm invasion and biocontrol". For those reviewers who give their assent, we will be publishing their names alongside the published article.  
+
+<|ref|>text<|/ref|><|det|>[[115, 592, 858, 698]]<|/det|>
+Nature Microbiology offers a Transparent Peer Review option for new original research manuscripts submitted after December 1st, 2019. As part of this initiative, we encourage our authors to support increased transparency into the peer review process by agreeing to have the reviewer comments, author rebuttal letters, and editorial decision letters published as a Supplementary item. When you submit your final files please clearly state in your cover letter whether or not you would like to participate in this initiative. Please note that failure to state your preference will result in delays in accepting your manuscript for publication.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 712, 268, 727]]<|/det|>
+## Cover suggestions  
+
+<|ref|>text<|/ref|><|det|>[[115, 741, 816, 772]]<|/det|>
+As you prepare your final files we encourage you to consider whether you have any images or illustrations that may be appropriate for use on the cover of Nature Microbiology.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[596, 45, 969, 88]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[115, 117, 880, 163]]<|/det|>
+Covers should be both aesthetically appealing and scientifically relevant, and should be supplied at the best quality available. Due to the prominence of these images, we do not generally select images featuring faces, children, text, graphs, schematic drawings, or collages on our covers.  
+
+<|ref|>text<|/ref|><|det|>[[115, 177, 844, 208]]<|/det|>
+We accept TIFF, JPEG, PNG or PSD file formats (a layered PSD file would be ideal), and the image should be at least 300ppi resolution (preferably 600- 1200 ppi), in CMYK colour mode.  
+
+<|ref|>text<|/ref|><|det|>[[115, 222, 877, 252]]<|/det|>
+If your image is selected, we may also use it on the journal website as a banner image, and may need to make artistic alterations to fit our journal style.  
+
+<|ref|>text<|/ref|><|det|>[[115, 266, 860, 297]]<|/det|>
+Please submit your suggestions, clearly labeled, along with your final files. We'll be in touch if more information is needed.  
+
+<|ref|>text<|/ref|><|det|>[[115, 325, 875, 415]]<|/det|>
+Nature Microbiology has now transitioned to a unified Rights Collection system which will allow our Author Services team to quickly and easily collect the rights and permissions required to publish your work. Approximately 10 days after your paper is formally accepted, you will receive an email in providing you with a link to complete the grant of rights. If your paper is eligible for Open Access, our Author Services team will also be in touch regarding any additional information that may be required to arrange payment for your article.  
+
+<|ref|>text<|/ref|><|det|>[[115, 430, 850, 460]]<|/det|>
+Please note that you will not receive your proofs until the publishing agreement has been received through our system.  
+
+<|ref|>text<|/ref|><|det|>[[115, 475, 868, 564]]<|/det|>
+Please note that <i>Nature Microbiology</i> is a Transformative Journal (TJ). Authors may publish their research with us through the traditional subscription access route or make their paper immediately open access through payment of an article- processing charge (APC). Authors will not be required to make a final decision about access to their article until it has been accepted. <a href="https://www.springernature.com/gp/open- research/transformative- journals"> Find out more about Transformative Journals</a>  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 579, 580, 594]]<|/det|>
+## Authors may need to take specific actions to achieve <a  
+
+<|ref|>text<|/ref|><|det|>[[115, 595, 880, 743]]<|/det|>
+href="https://www.springernature.com/gp/open- research/funding/policy- compliance- faqs"> compliance</a> with funder and institutional open access mandates. If your research is supported by a funder that requires immediate open access (e.g. according to <a href="https://www.springernature.com/gp/open- research/plan- s- compliance">Plan S principles</a>) then you should select the gold OA route, and we will direct you to the compliant route where possible. For authors selecting the subscription publication route, the journal's standard licensing terms will need to be accepted, including <a href="https://www.nature.com/nature- portfolio/editorial- policies/self- archiving- and- license- to- publish">self- archiving policies</a>. Those licensing terms will supersede any other terms that the author or any third party may assert apply to any version of the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 773, 680, 788]]<|/det|>
+For information regarding our different publishing models please see our <a
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[595, 45, 970, 88]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[115, 117, 866, 163]]<|/det|>
+href="https://www.springernature.com/gp/open- research/transformative- journals"> Transformative Journals </a> page. If you have any questions about costs, Open Access requirements, or our legal forms, please contact ASJournals@springernature.com.  
+
+<|ref|>text<|/ref|><|det|>[[115, 176, 554, 207]]<|/det|>
+Please use the following link for uploading these materials: {redacted}  
+
+<|ref|>text<|/ref|><|det|>[[115, 221, 601, 237]]<|/det|>
+If you have any further questions, please feel free to contact me.  
+
+<|ref|>text<|/ref|><|det|>[[115, 267, 215, 282]]<|/det|>
+Best regards,  
+
+<|ref|>text<|/ref|><|det|>[[115, 297, 200, 312]]<|/det|>
+{redacted}  
+
+<|ref|>text<|/ref|><|det|>[[115, 355, 292, 400]]<|/det|>
+Reviewer #1: Remarks to the Author: see below  
+
+<|ref|>text<|/ref|><|det|>[[115, 444, 217, 457]]<|/det|>
+Reviewer #3:  
+
+<|ref|>text<|/ref|><|det|>[[115, 459, 291, 472]]<|/det|>
+Remarks to the Author:  
+
+<|ref|>text<|/ref|><|det|>[[115, 474, 872, 504]]<|/det|>
+This revised version is an improvement as authors have carefully revised most of the points raised by the three reviewers.  
+
+<|ref|>text<|/ref|><|det|>[[115, 519, 555, 535]]<|/det|>
+Most additional experimenting and data was also provided:  
+
+<|ref|>text<|/ref|><|det|>[[113, 549, 879, 730]]<|/det|>
+- SDS PAGE analysis, the immunity protein was believed to be under-expressed compared to the toxin hence there was a lack of complementation.- The authors have performed a new plant experiment based on a soil infection model in order to test biocontrol activity under conditions which resemble more the wild. The results confirmed what was previously observed in more controlled conditions.- An additional biofilm experiment was also performed involving a longer time frame further evidencing the effects of bacterial invasion.- The genome sequence was properly deposited in data banks.- Performed a more complete phylogenetic search of the two ORFs (2332 and 2333)- More in silico protein information of the 2332 and 2333 ORFs- Have tested the killing model on C. elegans indicating that it does not appear to infect eukaryotic cells  
+
+<|ref|>text<|/ref|><|det|>[[115, 744, 866, 773]]<|/det|>
+In the revised text, these new additional results and figures are well integrated in the document and have no comments.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[596, 44, 970, 88]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[115, 117, 238, 131]]<|/det|>
+Minor comment:  
+
+<|ref|>text<|/ref|><|det|>[[115, 146, 880, 192]]<|/det|>
+Not clear the reasoning why 8 Tn5 mutants could not be mapped - nowadays with NGS it is ultimately possible to map any mutation in the genome. I suggest to either map these mutants or just remove this information and simply state that a subset was mapped.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 212, 272, 226]]<|/det|>
+## Final Decision Letter:  
+
+<|ref|>text<|/ref|><|det|>[[115, 247, 188, 261]]<|/det|>
+Dear Leo,  
+
+<|ref|>text<|/ref|><|det|>[[115, 275, 856, 321]]<|/det|>
+I am pleased to accept your Article "<i>Pseudomonas putida</i> mediates bacterial killing, biofilm invasion and biocontrol with a type IVB secretion system" for publication in Nature Microbiology. Thank you for having chosen to submit your work to us and many congratulations.  
+
+<|ref|>text<|/ref|><|det|>[[115, 335, 841, 381]]<|/det|>
+Over the next few weeks, your paper will be copyedited to ensure that it conforms to Nature Microbiology style. We look particularly carefully at the titles of all papers to ensure that they are relatively brief and understandable.  
+
+<|ref|>text<|/ref|><|det|>[[115, 395, 869, 456]]<|/det|>
+Once your paper is typeset, you will receive an email with a link to choose the appropriate publishing options for your paper and our Author Services team will be in touch regarding any additional information that may be required. Once your paper has been scheduled for online publication, the Nature press office will be in touch to confirm the details.  
+
+<|ref|>text<|/ref|><|det|>[[115, 470, 878, 531]]<|/det|>
+After the grant of rights is completed, you will receive a link to your electronic proof via email with a request to make any corrections within 48 hours. If, when you receive your proof, you cannot meet this deadline, please inform us at rjsproduction@springernature.com immediately. You will not receive your proofs until the publishing agreement has been received through our system  
+
+<|ref|>text<|/ref|><|det|>[[115, 544, 881, 604]]<|/det|>
+Due to the importance of these deadlines, we ask you please us know now whether you will be difficult to contact over the next month. If this is the case, we ask you provide us with the contact information (email, phone and fax) of someone who will be able to check the proofs on your behalf, and who will be available to address any last- minute problems.  
+
+<|ref|>text<|/ref|><|det|>[[115, 618, 879, 679]]<|/det|>
+Acceptance of your manuscript is conditional on all authors' agreement with our publication policies (see https://www.nature.com/nmicrobiol/editorial- policies). In particular your manuscript must not be published elsewhere and there must be no announcement of the work to any media outlet until the publication date (the day on which it is uploaded onto our website).  
+
+<|ref|>text<|/ref|><|det|>[[115, 693, 867, 784]]<|/det|>
+Please note that <i>Nature Microbiology</i> is a Transformative Journal (TJ). Authors may publish their research with us through the traditional subscription access route or make their paper immediately open access through payment of an article- processing charge (APC). Authors will not be required to make a final decision about access to their article until it has been accepted. <a href="https://www.springernature.com/gp/open- research/transformative- journals"> Find out more about Transformative Journals</a>
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[597, 44, 970, 88]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[113, 131, 881, 298]]<|/det|>
+Authors may need to take specific actions to achieve <a href="https://www.springernature.com/gp/open- research/funding/policy- compliance- faqs"> compliance</a> with funder and institutional open access mandates. If your research is supported by a funder that requires immediate open access (e.g. according to <a href="https://www.springernature.com/gp/open- research/plan- s- compliance">Plan S principles</a>) then you should select the gold OA route, and we will direct you to the compliant route where possible. For authors selecting the subscription publication route, the journal's standard licensing terms will need to be accepted, including <a href="https://www.nature.com/nature- portfolio/editorial- policies/self- archiving- and- license- to- publish">self- archiving policies</a>. Those licensing terms will supersede any other terms that the author or any third party may assert apply to any version of the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 326, 875, 356]]<|/det|>
+If you have any questions about our publishing options, costs, Open Access requirements, or our legal forms, please contact ASJournals@springernature.com  
+
+<|ref|>text<|/ref|><|det|>[[115, 370, 595, 385]]<|/det|>
+An online order form for reprints of your paper is available at <a  
+
+<|ref|>text<|/ref|><|det|>[[115, 386, 477, 400]]<|/det|>
+href="https://www.nature.com/reprints/author-  
+
+<|ref|>text<|/ref|><|det|>[[115, 400, 864, 446]]<|/det|>
+reprints.html">https://www.nature.com/reprints/author- reprints.html</a>. All co- authors, authors' institutions and authors' funding agencies can order reprints using the form appropriate to their geographical region.  
+
+<|ref|>text<|/ref|><|det|>[[115, 460, 876, 580]]<|/det|>
+We welcome the submission of potential cover material (including a short caption of around 40 words) related to your manuscript; suggestions should be sent to Nature Microbiology as electronic files (the image should be 300 dpi at \(210 \times 297 \text{mm}\) in either TIFF or JPEG format). Please note that such pictures should be selected more for their aesthetic appeal than for their scientific content, and that colour images work better than black and white or grayscale images. Please do not try to design a cover with the Nature Microbiology logo etc., and please do not submit composites of images related to your work. I am sure you will understand that we cannot make any promise as to whether any of your suggestions might be selected for the cover of the journal.  
+
+<|ref|>text<|/ref|><|det|>[[115, 594, 872, 639]]<|/det|>
+You can now use a single sign- on for all your accounts, view the status of all your manuscript submissions and reviews, access usage statistics for your published articles and download a record of your refereeing activity for the Nature journals.  
+
+<|ref|>text<|/ref|><|det|>[[115, 653, 877, 714]]<|/det|>
+To assist our authors in disseminating their research to the broader community, our Shared initiative provides you with a unique shareable link that will allow anyone (with or without a subscription) to read the published article. Recipients of the link with a subscription will also be able to download and print the PDF.  
+
+<|ref|>text<|/ref|><|det|>[[115, 742, 843, 758]]<|/det|>
+As soon as your article is published, you will receive an automated email with your shareable link.
+
+<--- Page Split --->

peer_reviews/fa0373ef1958c81223e0042c2e3a7e6c6ff19e4fe4da86bed30195de098881e7/supplementary_0_peer review file/images_list.json

@@ -0,0 +1,305 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1 has been updated in the revised manuscript as shown below.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 0
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Supplementary Fig. S13 | Discontinuity of \\(\\mathrm{NiO_2}\\) planes within the grains. a and b The red and yellow dotted boxes correspond to domains with the orientations of [100] and [001], respectively.",
+    "footnote": [],
+    "bbox": [
+      [
+        149,
+        95,
+        848,
+        354
+      ]
+    ],
+    "page_idx": 11
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Figure 4 has been updated in the revised manuscript as shown below.",
+    "footnote": [],
+    "bbox": [
+      [
+        148,
+        350,
+        850,
+        765
+      ]
+    ],
+    "page_idx": 11
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_1.jpg",
+    "caption": "Supplementary Fig. S10 | iDPC images simulations of the infinite layer structure. a Simulated iDPC images with different thicknesses and defocus values based on the atomic model along [100] orientation. The red box shows the conditions with better imaging contrast. b To better show the details, some of the images are zoomed in. The atom model of the infinite layer structure is displayed on the right-hand side for comparison.",
+    "footnote": [],
+    "bbox": [
+      [
+        145,
+        81,
+        848,
+        545
+      ]
+    ],
+    "page_idx": 15
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_2.jpg",
+    "caption": "Supplementary Fig. S13 | Discontinuity of \\(\\mathrm{NiO_2}\\) planes within the grains. a and b The red and yellow dotted boxes correspond to domains with the orientations of [100] and [001], respectively.",
+    "footnote": [],
+    "bbox": [
+      [
+        148,
+        536,
+        848,
+        800
+      ]
+    ],
+    "page_idx": 16
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_3.jpg",
+    "caption": "Supplementary Fig. S7 | The EELS data of \\(\\mathrm{Nd_{0.8}Sr_{0.2}NiO_2}\\) in this manuscript compared with those of Goodge, B. H [Proc. Natl Acad. Sci. USA 118, e2007683118, 2021. ]. The weak pre-peak of O \\(K\\) edge corresponds to the underdoped level of Sr, as indicated by the red arrows. The shift of",
+    "footnote": [],
+    "bbox": [
+      [
+        149,
+        686,
+        848,
+        840
+      ]
+    ],
+    "page_idx": 19
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_4.jpg",
+    "caption": "Supplementary Fig. S4 | Crystalline orientation measured by Transmission Kikuchi Diffraction (TKD). a The BF image of \\(\\mathrm{Nd_{0.8}Sr_{0.2}NiO_2}\\) including multiple grains. b The measured crystalline orientation map corresponding to the image in a.",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        233,
+        848,
+        455
+      ]
+    ],
+    "page_idx": 21
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_1e.jpg",
+    "caption": "Supplementary Fig. S1 | a and b Enlarged views of selected area electron diffraction (SAED) in Fig. 1e (left) and 1f (right), respectively.",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        124,
+        838,
+        375
+      ]
+    ],
+    "page_idx": 25
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_5.jpg",
+    "caption": "Supplementary Fig. S11 | Structure characterization of bulk \\(\\mathrm{Nd_{0.8}Sr_{0.2}NiO_3}\\) before topotactic reduction. a EDS mapping of \\(\\mathrm{Nd_{0.8}Sr_{0.2}NiO_3\\) polycrystalline at low magnification. b and d High-resolution BF-STEM image of \\(\\mathrm{Nd_{0.8}Sr_{0.2}NiO_3}\\) . c and e The magnified HAADF-STEM images of marked regions in b and d.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 26
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Fig. 3 | EDS and EELS measurements of \\(\\mathrm{Nd_{0.8}Sr_{0.2}NiO_2}\\) . a EDS elemental mappings of the stripe structure in \\(\\mathrm{Nd_{0.8}Sr_{0.2}NiO_2}.\\) b The intensity profile of elemental distributions in a. c EELS O K, Ni \\(L\\) , and Nd \\(M\\) of infinite-layer phase (black) and T'-type phase (red).",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 28
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_1e.jpg",
+    "caption": "Supplementary Fig. S1 | a and b Enlarged views of selected area electron diffraction (SAED) in Fig. 1e (left) and 1f (right), respectively.",
+    "footnote": [],
+    "bbox": [
+      [
+        148,
+        569,
+        844,
+        821
+      ]
+    ],
+    "page_idx": 29
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Fig. 2 | Atomic structure of the stripe domains in \\(\\mathrm{Nd_{0.8}Sr_{0.2}NiO_2}\\) . a Atom model of layered T'-type \\((\\mathrm{Nd_{0.8}Sr_{0.2})_4Ni_3O_8}\\) (n = 3). b Atomic-resolved iDPC-STEM image of the T'-type phase along [100] direction with the light elements in infinite and fluorite layers clearly resolved. The atomic model is overlaid to guide the eyes. c Atomic-resolved HAADF-STEM image of the stripes under [100] orientation, and the atomic models without oxygen are overlaid accordingly. To differentiate between T'-type phase and infinite layer structures, the latter are simply defined here as n > 10 layers. d The intensity profile across adjacent Nd atomic columns is extracted along the direction indicated by the green arrow in c. The orange arrows are used to mark the fluorite layers. e Averaged bond lengths between \\(R\\) -site atoms along [001] direction. f Strain \\(\\epsilon_{yy}\\) along the [001] calculated by GPA for c.",
+    "footnote": [],
+    "bbox": [
+      [
+        152,
+        85,
+        850,
+        395
+      ]
+    ],
+    "page_idx": 29
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4a.jpg",
+    "caption": "Supplementary Fig. S8 | Lattice parameter analysis for the stripe structure. a Atomic resolution HAADF-STEM images of the parallel and vertical stripes in Fig. 4a. b Averaged bond lengths of the parallel and vertical stripes measured from the red dotted box in a.",
+    "footnote": [],
+    "bbox": [
+      [
+        247,
+        92,
+        747,
+        519
+      ]
+    ],
+    "page_idx": 31
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_6.jpg",
+    "caption": "Supplementary Fig. S14 | TEM observations of samples by grinding. To rule out structural transitions or defects induced by FIB processing, we chose an alternative approach by grinding the sample, dispersing it into alcohol, and subsequently transferring it to the TEM grid. The BF images show the stripe structure and defects within \\(\\mathrm{Nd_{0.8}Sr_{0.2}NiO_2}\\) grains.",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        83,
+        844,
+        320
+      ]
+    ],
+    "page_idx": 32
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1 has been updated in the revised manuscript as shown below.",
+    "footnote": [],
+    "bbox": [
+      [
+        148,
+        384,
+        850,
+        654
+      ]
+    ],
+    "page_idx": 32
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1 and Supplementary Fig. S12 have been updated in the revised manuscript as shown below.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 33
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_7.jpg",
+    "caption": "Supplementary Fig. S12 | Structure characterization of bulk \\(\\mathrm{Nd_{0.8}Sr_{0.2}NiO_3}\\) before topotactic reduction. a EDS mapping of \\(\\mathrm{Nd_{0.8}Sr_{0.2}NiO_3\\) polycrystalline at low magnification. b Temperature dependence of resistivity for bulk \\(\\mathrm{Nd_{0.8}Sr_{0.2}NiO_3}\\) in the range of 2-300 K. c and e High-resolution BF-STEM image of \\(\\mathrm{Nd_{0.8}Sr_{0.2}NiO_3}\\) . d and f The magnified HAADF-STEM images of marked regions in c and e.",
+    "footnote": [],
+    "bbox": [
+      [
+        147,
+        85,
+        847,
+        320
+      ]
+    ],
+    "page_idx": 34
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3 has been updated in the revised manuscript as shown below.",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        646,
+        850,
+        878
+      ]
+    ],
+    "page_idx": 37
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Fig. 5 | Atomic iDPC-STEM imaging of infinite-layer phase. a Structure transition during the topotactic reduction. b iDPC-STEM image of infinite-layer phase at [100] orientation. c, d and e Enlarged iDPC-STEM images extracted from the orange dotted line box in b, where the orange arrow marks the residual oxygen atom in the Nd/Sr atomic plane. f Atom model of the distorted \\(\\mathrm{NiO_2}\\) layers projected in [100] direction. The bond angle and the deviation of Ni atom are marked as well. g The ratio of lattice parameter \\(c / a\\) according to b. h Statistical data of lattice constants \\(a\\) and \\(c\\) . i The calculated bond angle contour maps are plotted over the iDPC-STEM image. j The histogram displays the statistical data for bond angles measured across multiple regions, with a fitted value of \\(177.0^{\\circ}\\pm 5.3^{\\circ}\\) .",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 39
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Fig. 5 | Atomic iDPC-STEM imaging of infinite-layer phase. a Structure transition during the topotactic reduction. b iDPC-STEM image of infinite-layer phase at [100] orientation. c, d and e Enlarged iDPC-STEM images extracted from the orange dotted line box in b, where the orange arrow marks the residual oxygen atom in the Nd/Sr atomic plane. f Atom model of the distorted \\(\\mathrm{NiO_2}\\) layers projected in [100] direction. The bond angle and the deviation of Ni atom are marked as well. g The ratio of lattice parameter \\(c / a\\) according to b. h Statistical data of lattice constants \\(a\\) and \\(c\\) . i The calculated bond angle contour maps are plotted over the iDPC-STEM image. j The histogram displays the statistical data for bond angles measured across multiple regions, with a fitted value of \\(177.0^{\\circ} \\pm 5.3^{\\circ} \\circ\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        147,
+        85,
+        845,
+        395
+      ]
+    ],
+    "page_idx": 40
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_8.jpg",
+    "caption": "Supplementary Fig. S11 has been added to the revised manuscript as shown below.",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        610,
+        840,
+        800
+      ]
+    ],
+    "page_idx": 43
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Fig. 1 | Polycrystalline \\(\\mathrm{Nd_{0.8}Sr_{0.2}NiO_2}\\) imaged by HAADF-STEM. a Atomic structure model with direction vectors of perovskite \\(R\\mathrm{NiO}_3\\) and infinite layer \\(R\\mathrm{NiO}_2\\) (R: Nd/Sr). b EDS mappings of \\(\\mathrm{Nd_{0.8}Sr_{0.2}NiO_2}\\) at low magnification. c Temperature dependence of resistivity for bulk \\(\\mathrm{Nd_{0.8}Sr_{0.2}NiO_2}\\mathrm{in}\\) the range of 2-300 K. d High-resolution HAADF-STEM image of \\(\\mathrm{Nd_{0.8}Sr_{0.2}NiO_2}\\) e and f The magnified images of marked regions in d, with the corresponding SAED pattern in the insets.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 50
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_9.jpg",
+    "caption": "Supplementary Fig. S16 has been added to the revised manuscript as shown below.",
+    "footnote": [],
+    "bbox": [
+      [
+        160,
+        255,
+        835,
+        430
+      ]
+    ],
+    "page_idx": 52
+  }
+]

peer_reviews/fe509d0325912985ab92d1087d749f0c9f6964b0218025a9f98843ee48609c9e/supplementary_0_Peer review file./images_list.json

@@ -0,0 +1,56 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Figure 5. Overlap (Jaccard index) in the 1,000 highest ranked A) when using the same Method with different Resources (Blue) and B) when using the same Resource with different Methods (Red). Overlap is represented as the pairwise. The dashed lines represent the median when using different resources (red) and methods (blue); the lines overlap for the CMCBs dataset.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 0
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_6.jpg",
+    "caption": "Figure 6. Odds ratios of A) Active cytokines and B) colocalized cell types among the highest ranked interaction predictions, across a ranked range between 100 and 10,000. Odds ratios representing the association of preferentially ranked CCC predictions and A) cytokine activities and B) spatial adjacencies were calculated using Fisher's exact test. Consensus represents the aggregated ranks of all interactions predicted by all the methods. Dashed line is the baseline represented by an odds ratio of 1. The vertical lines represent the truncated ranges of CellChat, CellPhoneDB, and LogFC Mean, arising from their relatively stricter preprocessing steps.",
+    "footnote": [],
+    "bbox": [
+      [
+        120,
+        88,
+        880,
+        675
+      ]
+    ],
+    "page_idx": 10
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2. Dependencies and overlap between CCC resources. The lineages of CCC interaction database knowledge. General biological knowledge databases (blue), CCC-dedicated resources used in this work (magenta), manual literature curation effort (yellow), additional resources included in iTALK (cyan), and OmniPath (green). Arrows show the data transfers between resources. \\(\\bullet\\) indicates the manually-curation of resources, defined by explicitly mentioning that these resources are 'manually' or 'expert' curated. \\(\\boxed{ \\begin{array}{r l} \\end{array} }\\) indicates that the resource was included in the analyses presented here.",
+    "footnote": [],
+    "bbox": [
+      [
+        117,
+        288,
+        877,
+        593
+      ]
+    ],
+    "page_idx": 13
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Figure 4. CCC resources distributions in terms of number of interactions (A) and relative abundance (B) matched to the SignaLink database. C) Interactions categorised by CancerSEA cancer cell states, and D) Human Protein Atlas organ markers. Differentially represented (log2(Odds ratio) \\(>1\\) ) categories were marked according to FDR-corrected \\(p\\) -values \\(< 0.05\\) \\((\\ast)\\) , 0.01 \\((\\ast)\\) , and 0.001 \\((\\ast)\\) .",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 18
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Response Figure 3. Schematic of spatial analysis limitation example.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 20
+  }
+]

peer_reviews/supplementary_0_Transparent Peer Review file__e8722be8aaa17d8618f2767801861104661e43cf67eb08e0172c8561a3bc771c/supplementary_0_Transparent Peer Review file__e8722be8aaa17d8618f2767801861104661e43cf67eb08e0172c8561a3bc771c.mmd

@@ -0,0 +1,618 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+# Effect of discontinuous fair-share emissions allocations immediately based on equity  
+
+Corresponding Author: Dr Yann Robiou du Pont  
+
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+Version 0:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+This paper presents a welcome methodological development to the analysis of NDC based on fairness/equity principles. The avoidance of using transition periods from current/baseline levels to 'fair' allocations has considerable implications for the distribution fair mitigation and finance efforts. I find Figure 4 particularly illuminating.  
+
+The presentation of the analysis and the language needs considerable improvement. This drawback also makes it difficult to fully assess the underlying method, although it appears to be sound. I would recommend to have the whole text copy- edited, in addition to addressing the specific issues listed below.  
+
+I would also like to see some more discussion of the merits and drawbacks of the paper's methodological contribution.  
+
+Most important points  
+
+1. The two fairness approaches are difficult to understand. The section in the main text should be made more accessible and related clearly to fairness principles in the literature. The differences between the two approaches should be highlighted. Why are the two approaches opposite wrt positive vs negative emissions?  
+
+143: [approach 1] "allocates global positive emissions to equalize historical responsibilities". It is confusing that historical responsibility enters into approach 1, as I thought approach 1 was based on capability and approach 2 on responsibility. Also, historical responsibility over what time period?  
+
+147: "the capability considerations affects the use the emissions budgets over time, but not its total" Multiple typos here, and I do not understand the sentence. Affects the distribution over time, but not the total budget, maybe?  
+
+489: "This approach yields significant differences in emissions allocations across countries" Differences relative to what? Variation across countries might be a better term.  
+
+491- 492: I do not understand this sentence. "Important" seems not the correct word here.  
+
+2. I would suggest to spend some more words on the criticism past studies have received for including transition periods (e.g., Kartha et al (2018)) and the weak ethical basis for such periods (e.g., Flerbaev et al (2014)). On the other hand, possible drawbacks of removing transition periods could also be discussed. For example, this means that reduced and avoided emissions are treated symmetrically, but the costs of reducing emissions are likely larger than the costs of avoiding future increases. The argument in the paper and in the literature against transition periods relies on emissions trading (ITMOs), but the imperfections of current institutions for this should be mentioned. Perhaps also refer to Knight (2013) for a defense of moderate grandfathering.  
+
+Sentences that are difficult to understand and should be reformulated:  
+
+17: "increasingly do so in the future" I could not understand what was meant until reading the introduction. This is an important point, but it needs to be elaborated more to be understandable in the abstract.
+
+<--- Page Split --->
+
+45: "most of the recent approaches rely on allocations of emissions rights following a continuous trajectory starting at current emissions levels, sometimes using a transition period" Why the word sometimes? Do not all continuous trajectories starting at current levels use a transition period by necessity? Or are there exceptions?. See also line 69.  
+
+131: "The relevance of equity concepts and their implementations in effort-sharing formulae show various consistency with international law"  
+
+167: "Approach 1 is mainly driven by responsibility in the near term" Change to (if I understand correctly: In the near term, a1 is driven mainly by responsibility  
+
+189: "The absence of zero or negative allocations for some countries results from fairness indicators as well as the absence of negative emissions in the \(1.5^{\circ}C\) scenarios- set with strong near- term mitigation, excluding LULUCF emissions."  
+
+300: "The egalitarian approach..." Why is this introduced here? Difficult to follow.  
+
+343: "with few arbitrary" parameters. As few as possible? Or a few?  
+
+359: "Even equity- based budgets could theoretically be used mostly in the near- term by countries and not collectively reflect any of the global \(1.5^{\circ}C\) mitigation scenarios underpinning the global budget."  
+
+364: "Additionally, emissions budgets are not suitable for addressing the knowledge gaps identified in the IPCC AR6 of "extending equity frameworks to quantify equitable international support, as the difference between equity- based national emissions scenarios and national domestic emissions scenarios"  
+
+457: "that may result from better governance or potentially ill acquired wealth"  
+
+524: "reflect alignment with symbolic warming thresholds"  
+
+527: "can be considered dragging even the insufficient ambition current policies that do not track towards NDCs"  
+
+Typos:  
+
+159: 'nations' should be notions? 246 "delay climate action and near 2030" 412: "that country could"  
+
+Minor points:  
+
+251: "Based on Approach 2, the assessment of the NDCs of the UK, Sweden and Switzerland is more stringent" Not apparent in the figure. Perhaps because they are nevertheless given the same assessment category?  
+
+Figure 2: For which countries is approach 2 less stringent than approach 1? I cannot see any country with a greener color under this approach, but perhaps the African countries are overdelivering 1.5degrees ambition even more under approach 2?  
+
+387: "Compared to a previous warming assessment5 (visible on Paris- Equity- Check.org), Approach 1 finds NDCs to be more ambitious ( \(1.5^{\circ}C\) aligned) for a few countries (including India, Indonesia and Egypt) and less for Norway." Are these these the only changes? Using what equity principle?  
+
+Request from the CVF: This section does not clearly distinguish between the request itself and the authors' interpretation of it. The whole section is in quotation marks, but refers to "the CF requested" and "this paper", which causes confusion over who formulated the text. I suggest replicating or summarizing the request first, then presenting the interpretation.  
+
+## References  
+
+Fleurbaey, M., Kartha, S., Bolwig, S., Chee, Y.L., Chen, Y., Corbera, E., et al., 2014. Sustainable development and equity. In: Edenhofer, O., Pichs- Madruga, R., Sokona, Y., Farahani, E., Kadner, S., Seyboth, K. (Eds.), Climate Change 2014: Mitigation of Climate Change. Contribution of Working Group III to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, United Kingdom and New York, NY, pp. 283- 350.  
+
+Kartha, S., Athanasiou, T., Caney, S., Cripps, E., Dooley, K., Dubash, N.K., Harris, P., Holz, C., Lahn, B., Moellendorf, D., Müller, B., Roberts, J.T., Sagar, A.D., Shue, H., Singer, P., Winkler, H., 2018. Cascading biases against poorer parties. Nat. Clim. Change 8 (5), 348- 349.  
+
+Knight, C., 2013. What is grandfathering? Env. Polit. 22 (3), 410- 427.  
+
+Reviewer #2  
+
+(Remarks to the Author)
+
+<--- Page Split --->
+
+Robiou du Pont and co- authors note that most existing equitable mitigation assessments start from the most recent year when emissions are available for countries (termed "continuous allocations" by the authors). The authors indicate that this choice induces a "grandfathering" effect, which unintentionally rewards countries that have not reduced emissions with relatively less stringent emission reduction benchmarks in the near term. To address this, the authors propose an approach to derive equitable allocations for countries, which the authors suggest departs from previous literature in two ways: (1) the allocations start in a historical year (e.g., 1990), and hence capture historical (in- )action, and (2) the authors treat gross emission reductions and gross emission removals separately.  
+
+I have some comments and suggestions that I think are important to consider. I have focussed my review comments on the substantive content of the paper. However, I encourage the authors to try to reduce repetition between the "Introduction" and "Approach Rationale" sections, and to check for and correct typos and other errors in the text.  
+
+## Review comments  
+
+Accounting for past (in- )action: I agree with the authors that updating an equitable mitigation assessment with updated historical emissions, all else equal, may result in an inadvertent benefit to emitters that are not reducing emissions (the argument the authors start to present in L44- L48). However, there are two things I think the authors should consider addressing:  
+
+1. The concept of "carbon debt": I think the same issue has been identified previously in the literature, where it has sometimes been termed as "carbon debt" or, emissions above a counterfactual equitable pathway (Gignac & Matthews, 2015; van den Berg et al., 2020). This approach does not fit neatly within the "continuous" versus "discontinuous" dichotomy that the authors have introduced.  
+
+2. The potential scale of the "grandfathering effect": The authors suggest, in L83-L85, that "Such iterative updates of ambition assessments based on continuous emissions allocations would iteratively find an insufficient NDC closer and closer to a calculated fair allocation". As I noted in the introduction to this section, I would tend to agree with this, all else equal. However, since pathways to a given temperature target (e.g., 1.5°C) will become progressively steeper, the responsibility of major emitters (which could be one equity allocation consideration) will increase. Given these two additional effects, I think there is more ambiguity in the effect on an equity assessment of NDCs. I suggest that the authors present some illustrative calculations to help the reader understand the validity of this statement.  
+
+Justification for the equity approaches: I appreciate that the authors provide a first estimation of the equitable mitigation targets that apply to both gross emission reductions as well as removals. As the authors correctly note, this is a gap in the existing literature, and addressing it is important to guide policy discussions. However, I have a few questions and concerns that I hope the authors can address:  
+
+1. I think the underlying justification for an equity approach is just as important as the numerical quantifications. Keeping this in mind, I found the justification for the application of different equity principles to gross reductions and removals to be one of the less comprehensive parts of the manuscript. I didn't understand why it is appropriate to factor in historical responsibility for one quantity (e.g., gross reductions) while factoring in capability for the other (e.g., gross removals). I think the manuscript could benefit from a more comprehensive discussion of the reasoning behind these equity approaches.  
+
+2. I found the description of the methods (L140 - L152 in the main text, and L436-L509 in the methods section) quite difficult to follow. I encourage the authors to publish the code used to carry out the analysis to allow for replication and to consider writing out the equations for each step so that the reader can follow the specific implementation.  
+
+Clarity on the use of scenarios: I have several comments on the use of global mitigation scenarios that I think the authors should address:  
+
+1. In L520-L522, the authors indicate that "The reference to a 1.5°C alignment corresponds to an alignment with the distribution of emissions of the average of scenarios of the IPCC Categories C1 [...], itself averaged with the distribution of C2 [...]. How do the authors come up with a distribution of emissions if they average across scenarios belonging to these categories? I'm not sure it is appropriate to use averages of such a scenario ensemble (see, e.g., (Guivarch et al., 2022)), and would recommend that the authors avoid this.  
+
+2. I think the authors should justify why they group the C1 and C2 categories of pathways together but do not do this for any of the other categories of pathways. Mapping the textual elements of the Paris Agreement Long Term Temperature Goal (LTTG) to specific pathway characteristics is a non-trivial value judgment (see, e.g., the discussions in (Kikstra et al., 2022; Schleussner et al., 2022)). I suggest that the authors improve the discussion on their pathway categorisation choices, especially since they explicitly indicate that this work is meant to guide the Global Stocktake.  
+
+3. Figure 2: The authors deviate from the labels presented in the methods section, by labeling C1+C2 pathways as "Below 1.5 degrees", and C3 pathways as "Well below 2 degrees". Please align the labels across the sections, and reflect on my comment above.  
+
+4. The 43% reduction by 2030 assessment: In L207-L209, the authors suggest that "The IPCC indicates that, on average across a set of scenarios, a 43% reduction in global GHG emissions by 2030 (here taken below 2020 levels) would align with a 1.5°C trajectory with no or limited overshoot. This global target [...]. There are a couple of conceptual challenges here, that I think the authors should consider addressing. The first, is that this is, by no means an IPCC-endorsed "global target" – it is only a description of the median (not average) of the scenarios assessed by the IPCC in that category of pathways. This value is also computed relative to 2019 emission levels, and given the structural differences between 2019 and 2020 emissions, I think it is further not appropriate to apply the 43% reduction below 2020 emission levels. Further, excluding LULUCF emissions (which are included in the original values presented in the IPCC report), means that this estimate is no longer appropriate. I suggest that the authors consider revising the text describing this approach, and use the uncertainty band presented in the AR6 report for this category while describing the results presented in Figure 3.
+
+<--- Page Split --->
+
+Additional analyses need to be motivated better: The two additional analyses presented in the results section (the addition of a "20- year transition phase", and the presentation of equal per capita results) are not motivated sufficiently in the text. I was unsure why the authors chose to present these results and suggest the authors add more text before the results section to justify why they have presented them.  
+
+## References  
+
+Gignac, R., & Matthews, H. D. (2015). Allocating a 2 °C cumulative carbon budget to countries. Environmental Research Letters, 10(7), 075004. https://doi.org/10.1088/1748- 9326/10/7/075004Guivarch, C., Le Gallic, T., Bauer, N., Fragkos, P., Huppmann, D., Jaxa- Rozen, M., Keppo, I., Kriegler, E., Krisztin, T., Marangoni, G., Pye, S., Riahi, K., Schaeffer, R., Tavoni, M., Trutnevye, E., van Vuuren, D., & Wagner, F. (2022). Using large ensembles of climate change mitigation scenarios for robust insights. Nature Climate Change, 12(5), Article 5. https://doi.org/10.1038/s41558- 022- 01349- xKikstra, J. S., Nicholls, Z. R., Smith, C. J., Lewis, J., Lamboll, R. D., Byers, E., Sandstad, M., Meinshausen, M., Gidden, M. J., Rogelj, J., & others. (2022). The IPCC Sixth Assessment Report WGIII climate assessment of mitigation pathways: From emissions to global temperatures. Geoscientific Model Development, 15(24), 9075- 9109. Schleussner, C.- F., Ganti, G., Rogelj, J., & Gidden, M. J. (2022). An emission pathway classification reflecting the Paris Agreement climate objectives. Communications Earth & Environment, 3(1), https://doi.org/10.1038/s34247- 022- 00467- wvan den Berg, N. J., van Soest, H. L., Hof, A. F., den Elzen, M. G. J., van Vuuren, D. P., Chen, W., Drouet, L., Emmerling, J., Fujimori, S., Hohne, N., Koberle, A. C., McCollum, D., Schaeffer, R., Shekhar, S., Vishwanathan, S. S., Vrontisi, Z., & Blok, K. (2020). Implications of various effort- sharing approaches for national carbon budgets and emission pathways. Climatic Change, 162(4), 1805- 1822. https://doi.org/10.1007/s10584- 019- 02368- y  
+
+## Version 1:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+In the previous round, I noted that the presentation of the analysis and the language needed considerable improvement, and also recommended to have the whole text copy- edited. I regret to say that the improvement on these issues has not been satisfactory. It does not appear as the text has been copy- edited. It should not be the job of the reviewers to point out simple mistakes and help improve the language. While the authors have corrected the specific mistakes I pointed out, there are still many language problems, including in the newly introduced text. Some appear as sloppy mistakes, such as incomplete sentences, while a more extensive problem is lack of clear and structured presentation. I believe the underlying model development would be a valuable addition to the literature, but I deem the progress on its presentation from the first version as insufficient to warrant another 'revise and resubmit'.  
+
+Some concrete issues:  
+
+Re. my first point that the two fairness approaches are difficult to understand, which was also brought up by R2: The motivation for including two different approaches is still not clear to me. The new text is quite technical. Do the approaches reflect different ethical assumptions, or different approaches to a more technical modeling choice for which there is no clear criterion for choosing one over the other?  
+
+The figures and captions appear in a mess. The same figures appear on multiple pages, sometimes with and sometimes without captions.  
+
+The discussion has no structure, which makes it difficult to follow. There is no conclusion.  
+
+The methods section appears to contain considerable overlap with the main text (partly reflecting that the main text is very technical).  
+
+Re. my suggestion to "spend some more words on the criticism past studies have received for including transition periods (e.g., Kartha et al (2018)) and the weak ethical basis for such periods (e.g., Flerbaev et al (2014)). On the other hand, possible drawbacks of removing transition periods could also be discussed. For example, this means that reduced and avoided emissions are treated symmetrically, but the costs of reducing emissions are likely larger than the costs of avoiding future increases. The argument in the paper and in the literature against transition periods relies on emissions trading (ITMOs), but the imperfections of current institutions for this should be mentioned. Perhaps also refer to Knight (2013) for a defense of moderate grandfathering."  
+
+The authors responded 'Thank you for the reference. We added it in the following sentence as follows with references to Fleurbaey and Knight:  
+
+"Considering continuous emissions trajectories that look realistic22 implies that present- day levels of domestic emissions are an acceptable starting point in terms of mitigation effort with a utilitarian perspective25." I would have liked to see a more engagement with the suggestion than just adding one sentence.
+
+<--- Page Split --->
+
+(Remarks on code availability)  
+
+Reviewer #2  
+
+(Remarks to the Author)Thank you for the opportunity to review the revised manuscript.  
+
+I have reviewed the revisions made by the authors in response to my previous round of review comments. I am satisfied with these revisions.  
+
+(Remarks on code availability)  
+
+Version 2:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+(Remarks on code availability)  
+
+Open Access This Peer Review 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 cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+The images or other third party material in this Peer Review 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 https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+This paper presents a welcome methodological development to the analysis of NDC based on fairness/equity principles. The avoidance of using transition periods from current/baseline levels to 'fair' allocations has considerable implications for the distribution fair mitigation and finance efforts. I find Figure 4 particularly illuminating.  
+
+The presentation of the analysis and the language needs considerable improvement. This drawback also makes it difficult to fully assess the underlying method, although it appears to be sound. I would recommend to have the whole text copy- edited, in addition to addressing the specific issues listed below.  
+
+I would also like to see some more discussion of the merits and drawbacks of the paper's methodological contribution.  
+
+Most important points  
+
+1. The two fairness approaches are difficult to understand. The section in the main text should be made more accessible and related clearly to fairness principles in the literature. The differences between the two approaches should be highlighted.  
+
+Why are the two approaches opposite wrt positive vs negative emissions?  
+
+Thank you. We extended the discussion to address these shortcomings:  
+
+"Here, we suggest two extensions, one for each approach of Fyson et al. to derive two allocations of economy- wide emissions to countries. Each new approach combines concepts of capability and responsibility, where each concept is applied to global positive or negative emissions distinctively. This study offers two conceptual combinations of the responsibility and capability concepts referred to in the Paris Agreement's CBDR- RC, with a differentiated treatment of negative emissions, which require costly and uncertain technologies, and are made necessary because of insufficient global emissions reductions to date. These approaches enable the assessment of the ambition of countries' NDC, in light of their responsibility and capability, with special considerations for negative emissions often used to enable and justify potentially dangerous warming overshoot. A first extension, named Approach 1, first allocates global negative emissions across countries based on their capability, assessed through GDP or Human Development Index (HDI, in Supplementary Information), and then allocates global positive emissions to equalize historical responsibilities over the net emissions (positive + negative, see Methods). Under this approach, rich countries are required to fund most of the negative emissions that require important research and development costs with high uncertainty and without local co- benefits31. Approach 1 also ensures equal cumulative per capita emissions by 2100 through the allocation of the positive emissions space. Here, the capability allocation affects the distribution of emissions over time, but not the total budget. Richer countries then have more important negative allocations in the future, and less stringent allocations in the near term (see Methods). The second extension, Approach 2, conversely first allocates global positive emissions based on countries' capabilities and then global negative emissions based on their responsibilities. There, all countries contribute to emissions reductions based on their wealth. Negative emissions, needed because of the world's important historical
+
+<--- Page Split --->
+
+emissions, are allocated proportionally to countries' individual historical responsibilities. In Approach 2, historical responsibility does not define countries' cumulative emissions alone. Looking at the global emissions scenarios, the positive emissions refer here to the projected physical emissions (e.g., fossil fuels, agriculture). The negative emissions here refer to emissions captured through Carbon Dioxide Removal (excluding those from LULUCF, unlike Fyson et al. 2020) and Direct Air Capture.  
+
+143: [approach 1] "allocates global positive emissions to equalize historical responsibilities". It is confusing that historical responsibility enters into approach 1, as I thought approach 1 was based on capability and approach 2 on responsibility. Also, historical responsibility over what time period?  
+
+Indeed, the text was misleading. We added a sentence explaining the rationale of the paper that is to suggest two extensions of the Fyson et al. paper, each of which combines responsibility (since 1990 and 1950) and capability. Extending each of the two approaches of Fyson et al. For clarity, we amended the previous paragraph and added the sentence line 143:  
+
+"Each new approach combines concepts of capability and responsibility, where each concept is applied to global positive or negative emissions distinctively."  
+
+And in the following paragraph on the parameterization:  
+
+"Here, we present results based on a parameterization that uses GDP for capability and accounts for responsibility through emissions since 1990 (with 1950 in the Supplementary Information)."  
+
+147: "the capability considerations affects the use the emissions budgets over time, but not its total" Multiple typos here, and I do not understand the sentence. Affects the distribution over time, but not the total budget, maybe?  
+
+Thanks for pointing this out and the concrete suggestion. The sentence was modified according to the suggestion:  
+
+"Approach 1 also ensures equal cumulative per capita emissions by 2100 through the allocation of the positive emissions space. Here, the capability allocation affects the distribution of emissions over time, but not the total budget. Richer countries then have more important negative allocations in the future, and less stringent allocations in the near term (see Methods)."  
+
+489: "This approach yields significant differences in emissions allocations across countries" Differences relative to what? Variation across countries might be a better term.  
+
+Thank you. I had not realized the difference in meanings of these two terms. The text was modified according to this suggestion.  
+
+491- 492: I do not understand this sentence. "Important" seems not the correct word here.
+
+<--- Page Split --->
+
+Thank you. Would this phrasing be clearer?  
+
+"Thank you. Would this phrasing be clearer? "This approach yields significant variations in emissions allocations across countries, which reflects the large differences across countries' GDPs (often proportionally greater than the differences of their historical contributions)."  
+
+2. I would suggest to spend some more words on the criticism past studies have received for including transition periods (e.g., Kartha et al (2018)) and the weak ethical basis for such periods (e.g., Flerbaev et al (2014)). On the other hand, possible drawbacks of removing transition periods could also be discussed. For example, this means that reduced and avoided emissions are treated symmetrically, but the costs of reducing emissions are likely larger than the costs of avoiding future increases. The argument in the paper and in the literature against transition periods relies on emissions trading (ITMOs), but the imperfections of current institutions for this should be mentioned. Perhaps also refer to Knight (2013) for a defense of moderate grandfathering.  
+
+Thank you for the reference. We added it in the following sentence as follows with references to Fleurbaev and Knight:  
+
+"Considering continuous emissions trajectories that look realistic22 implies that present- day levels of domestic emissions are an acceptable starting point in terms of mitigation effort with a utilitarian perspective25."  
+
+Sentences that are difficult to understand and should be reformulated:  
+
+17: "increasingly do so in the future" I could not understand what was meant until reading the introduction. This is an important point, but it needs to be elaborated more to be understandable in the abstract.  
+
+Certainly. We reformulate to:  
+
+"Ambition assessments based on trajectories that start at present- day emissions levels inherently reward past inaction, and increasingly do so with their iterative updates."  
+
+45: "most of the recent approaches rely on allocations of emissions rights following a continuous trajectory starting at current emissions levels, sometimes using a transition period" Why the word sometimes? Do not all continuous trajectories starting at current levels use a transition period by necessity? Or are there exceptions?. See also line 69.  
+
+Thanks for raising this. Indeed, the continuity of emissions allocations does not require a transition period. The formula can be designed in other ways. To clarify this point, we suggest changing the text shortly after (originally on line 73):  
+
+"Effort- sharing formulas can be designed to directly achieve such continuity2,3 or an ad- hoc transition period can be added to ensure continuity3,14,16 between current emissions and future allocations only based on fairness considerations."  
+
+This is also further explained in the discussion section:
+
+<--- Page Split --->
+
+"In other models, the continuity of the emissions trajectory results from the choice of allocating mitigation burden to depart from a reference trajectory rather than allocating the remaining emissions space2. Such an approach based on reference trajectories can be adequate to assess the ambition of an emissions target when provided with a corresponding reference scenario, e.g. pledges taken in 2015 against allocations starting in 2015 (ref. 2)."  
+
+131: "The relevance of equity concepts and their implementations in effort- sharing formulae show various consistency with international law"  
+
+Thank you. We suggested this formulation:  
+
+"Some of the equity concepts quantified in the literature are not backed by principles of international law that require excluding approaches based on grandfathering6."  
+
+167: "Approach 1 is mainly driven by responsibility in the near term" Change to (if I understand correctly: In the near term, a1 is driven mainly by responsibility  
+
+Thank you for the concrete suggestion. We decided to delete this sentence.  
+
+189: "The absence of zero or negative allocations for some countries results from fairness indicators as well as the absence of negative emissions in the \(1.5^{\circ}\mathrm{C}\) scenarios- set with strong near- term mitigation, excluding LULUCF emissions."  
+
+We suggest reformulating to two sentences:  
+
+"Looking at when allocations reach net- zero, some countries with relatively low responsibility and capability have emissions allocations that are positive throughout the century under a \(1.5^{\circ}\mathrm{C}\) objective. It is important to note that some of the selected global \(1.5^{\circ}\mathrm{C}\) scenarios with strong near- term mitigation also have positive emissions throughout the century since we excluded the LULUCF sector."  
+
+300: "The egalitarian approach..." Why is this introduced here? Difficult to follow.  
+
+Thanks for highlighting this. We added an introductory sentence and modified the previous one to:  
+
+"In addition to capability and responsibility, equality is the third equity principle described in the IPCC AR517,22. IPCC reports do not present equity- based emissions allocations since AR5, despite available studies and its importance for courts of law12. The egalitarian approach modelled as equal per capita emissions is not directly anchored in the Paris Agreement or international environmental law6, but can reveal the inequalities of emissions spaces claimed through NDCs."
+
+<--- Page Split --->
+
+343: "with few arbitrary" parameters. As few as possible? Or a few?  
+
+We changed wording to: "as few parameters as possible"  
+
+359: "Even equity- based budgets could theoretically be used mostly in the near- term by countries and not collectively reflect any of the global \(1.5^{\circ}\mathrm{C}\) mitigation scenarios underpinning the global budget."  
+
+We suggest simplifying the sentence to:  
+
+"Theoretically, countries could choose to use a budget mostly in the near- term to justify insufficient emissions objectives, which raises issues of intergenerational justice."  
+
+364: "Additionally, emissions budgets are not suitable for addressing the knowledge gaps identified in the IPCC AR6 of "extending equity frameworks to quantify equitable international support, as the difference between equity- based national emissions scenarios and national domestic emissions scenarios""  
+
+In response, we modified the previous few sentences for clarity and amended these two sentences to:  
+
+"Theoretically, countries could choose to use a budget mostly in the near- term to justify insufficient emissions objectives, which raises issues of intergenerational justice. The "flexibility" provided by carbon budgets over emissions pathways14 comes at the expense of the ability to assess the ambition of time- defined objectives, without additional assumptions18. Time- defined emissions allocations are also needed to address the knowledge gaps identified in the IPCC AR6 of "extending equity frameworks to quantify equitable international support, as the difference between equity- based national emissions scenarios and national domestic emissions scenarios"52."  
+
+457: "that may result from better governance or potentially ill acquired wealth"  
+
+We suggest reformulating to:  
+
+"Comparing two countries with equal populations and equal GDPs, the country with higher HDI will have greater effort to provide when using HDI as the capacity indicator rather than GDP. Using HDI as a capacity indicator may then penalize good governance, compared to using GDP. Results based on HDI are available in the supplementary information."
+
+<--- Page Split --->
+
+524: "reflect alignment with symbolic warming thresholds"  
+
+We simplified the phrasing to:  
+
+"Additionally, the effort- sharing formulas are applied to the scenario categories C6 ('below \(3^{\circ}\mathrm{C}\) ) and C7 (below \(4^{\circ}\mathrm{C}\) ) that reflect current policies."  
+
+527: "can be considered dragging even the insufficient ambition current policies that do not track towards NDCs"  
+
+We simplified to:  
+
+"Countries with NDCs that do not align with their fair allocation of C7 scenarios can be considered to be dragging global decarbonization efforts."  
+
+Typos:  
+
+159: 'nations' should be notions?  
+
+Yes. This typo is corrected.  
+
+246 "delay climate action and near 2030"  
+
+We clarified the sentence to:  
+
+"This effect increases as we delay climate action and as we near 2030."  
+
+412: "that country could"  
+
+Indeed, corrected to "countries".  
+
+Minor points:  
+
+251: "Based on Approach 2, the assessment of the NDCs of the UK, Sweden and Switzerland is more stringent" Not apparent in the figure. Perhaps because they are nevertheless given the same assessment category?  
+
+Thank you. We have now corrected to:  
+
+"Based on Approach 2, the assessment of the NDCs of the UK and Switzerland is more stringent given their relatively low historical responsibility compared to their relatively high capability."  
+
+Figure 2: For which countries is approach 2 less stringent than approach 1? I cannot see any country with a greener color under this approach, but perhaps the African countries are overdelivering 1.5degrees ambition even more under approach 2?
+
+<--- Page Split --->
+
+Indeed, Figure 3 (formerly figure 2) provides a rating of NDC ambition. The full allocations are provided in the supplementary data that will be made available publicly for all countries (the version initially submitted is available here https://zenodo.org/record/8003393). We added a new sentence after the two sentences that discussed the relative stringencies of approaches 1 & 2:  
+
+"While Approach 1 constrains countries' cumulative emissions based on their responsibility, which often overlaps with high capability38, Approach 2 is more stringent in the near term for countries with high GDP. Approach 2 is less stringent for countries with very low capability (sub- Saharan African countries), and for countries with high historical responsibility (USA, Russia, Qatar and other fossil fuel extracting countries, Figure 2 and Supplementary Information). These different stringencies of emissions allocations do not always change the warming assessment of countries' NDCs (see country- level allocations in SI)."  
+
+387: Compared to a previous warming assessment5 (visible on Paris- Equity- Check.org), Approach 1 finds NDCs to be more ambitious (1.5°C aligned) for a few countries (including India, Indonesia and Egypt) and less for Norway." Are these these the only changes? Using what equity principle?  
+
+This previous assessment (which I am also author of), had different methods combining different equity principles. Here, we choose to report the most prominent differences. We suggest clarifying that this previous assessment relied on three equity principles:  
+
+"Compared to a previous warming assessment5 (visible on Paris- Equity- Check.org) where each country follows the least stringent of three equity principles (capability, responsibility, and equality), the present approaches find NDCs to be more ambitious (1.5°C aligned) for a few countries (including India, Indonesia, and Egypt depending on the approach) and less for countries in the Global North and Latin America."  
+
+Request from the CVF: This section does not clearly distinguish between the request itself and the authors' interpretation of it. The whole section is in quotation marks, but refers to "the CF requested" and "this paper", which causes confusion over who formulated the text. I suggest replicating or summarizing the request first, then presenting the interpretation.  
+
+Thank you. We seek to provide transparency with this section. Your input is helpful in that regard. We changed the titles of the two sections as follows:  
+
+Request as explained by the CVF:  
+
+Interpretation and discussion of CVF's request in light of the available literature:  
+
+## References  
+
+Fleurbaey, M., Kartha, S., Bolwig, S., Chee, Y.L., Chen, Y., Corbera, E., et al., 2014. Sustainable development and equity. In: Edenhofer, O., Pichs- Madruga, R., Sokona, Y., Farahani, E., Kadner, S., Seyboth, K. (Eds.), Climate Change 2014: Mitigation of Climate Change. Contribution of Working Group III to the Fifth Assessment Report of the
+
+<--- Page Split --->
+
+Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, United Kingdom and New York, NY, pp. 283- 350.  
+
+Kartha, S., Athanasiou, T., Caney, S., Cripps, E., Dooley, K., Dubash, N.K., Harris, P., Holz, C., Lahn, B., Moellendorf, D., Müller, B., Roberts, J.T., Sagar, A.D., Shue, H., Singer, P., Winkler, H., 2018. Cascading biases against poorer parties. Nat. Clim. Change 8 (5), 348- 349.  
+
+Knight, C., 2013. What is grandfathering? Env. Polit. 22 (3), 410- 427.  
+
+Reviewer #2 (Remarks to the Author):  
+
+Robiou du Pont and co- authors note that most existing equitable mitigation assessments start from the most recent year when emissions are available for countries (termed "continuous allocations" by the authors). The authors indicate that this choice induces a "grandfathering" effect, which unintentionally rewards countries that have not reduced emissions with relatively less stringent emission reduction benchmarks in the near term. To address this, the authors propose an approach to derive equitable allocations for countries, which the authors suggest departs from previous literature in two ways: (1) the allocations start in a historical year (e.g., 1990), and hence capture historical (in- )action, and (2) the authors treat gross emission reductions and gross emission removals separately.  
+
+I have some comments and suggestions that I think are important to consider. I have focussed my review comments on the substantive content of the paper  
+
+Review comments  
+
+Accounting for past (in- )action: I agree with the authors that updating an equitable mitigation assessment with updated historical emissions, all else equal, may result in an inadvertent benefit to emitters that are not reducing emissions (the argument the authors start to present in L44- L48). However, there are two things I think the authors should consider addressing: 1. The concept of "carbon debt": I think the same issue has been identified previously in the literature, where it has sometimes been termed as "carbon debt" or, emissions above a counterfactual equitable pathway (Gignac & Matthews, 2015; van den Berg et al., 2020). This approach does not fit neatly within the "continuous" versus "discontinuous" dichotomy that the authors have introduced.  
+
+I understand the carbon debt as related to the accounting of historical emissions. The disparities in historical emissions across countries can be seen as a carbon debt from the high historical emitters, including to the lower emitting countries. Previous studies, including (Gignac & Matthews, 2015; van den Berg et al., 2020), suggested that future emissions allocations can account for this debt. The debt could be compensated over the period for which the future emissions budget is allocated, or possibly through adaptation and loss- and- damage finance. In a recent submission, Pelz et al. (http://dx.doi.org/10.21203/rs.3.rs- 4394688/v1, under review) highlight the importance of this debt to "') responsibility for overshoot, ii) exceedance drawdown obligations, and iii) increase in extreme climate exposure if drawdown does not occur." While our approach allocates immediate effort to
+
+<--- Page Split --->
+
+compensate inequities, the "carbon debt" tracks overshoot but does not suggest how this debt should be compensated over time.  
+
+Our present study and its use of discontinuous dynamic emissions allocations does not differ in that aspect. The carbon debt is accounted for and influences future emissions allocations, even resulting in equal cumulative per capita emissions in Approach 1. In this case, the 'discontinuity' feature only affects how the emissions budget is used over time.  
+
+We clarify this point in the text as:  
+
+"Other approaches use the concept of 'carbon debt'55,56 to assess countries' responsibility for overshooting their fair shares of a global carbon budget, including through their future objectives, and compensate through future negative emissions57. This budget- based approach can be used to characterize breaches by courts and be complementary to dynamic approaches, as in this study, immediately allocating feasible socio- economic scenarios, which can inform Paris- aligned emissions targets."  
+
+2. The potential scale of the "grandfathering effect": The authors suggest, in L83-L85, that "Such iterative updates of ambition assessments based on continuous emissions allocations would iteratively find an insufficient NDC closer and closer to a calculated fair allocation". As I noted in the introduction to this section, I would tend to agree with this, all else equal. However, since pathways to a given temperature target (e.g., \(1.5^{\circ}\mathrm{C}\) ) will become progressively steeper, the responsibility of major emitters (which could be one equity allocation consideration) will increase. Given these two additional effects, I think there is more ambiguity in the effect on an equity assessment of NDCs. I suggest that the authors present some illustrative calculations to help the reader understand the validity of this statement.  
+
+Thank you. We updated Figure 1 to use actual data instead of the schematic representation in the first submission. We modelled allocation starting in 2015 and 2020. As you guessed, the slope is steeper in the later allocation, but the grandfathering effect remains visible. Thank you for the suggestion.  
+
+Justification for the equity approaches: I appreciate that the authors provide a first estimation of the equitable mitigation targets that apply to both gross emission reductions as well as removals. As the authors correctly note, this is a gap in the existing literature, and addressing it is important to guide policy discussions. However, I have a few questions and concerns that I hope the authors can address:  
+
+1. I think the underlying justification for an equity approach is just as important as the numerical quantifications. Keeping this in mind, I found the justification for the application of different equity principles to gross reductions and removals to be one of the less comprehensive parts of the manuscript. I didn't understand why it is appropriate to factor in historical responsibility for one quantity (e.g., gross reductions) while factoring in capability for the other (e.g., gross removals). I think the manuscript could benefit from a more comprehensive discussion of the reasoning behind these equity approaches.  
+
+Thank you, the lack of clarity of this section was also raised by reviewer 1. We agree with the importance of the justification updated the description as follows:
+
+<--- Page Split --->
+
+"Here, we suggest two extensions, one for each approach of Fyson et al. to derive two allocations of economy- wide emissions to countries. Each new approach combines concepts of capability and responsibility, where each concept is applied to global positive or negative emissions distinctively. This study offers two conceptual combinations of the responsibility and capability concepts referred to in the Paris Agreement's CBDR- RC, with a differentiated treatment of negative emissions, which require costly and uncertain technologies, and are made necessary because of insufficient global emissions reductions to date. These approaches enable the assessment of the ambition of countries' NDC, in light of their responsibility and capability, with special considerations for negative emissions often used to enable and justify potentially dangerous warming overshoot. A first extension, named Approach 1, first allocates global negative emissions across countries based on their capability, assessed through GDP or Human Development Index (HDI, in Supplementary Information), and then allocates global positive emissions to equalize historical responsibilities over the net emissions (positive + negative, see Methods). Under this approach, rich countries are required to fund most of the negative emissions that require important research and development costs with high uncertainty and without local co- benefits31. Approach 1 also ensures equal cumulative per capita emissions by 2100 through the allocation of the positive emissions space. Here, the capability allocation affects the distribution of emissions over time, but not the total budget. Richer countries then have more important negative allocations in the future, and less stringent allocations in the near term (see Methods). The second extension, Approach 2, conversely first allocates global positive emissions based on countries' capabilities and then global negative emissions based on their responsibilities. There, all countries contribute to emissions reductions based on their wealth. Negative emissions, needed because of the world's important historical emissions, are allocated proportionally to countries' individual historical responsibilities. In Approach 2, historical responsibility does not define countries' cumulative emissions alone. Looking at the global emissions scenarios, the positive emissions refer here to the projected physical emissions (e.g., fossil fuels, agriculture). The negative emissions here refer to emissions captured through Carbon Dioxide Removal (excluding those from LULUCF, unlike Fyson et al. 2020) and Direct Air Capture."  
+
+2. I found the description of the methods (L140 – L152 in the main text, and L436-L509 in the methods section) quite difficult to follow. I encourage the authors to publish the code used to carry out the analysis to allow for replication and to consider writing out the equations for each step so that the reader can follow the specific implementation.  
+
+Thank you. We will make the code publicly available with the manuscript on a Github. It is shared with this version of the manuscript through Code Ocean and at: https://github.com/imagepbl/effort-sharing. We also added a description of the formulas used in the Methods.  
+
+Clarity on the use of scenarios: I have several comments on the use of global mitigation scenarios that I think the authors should address:  
+
+1. In L520-L522, the authors indicate that "The reference to a \(1.5^{\circ}\mathrm{C}\) alignment corresponds to an alignment with the distribution of emissions of the average of scenarios of the IPCC
+
+<--- Page Split --->
+
+Categories C1 [...], itself averaged with the distribution of C2 [...]. How do the authors come up with a distribution of emissions if they average across scenarios belonging to these categories? I'm not sure it is appropriate to use averages of such a scenario ensemble (see, e.g., (Guivarch et al., 2022)), and would recommend that the authors avoid this.  
+
+Thank you for raising this. Guivarch et al. 2022 (co- authored by one of the authors of the present manuscript) highlight: "Although scenario ensembles are designed to explore the possibility space, neither type of ensemble can be interpreted as a perfect statistical sample. Given the unknown unknowns, the scenarios' outcomes cannot be interpreted in terms of likelihoods, and even large scenario ensembles do not fully or equally explore the space of possibilities".  
+
+We agree that averaging across the ensemble of scenarios in C- categories does not provide a representative view of the literature as some modeling choices may be over/under- represented. However, our analysis uses global scenarios only for their global emissions bound by common physical considerations. The socio- economic assumptions of these scenarios have limited effect on the total emissions profile, mostly driven by considerations regarding total negative emissions. This approach is consistent with previous studies, including:  
+
+Robiou du Pont, Y., Jeffery, M., Gutschew, J. et al. Equitable mitigation to achieve the Paris Agreement goals. Nature Clim Change 7, 38- 43 (2017).  
+
+Xunzhang Pan, Michel den Elzen, Niklas Hohne, Fei Teng, Lining Wang, Exploring fair and ambitious mitigation contributions under the Paris Agreement goals, Environmental Science & Policy, Volume 74, 2017, Pages 49- 56, ISSN 1462- 9011, https://doi.org/10.1016/j.envsci.2017.04.020.  
+
+Note that part of the author team is working on another article, under review, suggesting custom- made global emissions scenario representatives of various warming categories, that enables the user to check the influence of modelling parameters separately:  
+
+Mark Dekker, Andries Hof, Yann Robiou du Pont et al. Navigating the black box of fair national emissions targets, 19 September 2024, PREPRINT (Version 1) available at Research Square [https://doi.org/10.21203/rs.3.rs- 5023350/v1]  
+
+2. I think the authors should justify why they group the C1 and C2 categories of pathways together but do not do this for any of the other categories of pathways. Mapping the textual elements of the Paris Agreement Long Term Temperature Goal (LTTG) to specific pathway characteristics is a non-trivial value judgment (see, e.g., the discussions in (Kikstra et al., 2022; Schleussner et al., 2022)). I suggest that the authors improve the discussion on their pathway categorisation choices, especially since they explicitly indicate that this work is meant to guide the Global Stocktake.  
+
+Thank you. We understand the critics of Schleussner et al. regarding the potential misinterpretation of scenario categories, their absence of overlaps and their representativeness of the Paris Agreement goal. We now provide the results for C1 and C2 separately and amended the text in the following manner:  
+
+"The reference to a \(1.5^{\circ}\mathrm{C}\) alignment corresponds here to an alignment with the
+
+<--- Page Split --->
+
+distribution of emissions of the average of scenarios of the IPCC Categories C1 ('below \(1.5^{\circ}\mathrm{C}\) with no or limited overshoot'). The distribution of C2 conveys a warming 'below \(1.5^{\circ}\mathrm{C}\) with high overshoot'. The upper threshold of \(2^{\circ}\mathrm{C}\) alignment here follows the definition based on emissions scenarios C3 ('likely below \(2^{\circ}\mathrm{C}\)') category. The consistency of such low emissions scenarios with the Paris Agreement temperature goal is discussed based on warming responses and levels of negative emissions62,63."  
+
+3. Figure 2: The authors deviate from the labels presented in the methods section, by labeling C1+C2 pathways as "Below 1.5 degrees", and C3 pathways as "Well below 2 degrees". Please align the labels across the sections, and reflect on my comment above.  
+
+Thank you. We have corrected the labelling.  
+
+4. The \(43\%\) reduction by 2030 assessment: In L207-L209, the authors suggest that "The IPCC indicates that, on average across a set of scenarios, a \(43\%\) reduction in global GHG emissions by 2030 (here taken below 2020 levels) would align with a \(1.5^{\circ}\mathrm{C}\) trajectory with no or limited overshoot. This global target [...]". There are a couple of conceptual challenges here, that I think the authors should consider addressing. The first, is that this is, by no means an IPCC-endorsed "global target" – it is only a description of the median (not average) of the scenarios assessed by the IPCC in that category of pathways. This value is also computed relative to 2019 emission levels, and given the structural differences between 2019 and 2020 emissions, I think it is further not appropriate to apply the \(43\%\) reduction below 2020 emission levels. Further, excluding LULUCF emissions (which are included in the original values presented in the IPCC report), means that this estimate is no longer appropriate. I suggest that the authors consider revising the text describing this approach, and use the uncertainty band presented in the AR6 report for this category while describing the results presented in Figure 3.  
+
+Thank you. We updated the data to show a \(50\%\) reduction below 2020 levels and no longer refer to the IPCC figure to simply illustrate national allocation at a midway point of global decarbonization. Figure 2 (formerly Figure 3) shows this new parameterization. Thank you for the suggestion that avoids misinterpretation.  
+
+Additional analyses need to be motivated better: The two additional analyses presented in the results section (the addition of a "20- year transition phase", and the presentation of equal per capita results) are not motivated sufficiently in the text. I was unsure why the authors chose to present these results and suggest the authors add more text before the results section to justify why they have presented them.  
+
+Thank you. We added the following explanations.  
+
+First we show equal per capita allocations as a point of comparison, revealing some lack of ambition even in the absence of equity considerations.  
+
+"In addition to capability and responsibility, equality is the third equity principle described in the IPCC AR517,22. IPCC reports do not present equity- based emissions allocations since AR5, despite available studies and its importance for courts of
+
+<--- Page Split --->
+
+law12. The egalitarian approach modelled as equal per capita emissions is not directly anchored in the Paris Agreement or international environmental law6, but can reveal the inequalities of emissions spaces claimed through NDCs."  
+
+Secondly, we explain how modelling discontinuous emissions allocations changes not simply countries' fair share, but also the ranking of countries in terms of the relative share of global mitigation finance that they could be expected to contribute to. It is not only important in terms of which country is doing enough or not, but also to determine the distribution of expected financial effort across countries. We modify the introductory text of Figure 4's results to:  
+
+"We show that modelling continuous emissions allocations, here exemplified by adding a 20- year transition period, affects countries' emissions gaps between their allocations and NDCs, unequally in 2030. Compared to a traditional continuous approach, applying a discontinuous approach implies here a much higher obligation to contribute to international finance for all G20 countries except from India. In terms of the ranking of the emissions gap between NDC and allocation, assuming a transition period benefits Canada and Australia (moving down 9 positions), the USA and South Korea (each 8 positions). This shows that continuous pathways 'reward' such countries for their history of comparably low mitigation efforts, lowering their implied contribution to international climate finance. Other countries, including China, Türkiye, South Africa and the EU move down in the ranking of the ambition gaps, when removing the transition period."  
+
+## References  
+
+Gignac, R., & Matthews, H. D. (2015). Allocating a \(2^{\circ}\mathrm{C}\) cumulative carbon budget to countries. Environmental Research Letters, 10(7), 075004. https://doi.org/10.1088/1748- 9326/10/7/075004Guivarch, C., Le Gallic, T., Bauer, N., Fragkos, P., Huppmann, D., Jaxa- Rozen, M., Keppo, I., Kriegler, E., Krisztin, T., Marangoni, G., Pye, S., Riahi, K., Schaeffer, R., Tavoni, M., Trutnevytse, E., van Vuuren, D., & Wagner, F. (2022). Using large ensembles of climate change mitigation scenarios for robust insights. Nature Climate Change, 12(5), Article 5. https://doi.org/10.1038/s41558- 022- 01349-x  
+
+Kikstra, J. S., Nicholls, Z. R., Smith, C. J., Lewis, J., Lamboll, R. D., Byers, E., Sandstad, M., Meinshausen, M., Gidden, M. J., Rogelj, J., & others. (2022). The IPCC Sixth Assessment Report WGIII climate assessment of mitigation pathways: From emissions to global temperatures. Geoscientific Model Development, 15(24), 9075- 9109.  
+
+Schleussner, C.- F., Ganti, G., Rogelj, J., & Gidden, M. J. (2022). An emission pathway classification reflecting the Paris Agreement climate objectives. Communications Earth & Environment, 3(1), https://doi.org/10.1038/s43247- 022- 00467- w  
+
+van den Berg, N. J., van Soest, H. L., Hof, A. F., den Elzen, M. G. J., van Vuuren, D. P., Chen, W., Drouet, L., Emmerling, J., Fujimori, S., Hohne, N., Koberle, A. C., McCollum, D., Schaeffer, R., Shekhar, S., Vishwanathan, S. S., Vrontisi, Z., & Blok, K. (2020). Implications of various effort- sharing approaches for national carbon budgets and emission pathways.
+
+<--- Page Split --->
+
+Climatic Change, 162(4), 1805–1822. https://doi.org/10.1007/s10584-019-02368-y
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+In the previous round, I noted that the presentation of the analysis and the language needed considerable improvement, and also recommended to have the whole text copy- edited. I regret to say that the improvement on these issues has not been satisfactory. It does not appear as the text has been copy- edited. It should not be the job of the reviewers to point out simple mistakes and help improve the language. While the authors have corrected the specific mistakes I pointed out, there are still many language problems, including in the newly introduced text. Some appear as sloppy mistakes, such as incomplete sentences, while a more extensive problem is lack of clear and structured presentation. I believe the underlying model development would be a valuable addition to the literature, but I deem the progress on its presentation from the first version as insufficient to warrant another 'revise and resubmit'.  
+
+We appreciate that improvements to the text were necessary and are thankful that you took the time to go over it one more time. We also apologise the display issues due to the conversion of the word document into a pdf file (in particular figure captions).  
+
+We now copy- edited the entire manuscript, reorganised sections for clarity, and removed redundant statements.  
+
+Some concrete issues:  
+
+Re. my first point that the two fairness approaches are difficult to understand, which was also brought up by R2: The motivation for including two different approaches is still not clear to me. The new text is quite technical. Do the approaches reflect different ethical assumptions, or different approaches to a more technical modeling choice for which there is no clear criterion for choosing one over the other?  
+
+Thank you. There are indeed several motivations behind these modelling choices.  
+
+1) combining capability and responsibility without relying on averages or statistical combinations,  
+
+2) differentiating the allocation of positive and negative emissions, to complement the study of Fyson et al. that focused on the allocation of global negative emissions only (with either responsibility or capability)  
+
+Given these choices, we chose simple allocation methods to represent the capability and the responsibility principles. We model the two manners to combine the responsibility and capability allocations of positive and negative emissions separately. We explain the differences (e.g., Approach 1 achieves equal cumulative per capita emissions) but do not suggest that one should be used over the other.  
+
+We have updated the manuscript to clarify the rationale of the methodology:  
+
+"Here we quantify two sets of emissions trajectories immediately based on equity principles and that do not start at current emissions levels (see Methods). The two
+
+<--- Page Split --->
+
+methods combine the equity principles of capability and responsibility<sup>1</sup> to reflect the principles of the UNFCCC and the Paris Agreement, notably CBDR- RC. The literature suggests several approaches, conceptual or statistical, for combining different equity principles into a single allocation method (see Discussion). Here we apply each of the two equity principles to allocate global positive or negative emissions separately. This differentiated treatment of negative emissions extends a study from Fyson et al.<sup>2</sup> that allocated negative emissions only, based on responsibility or capability. Fyson et al. explain that obligations to deliver negative emissions require uncertain technologies made necessary because of insufficient global emissions reductions to date. That study alone could not be used to inform economy- wide emissions targets, and thus not assess the ambition of NDCs, as it only allocated negative emissions and “assume[d] that positive emissions follow least- cost pathways (that is, no equity principle is applied to gross emissions)”<sup>2</sup>.  
+
+The figures and captions appear in a mess. The same figures appear on multiple pages, sometimes with and sometimes without captions.  
+
+Indeed, our apologies for that. I believe that the figure referencing system of word is not well handled by the pdf conversion tool on the journal’s platform.  
+
+The discussion has no structure, which makes it difficult to follow. There is no conclusion.  
+
+We have thoroughly revised the discussion that now focusses on 1) how continuity assumptions are present in the literature, 2) how our combination of equity principles compares to the literature, 3) how our allocation results compare to the literature. Note that Nature Communication’s format does not allow for a conclusion section. The last paragraph was revised and clearly introduced as a conclusion.  
+
+We introduce the discussion section with:  
+
+“Here we discuss how this study’s modelling choices compare to the literature regarding the continuity assumption and the combination of equity principles. Then, we compare results.”  
+
+The methods section appears to contain considerable overlap with the main text (partly reflecting that the main text is very technical).  
+
+Thank you for the thorough review. We removed content from the manuscript that already appeared in the methods. Additionally, we moved some content to the methods.  
+
+Re. my suggestion to “spend some more words on the criticism past studies have received for including transition periods (e.g., Kartha et al (2018)) and the weak ethical basis for such periods (e.g., Flerbaej et al (2014)). On the other hand, possible drawbacks of removing transition periods could also be discussed. For example, this means that reduced and avoided emissions are treated symmetrically, but the costs of reducing emissions are likely larger than the costs of avoiding future increases. The argument in the paper and in the literature against transition periods relies on emissions trading (ITMOs), but the imperfections of current institutions for this should be mentioned. Perhaps also refer to Knight (2013) for a defense of
+
+<--- Page Split --->
+
+moderate grandfathering."  
+
+The authors responded 'Thank you for the reference. We added it in the following sentence as follows with references to Fleurbaey and Knight:  
+
+"Considering continuous emissions trajectories that look realistic22 implies that present- day levels of domestic emissions are an acceptable starting point in terms of mitigation effort with a utilitarian perspective25."  
+
+I would have liked to see a more engagement with the suggestion than just adding one sentence.  
+
+Thank you for standing for your point.  
+
+In the fair- share literature, the allocation of emissions space does not distinguish treat reduced and avoided emissions similarly, regardless of whether a transition period is used or not. A transition period does not account for reduction potential, as opposed to IAM scenarios for example. For this reason, we do not find it possible to relate the inclusion of a transition period to the distinction between reduced and avoided emissions. However, the allocation of emissions space into fair- shares may result in 'hot air' as the emissions allocation of a country may exceed the emissions of its business as usual scenario. While a transition period mitigated this effect, it does not avoid hot air. Even a pure grandfathering approach could theoretically result in hot air allocations.  
+
+We raise the issue of hot air, made more visible through discontinuous allocations in the following paragraph:  
+
+"The near- term allocation of some countries, mostly sub- Saharan countries, may exceed their current emissions and business- as- usual trajectory beyond 2030, implying mitigation efforts only later<sup>3</sup>. However, staying within such decreasing allocations beyond 2030 implies immediate investments, possibly with international support. International support can enable recipient countries to implement mitigation measures in line with the underlying global socio- economic scenario in the near term. Approach 2 uses allocations inversely proportional to GDP per capita<sup>4,5</sup> (see methods), resulting in high emissions allocations compared to current emissions and allocations based on business- as- usual trajectories<sup>3,6</sup> for countries with very low GDP per capita (e.g., Ethiopia, Democratic Republic of Congo). These allocations theoretically imply financial transfers that may go beyond needs- based considerations and contribute to poverty reduction through climate action<sup>7</sup>.  
+
+Regarding the arguments raised by Knight for the justification of a moderate grandfathering, we discuss the realist justification and the utilitarian justification, with references to Knight and Fleurbaey. Indeed, all international agreements reflect some inertia that give a realist justification to grandfathering. However, we focus on the modelling of equity- based allocations to inform processes not limited to negotiating agreements, and that include climate litigation. We seek to rely directly on concepts of international law and the Paris Agreement that do not support grandfathering. We hope that the following modifications clarify this point:  
+
+"The legacy influence of current emissions levels on near- term emissions allocations is described here as a 'grandfathering' effect<sup>8</sup>. This grandfathering influence on equity- based emissions allocation is strongest in the near term and increasingly affects
+
+<--- Page Split --->
+
+the ambition assessment of NDCs in 2030. As we near 2030, a given NDC's emissions target will be closer and closer to a continuous emissions allocation that is iteratively updated (Figure 1). The grandfathering allocation is criticized for its lack of ethical basis \(^{9 - 11}\) and has been shown to penalize the poorest countries \(^{12}\) as it preserves a status- quo, including current inequalities. Prior to the Paris Agreement, a study highlighted the value of a 'moderate grandfathering' \(^{13}\) , from a political theory perspective, with a realist justification for negotiations and a utilitarian justification. Indeed, the pledges of many high- emitters only align with a grandfathering allocation \(^{4}\) . However, the IPCC has highlighted the need for a fair distribution of mitigation efforts, excluding grandfathering, in order to achieve an effective global agreement on emissions reductions \(^{1,10}\) . Likewise, recent reports of scientific advisory bodies have disapproved grandfathering when presenting fair- share emissions allocation \(^{14,15}\) . The Paris Agreement now requires NDCs of the 'highest possible ambition' that reflect equity. A recent study \(^{8}\) described grandfathering allocations as not in line with international law. It identified that all continuous allocations entail elements of grandfathering but did not offer a solution.  
+
+Regarding the justification of utilitarianism, we added the following sentence in the next paragraph:  
+
+"The utilitarian justification \(^{13}\) for a moderate grandfathering relies domestic mitigation costs and is no longer relevant when allocations can be traded to achieve a globally cost- effective pathway \(^{10}\) ."  
+
+This interpretation is based on p. 417 of Knight's analysis \(^{13}\) stating: "On welfare views such as utilitarianism, the relevant costs are welfare costs, rather than the monetary marginal abatement costs familiar from economics. According to utilitarianism, welfare costs are more important the greater they are. It might be claimed that high emitters face high marginal abatement costs. The marginal abatement cost is the welfare cost of one extra unit of emissions reduction. Thus, the main claim of the marginal cost argument is: one extra unit of emission reductions from a baseline of actual prior emissions decreases welfare to a greater extent where it is assigned to a high emitter than where it is assigned to a low emitter. If this is correct, utilitarianism will maintain that high emitters have greater entitlements, as this will save them – and the global economy – from the severe effects of deeper cuts (cf. Wesley and Peterson 1999, p. 186)."  
+
+We also highlight the possible important shortcoming of Article 6:  
+
+"As a novel mechanism, the international trading of mitigation outcomes raises implementation issues regarding the additionality of the finance and of the funded mitigation measures. Scrutiny will be needed to ensure the integrity of mitigation measures under Article 6 whose implementation rules were just adopted at COP29, with safeguards on human rights and the additionality of emissions reductions \(^{16 - 18}\) ."  
+
+Reviewer #2 (Remarks to the Author):  
+
+Thank you for the opportunity to review the revised manuscript.
+
+<--- Page Split --->
+
+I have reviewed the revisions made by the authors in response to my previous round of review comments. I am satisfied with these revisions.  
+
+Thank you for your time and reviews.  
+
+References:  
+
+1. Clarke, L. et al. Chapter 6 Assessing Transformation Pathways. In: IPCC AR5 WGIII. 413–510 (2014).  
+2. Fyson, C. L., Baur, S., Gidden, M. & Schleussner, C. F. Fair-share carbon dioxide removal increases major emitter responsibility. Nature Climate Change 10, 836–841 (2020).  
+3. van den Berg, N. J. et al. Implications of various effort-sharing approaches for national carbon budgets and emission pathways. Climatic Change 162, 1805–1822 (2020).  
+4. Robiou du Pont, Y. et al. Equitable mitigation to achieve the Paris Agreement goals. Nature Climate Change 7, 38–43 (2017).  
+5. Jacoby, H. D., Babiker, M. H., Paltsev, S. & Reilly, J. M. Sharing the Burden of GHG Reductions. MIT Joint Program on the Science and Policy of Global Change 1–28 https://globalchange.mit.edu/publication/14428 (2008).  
+6. Holz, C., Kartha, S. & Athanasiou, T. Fairly sharing 1.5: national fair shares of a 1.5 °C-compliant global mitigation effort. International Environmental Agreements: Politics, Law and Economics 18, 117–134 (2017).  
+7. Budolfson, M. B. et al. Utilitarian benchmarks for emissions and pledges promote equity, climate and development. Nature Climate Change 11, 827–833 (2021).  
+8. Rajamani, L. et al. National ‘fair shares’ in reducing greenhouse gas emissions within the principled framework of international environmental law. Climate Policy 21, 1–22 (2021).  
+9. Caney, S. Justice and the distribution of greenhouse gas emissions. Journal of Global Ethics 5, 125–146 (2009).  
+10. Fleurbaey, M. et al. Chapter 4. Sustainable Development and Equity. In: Climate Change 2014: Mitigation of Climate Change. Contribution of Working Group III to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change 283–350 (2014).  
+11. Dooley, K. et al. Ethical choices behind quantifications of fair contributions under the Paris Agreement. Nature Climate Change 11, (2021).  
+12. Kartha, S. et al. Cascading biases against poorer countries. Nature Climate Change 8, 348–349 (2018).  
+13. Knight, C. What is grandfathering? Environmental Politics (2013) doi:10.1080/09644016.2012.740937.
+
+<--- Page Split --->
+
+14. European Scientific Advisory Board on Climate Change. Scientific Advice for the Determination of an EU-Wide 2040 Climate Target and a Greenhouse Gas Budget for 2030-2050. https://doi.org/10.2800/609405 (2023).  
+
+15. A Justified Ceiling to Germany's CO₂ Emissions: Questions and Answers on Its CO₂ Budget. https://www.umweltrat.de/SharedDocs/Downloads/EN/04_Statements/2020_2024/2022_09_The_CO2_bud get_approach.html (2022).  
+
+16. COP29: Key outcomes agreed at the UN climate talks in Baku. Carbon Brief https://www.carbonbrief.org/cop29-key-outcomes-agreed-at-the-un-climate-talks-in-baku/#6 (2024).  
+
+17. Songwe, V., Stern, N. & Bhattacharya, A. Finance for Climate Action: Scaling up Investment for Climate and Development. (2022).  
+
+18. Haynes, R. & Benjamin, L. Ambition-raising and ambition-reducing features of the Paris Agreement. in Research Handbook on the Law of the Paris Agreement (ed. Zahar, A.) 126-142 (Edward Elgar Publishing, 2024). doi:10.4337/9781800886742.00012.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Transparent Peer Review file__e8722be8aaa17d8618f2767801861104661e43cf67eb08e0172c8561a3bc771c/supplementary_0_Transparent Peer Review file__e8722be8aaa17d8618f2767801861104661e43cf67eb08e0172c8561a3bc771c_det.mmd

@@ -0,0 +1,871 @@
+<|ref|>title<|/ref|><|det|>[[73, 53, 295, 80]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[74, 96, 296, 119]]<|/det|>
+Peer Review File  
+
+<|ref|>title<|/ref|><|det|>[[73, 161, 847, 211]]<|/det|>
+# Effect of discontinuous fair-share emissions allocations immediately based on equity  
+
+<|ref|>text<|/ref|><|det|>[[73, 224, 485, 241]]<|/det|>
+Corresponding Author: Dr Yann Robiou du Pont  
+
+<|ref|>text<|/ref|><|det|>[[70, 274, 864, 289]]<|/det|>
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+<|ref|>text<|/ref|><|det|>[[73, 327, 144, 341]]<|/det|>
+Version 0:  
+
+<|ref|>text<|/ref|><|det|>[[73, 354, 219, 368]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[73, 380, 160, 393]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[73, 405, 238, 419]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[73, 419, 911, 459]]<|/det|>
+This paper presents a welcome methodological development to the analysis of NDC based on fairness/equity principles. The avoidance of using transition periods from current/baseline levels to 'fair' allocations has considerable implications for the distribution fair mitigation and finance efforts. I find Figure 4 particularly illuminating.  
+
+<|ref|>text<|/ref|><|det|>[[72, 470, 920, 510]]<|/det|>
+The presentation of the analysis and the language needs considerable improvement. This drawback also makes it difficult to fully assess the underlying method, although it appears to be sound. I would recommend to have the whole text copy- edited, in addition to addressing the specific issues listed below.  
+
+<|ref|>text<|/ref|><|det|>[[70, 510, 879, 524]]<|/det|>
+I would also like to see some more discussion of the merits and drawbacks of the paper's methodological contribution.  
+
+<|ref|>text<|/ref|><|det|>[[73, 537, 221, 550]]<|/det|>
+Most important points  
+
+<|ref|>text<|/ref|><|det|>[[72, 550, 923, 590]]<|/det|>
+1. The two fairness approaches are difficult to understand. The section in the main text should be made more accessible and related clearly to fairness principles in the literature. The differences between the two approaches should be highlighted. Why are the two approaches opposite wrt positive vs negative emissions?  
+
+<|ref|>text<|/ref|><|det|>[[72, 601, 905, 641]]<|/det|>
+143: [approach 1] "allocates global positive emissions to equalize historical responsibilities". It is confusing that historical responsibility enters into approach 1, as I thought approach 1 was based on capability and approach 2 on responsibility. Also, historical responsibility over what time period?  
+
+<|ref|>text<|/ref|><|det|>[[70, 653, 923, 679]]<|/det|>
+147: "the capability considerations affects the use the emissions budgets over time, but not its total" Multiple typos here, and I do not understand the sentence. Affects the distribution over time, but not the total budget, maybe?  
+
+<|ref|>text<|/ref|><|det|>[[70, 692, 900, 719]]<|/det|>
+489: "This approach yields significant differences in emissions allocations across countries" Differences relative to what? Variation across countries might be a better term.  
+
+<|ref|>text<|/ref|><|det|>[[72, 731, 673, 744]]<|/det|>
+491- 492: I do not understand this sentence. "Important" seems not the correct word here.  
+
+<|ref|>text<|/ref|><|det|>[[72, 769, 914, 862]]<|/det|>
+2. I would suggest to spend some more words on the criticism past studies have received for including transition periods (e.g., Kartha et al (2018)) and the weak ethical basis for such periods (e.g., Flerbaev et al (2014)). On the other hand, possible drawbacks of removing transition periods could also be discussed. For example, this means that reduced and avoided emissions are treated symmetrically, but the costs of reducing emissions are likely larger than the costs of avoiding future increases. The argument in the paper and in the literature against transition periods relies on emissions trading (ITMOs), but the imperfections of current institutions for this should be mentioned. Perhaps also refer to Knight (2013) for a defense of moderate grandfathering.  
+
+<|ref|>text<|/ref|><|det|>[[72, 887, 550, 900]]<|/det|>
+Sentences that are difficult to understand and should be reformulated:  
+
+<|ref|>text<|/ref|><|det|>[[70, 900, 865, 926]]<|/det|>
+17: "increasingly do so in the future" I could not understand what was meant until reading the introduction. This is an important point, but it needs to be elaborated more to be understandable in the abstract.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 46, 920, 88]]<|/det|>
+45: "most of the recent approaches rely on allocations of emissions rights following a continuous trajectory starting at current emissions levels, sometimes using a transition period" Why the word sometimes? Do not all continuous trajectories starting at current levels use a transition period by necessity? Or are there exceptions?. See also line 69.  
+
+<|ref|>text<|/ref|><|det|>[[72, 99, 911, 127]]<|/det|>
+131: "The relevance of equity concepts and their implementations in effort-sharing formulae show various consistency with international law"  
+
+<|ref|>text<|/ref|><|det|>[[72, 138, 916, 166]]<|/det|>
+167: "Approach 1 is mainly driven by responsibility in the near term" Change to (if I understand correctly: In the near term, a1 is driven mainly by responsibility  
+
+<|ref|>text<|/ref|><|det|>[[72, 177, 916, 205]]<|/det|>
+189: "The absence of zero or negative allocations for some countries results from fairness indicators as well as the absence of negative emissions in the \(1.5^{\circ}C\) scenarios- set with strong near- term mitigation, excluding LULUCF emissions."  
+
+<|ref|>text<|/ref|><|det|>[[72, 216, 625, 230]]<|/det|>
+300: "The egalitarian approach..." Why is this introduced here? Difficult to follow.  
+
+<|ref|>text<|/ref|><|det|>[[72, 242, 531, 256]]<|/det|>
+343: "with few arbitrary" parameters. As few as possible? Or a few?  
+
+<|ref|>text<|/ref|><|det|>[[72, 268, 918, 296]]<|/det|>
+359: "Even equity- based budgets could theoretically be used mostly in the near- term by countries and not collectively reflect any of the global \(1.5^{\circ}C\) mitigation scenarios underpinning the global budget."  
+
+<|ref|>text<|/ref|><|det|>[[72, 307, 904, 348]]<|/det|>
+364: "Additionally, emissions budgets are not suitable for addressing the knowledge gaps identified in the IPCC AR6 of "extending equity frameworks to quantify equitable international support, as the difference between equity- based national emissions scenarios and national domestic emissions scenarios"  
+
+<|ref|>text<|/ref|><|det|>[[72, 359, 600, 373]]<|/det|>
+457: "that may result from better governance or potentially ill acquired wealth"  
+
+<|ref|>text<|/ref|><|det|>[[72, 385, 470, 399]]<|/det|>
+524: "reflect alignment with symbolic warming thresholds"  
+
+<|ref|>text<|/ref|><|det|>[[72, 411, 839, 425]]<|/det|>
+527: "can be considered dragging even the insufficient ambition current policies that do not track towards NDCs"  
+
+<|ref|>text<|/ref|><|det|>[[72, 438, 122, 450]]<|/det|>
+Typos:  
+
+<|ref|>text<|/ref|><|det|>[[72, 451, 355, 490]]<|/det|>
+159: 'nations' should be notions? 246 "delay climate action and near 2030" 412: "that country could"  
+
+<|ref|>text<|/ref|><|det|>[[72, 503, 163, 515]]<|/det|>
+Minor points:  
+
+<|ref|>text<|/ref|><|det|>[[72, 515, 866, 543]]<|/det|>
+251: "Based on Approach 2, the assessment of the NDCs of the UK, Sweden and Switzerland is more stringent" Not apparent in the figure. Perhaps because they are nevertheless given the same assessment category?  
+
+<|ref|>text<|/ref|><|det|>[[72, 554, 905, 593]]<|/det|>
+Figure 2: For which countries is approach 2 less stringent than approach 1? I cannot see any country with a greener color under this approach, but perhaps the African countries are overdelivering 1.5degrees ambition even more under approach 2?  
+
+<|ref|>text<|/ref|><|det|>[[72, 605, 901, 646]]<|/det|>
+387: "Compared to a previous warming assessment5 (visible on Paris- Equity- Check.org), Approach 1 finds NDCs to be more ambitious ( \(1.5^{\circ}C\) aligned) for a few countries (including India, Indonesia and Egypt) and less for Norway." Are these these the only changes? Using what equity principle?  
+
+<|ref|>text<|/ref|><|det|>[[72, 657, 912, 698]]<|/det|>
+Request from the CVF: This section does not clearly distinguish between the request itself and the authors' interpretation of it. The whole section is in quotation marks, but refers to "the CF requested" and "this paper", which causes confusion over who formulated the text. I suggest replicating or summarizing the request first, then presenting the interpretation.  
+
+<|ref|>sub_title<|/ref|><|det|>[[72, 724, 155, 736]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[72, 737, 920, 789]]<|/det|>
+Fleurbaey, M., Kartha, S., Bolwig, S., Chee, Y.L., Chen, Y., Corbera, E., et al., 2014. Sustainable development and equity. In: Edenhofer, O., Pichs- Madruga, R., Sokona, Y., Farahani, E., Kadner, S., Seyboth, K. (Eds.), Climate Change 2014: Mitigation of Climate Change. Contribution of Working Group III to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, United Kingdom and New York, NY, pp. 283- 350.  
+
+<|ref|>text<|/ref|><|det|>[[72, 801, 899, 840]]<|/det|>
+Kartha, S., Athanasiou, T., Caney, S., Cripps, E., Dooley, K., Dubash, N.K., Harris, P., Holz, C., Lahn, B., Moellendorf, D., Müller, B., Roberts, J.T., Sagar, A.D., Shue, H., Singer, P., Winkler, H., 2018. Cascading biases against poorer parties. Nat. Clim. Change 8 (5), 348- 349.  
+
+<|ref|>text<|/ref|><|det|>[[72, 852, 546, 866]]<|/det|>
+Knight, C., 2013. What is grandfathering? Env. Polit. 22 (3), 410- 427.  
+
+<|ref|>text<|/ref|><|det|>[[72, 905, 161, 918]]<|/det|>
+Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[72, 931, 238, 944]]<|/det|>
+(Remarks to the Author)
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 46, 914, 140]]<|/det|>
+Robiou du Pont and co- authors note that most existing equitable mitigation assessments start from the most recent year when emissions are available for countries (termed "continuous allocations" by the authors). The authors indicate that this choice induces a "grandfathering" effect, which unintentionally rewards countries that have not reduced emissions with relatively less stringent emission reduction benchmarks in the near term. To address this, the authors propose an approach to derive equitable allocations for countries, which the authors suggest departs from previous literature in two ways: (1) the allocations start in a historical year (e.g., 1990), and hence capture historical (in- )action, and (2) the authors treat gross emission reductions and gross emission removals separately.  
+
+<|ref|>text<|/ref|><|det|>[[72, 151, 918, 192]]<|/det|>
+I have some comments and suggestions that I think are important to consider. I have focussed my review comments on the substantive content of the paper. However, I encourage the authors to try to reduce repetition between the "Introduction" and "Approach Rationale" sections, and to check for and correct typos and other errors in the text.  
+
+<|ref|>sub_title<|/ref|><|det|>[[73, 204, 201, 217]]<|/det|>
+## Review comments  
+
+<|ref|>text<|/ref|><|det|>[[72, 229, 888, 282]]<|/det|>
+Accounting for past (in- )action: I agree with the authors that updating an equitable mitigation assessment with updated historical emissions, all else equal, may result in an inadvertent benefit to emitters that are not reducing emissions (the argument the authors start to present in L44- L48). However, there are two things I think the authors should consider addressing:  
+
+<|ref|>text<|/ref|><|det|>[[72, 281, 920, 334]]<|/det|>
+1. The concept of "carbon debt": I think the same issue has been identified previously in the literature, where it has sometimes been termed as "carbon debt" or, emissions above a counterfactual equitable pathway (Gignac & Matthews, 2015; van den Berg et al., 2020). This approach does not fit neatly within the "continuous" versus "discontinuous" dichotomy that the authors have introduced.  
+
+<|ref|>text<|/ref|><|det|>[[72, 333, 920, 425]]<|/det|>
+2. The potential scale of the "grandfathering effect": The authors suggest, in L83-L85, that "Such iterative updates of ambition assessments based on continuous emissions allocations would iteratively find an insufficient NDC closer and closer to a calculated fair allocation". As I noted in the introduction to this section, I would tend to agree with this, all else equal. However, since pathways to a given temperature target (e.g., 1.5°C) will become progressively steeper, the responsibility of major emitters (which could be one equity allocation consideration) will increase. Given these two additional effects, I think there is more ambiguity in the effect on an equity assessment of NDCs. I suggest that the authors present some illustrative calculations to help the reader understand the validity of this statement.  
+
+<|ref|>text<|/ref|><|det|>[[72, 436, 920, 489]]<|/det|>
+Justification for the equity approaches: I appreciate that the authors provide a first estimation of the equitable mitigation targets that apply to both gross emission reductions as well as removals. As the authors correctly note, this is a gap in the existing literature, and addressing it is important to guide policy discussions. However, I have a few questions and concerns that I hope the authors can address:  
+
+<|ref|>text<|/ref|><|det|>[[72, 488, 920, 555]]<|/det|>
+1. I think the underlying justification for an equity approach is just as important as the numerical quantifications. Keeping this in mind, I found the justification for the application of different equity principles to gross reductions and removals to be one of the less comprehensive parts of the manuscript. I didn't understand why it is appropriate to factor in historical responsibility for one quantity (e.g., gross reductions) while factoring in capability for the other (e.g., gross removals). I think the manuscript could benefit from a more comprehensive discussion of the reasoning behind these equity approaches.  
+
+<|ref|>text<|/ref|><|det|>[[72, 554, 916, 595]]<|/det|>
+2. I found the description of the methods (L140 - L152 in the main text, and L436-L509 in the methods section) quite difficult to follow. I encourage the authors to publish the code used to carry out the analysis to allow for replication and to consider writing out the equations for each step so that the reader can follow the specific implementation.  
+
+<|ref|>text<|/ref|><|det|>[[72, 606, 896, 633]]<|/det|>
+Clarity on the use of scenarios: I have several comments on the use of global mitigation scenarios that I think the authors should address:  
+
+<|ref|>text<|/ref|><|det|>[[72, 645, 910, 711]]<|/det|>
+1. In L520-L522, the authors indicate that "The reference to a 1.5°C alignment corresponds to an alignment with the distribution of emissions of the average of scenarios of the IPCC Categories C1 [...], itself averaged with the distribution of C2 [...]. How do the authors come up with a distribution of emissions if they average across scenarios belonging to these categories? I'm not sure it is appropriate to use averages of such a scenario ensemble (see, e.g., (Guivarch et al., 2022)), and would recommend that the authors avoid this.  
+
+<|ref|>text<|/ref|><|det|>[[72, 710, 912, 775]]<|/det|>
+2. I think the authors should justify why they group the C1 and C2 categories of pathways together but do not do this for any of the other categories of pathways. Mapping the textual elements of the Paris Agreement Long Term Temperature Goal (LTTG) to specific pathway characteristics is a non-trivial value judgment (see, e.g., the discussions in (Kikstra et al., 2022; Schleussner et al., 2022)). I suggest that the authors improve the discussion on their pathway categorisation choices, especially since they explicitly indicate that this work is meant to guide the Global Stocktake.  
+
+<|ref|>text<|/ref|><|det|>[[72, 775, 910, 815]]<|/det|>
+3. Figure 2: The authors deviate from the labels presented in the methods section, by labeling C1+C2 pathways as "Below 1.5 degrees", and C3 pathways as "Well below 2 degrees". Please align the labels across the sections, and reflect on my comment above.  
+
+<|ref|>text<|/ref|><|det|>[[72, 815, 912, 945]]<|/det|>
+4. The 43% reduction by 2030 assessment: In L207-L209, the authors suggest that "The IPCC indicates that, on average across a set of scenarios, a 43% reduction in global GHG emissions by 2030 (here taken below 2020 levels) would align with a 1.5°C trajectory with no or limited overshoot. This global target [...]. There are a couple of conceptual challenges here, that I think the authors should consider addressing. The first, is that this is, by no means an IPCC-endorsed "global target" – it is only a description of the median (not average) of the scenarios assessed by the IPCC in that category of pathways. This value is also computed relative to 2019 emission levels, and given the structural differences between 2019 and 2020 emissions, I think it is further not appropriate to apply the 43% reduction below 2020 emission levels. Further, excluding LULUCF emissions (which are included in the original values presented in the IPCC report), means that this estimate is no longer appropriate. I suggest that the authors consider revising the text describing this approach, and use the uncertainty band presented in the AR6 report for this category while describing the results presented in Figure 3.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 58, 925, 112]]<|/det|>
+Additional analyses need to be motivated better: The two additional analyses presented in the results section (the addition of a "20- year transition phase", and the presentation of equal per capita results) are not motivated sufficiently in the text. I was unsure why the authors chose to present these results and suggest the authors add more text before the results section to justify why they have presented them.  
+
+<|ref|>sub_title<|/ref|><|det|>[[72, 125, 154, 138]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[70, 150, 920, 346]]<|/det|>
+Gignac, R., & Matthews, H. D. (2015). Allocating a 2 °C cumulative carbon budget to countries. Environmental Research Letters, 10(7), 075004. https://doi.org/10.1088/1748- 9326/10/7/075004Guivarch, C., Le Gallic, T., Bauer, N., Fragkos, P., Huppmann, D., Jaxa- Rozen, M., Keppo, I., Kriegler, E., Krisztin, T., Marangoni, G., Pye, S., Riahi, K., Schaeffer, R., Tavoni, M., Trutnevye, E., van Vuuren, D., & Wagner, F. (2022). Using large ensembles of climate change mitigation scenarios for robust insights. Nature Climate Change, 12(5), Article 5. https://doi.org/10.1038/s41558- 022- 01349- xKikstra, J. S., Nicholls, Z. R., Smith, C. J., Lewis, J., Lamboll, R. D., Byers, E., Sandstad, M., Meinshausen, M., Gidden, M. J., Rogelj, J., & others. (2022). The IPCC Sixth Assessment Report WGIII climate assessment of mitigation pathways: From emissions to global temperatures. Geoscientific Model Development, 15(24), 9075- 9109. Schleussner, C.- F., Ganti, G., Rogelj, J., & Gidden, M. J. (2022). An emission pathway classification reflecting the Paris Agreement climate objectives. Communications Earth & Environment, 3(1), https://doi.org/10.1038/s34247- 022- 00467- wvan den Berg, N. J., van Soest, H. L., Hof, A. F., den Elzen, M. G. J., van Vuuren, D. P., Chen, W., Drouet, L., Emmerling, J., Fujimori, S., Hohne, N., Koberle, A. C., McCollum, D., Schaeffer, R., Shekhar, S., Vishwanathan, S. S., Vrontisi, Z., & Blok, K. (2020). Implications of various effort- sharing approaches for national carbon budgets and emission pathways. Climatic Change, 162(4), 1805- 1822. https://doi.org/10.1007/s10584- 019- 02368- y  
+
+<|ref|>sub_title<|/ref|><|det|>[[72, 371, 144, 384]]<|/det|>
+## Version 1:  
+
+<|ref|>text<|/ref|><|det|>[[72, 397, 218, 411]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[72, 423, 160, 436]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[72, 449, 238, 462]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 462, 919, 567]]<|/det|>
+In the previous round, I noted that the presentation of the analysis and the language needed considerable improvement, and also recommended to have the whole text copy- edited. I regret to say that the improvement on these issues has not been satisfactory. It does not appear as the text has been copy- edited. It should not be the job of the reviewers to point out simple mistakes and help improve the language. While the authors have corrected the specific mistakes I pointed out, there are still many language problems, including in the newly introduced text. Some appear as sloppy mistakes, such as incomplete sentences, while a more extensive problem is lack of clear and structured presentation. I believe the underlying model development would be a valuable addition to the literature, but I deem the progress on its presentation from the first version as insufficient to warrant another 'revise and resubmit'.  
+
+<|ref|>text<|/ref|><|det|>[[72, 580, 234, 593]]<|/det|>
+Some concrete issues:  
+
+<|ref|>text<|/ref|><|det|>[[72, 593, 912, 646]]<|/det|>
+Re. my first point that the two fairness approaches are difficult to understand, which was also brought up by R2: The motivation for including two different approaches is still not clear to me. The new text is quite technical. Do the approaches reflect different ethical assumptions, or different approaches to a more technical modeling choice for which there is no clear criterion for choosing one over the other?  
+
+<|ref|>text<|/ref|><|det|>[[72, 657, 899, 684]]<|/det|>
+The figures and captions appear in a mess. The same figures appear on multiple pages, sometimes with and sometimes without captions.  
+
+<|ref|>text<|/ref|><|det|>[[72, 696, 682, 710]]<|/det|>
+The discussion has no structure, which makes it difficult to follow. There is no conclusion.  
+
+<|ref|>text<|/ref|><|det|>[[72, 722, 904, 749]]<|/det|>
+The methods section appears to contain considerable overlap with the main text (partly reflecting that the main text is very technical).  
+
+<|ref|>text<|/ref|><|det|>[[72, 761, 912, 853]]<|/det|>
+Re. my suggestion to "spend some more words on the criticism past studies have received for including transition periods (e.g., Kartha et al (2018)) and the weak ethical basis for such periods (e.g., Flerbaev et al (2014)). On the other hand, possible drawbacks of removing transition periods could also be discussed. For example, this means that reduced and avoided emissions are treated symmetrically, but the costs of reducing emissions are likely larger than the costs of avoiding future increases. The argument in the paper and in the literature against transition periods relies on emissions trading (ITMOs), but the imperfections of current institutions for this should be mentioned. Perhaps also refer to Knight (2013) for a defense of moderate grandfathering."  
+
+<|ref|>text<|/ref|><|det|>[[72, 853, 900, 879]]<|/det|>
+The authors responded 'Thank you for the reference. We added it in the following sentence as follows with references to Fleurbaey and Knight:  
+
+<|ref|>text<|/ref|><|det|>[[72, 879, 901, 919]]<|/det|>
+"Considering continuous emissions trajectories that look realistic22 implies that present- day levels of domestic emissions are an acceptable starting point in terms of mitigation effort with a utilitarian perspective25." I would have liked to see a more engagement with the suggestion than just adding one sentence.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[73, 73, 283, 87]]<|/det|>
+(Remarks on code availability)  
+
+<|ref|>text<|/ref|><|det|>[[73, 112, 162, 126]]<|/det|>
+Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[73, 138, 504, 166]]<|/det|>
+(Remarks to the Author)Thank you for the opportunity to review the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[70, 177, 920, 204]]<|/det|>
+I have reviewed the revisions made by the authors in response to my previous round of review comments. I am satisfied with these revisions.  
+
+<|ref|>text<|/ref|><|det|>[[73, 217, 283, 231]]<|/det|>
+(Remarks on code availability)  
+
+<|ref|>text<|/ref|><|det|>[[73, 256, 144, 270]]<|/det|>
+Version 2:  
+
+<|ref|>text<|/ref|><|det|>[[73, 282, 220, 295]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[73, 307, 160, 320]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[73, 333, 238, 347]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[73, 372, 283, 386]]<|/det|>
+(Remarks on code availability)  
+
+<|ref|>text<|/ref|><|det|>[[72, 752, 916, 805]]<|/det|>
+Open Access This Peer Review 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.  
+
+<|ref|>text<|/ref|><|det|>[[72, 805, 796, 819]]<|/det|>
+In cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+<|ref|>text<|/ref|><|det|>[[72, 819, 910, 870]]<|/det|>
+The images or other third party material in this Peer Review 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.  
+
+<|ref|>text<|/ref|><|det|>[[72, 870, 618, 884]]<|/det|>
+To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[119, 102, 358, 119]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[119, 135, 430, 152]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[119, 167, 875, 235]]<|/det|>
+This paper presents a welcome methodological development to the analysis of NDC based on fairness/equity principles. The avoidance of using transition periods from current/baseline levels to 'fair' allocations has considerable implications for the distribution fair mitigation and finance efforts. I find Figure 4 particularly illuminating.  
+
+<|ref|>text<|/ref|><|det|>[[119, 250, 870, 317]]<|/det|>
+The presentation of the analysis and the language needs considerable improvement. This drawback also makes it difficult to fully assess the underlying method, although it appears to be sound. I would recommend to have the whole text copy- edited, in addition to addressing the specific issues listed below.  
+
+<|ref|>text<|/ref|><|det|>[[119, 316, 842, 349]]<|/det|>
+I would also like to see some more discussion of the merits and drawbacks of the paper's methodological contribution.  
+
+<|ref|>text<|/ref|><|det|>[[119, 365, 300, 380]]<|/det|>
+Most important points  
+
+<|ref|>text<|/ref|><|det|>[[119, 382, 878, 432]]<|/det|>
+1. The two fairness approaches are difficult to understand. The section in the main text should be made more accessible and related clearly to fairness principles in the literature. The differences between the two approaches should be highlighted.  
+
+<|ref|>text<|/ref|><|det|>[[120, 432, 714, 449]]<|/det|>
+Why are the two approaches opposite wrt positive vs negative emissions?  
+
+<|ref|>text<|/ref|><|det|>[[120, 462, 692, 478]]<|/det|>
+Thank you. We extended the discussion to address these shortcomings:  
+
+<|ref|>text<|/ref|><|det|>[[178, 494, 880, 904]]<|/det|>
+"Here, we suggest two extensions, one for each approach of Fyson et al. to derive two allocations of economy- wide emissions to countries. Each new approach combines concepts of capability and responsibility, where each concept is applied to global positive or negative emissions distinctively. This study offers two conceptual combinations of the responsibility and capability concepts referred to in the Paris Agreement's CBDR- RC, with a differentiated treatment of negative emissions, which require costly and uncertain technologies, and are made necessary because of insufficient global emissions reductions to date. These approaches enable the assessment of the ambition of countries' NDC, in light of their responsibility and capability, with special considerations for negative emissions often used to enable and justify potentially dangerous warming overshoot. A first extension, named Approach 1, first allocates global negative emissions across countries based on their capability, assessed through GDP or Human Development Index (HDI, in Supplementary Information), and then allocates global positive emissions to equalize historical responsibilities over the net emissions (positive + negative, see Methods). Under this approach, rich countries are required to fund most of the negative emissions that require important research and development costs with high uncertainty and without local co- benefits31. Approach 1 also ensures equal cumulative per capita emissions by 2100 through the allocation of the positive emissions space. Here, the capability allocation affects the distribution of emissions over time, but not the total budget. Richer countries then have more important negative allocations in the future, and less stringent allocations in the near term (see Methods). The second extension, Approach 2, conversely first allocates global positive emissions based on countries' capabilities and then global negative emissions based on their responsibilities. There, all countries contribute to emissions reductions based on their wealth. Negative emissions, needed because of the world's important historical
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[178, 86, 880, 188]]<|/det|>
+emissions, are allocated proportionally to countries' individual historical responsibilities. In Approach 2, historical responsibility does not define countries' cumulative emissions alone. Looking at the global emissions scenarios, the positive emissions refer here to the projected physical emissions (e.g., fossil fuels, agriculture). The negative emissions here refer to emissions captured through Carbon Dioxide Removal (excluding those from LULUCF, unlike Fyson et al. 2020) and Direct Air Capture.  
+
+<|ref|>text<|/ref|><|det|>[[118, 233, 879, 305]]<|/det|>
+143: [approach 1] "allocates global positive emissions to equalize historical responsibilities". It is confusing that historical responsibility enters into approach 1, as I thought approach 1 was based on capability and approach 2 on responsibility. Also, historical responsibility over what time period?  
+
+<|ref|>text<|/ref|><|det|>[[118, 314, 879, 382]]<|/det|>
+Indeed, the text was misleading. We added a sentence explaining the rationale of the paper that is to suggest two extensions of the Fyson et al. paper, each of which combines responsibility (since 1990 and 1950) and capability. Extending each of the two approaches of Fyson et al. For clarity, we amended the previous paragraph and added the sentence line 143:  
+
+<|ref|>text<|/ref|><|det|>[[178, 394, 864, 428]]<|/det|>
+"Each new approach combines concepts of capability and responsibility, where each concept is applied to global positive or negative emissions distinctively."  
+
+<|ref|>text<|/ref|><|det|>[[118, 441, 575, 458]]<|/det|>
+And in the following paragraph on the parameterization:  
+
+<|ref|>text<|/ref|><|det|>[[178, 471, 878, 522]]<|/det|>
+"Here, we present results based on a parameterization that uses GDP for capability and accounts for responsibility through emissions since 1990 (with 1950 in the Supplementary Information)."  
+
+<|ref|>text<|/ref|><|det|>[[118, 551, 876, 602]]<|/det|>
+147: "the capability considerations affects the use the emissions budgets over time, but not its total" Multiple typos here, and I do not understand the sentence. Affects the distribution over time, but not the total budget, maybe?  
+
+<|ref|>text<|/ref|><|det|>[[118, 615, 805, 649]]<|/det|>
+Thanks for pointing this out and the concrete suggestion. The sentence was modified according to the suggestion:  
+
+<|ref|>text<|/ref|><|det|>[[178, 661, 867, 744]]<|/det|>
+"Approach 1 also ensures equal cumulative per capita emissions by 2100 through the allocation of the positive emissions space. Here, the capability allocation affects the distribution of emissions over time, but not the total budget. Richer countries then have more important negative allocations in the future, and less stringent allocations in the near term (see Methods)."  
+
+<|ref|>text<|/ref|><|det|>[[118, 774, 864, 808]]<|/det|>
+489: "This approach yields significant differences in emissions allocations across countries" Differences relative to what? Variation across countries might be a better term.  
+
+<|ref|>text<|/ref|><|det|>[[118, 821, 844, 855]]<|/det|>
+Thank you. I had not realized the difference in meanings of these two terms. The text was modified according to this suggestion.  
+
+<|ref|>text<|/ref|><|det|>[[115, 870, 836, 888]]<|/det|>
+491- 492: I do not understand this sentence. "Important" seems not the correct word here.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[119, 86, 474, 102]]<|/det|>
+Thank you. Would this phrasing be clearer?  
+
+<|ref|>text<|/ref|><|det|>[[178, 117, 870, 166]]<|/det|>
+"Thank you. Would this phrasing be clearer? "This approach yields significant variations in emissions allocations across countries, which reflects the large differences across countries' GDPs (often proportionally greater than the differences of their historical contributions)."  
+
+<|ref|>text<|/ref|><|det|>[[118, 196, 876, 345]]<|/det|>
+2. I would suggest to spend some more words on the criticism past studies have received for including transition periods (e.g., Kartha et al (2018)) and the weak ethical basis for such periods (e.g., Flerbaev et al (2014)). On the other hand, possible drawbacks of removing transition periods could also be discussed. For example, this means that reduced and avoided emissions are treated symmetrically, but the costs of reducing emissions are likely larger than the costs of avoiding future increases. The argument in the paper and in the literature against transition periods relies on emissions trading (ITMOs), but the imperfections of current institutions for this should be mentioned. Perhaps also refer to Knight (2013) for a defense of moderate grandfathering.  
+
+<|ref|>text<|/ref|><|det|>[[118, 360, 876, 393]]<|/det|>
+Thank you for the reference. We added it in the following sentence as follows with references to Fleurbaev and Knight:  
+
+<|ref|>text<|/ref|><|det|>[[178, 407, 863, 456]]<|/det|>
+"Considering continuous emissions trajectories that look realistic22 implies that present- day levels of domestic emissions are an acceptable starting point in terms of mitigation effort with a utilitarian perspective25."  
+
+<|ref|>text<|/ref|><|det|>[[118, 487, 680, 503]]<|/det|>
+Sentences that are difficult to understand and should be reformulated:  
+
+<|ref|>text<|/ref|><|det|>[[119, 504, 877, 552]]<|/det|>
+17: "increasingly do so in the future" I could not understand what was meant until reading the introduction. This is an important point, but it needs to be elaborated more to be understandable in the abstract.  
+
+<|ref|>text<|/ref|><|det|>[[119, 567, 358, 583]]<|/det|>
+Certainly. We reformulate to:  
+
+<|ref|>text<|/ref|><|det|>[[178, 598, 871, 630]]<|/det|>
+"Ambition assessments based on trajectories that start at present- day emissions levels inherently reward past inaction, and increasingly do so with their iterative updates."  
+
+<|ref|>text<|/ref|><|det|>[[118, 660, 875, 726]]<|/det|>
+45: "most of the recent approaches rely on allocations of emissions rights following a continuous trajectory starting at current emissions levels, sometimes using a transition period" Why the word sometimes? Do not all continuous trajectories starting at current levels use a transition period by necessity? Or are there exceptions?. See also line 69.  
+
+<|ref|>text<|/ref|><|det|>[[119, 741, 839, 789]]<|/det|>
+Thanks for raising this. Indeed, the continuity of emissions allocations does not require a transition period. The formula can be designed in other ways. To clarify this point, we suggest changing the text shortly after (originally on line 73):  
+
+<|ref|>text<|/ref|><|det|>[[178, 804, 868, 852]]<|/det|>
+"Effort- sharing formulas can be designed to directly achieve such continuity2,3 or an ad- hoc transition period can be added to ensure continuity3,14,16 between current emissions and future allocations only based on fairness considerations."  
+
+<|ref|>text<|/ref|><|det|>[[118, 867, 562, 883]]<|/det|>
+This is also further explained in the discussion section:
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[179, 85, 880, 185]]<|/det|>
+"In other models, the continuity of the emissions trajectory results from the choice of allocating mitigation burden to depart from a reference trajectory rather than allocating the remaining emissions space2. Such an approach based on reference trajectories can be adequate to assess the ambition of an emissions target when provided with a corresponding reference scenario, e.g. pledges taken in 2015 against allocations starting in 2015 (ref. 2)."  
+
+<|ref|>text<|/ref|><|det|>[[118, 230, 866, 264]]<|/det|>
+131: "The relevance of equity concepts and their implementations in effort- sharing formulae show various consistency with international law"  
+
+<|ref|>text<|/ref|><|det|>[[119, 278, 469, 295]]<|/det|>
+Thank you. We suggested this formulation:  
+
+<|ref|>text<|/ref|><|det|>[[177, 309, 866, 342]]<|/det|>
+"Some of the equity concepts quantified in the literature are not backed by principles of international law that require excluding approaches based on grandfathering6."  
+
+<|ref|>text<|/ref|><|det|>[[119, 372, 806, 406]]<|/det|>
+167: "Approach 1 is mainly driven by responsibility in the near term" Change to (if I understand correctly: In the near term, a1 is driven mainly by responsibility  
+
+<|ref|>text<|/ref|><|det|>[[119, 419, 722, 436]]<|/det|>
+Thank you for the concrete suggestion. We decided to delete this sentence.  
+
+<|ref|>text<|/ref|><|det|>[[119, 465, 870, 515]]<|/det|>
+189: "The absence of zero or negative allocations for some countries results from fairness indicators as well as the absence of negative emissions in the \(1.5^{\circ}\mathrm{C}\) scenarios- set with strong near- term mitigation, excluding LULUCF emissions."  
+
+<|ref|>text<|/ref|><|det|>[[119, 530, 473, 546]]<|/det|>
+We suggest reformulating to two sentences:  
+
+<|ref|>text<|/ref|><|det|>[[178, 560, 866, 643]]<|/det|>
+"Looking at when allocations reach net- zero, some countries with relatively low responsibility and capability have emissions allocations that are positive throughout the century under a \(1.5^{\circ}\mathrm{C}\) objective. It is important to note that some of the selected global \(1.5^{\circ}\mathrm{C}\) scenarios with strong near- term mitigation also have positive emissions throughout the century since we excluded the LULUCF sector."  
+
+<|ref|>text<|/ref|><|det|>[[118, 688, 793, 706]]<|/det|>
+300: "The egalitarian approach..." Why is this introduced here? Difficult to follow.  
+
+<|ref|>text<|/ref|><|det|>[[118, 720, 860, 752]]<|/det|>
+Thanks for highlighting this. We added an introductory sentence and modified the previous one to:  
+
+<|ref|>text<|/ref|><|det|>[[178, 766, 876, 866]]<|/det|>
+"In addition to capability and responsibility, equality is the third equity principle described in the IPCC AR517,22. IPCC reports do not present equity- based emissions allocations since AR5, despite available studies and its importance for courts of law12. The egalitarian approach modelled as equal per capita emissions is not directly anchored in the Paris Agreement or international environmental law6, but can reveal the inequalities of emissions spaces claimed through NDCs."
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 101, 672, 120]]<|/det|>
+343: "with few arbitrary" parameters. As few as possible? Or a few?  
+
+<|ref|>text<|/ref|><|det|>[[118, 133, 580, 150]]<|/det|>
+We changed wording to: "as few parameters as possible"  
+
+<|ref|>text<|/ref|><|det|>[[118, 195, 829, 246]]<|/det|>
+359: "Even equity- based budgets could theoretically be used mostly in the near- term by countries and not collectively reflect any of the global \(1.5^{\circ}\mathrm{C}\) mitigation scenarios underpinning the global budget."  
+
+<|ref|>text<|/ref|><|det|>[[118, 277, 441, 293]]<|/det|>
+We suggest simplifying the sentence to:  
+
+<|ref|>text<|/ref|><|det|>[[178, 307, 879, 356]]<|/det|>
+"Theoretically, countries could choose to use a budget mostly in the near- term to justify insufficient emissions objectives, which raises issues of intergenerational justice."  
+
+<|ref|>text<|/ref|><|det|>[[118, 387, 847, 453]]<|/det|>
+364: "Additionally, emissions budgets are not suitable for addressing the knowledge gaps identified in the IPCC AR6 of "extending equity frameworks to quantify equitable international support, as the difference between equity- based national emissions scenarios and national domestic emissions scenarios""  
+
+<|ref|>text<|/ref|><|det|>[[118, 467, 825, 499]]<|/det|>
+In response, we modified the previous few sentences for clarity and amended these two sentences to:  
+
+<|ref|>text<|/ref|><|det|>[[178, 514, 879, 660]]<|/det|>
+"Theoretically, countries could choose to use a budget mostly in the near- term to justify insufficient emissions objectives, which raises issues of intergenerational justice. The "flexibility" provided by carbon budgets over emissions pathways14 comes at the expense of the ability to assess the ambition of time- defined objectives, without additional assumptions18. Time- defined emissions allocations are also needed to address the knowledge gaps identified in the IPCC AR6 of "extending equity frameworks to quantify equitable international support, as the difference between equity- based national emissions scenarios and national domestic emissions scenarios"52."  
+
+<|ref|>text<|/ref|><|det|>[[118, 691, 759, 708]]<|/det|>
+457: "that may result from better governance or potentially ill acquired wealth"  
+
+<|ref|>text<|/ref|><|det|>[[119, 722, 355, 738]]<|/det|>
+We suggest reformulating to:  
+
+<|ref|>text<|/ref|><|det|>[[178, 752, 870, 835]]<|/det|>
+"Comparing two countries with equal populations and equal GDPs, the country with higher HDI will have greater effort to provide when using HDI as the capacity indicator rather than GDP. Using HDI as a capacity indicator may then penalize good governance, compared to using GDP. Results based on HDI are available in the supplementary information."
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 85, 601, 103]]<|/det|>
+524: "reflect alignment with symbolic warming thresholds"  
+
+<|ref|>text<|/ref|><|det|>[[119, 117, 366, 134]]<|/det|>
+We simplified the phrasing to:  
+
+<|ref|>text<|/ref|><|det|>[[177, 147, 852, 181]]<|/det|>
+"Additionally, the effort- sharing formulas are applied to the scenario categories C6 ('below \(3^{\circ}\mathrm{C}\) ) and C7 (below \(4^{\circ}\mathrm{C}\) ) that reflect current policies."  
+
+<|ref|>text<|/ref|><|det|>[[118, 227, 860, 260]]<|/det|>
+527: "can be considered dragging even the insufficient ambition current policies that do not track towards NDCs"  
+
+<|ref|>text<|/ref|><|det|>[[119, 274, 261, 291]]<|/det|>
+We simplified to:  
+
+<|ref|>text<|/ref|><|det|>[[177, 304, 864, 338]]<|/det|>
+"Countries with NDCs that do not align with their fair allocation of C7 scenarios can be considered to be dragging global decarbonization efforts."  
+
+<|ref|>text<|/ref|><|det|>[[118, 368, 179, 384]]<|/det|>
+Typos:  
+
+<|ref|>text<|/ref|><|det|>[[120, 385, 390, 401]]<|/det|>
+159: 'nations' should be notions?  
+
+<|ref|>text<|/ref|><|det|>[[119, 415, 342, 432]]<|/det|>
+Yes. This typo is corrected.  
+
+<|ref|>text<|/ref|><|det|>[[118, 461, 456, 479]]<|/det|>
+246 "delay climate action and near 2030"  
+
+<|ref|>text<|/ref|><|det|>[[119, 493, 351, 509]]<|/det|>
+We clarified the sentence to:  
+
+<|ref|>text<|/ref|><|det|>[[177, 522, 757, 540]]<|/det|>
+"This effect increases as we delay climate action and as we near 2030."  
+
+<|ref|>text<|/ref|><|det|>[[118, 570, 327, 586]]<|/det|>
+412: "that country could"  
+
+<|ref|>text<|/ref|><|det|>[[119, 600, 383, 617]]<|/det|>
+Indeed, corrected to "countries".  
+
+<|ref|>text<|/ref|><|det|>[[118, 648, 230, 664]]<|/det|>
+Minor points:  
+
+<|ref|>text<|/ref|><|det|>[[118, 664, 876, 714]]<|/det|>
+251: "Based on Approach 2, the assessment of the NDCs of the UK, Sweden and Switzerland is more stringent" Not apparent in the figure. Perhaps because they are nevertheless given the same assessment category?  
+
+<|ref|>text<|/ref|><|det|>[[119, 728, 434, 744]]<|/det|>
+Thank you. We have now corrected to:  
+
+<|ref|>text<|/ref|><|det|>[[178, 758, 850, 808]]<|/det|>
+"Based on Approach 2, the assessment of the NDCs of the UK and Switzerland is more stringent given their relatively low historical responsibility compared to their relatively high capability."  
+
+<|ref|>text<|/ref|><|det|>[[118, 837, 862, 888]]<|/det|>
+Figure 2: For which countries is approach 2 less stringent than approach 1? I cannot see any country with a greener color under this approach, but perhaps the African countries are overdelivering 1.5degrees ambition even more under approach 2?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 85, 864, 168]]<|/det|>
+Indeed, Figure 3 (formerly figure 2) provides a rating of NDC ambition. The full allocations are provided in the supplementary data that will be made available publicly for all countries (the version initially submitted is available here https://zenodo.org/record/8003393). We added a new sentence after the two sentences that discussed the relative stringencies of approaches 1 & 2:  
+
+<|ref|>text<|/ref|><|det|>[[178, 182, 877, 313]]<|/det|>
+"While Approach 1 constrains countries' cumulative emissions based on their responsibility, which often overlaps with high capability38, Approach 2 is more stringent in the near term for countries with high GDP. Approach 2 is less stringent for countries with very low capability (sub- Saharan African countries), and for countries with high historical responsibility (USA, Russia, Qatar and other fossil fuel extracting countries, Figure 2 and Supplementary Information). These different stringencies of emissions allocations do not always change the warming assessment of countries' NDCs (see country- level allocations in SI)."  
+
+<|ref|>text<|/ref|><|det|>[[118, 343, 866, 411]]<|/det|>
+387: Compared to a previous warming assessment5 (visible on Paris- Equity- Check.org), Approach 1 finds NDCs to be more ambitious (1.5°C aligned) for a few countries (including India, Indonesia and Egypt) and less for Norway." Are these these the only changes? Using what equity principle?  
+
+<|ref|>text<|/ref|><|det|>[[118, 423, 844, 474]]<|/det|>
+This previous assessment (which I am also author of), had different methods combining different equity principles. Here, we choose to report the most prominent differences. We suggest clarifying that this previous assessment relied on three equity principles:  
+
+<|ref|>text<|/ref|><|det|>[[178, 487, 868, 570]]<|/det|>
+"Compared to a previous warming assessment5 (visible on Paris- Equity- Check.org) where each country follows the least stringent of three equity principles (capability, responsibility, and equality), the present approaches find NDCs to be more ambitious (1.5°C aligned) for a few countries (including India, Indonesia, and Egypt depending on the approach) and less for countries in the Global North and Latin America."  
+
+<|ref|>text<|/ref|><|det|>[[118, 599, 868, 667]]<|/det|>
+Request from the CVF: This section does not clearly distinguish between the request itself and the authors' interpretation of it. The whole section is in quotation marks, but refers to "the CF requested" and "this paper", which causes confusion over who formulated the text. I suggest replicating or summarizing the request first, then presenting the interpretation.  
+
+<|ref|>text<|/ref|><|det|>[[118, 680, 855, 713]]<|/det|>
+Thank you. We seek to provide transparency with this section. Your input is helpful in that regard. We changed the titles of the two sections as follows:  
+
+<|ref|>text<|/ref|><|det|>[[179, 727, 471, 744]]<|/det|>
+Request as explained by the CVF:  
+
+<|ref|>text<|/ref|><|det|>[[178, 757, 788, 790]]<|/det|>
+Interpretation and discussion of CVF's request in light of the available literature:  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 822, 210, 837]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[118, 838, 848, 904]]<|/det|>
+Fleurbaey, M., Kartha, S., Bolwig, S., Chee, Y.L., Chen, Y., Corbera, E., et al., 2014. Sustainable development and equity. In: Edenhofer, O., Pichs- Madruga, R., Sokona, Y., Farahani, E., Kadner, S., Seyboth, K. (Eds.), Climate Change 2014: Mitigation of Climate Change. Contribution of Working Group III to the Fifth Assessment Report of the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 85, 822, 120]]<|/det|>
+Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, United Kingdom and New York, NY, pp. 283- 350.  
+
+<|ref|>text<|/ref|><|det|>[[118, 135, 875, 201]]<|/det|>
+Kartha, S., Athanasiou, T., Caney, S., Cripps, E., Dooley, K., Dubash, N.K., Harris, P., Holz, C., Lahn, B., Moellendorf, D., Müller, B., Roberts, J.T., Sagar, A.D., Shue, H., Singer, P., Winkler, H., 2018. Cascading biases against poorer parties. Nat. Clim. Change 8 (5), 348- 349.  
+
+<|ref|>text<|/ref|><|det|>[[118, 216, 689, 234]]<|/det|>
+Knight, C., 2013. What is grandfathering? Env. Polit. 22 (3), 410- 427.  
+
+<|ref|>text<|/ref|><|det|>[[118, 299, 430, 315]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 332, 870, 479]]<|/det|>
+Robiou du Pont and co- authors note that most existing equitable mitigation assessments start from the most recent year when emissions are available for countries (termed "continuous allocations" by the authors). The authors indicate that this choice induces a "grandfathering" effect, which unintentionally rewards countries that have not reduced emissions with relatively less stringent emission reduction benchmarks in the near term. To address this, the authors propose an approach to derive equitable allocations for countries, which the authors suggest departs from previous literature in two ways: (1) the allocations start in a historical year (e.g., 1990), and hence capture historical (in- )action, and (2) the authors treat gross emission reductions and gross emission removals separately.  
+
+<|ref|>text<|/ref|><|det|>[[118, 495, 874, 529]]<|/det|>
+I have some comments and suggestions that I think are important to consider. I have focussed my review comments on the substantive content of the paper  
+
+<|ref|>text<|/ref|><|det|>[[118, 545, 269, 561]]<|/det|>
+Review comments  
+
+<|ref|>text<|/ref|><|det|>[[118, 577, 875, 730]]<|/det|>
+Accounting for past (in- )action: I agree with the authors that updating an equitable mitigation assessment with updated historical emissions, all else equal, may result in an inadvertent benefit to emitters that are not reducing emissions (the argument the authors start to present in L44- L48). However, there are two things I think the authors should consider addressing: 1. The concept of "carbon debt": I think the same issue has been identified previously in the literature, where it has sometimes been termed as "carbon debt" or, emissions above a counterfactual equitable pathway (Gignac & Matthews, 2015; van den Berg et al., 2020). This approach does not fit neatly within the "continuous" versus "discontinuous" dichotomy that the authors have introduced.  
+
+<|ref|>text<|/ref|><|det|>[[118, 739, 880, 904]]<|/det|>
+I understand the carbon debt as related to the accounting of historical emissions. The disparities in historical emissions across countries can be seen as a carbon debt from the high historical emitters, including to the lower emitting countries. Previous studies, including (Gignac & Matthews, 2015; van den Berg et al., 2020), suggested that future emissions allocations can account for this debt. The debt could be compensated over the period for which the future emissions budget is allocated, or possibly through adaptation and loss- and- damage finance. In a recent submission, Pelz et al. (http://dx.doi.org/10.21203/rs.3.rs- 4394688/v1, under review) highlight the importance of this debt to "') responsibility for overshoot, ii) exceedance drawdown obligations, and iii) increase in extreme climate exposure if drawdown does not occur." While our approach allocates immediate effort to
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 86, 880, 120]]<|/det|>
+compensate inequities, the "carbon debt" tracks overshoot but does not suggest how this debt should be compensated over time.  
+
+<|ref|>text<|/ref|><|det|>[[118, 133, 880, 200]]<|/det|>
+Our present study and its use of discontinuous dynamic emissions allocations does not differ in that aspect. The carbon debt is accounted for and influences future emissions allocations, even resulting in equal cumulative per capita emissions in Approach 1. In this case, the 'discontinuity' feature only affects how the emissions budget is used over time.  
+
+<|ref|>text<|/ref|><|det|>[[120, 214, 400, 230]]<|/det|>
+We clarify this point in the text as:  
+
+<|ref|>text<|/ref|><|det|>[[178, 243, 873, 343]]<|/det|>
+"Other approaches use the concept of 'carbon debt'55,56 to assess countries' responsibility for overshooting their fair shares of a global carbon budget, including through their future objectives, and compensate through future negative emissions57. This budget- based approach can be used to characterize breaches by courts and be complementary to dynamic approaches, as in this study, immediately allocating feasible socio- economic scenarios, which can inform Paris- aligned emissions targets."  
+
+<|ref|>text<|/ref|><|det|>[[117, 372, 860, 537]]<|/det|>
+2. The potential scale of the "grandfathering effect": The authors suggest, in L83-L85, that "Such iterative updates of ambition assessments based on continuous emissions allocations would iteratively find an insufficient NDC closer and closer to a calculated fair allocation". As I noted in the introduction to this section, I would tend to agree with this, all else equal. However, since pathways to a given temperature target (e.g., \(1.5^{\circ}\mathrm{C}\) ) will become progressively steeper, the responsibility of major emitters (which could be one equity allocation consideration) will increase. Given these two additional effects, I think there is more ambiguity in the effect on an equity assessment of NDCs. I suggest that the authors present some illustrative calculations to help the reader understand the validity of this statement.  
+
+<|ref|>text<|/ref|><|det|>[[118, 551, 870, 617]]<|/det|>
+Thank you. We updated Figure 1 to use actual data instead of the schematic representation in the first submission. We modelled allocation starting in 2015 and 2020. As you guessed, the slope is steeper in the later allocation, but the grandfathering effect remains visible. Thank you for the suggestion.  
+
+<|ref|>text<|/ref|><|det|>[[117, 646, 875, 728]]<|/det|>
+Justification for the equity approaches: I appreciate that the authors provide a first estimation of the equitable mitigation targets that apply to both gross emission reductions as well as removals. As the authors correctly note, this is a gap in the existing literature, and addressing it is important to guide policy discussions. However, I have a few questions and concerns that I hope the authors can address:  
+
+<|ref|>text<|/ref|><|det|>[[118, 729, 867, 844]]<|/det|>
+1. I think the underlying justification for an equity approach is just as important as the numerical quantifications. Keeping this in mind, I found the justification for the application of different equity principles to gross reductions and removals to be one of the less comprehensive parts of the manuscript. I didn't understand why it is appropriate to factor in historical responsibility for one quantity (e.g., gross reductions) while factoring in capability for the other (e.g., gross removals). I think the manuscript could benefit from a more comprehensive discussion of the reasoning behind these equity approaches.  
+
+<|ref|>text<|/ref|><|det|>[[118, 858, 875, 891]]<|/det|>
+Thank you, the lack of clarity of this section was also raised by reviewer 1. We agree with the importance of the justification updated the description as follows:
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[177, 85, 880, 604]]<|/det|>
+"Here, we suggest two extensions, one for each approach of Fyson et al. to derive two allocations of economy- wide emissions to countries. Each new approach combines concepts of capability and responsibility, where each concept is applied to global positive or negative emissions distinctively. This study offers two conceptual combinations of the responsibility and capability concepts referred to in the Paris Agreement's CBDR- RC, with a differentiated treatment of negative emissions, which require costly and uncertain technologies, and are made necessary because of insufficient global emissions reductions to date. These approaches enable the assessment of the ambition of countries' NDC, in light of their responsibility and capability, with special considerations for negative emissions often used to enable and justify potentially dangerous warming overshoot. A first extension, named Approach 1, first allocates global negative emissions across countries based on their capability, assessed through GDP or Human Development Index (HDI, in Supplementary Information), and then allocates global positive emissions to equalize historical responsibilities over the net emissions (positive + negative, see Methods). Under this approach, rich countries are required to fund most of the negative emissions that require important research and development costs with high uncertainty and without local co- benefits31. Approach 1 also ensures equal cumulative per capita emissions by 2100 through the allocation of the positive emissions space. Here, the capability allocation affects the distribution of emissions over time, but not the total budget. Richer countries then have more important negative allocations in the future, and less stringent allocations in the near term (see Methods). The second extension, Approach 2, conversely first allocates global positive emissions based on countries' capabilities and then global negative emissions based on their responsibilities. There, all countries contribute to emissions reductions based on their wealth. Negative emissions, needed because of the world's important historical emissions, are allocated proportionally to countries' individual historical responsibilities. In Approach 2, historical responsibility does not define countries' cumulative emissions alone. Looking at the global emissions scenarios, the positive emissions refer here to the projected physical emissions (e.g., fossil fuels, agriculture). The negative emissions here refer to emissions captured through Carbon Dioxide Removal (excluding those from LULUCF, unlike Fyson et al. 2020) and Direct Air Capture."  
+
+<|ref|>text<|/ref|><|det|>[[118, 629, 856, 697]]<|/det|>
+2. I found the description of the methods (L140 – L152 in the main text, and L436-L509 in the methods section) quite difficult to follow. I encourage the authors to publish the code used to carry out the analysis to allow for replication and to consider writing out the equations for each step so that the reader can follow the specific implementation.  
+
+<|ref|>text<|/ref|><|det|>[[118, 711, 870, 775]]<|/det|>
+Thank you. We will make the code publicly available with the manuscript on a Github. It is shared with this version of the manuscript through Code Ocean and at: https://github.com/imagepbl/effort-sharing. We also added a description of the formulas used in the Methods.  
+
+<|ref|>text<|/ref|><|det|>[[118, 821, 828, 855]]<|/det|>
+Clarity on the use of scenarios: I have several comments on the use of global mitigation scenarios that I think the authors should address:  
+
+<|ref|>text<|/ref|><|det|>[[118, 870, 878, 904]]<|/det|>
+1. In L520-L522, the authors indicate that "The reference to a \(1.5^{\circ}\mathrm{C}\) alignment corresponds to an alignment with the distribution of emissions of the average of scenarios of the IPCC
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 85, 863, 152]]<|/det|>
+Categories C1 [...], itself averaged with the distribution of C2 [...]. How do the authors come up with a distribution of emissions if they average across scenarios belonging to these categories? I'm not sure it is appropriate to use averages of such a scenario ensemble (see, e.g., (Guivarch et al., 2022)), and would recommend that the authors avoid this.  
+
+<|ref|>text<|/ref|><|det|>[[118, 165, 858, 265]]<|/det|>
+Thank you for raising this. Guivarch et al. 2022 (co- authored by one of the authors of the present manuscript) highlight: "Although scenario ensembles are designed to explore the possibility space, neither type of ensemble can be interpreted as a perfect statistical sample. Given the unknown unknowns, the scenarios' outcomes cannot be interpreted in terms of likelihoods, and even large scenario ensembles do not fully or equally explore the space of possibilities".  
+
+<|ref|>text<|/ref|><|det|>[[118, 278, 866, 393]]<|/det|>
+We agree that averaging across the ensemble of scenarios in C- categories does not provide a representative view of the literature as some modeling choices may be over/under- represented. However, our analysis uses global scenarios only for their global emissions bound by common physical considerations. The socio- economic assumptions of these scenarios have limited effect on the total emissions profile, mostly driven by considerations regarding total negative emissions. This approach is consistent with previous studies, including:  
+
+<|ref|>text<|/ref|><|det|>[[177, 406, 855, 440]]<|/det|>
+Robiou du Pont, Y., Jeffery, M., Gutschew, J. et al. Equitable mitigation to achieve the Paris Agreement goals. Nature Clim Change 7, 38- 43 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[178, 453, 864, 519]]<|/det|>
+Xunzhang Pan, Michel den Elzen, Niklas Hohne, Fei Teng, Lining Wang, Exploring fair and ambitious mitigation contributions under the Paris Agreement goals, Environmental Science & Policy, Volume 74, 2017, Pages 49- 56, ISSN 1462- 9011, https://doi.org/10.1016/j.envsci.2017.04.020.  
+
+<|ref|>text<|/ref|><|det|>[[119, 533, 860, 583]]<|/det|>
+Note that part of the author team is working on another article, under review, suggesting custom- made global emissions scenario representatives of various warming categories, that enables the user to check the influence of modelling parameters separately:  
+
+<|ref|>text<|/ref|><|det|>[[178, 597, 868, 647]]<|/det|>
+Mark Dekker, Andries Hof, Yann Robiou du Pont et al. Navigating the black box of fair national emissions targets, 19 September 2024, PREPRINT (Version 1) available at Research Square [https://doi.org/10.21203/rs.3.rs- 5023350/v1]  
+
+<|ref|>text<|/ref|><|det|>[[118, 676, 864, 792]]<|/det|>
+2. I think the authors should justify why they group the C1 and C2 categories of pathways together but do not do this for any of the other categories of pathways. Mapping the textual elements of the Paris Agreement Long Term Temperature Goal (LTTG) to specific pathway characteristics is a non-trivial value judgment (see, e.g., the discussions in (Kikstra et al., 2022; Schleussner et al., 2022)). I suggest that the authors improve the discussion on their pathway categorisation choices, especially since they explicitly indicate that this work is meant to guide the Global Stocktake.  
+
+<|ref|>text<|/ref|><|det|>[[118, 805, 852, 871]]<|/det|>
+Thank you. We understand the critics of Schleussner et al. regarding the potential misinterpretation of scenario categories, their absence of overlaps and their representativeness of the Paris Agreement goal. We now provide the results for C1 and C2 separately and amended the text in the following manner:  
+
+<|ref|>text<|/ref|><|det|>[[177, 885, 816, 902]]<|/det|>
+"The reference to a \(1.5^{\circ}\mathrm{C}\) alignment corresponds here to an alignment with the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[179, 85, 867, 202]]<|/det|>
+distribution of emissions of the average of scenarios of the IPCC Categories C1 ('below \(1.5^{\circ}\mathrm{C}\) with no or limited overshoot'). The distribution of C2 conveys a warming 'below \(1.5^{\circ}\mathrm{C}\) with high overshoot'. The upper threshold of \(2^{\circ}\mathrm{C}\) alignment here follows the definition based on emissions scenarios C3 ('likely below \(2^{\circ}\mathrm{C}\)') category. The consistency of such low emissions scenarios with the Paris Agreement temperature goal is discussed based on warming responses and levels of negative emissions62,63."  
+
+<|ref|>text<|/ref|><|det|>[[118, 230, 874, 281]]<|/det|>
+3. Figure 2: The authors deviate from the labels presented in the methods section, by labeling C1+C2 pathways as "Below 1.5 degrees", and C3 pathways as "Well below 2 degrees". Please align the labels across the sections, and reflect on my comment above.  
+
+<|ref|>text<|/ref|><|det|>[[120, 295, 477, 312]]<|/det|>
+Thank you. We have corrected the labelling.  
+
+<|ref|>text<|/ref|><|det|>[[117, 341, 875, 573]]<|/det|>
+4. The \(43\%\) reduction by 2030 assessment: In L207-L209, the authors suggest that "The IPCC indicates that, on average across a set of scenarios, a \(43\%\) reduction in global GHG emissions by 2030 (here taken below 2020 levels) would align with a \(1.5^{\circ}\mathrm{C}\) trajectory with no or limited overshoot. This global target [...]". There are a couple of conceptual challenges here, that I think the authors should consider addressing. The first, is that this is, by no means an IPCC-endorsed "global target" – it is only a description of the median (not average) of the scenarios assessed by the IPCC in that category of pathways. This value is also computed relative to 2019 emission levels, and given the structural differences between 2019 and 2020 emissions, I think it is further not appropriate to apply the \(43\%\) reduction below 2020 emission levels. Further, excluding LULUCF emissions (which are included in the original values presented in the IPCC report), means that this estimate is no longer appropriate. I suggest that the authors consider revising the text describing this approach, and use the uncertainty band presented in the AR6 report for this category while describing the results presented in Figure 3.  
+
+<|ref|>text<|/ref|><|det|>[[118, 585, 857, 652]]<|/det|>
+Thank you. We updated the data to show a \(50\%\) reduction below 2020 levels and no longer refer to the IPCC figure to simply illustrate national allocation at a midway point of global decarbonization. Figure 2 (formerly Figure 3) shows this new parameterization. Thank you for the suggestion that avoids misinterpretation.  
+
+<|ref|>text<|/ref|><|det|>[[118, 681, 874, 765]]<|/det|>
+Additional analyses need to be motivated better: The two additional analyses presented in the results section (the addition of a "20- year transition phase", and the presentation of equal per capita results) are not motivated sufficiently in the text. I was unsure why the authors chose to present these results and suggest the authors add more text before the results section to justify why they have presented them.  
+
+<|ref|>text<|/ref|><|det|>[[118, 778, 522, 795]]<|/det|>
+Thank you. We added the following explanations.  
+
+<|ref|>text<|/ref|><|det|>[[118, 809, 857, 841]]<|/det|>
+First we show equal per capita allocations as a point of comparison, revealing some lack of ambition even in the absence of equity considerations.  
+
+<|ref|>text<|/ref|><|det|>[[178, 855, 875, 904]]<|/det|>
+"In addition to capability and responsibility, equality is the third equity principle described in the IPCC AR517,22. IPCC reports do not present equity- based emissions allocations since AR5, despite available studies and its importance for courts of
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[179, 85, 875, 135]]<|/det|>
+law12. The egalitarian approach modelled as equal per capita emissions is not directly anchored in the Paris Agreement or international environmental law6, but can reveal the inequalities of emissions spaces claimed through NDCs."  
+
+<|ref|>text<|/ref|><|det|>[[118, 149, 875, 248]]<|/det|>
+Secondly, we explain how modelling discontinuous emissions allocations changes not simply countries' fair share, but also the ranking of countries in terms of the relative share of global mitigation finance that they could be expected to contribute to. It is not only important in terms of which country is doing enough or not, but also to determine the distribution of expected financial effort across countries. We modify the introductory text of Figure 4's results to:  
+
+<|ref|>text<|/ref|><|det|>[[178, 262, 877, 459]]<|/det|>
+"We show that modelling continuous emissions allocations, here exemplified by adding a 20- year transition period, affects countries' emissions gaps between their allocations and NDCs, unequally in 2030. Compared to a traditional continuous approach, applying a discontinuous approach implies here a much higher obligation to contribute to international finance for all G20 countries except from India. In terms of the ranking of the emissions gap between NDC and allocation, assuming a transition period benefits Canada and Australia (moving down 9 positions), the USA and South Korea (each 8 positions). This shows that continuous pathways 'reward' such countries for their history of comparably low mitigation efforts, lowering their implied contribution to international climate finance. Other countries, including China, Türkiye, South Africa and the EU move down in the ranking of the ambition gaps, when removing the transition period."  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 490, 210, 505]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[117, 522, 860, 655]]<|/det|>
+Gignac, R., & Matthews, H. D. (2015). Allocating a \(2^{\circ}\mathrm{C}\) cumulative carbon budget to countries. Environmental Research Letters, 10(7), 075004. https://doi.org/10.1088/1748- 9326/10/7/075004Guivarch, C., Le Gallic, T., Bauer, N., Fragkos, P., Huppmann, D., Jaxa- Rozen, M., Keppo, I., Kriegler, E., Krisztin, T., Marangoni, G., Pye, S., Riahi, K., Schaeffer, R., Tavoni, M., Trutnevytse, E., van Vuuren, D., & Wagner, F. (2022). Using large ensembles of climate change mitigation scenarios for robust insights. Nature Climate Change, 12(5), Article 5. https://doi.org/10.1038/s41558- 022- 01349-x  
+
+<|ref|>text<|/ref|><|det|>[[118, 667, 877, 734]]<|/det|>
+Kikstra, J. S., Nicholls, Z. R., Smith, C. J., Lewis, J., Lamboll, R. D., Byers, E., Sandstad, M., Meinshausen, M., Gidden, M. J., Rogelj, J., & others. (2022). The IPCC Sixth Assessment Report WGIII climate assessment of mitigation pathways: From emissions to global temperatures. Geoscientific Model Development, 15(24), 9075- 9109.  
+
+<|ref|>text<|/ref|><|det|>[[118, 763, 848, 813]]<|/det|>
+Schleussner, C.- F., Ganti, G., Rogelj, J., & Gidden, M. J. (2022). An emission pathway classification reflecting the Paris Agreement climate objectives. Communications Earth & Environment, 3(1), https://doi.org/10.1038/s43247- 022- 00467- w  
+
+<|ref|>text<|/ref|><|det|>[[118, 843, 876, 909]]<|/det|>
+van den Berg, N. J., van Soest, H. L., Hof, A. F., den Elzen, M. G. J., van Vuuren, D. P., Chen, W., Drouet, L., Emmerling, J., Fujimori, S., Hohne, N., Koberle, A. C., McCollum, D., Schaeffer, R., Shekhar, S., Vishwanathan, S. S., Vrontisi, Z., & Blok, K. (2020). Implications of various effort- sharing approaches for national carbon budgets and emission pathways.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 85, 792, 103]]<|/det|>
+Climatic Change, 162(4), 1805–1822. https://doi.org/10.1007/s10584-019-02368-y
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[119, 84, 358, 101]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[119, 117, 430, 133]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[117, 149, 878, 315]]<|/det|>
+In the previous round, I noted that the presentation of the analysis and the language needed considerable improvement, and also recommended to have the whole text copy- edited. I regret to say that the improvement on these issues has not been satisfactory. It does not appear as the text has been copy- edited. It should not be the job of the reviewers to point out simple mistakes and help improve the language. While the authors have corrected the specific mistakes I pointed out, there are still many language problems, including in the newly introduced text. Some appear as sloppy mistakes, such as incomplete sentences, while a more extensive problem is lack of clear and structured presentation. I believe the underlying model development would be a valuable addition to the literature, but I deem the progress on its presentation from the first version as insufficient to warrant another 'revise and resubmit'.  
+
+<|ref|>text<|/ref|><|det|>[[119, 344, 855, 394]]<|/det|>
+We appreciate that improvements to the text were necessary and are thankful that you took the time to go over it one more time. We also apologise the display issues due to the conversion of the word document into a pdf file (in particular figure captions).  
+
+<|ref|>text<|/ref|><|det|>[[119, 408, 840, 441]]<|/det|>
+We now copy- edited the entire manuscript, reorganised sections for clarity, and removed redundant statements.  
+
+<|ref|>text<|/ref|><|det|>[[119, 472, 297, 487]]<|/det|>
+Some concrete issues:  
+
+<|ref|>text<|/ref|><|det|>[[118, 488, 875, 570]]<|/det|>
+Re. my first point that the two fairness approaches are difficult to understand, which was also brought up by R2: The motivation for including two different approaches is still not clear to me. The new text is quite technical. Do the approaches reflect different ethical assumptions, or different approaches to a more technical modeling choice for which there is no clear criterion for choosing one over the other?  
+
+<|ref|>text<|/ref|><|det|>[[118, 599, 777, 616]]<|/det|>
+Thank you. There are indeed several motivations behind these modelling choices.  
+
+<|ref|>text<|/ref|><|det|>[[175, 630, 857, 664]]<|/det|>
+1) combining capability and responsibility without relying on averages or statistical combinations,  
+
+<|ref|>text<|/ref|><|det|>[[178, 678, 875, 728]]<|/det|>
+2) differentiating the allocation of positive and negative emissions, to complement the study of Fyson et al. that focused on the allocation of global negative emissions only (with either responsibility or capability)  
+
+<|ref|>text<|/ref|><|det|>[[118, 741, 861, 824]]<|/det|>
+Given these choices, we chose simple allocation methods to represent the capability and the responsibility principles. We model the two manners to combine the responsibility and capability allocations of positive and negative emissions separately. We explain the differences (e.g., Approach 1 achieves equal cumulative per capita emissions) but do not suggest that one should be used over the other.  
+
+<|ref|>text<|/ref|><|det|>[[118, 837, 736, 854]]<|/det|>
+We have updated the manuscript to clarify the rationale of the methodology:  
+
+<|ref|>text<|/ref|><|det|>[[178, 868, 841, 902]]<|/det|>
+"Here we quantify two sets of emissions trajectories immediately based on equity principles and that do not start at current emissions levels (see Methods). The two
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[178, 84, 870, 299]]<|/det|>
+methods combine the equity principles of capability and responsibility<sup>1</sup> to reflect the principles of the UNFCCC and the Paris Agreement, notably CBDR- RC. The literature suggests several approaches, conceptual or statistical, for combining different equity principles into a single allocation method (see Discussion). Here we apply each of the two equity principles to allocate global positive or negative emissions separately. This differentiated treatment of negative emissions extends a study from Fyson et al.<sup>2</sup> that allocated negative emissions only, based on responsibility or capability. Fyson et al. explain that obligations to deliver negative emissions require uncertain technologies made necessary because of insufficient global emissions reductions to date. That study alone could not be used to inform economy- wide emissions targets, and thus not assess the ambition of NDCs, as it only allocated negative emissions and “assume[d] that positive emissions follow least- cost pathways (that is, no equity principle is applied to gross emissions)”<sup>2</sup>.  
+
+<|ref|>text<|/ref|><|det|>[[118, 328, 820, 361]]<|/det|>
+The figures and captions appear in a mess. The same figures appear on multiple pages, sometimes with and sometimes without captions.  
+
+<|ref|>text<|/ref|><|det|>[[118, 391, 870, 425]]<|/det|>
+Indeed, our apologies for that. I believe that the figure referencing system of word is not well handled by the pdf conversion tool on the journal’s platform.  
+
+<|ref|>text<|/ref|><|det|>[[120, 454, 850, 472]]<|/det|>
+The discussion has no structure, which makes it difficult to follow. There is no conclusion.  
+
+<|ref|>text<|/ref|><|det|>[[118, 487, 864, 570]]<|/det|>
+We have thoroughly revised the discussion that now focusses on 1) how continuity assumptions are present in the literature, 2) how our combination of equity principles compares to the literature, 3) how our allocation results compare to the literature. Note that Nature Communication’s format does not allow for a conclusion section. The last paragraph was revised and clearly introduced as a conclusion.  
+
+<|ref|>text<|/ref|><|det|>[[119, 584, 456, 600]]<|/det|>
+We introduce the discussion section with:  
+
+<|ref|>text<|/ref|><|det|>[[178, 615, 860, 664]]<|/det|>
+“Here we discuss how this study’s modelling choices compare to the literature regarding the continuity assumption and the combination of equity principles. Then, we compare results.”  
+
+<|ref|>text<|/ref|><|det|>[[118, 679, 822, 712]]<|/det|>
+The methods section appears to contain considerable overlap with the main text (partly reflecting that the main text is very technical).  
+
+<|ref|>text<|/ref|><|det|>[[118, 726, 850, 758]]<|/det|>
+Thank you for the thorough review. We removed content from the manuscript that already appeared in the methods. Additionally, we moved some content to the methods.  
+
+<|ref|>text<|/ref|><|det|>[[118, 774, 876, 905]]<|/det|>
+Re. my suggestion to “spend some more words on the criticism past studies have received for including transition periods (e.g., Kartha et al (2018)) and the weak ethical basis for such periods (e.g., Flerbaej et al (2014)). On the other hand, possible drawbacks of removing transition periods could also be discussed. For example, this means that reduced and avoided emissions are treated symmetrically, but the costs of reducing emissions are likely larger than the costs of avoiding future increases. The argument in the paper and in the literature against transition periods relies on emissions trading (ITMOs), but the imperfections of current institutions for this should be mentioned. Perhaps also refer to Knight (2013) for a defense of
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[119, 85, 333, 100]]<|/det|>
+moderate grandfathering."  
+
+<|ref|>text<|/ref|><|det|>[[119, 101, 880, 134]]<|/det|>
+The authors responded 'Thank you for the reference. We added it in the following sentence as follows with references to Fleurbaey and Knight:  
+
+<|ref|>text<|/ref|><|det|>[[119, 134, 880, 183]]<|/det|>
+"Considering continuous emissions trajectories that look realistic22 implies that present- day levels of domestic emissions are an acceptable starting point in terms of mitigation effort with a utilitarian perspective25."  
+
+<|ref|>text<|/ref|><|det|>[[119, 183, 816, 216]]<|/det|>
+I would have liked to see a more engagement with the suggestion than just adding one sentence.  
+
+<|ref|>text<|/ref|><|det|>[[120, 232, 430, 249]]<|/det|>
+Thank you for standing for your point.  
+
+<|ref|>text<|/ref|><|det|>[[118, 262, 877, 410]]<|/det|>
+In the fair- share literature, the allocation of emissions space does not distinguish treat reduced and avoided emissions similarly, regardless of whether a transition period is used or not. A transition period does not account for reduction potential, as opposed to IAM scenarios for example. For this reason, we do not find it possible to relate the inclusion of a transition period to the distinction between reduced and avoided emissions. However, the allocation of emissions space into fair- shares may result in 'hot air' as the emissions allocation of a country may exceed the emissions of its business as usual scenario. While a transition period mitigated this effect, it does not avoid hot air. Even a pure grandfathering approach could theoretically result in hot air allocations.  
+
+<|ref|>text<|/ref|><|det|>[[119, 424, 833, 457]]<|/det|>
+We raise the issue of hot air, made more visible through discontinuous allocations in the following paragraph:  
+
+<|ref|>text<|/ref|><|det|>[[179, 471, 874, 668]]<|/det|>
+"The near- term allocation of some countries, mostly sub- Saharan countries, may exceed their current emissions and business- as- usual trajectory beyond 2030, implying mitigation efforts only later<sup>3</sup>. However, staying within such decreasing allocations beyond 2030 implies immediate investments, possibly with international support. International support can enable recipient countries to implement mitigation measures in line with the underlying global socio- economic scenario in the near term. Approach 2 uses allocations inversely proportional to GDP per capita<sup>4,5</sup> (see methods), resulting in high emissions allocations compared to current emissions and allocations based on business- as- usual trajectories<sup>3,6</sup> for countries with very low GDP per capita (e.g., Ethiopia, Democratic Republic of Congo). These allocations theoretically imply financial transfers that may go beyond needs- based considerations and contribute to poverty reduction through climate action<sup>7</sup>.  
+
+<|ref|>text<|/ref|><|det|>[[118, 712, 867, 844]]<|/det|>
+Regarding the arguments raised by Knight for the justification of a moderate grandfathering, we discuss the realist justification and the utilitarian justification, with references to Knight and Fleurbaey. Indeed, all international agreements reflect some inertia that give a realist justification to grandfathering. However, we focus on the modelling of equity- based allocations to inform processes not limited to negotiating agreements, and that include climate litigation. We seek to rely directly on concepts of international law and the Paris Agreement that do not support grandfathering. We hope that the following modifications clarify this point:  
+
+<|ref|>text<|/ref|><|det|>[[179, 858, 877, 907]]<|/det|>
+"The legacy influence of current emissions levels on near- term emissions allocations is described here as a 'grandfathering' effect<sup>8</sup>. This grandfathering influence on equity- based emissions allocation is strongest in the near term and increasingly affects
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[178, 84, 866, 347]]<|/det|>
+the ambition assessment of NDCs in 2030. As we near 2030, a given NDC's emissions target will be closer and closer to a continuous emissions allocation that is iteratively updated (Figure 1). The grandfathering allocation is criticized for its lack of ethical basis \(^{9 - 11}\) and has been shown to penalize the poorest countries \(^{12}\) as it preserves a status- quo, including current inequalities. Prior to the Paris Agreement, a study highlighted the value of a 'moderate grandfathering' \(^{13}\) , from a political theory perspective, with a realist justification for negotiations and a utilitarian justification. Indeed, the pledges of many high- emitters only align with a grandfathering allocation \(^{4}\) . However, the IPCC has highlighted the need for a fair distribution of mitigation efforts, excluding grandfathering, in order to achieve an effective global agreement on emissions reductions \(^{1,10}\) . Likewise, recent reports of scientific advisory bodies have disapproved grandfathering when presenting fair- share emissions allocation \(^{14,15}\) . The Paris Agreement now requires NDCs of the 'highest possible ambition' that reflect equity. A recent study \(^{8}\) described grandfathering allocations as not in line with international law. It identified that all continuous allocations entail elements of grandfathering but did not offer a solution.  
+
+<|ref|>text<|/ref|><|det|>[[118, 360, 835, 394]]<|/det|>
+Regarding the justification of utilitarianism, we added the following sentence in the next paragraph:  
+
+<|ref|>text<|/ref|><|det|>[[178, 407, 870, 457]]<|/det|>
+"The utilitarian justification \(^{13}\) for a moderate grandfathering relies domestic mitigation costs and is no longer relevant when allocations can be traded to achieve a globally cost- effective pathway \(^{10}\) ."  
+
+<|ref|>text<|/ref|><|det|>[[118, 470, 877, 653]]<|/det|>
+This interpretation is based on p. 417 of Knight's analysis \(^{13}\) stating: "On welfare views such as utilitarianism, the relevant costs are welfare costs, rather than the monetary marginal abatement costs familiar from economics. According to utilitarianism, welfare costs are more important the greater they are. It might be claimed that high emitters face high marginal abatement costs. The marginal abatement cost is the welfare cost of one extra unit of emissions reduction. Thus, the main claim of the marginal cost argument is: one extra unit of emission reductions from a baseline of actual prior emissions decreases welfare to a greater extent where it is assigned to a high emitter than where it is assigned to a low emitter. If this is correct, utilitarianism will maintain that high emitters have greater entitlements, as this will save them – and the global economy – from the severe effects of deeper cuts (cf. Wesley and Peterson 1999, p. 186)."  
+
+<|ref|>text<|/ref|><|det|>[[118, 666, 657, 683]]<|/det|>
+We also highlight the possible important shortcoming of Article 6:  
+
+<|ref|>text<|/ref|><|det|>[[178, 696, 857, 779]]<|/det|>
+"As a novel mechanism, the international trading of mitigation outcomes raises implementation issues regarding the additionality of the finance and of the funded mitigation measures. Scrutiny will be needed to ensure the integrity of mitigation measures under Article 6 whose implementation rules were just adopted at COP29, with safeguards on human rights and the additionality of emissions reductions \(^{16 - 18}\) ."  
+
+<|ref|>text<|/ref|><|det|>[[118, 826, 431, 842]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 858, 636, 875]]<|/det|>
+Thank you for the opportunity to review the revised manuscript.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 84, 825, 118]]<|/det|>
+I have reviewed the revisions made by the authors in response to my previous round of review comments. I am satisfied with these revisions.  
+
+<|ref|>text<|/ref|><|det|>[[118, 131, 426, 148]]<|/det|>
+Thank you for your time and reviews.  
+
+<|ref|>text<|/ref|><|det|>[[118, 181, 215, 197]]<|/det|>
+References:  
+
+<|ref|>text<|/ref|><|det|>[[115, 197, 870, 900]]<|/det|>
+1. Clarke, L. et al. Chapter 6 Assessing Transformation Pathways. In: IPCC AR5 WGIII. 413–510 (2014).  
+2. Fyson, C. L., Baur, S., Gidden, M. & Schleussner, C. F. Fair-share carbon dioxide removal increases major emitter responsibility. Nature Climate Change 10, 836–841 (2020).  
+3. van den Berg, N. J. et al. Implications of various effort-sharing approaches for national carbon budgets and emission pathways. Climatic Change 162, 1805–1822 (2020).  
+4. Robiou du Pont, Y. et al. Equitable mitigation to achieve the Paris Agreement goals. Nature Climate Change 7, 38–43 (2017).  
+5. Jacoby, H. D., Babiker, M. H., Paltsev, S. & Reilly, J. M. Sharing the Burden of GHG Reductions. MIT Joint Program on the Science and Policy of Global Change 1–28 https://globalchange.mit.edu/publication/14428 (2008).  
+6. Holz, C., Kartha, S. & Athanasiou, T. Fairly sharing 1.5: national fair shares of a 1.5 °C-compliant global mitigation effort. International Environmental Agreements: Politics, Law and Economics 18, 117–134 (2017).  
+7. Budolfson, M. B. et al. Utilitarian benchmarks for emissions and pledges promote equity, climate and development. Nature Climate Change 11, 827–833 (2021).  
+8. Rajamani, L. et al. National ‘fair shares’ in reducing greenhouse gas emissions within the principled framework of international environmental law. Climate Policy 21, 1–22 (2021).  
+9. Caney, S. Justice and the distribution of greenhouse gas emissions. Journal of Global Ethics 5, 125–146 (2009).  
+10. Fleurbaey, M. et al. Chapter 4. Sustainable Development and Equity. In: Climate Change 2014: Mitigation of Climate Change. Contribution of Working Group III to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change 283–350 (2014).  
+11. Dooley, K. et al. Ethical choices behind quantifications of fair contributions under the Paris Agreement. Nature Climate Change 11, (2021).  
+12. Kartha, S. et al. Cascading biases against poorer countries. Nature Climate Change 8, 348–349 (2018).  
+13. Knight, C. What is grandfathering? Environmental Politics (2013) doi:10.1080/09644016.2012.740937.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 82, 877, 155]]<|/det|>
+14. European Scientific Advisory Board on Climate Change. Scientific Advice for the Determination of an EU-Wide 2040 Climate Target and a Greenhouse Gas Budget for 2030-2050. https://doi.org/10.2800/609405 (2023).  
+
+<|ref|>text<|/ref|><|det|>[[118, 165, 880, 237]]<|/det|>
+15. A Justified Ceiling to Germany's CO₂ Emissions: Questions and Answers on Its CO₂ Budget. https://www.umweltrat.de/SharedDocs/Downloads/EN/04_Statements/2020_2024/2022_09_The_CO2_bud get_approach.html (2022).  
+
+<|ref|>text<|/ref|><|det|>[[118, 247, 833, 292]]<|/det|>
+16. COP29: Key outcomes agreed at the UN climate talks in Baku. Carbon Brief https://www.carbonbrief.org/cop29-key-outcomes-agreed-at-the-un-climate-talks-in-baku/#6 (2024).  
+
+<|ref|>text<|/ref|><|det|>[[118, 302, 866, 346]]<|/det|>
+17. Songwe, V., Stern, N. & Bhattacharya, A. Finance for Climate Action: Scaling up Investment for Climate and Development. (2022).  
+
+<|ref|>text<|/ref|><|det|>[[118, 356, 878, 428]]<|/det|>
+18. Haynes, R. & Benjamin, L. Ambition-raising and ambition-reducing features of the Paris Agreement. in Research Handbook on the Law of the Paris Agreement (ed. Zahar, A.) 126-142 (Edward Elgar Publishing, 2024). doi:10.4337/9781800886742.00012.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Transparent Peer Review file__e8edc04180bfdf0eefaef4278be6e83a9bef2cad24998f6d65e53837953baf6f/images_list.json

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

peer_reviews/supplementary_0_Transparent Peer Review file__e8edc04180bfdf0eefaef4278be6e83a9bef2cad24998f6d65e53837953baf6f/supplementary_0_Transparent Peer Review file__e8edc04180bfdf0eefaef4278be6e83a9bef2cad24998f6d65e53837953baf6f.mmd

@@ -0,0 +1,456 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+# Transferable polychromatic optical encoder for neural networks  
+
+Corresponding Author: Professor Arka Majumdar  
+
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+Version 0:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+This work demonstrates some results by developing an optical encoder that can perform convolution simultaneously in three color channels during the image capture, implementing several initial convolutional layers of a ANN. However, although the optical encoding can decrease computational operations, the classification accuracy ( \(\sim 73.2\%\) ) is too low for real applications. I donot think this manuscript reach the level of Nature Communications. Here are comments.  
+
+1. Many similar hybrid optical/digital architectures have been demonstrated. I donot find something new for the current one. Is there any new for the design? Just because they use metasurfaces? If fact, the testing accuracy drops significantly with their meta-optics.  
+
+2. It is strange that the authors said their architectures are integrated. But they compare the classification accuracy with the free space optical systems. It is possible to improve classification accuracy? How? At least, it should be comparable with integrated optical neural network architectures.  
+
+3. Low energy consumption may one of the advantages of the developed architectures, it should be reasonable due to the less operations. However, it looks meaningless as they sacrifice the classification accuracy. In addition, they overlooked the energy consumption of light source. Can they estimate the real energy consumption if the really mean "energy consumption"?  
+
+4. Why the classification accuracy drops more significant for the CIFAR-10 dataset compared to the MNIST dataset? How about the classification accuracy for the MNIST dataset?  
+
+5. Physical sizes about the metasurfaces should be provided for the testing samples.  
+
+Reviewer #2  
+
+(Remarks to the Author)  
+
+This work presents a novel approach for a hybrid optical/digital neural network, performing the computationally expensive convolution operation in optics. The manuscript introduces an innovative design that demonstrates significant computational efficiency while maintaining competitive classification accuracy. The work can be considered for publication; however, the following concerns should be addressed:  
+
+Phase Coverage of Meta- Atoms: Based on Figure 2b, the phase coverage of the meta- atoms appears to be around 1 radian, even when assuming impractically small feature sizes. For designing an effective metasurface, a phase coverage is typically required. The authors should clarify how they achieve the necessary phase modulation with this limited range or discuss the implications of this constraint on device performance.  
+
+Kernel- Image Interaction: The interaction mechanism between the kernels and the image is unclear. The kernels are spatially distributed in different locations, yet they seem to interact with the entire image. The authors should elaborate on how this is physically implemented, particularly addressing whether there are optical multiplexing effects or specific alignment strategies that enable this global interaction.  
+
+Scope for Improvement in Hybrid CNN: The authors claim that "Our hybrid optical/digital CNN can be further improved by using complex meta- atoms to reproduce better PSFs optically." However, the current performance of the hybrid neural network is already close to that of the compressed CNN, suggesting limited room for further enhancement. It would be beneficial for the authors to expand on how complex meta- atoms could offer substantial performance gains and to discuss
+
+<--- Page Split --->
+
+other potential strategies for improving compressed CNN performance, perhaps focusing on advanced knowledge distillation or transfer learning techniques.  
+
+Generalization Performance: While the hybrid system achieves reasonable transferability from CIFAR- 10 to High- 10, the performance drop is notable. The authors could discuss potential design modifications or training strategies that might enhance the generalization capability of the optical encoder without the need for extensive backend retraining.  
+
+Physical Realizability of PSFs: The paper acknowledges discrepancies between the ground- truth and experimentally measured PSFs (Figure 3c), partly due to fabrication imperfections and the inherent limitations of phase control across RGB channels. A more detailed analysis of how these imperfections affect system robustness and potential mitigation strategies would strengthen the paper.  
+
+Energy Efficiency vs. Sensor Overhead: The authors highlight a significant reduction in computational energy consumption but mention that the optical encoder requires capturing more pixels than the original CNN, increasing the sensor's energy consumption. A deeper quantitative comparison of total system energy consumption, considering both sensor and backend, would provide a clearer picture of practical efficiency gains.  
+
+Scalability and Practical Deployment: The current implementation focuses on relatively small datasets (CIFAR- 10, High- 10). The scalability of the system to more complex, real- world datasets (e.g., full ImageNet) remains unclear. It would be valuable if the authors discussed potential bottlenecks in scaling the optical encoder, such as fabrication complexity, alignment issues, or increased noise sensitivity.  
+
+Typos and Minor Errors: There are several typographical errors throughout the manuscript that need correction. A thorough proofreading is recommended to improve the manuscript's readability.  
+
+Overall, this paper introduces an exciting advancement in hybrid optical/digital neural networks, but addressing the above concerns will significantly enhance its clarity, robustness, and impact.  
+
+## Reviewer #3  
+
+(Remarks to the Author) This paper introduces a hybrid optical/digital neural network architecture that leverages polychromatic meta- optics to perform convolutional operations during image capture, reducing computational load. The system achieves \(73.17\%\) accuracy on CIFAR- 10 and demonstrates transferability to an ImageNet subset (High- 10) with moderate accuracy. The work addresses critical challenges in real- time, energy- efficient computer vision and offers a novel integration of optical encoding with digital backends. However, I feel the authors should clarify a few points below before being considered for publication in Nature Communications.  
+
+1. Regarding the accuracy Trade-offs, the accuracy drop ( \(\sim 8\%\) from AlexNet on CIFAR-10, \(\sim 25\%\) on High-10) may limit adoption in "safety-critical" applications. Could the authors describe how one could potentially achieve better accuracy?  
+
+2. Could the author comment on fabrication Limitations: The impact of PSF discrepancies (e.g., \(\eta = 0.56\) for green) is under- analysed. Quantifying fabrication tolerances and robustness to misalignment would strengthen the results.  
+
+Scalability: The system uses 32 meta- optics for 16 kernels. How does this scale to deeper networks or larger datasets? A discussion on physical size constraints is missing.  
+
+Calibration Layer Dependency: The reliance on a digital calibration layer to correct optical imperfections partially offsets the analog advantage. Clarify if this layer adds computational overhead.  
+
+2. Could the authors comment on real-World applicability by testing under variable illumination/backgrounds (not just controlled lab conditions) as this way would better demonstrate practical utility?  
+
+3. The proxy function(i.e., Eq. 1) ignores resonance effects. How does this approximation affect PSF fidelity, especially for green ( \(\eta = 0.56\) )? The authors are suggested to include an error analysis across all kernels.  
+
+4. The authors wrote "Thus the energy consumption for a single object classification task for the hybrid CNN is about 150nJ, which is more than four orders of magnitude smaller than that of the original CNN." Does the backend energy savings justify increased sensor costs?  
+
+5. The added "transfer learning layer" is not detailed in the main text. Could the authors provide architecture specifics and ablation studies?  
+
+6. By comparing accuracy/MAC reductions with recent works, could the author highlight how polychromatic encoding advances the field?  
+
+7. Since the authors used the system to processes static images, could they discuss feasibility for video streams and temporal feature extraction?
+
+<--- Page Split --->
+
+Version 1:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author) As the authors addressed all my concerns, I recommend it can be accepted for publication.  
+
+Reviewer #3  
+
+(Remarks to the Author)  
+
+The authors have properly addressed all my comments from the previous round of review. I now recommend accepting this manuscript for publishing in Nature Communications.  
+
+Open Access This Peer Review 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 cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+The images or other third party material in this Peer Review 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 https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+## Reviewer #1 (Remarks to the Author):  
+
+This work demonstrates some results by developing an optical encoder that can perform convolution simultaneously in three color channels during the image capture, implementing several initial convolutional layers of a ANN. However, although the optical encoding can decrease computational operations, the classification accuracy ( \(\sim 73.2\%\) ) is too low for real applications. I do not think this manuscript reaches the level of Nature Communications. Here are comments.  
+
+Response: We thank the reviewer for carefully reviewing the paper and providing valuable feedback. We have provided a point- by- point response and modified the paper accordingly. We respectfully disagree with the comments that our classification accuracy is too low. Our accuracy on CIFAR- 10 is better than any other free- space optical implementation of neural networks. And there are usages of a moderate accuracy AI (but with high energy efficiency and low latency) for non- safety- critical situations as we describe in the Discussion- Applications section. We emphasize that most optical neural networks work on MNIST or some toy problems [1- 5]. Majority of works cannot handle a colorful dataset like CIFAR- 10, while polychromatic information contains significant information in a real scenario. Moreover, we report a system level benefit in energy consumption (more than two orders of magnitude), while none of the other works describe the system level energy consumption including optoelectronic devices (e.g., laser, modulators, etc). In these regards, we believe that our work significantly extends the capability of existing optical neural networks.  
+
+[1] Chen, Y., Nazhamaiti, M., Xu, H., Meng, Y., Zhou, T., Li, G., Fan, J., Wei, Q., Wu, J., Qiao, F., et al.: All- analog photoelectronic chip for high- speed vision tasks. Nature 623(7985), 48- 57 (2023) [2] Lin, X., Rivenson, Y., Yardimci, N.T., Veli, M., Luo, Y., Jarrahi, M., Ozcan, A.: All- optical machine learning using diffractive deep neural networks. Science 361(6406), 1004- 1008 (2018) [3] Chen, Z., Sludds, A., Davis III, R., Christen, I., Bernstein, L., Ateshian, L., Heuser, T., Heermeier, N., Lott, J.A., Reitzenstein, S., et al.: Deep learning with coherent vcsel neural networks. Nature Photonics 17(8), 723- 730 (2023) [4] Xia, F., Kim, K., Eliezer, Y., Han, S., Shaughnessy, L., Gigan, S., Cao, H.: Non- linear optical encoding enabled by recurrent linear scattering. Nature Photonics, 1- 9 (2024) [5] Xue, Z., Zhou, T., Xu, Z., Yu, S., Dai, Q., Fang, L.: Fully forward mode training for optical neural networks. Nature 632(8024), 280- 286 (2024)  
+
+1. Many similar hybrid optical/digital architectures have been demonstrated. I do not find something new for the current one. Is there any new for the design? Just because they use metasurfaces? In fact, the testing accuracy drops significantly with their meta-optics.  
+
+Response: Thank you for the valuable comment. However, we respectfully disagree with the reviewer. First of all, we would like to emphasize that our work is the first optical/digital architecture which utilizes polychromatic meta-optics (a single metasurface performs different convolutions for three different colors: RGB), and this is an important feature to minimize the optical system as well as the system level energy consumption as we need less number of pixels to capture the image. This property is very difficult to realize using any other engineered optics, and critically relies on the complex dependence of the optical phase on the geometric parameters of the sub- wavelength meta- atoms. This is a completely new design, and indeed this can be achieved because we used sub- wavelength diffractive metasurface. Secondly, we will encourage the reviewers to provide references on hybrid architecture that has worked with complex datasets like CIFAR- 10. Optical structures have predominantly focused on MNIST, which is linearly separable and as such is extremely simple. Bringing optical structures to CIFAR- 10, and a subset of ImageNet (as we demonstrated) is
+
+<--- Page Split --->
+
+non- trivial.  
+
+The Reviewer also mentioned the significant testing accuracy drop during the optical implementation. However, accuracy drop is relatively small, from \(76.59\%\) to \(72.06\%\) (73.17% after retraining the backend), compared to the digital compression variant. Furthermore, our result is state- of- the- art compared to other works published very recently [5- 8]. Again, we emphasize that demonstration with CIFAR- 10 and ImageNet is significantly more complex than with MNIST datasets, and to the best of our knowledge, our work is at par, if not better than most existing demonstrations of hybrid optical networks on CIFAR- 10. We explicitly describe the advantages of our system in energy consumption (more than two orders of magnitude), while none of the other works describe the system level energy consumption. Reported works that we are aware of have ignored the power of the camera, or intensifier or some other components. We will be happy to modify our claim if the reviewer can provide a reference that refutes it. In fact, the reviewer is correct to say that there is indeed a large body of work on the optical neural network, however, none of the works have shown system level energy consumption improvement (including both optoelectronic and digital energy consumption) as we show in our work. Majority of the work shows improvement in linear operations, but when the whole system is considered, it is not clear if the energy benefits are carried through.  
+
+[5] Xue, Z., Zhou, T., Xu, Z., Yu, S., Dai, Q., Fang, L.: Fully forward mode training for optical neural networks. Nature 632(8024), 280- 286 (2024) [6] Huo, Y., Bao, H., Peng, Y., Gao, C., Hua, W., Yang, Q., Li, H., Wang, R., Yoon, S.- E.: Optical neural network via loose neuron array and functional learning. Nature Communications 14(1), 2535 (2023) [7] Rahman, M.S.S., Ozcan, A.: Time- lapse image classification using a diffractive neural network. Advanced Intelligent Systems 5(5), 2200387 (2023) [8] Wei, K., Li, X., Froech, J., Chakravarthula, P., Whitehead, J., Tseng, E., Majumdar, A., Heide, F.: Spatially varying nanophotonic neural networks. Science Advances 10(45), 0391 (2024)  
+
+2. It is strange that the authors said their architectures are integrated. But they compare the classification accuracy with the free space optical systems. It is possible to improve classification accuracy? How? At least, it should be comparable with integrated optical neural network architectures.  
+
+Response: Thank you for the valuable comment. We claimed our system as a free- space optical encoder, not as integrated photonics. What might have been the source of confusion, is that we are using solid- state meta- optics, which can be manufactured using semiconductor manufacturing. They can also be integrated with a camera. Thus they are indeed an integrated system, but not a photonic integrated circuit (where the light flows via waveguide in the plane of the chip). In our system, the light travels perpendicular to the chip.  
+
+Free- space integrated systems can handle two- dimensional data and can directly interface with ambient incoherent light, unlike photonic integrated circuits. Our demonstrated free- space encoder can be easily integrated with conventional imaging systems with minimized modification of the system hardware. We did not compare our system with photonic integrated circuits since our performance in terms of space- bandwidth product is significantly better than those circuits.  
+
+For the other comment of the Reviewer about "improving classification accuracy", we can use a stronger knowledge distillation loss. Our current version is based on Geoffrey Hinton's 'Distilling the Knowledge in a Neural Network.' Recently, the community has introduced more advanced versions of knowledge distillation loss [9- 11]. Another possibility is that we can increase the number of kernels to enhance optical feature engineering; however it will increase the number of pixels required for image
+
+<--- Page Split --->
+
+capturing, then increase the energy consumption for the optical side. Having said that, as explained above, our reported accuracy is better than current state- of- the- art on complex datasets like CIFAR- 10 using hybrid optical/digital neural networks. We added further information in the manuscript under Discussion- Opportunities for improvement addressing improvement in accuracy:  
+
+"While the current hybrid CNN achieves competitive performance, a noticeable gap ( \(\sim 4.4\%\) ) remains compared to the compressed CNN. This discrepancy can be mitigated by employing a more sophisticated PSF design, which enhances optical processing capabilities and reduces information loss. By leveraging complex meta- atoms with improved phase and amplitude control, the optical system can more accurately approximate ideal convolutional operations, thereby closing the performance gap. Additionally, performance discrepancy exists between the original CNN and its compressed counterpart. To address this, we could leverage advanced knowledge distillation techniques to enhance the compressed model's learning efficiency. By integrating both improved PSF design and advanced knowledge distillation methods, our approach can effectively bridge these gaps. We can also increase the number of kernels; however, this will increase the number of pixels required for image capture.  
+
+Figure 3c shows a clear discrepancy between the ground- truth convolutional kernels and the experimentally measured PSFs. Other than fabrication imperfections and optical misalignment, we identify three other reasons for the discrepancy.  
+
+1. Imperfect fitting function for the phase over scatterer: We assumed the scatterers have constant transmission and do not have any resonant features in their relative phases, which is entirely accurate (Fig. 2b).  
+
+2. Limited degree of freedom of the metasurface for multicolor PSFs: We optimized each of the metasurfaces targeting three different PSFs in red, green, and blue colors. The phase profiles of one rectangular scatterer for three different colors are not independent of each other. One solution is to use complex-shaped scatterers or super cells to have more degrees of freedom.  
+
+3. Broadband light sources: We simulated the polychromatic metasurfaces at three discrete wavelengths (450, 532, and 635 nm). However, the OLED pixels have much broader wavelengths. We can reduce this discrepancy if we optimize the metasurface for more representative wavelengths.  
+
+However, the classification accuracy reported here for the CIFAR- 10 dataset is considerable compared to the other reported results in optical neural networks. There will be inevitable imperfection in optical implementation as the spectral information of the scene is always changing depending on the daylight, cloud, aerial, and many other conditions. Our results show that the digital backend can compensate for the non- ideal optical implementation and achieve a relatively high classification accuracy."  
+
+[9] Tian, Y., Krishnan, D., & Isola, P. (2020). Contrastive representation distillation. International Conference on Learning Representations (ICLR). https://arxiv.org/abs/1910.10699 [10] Zagoruyko, S., & Komodakis, N. (2017). Paying more attention to attention: Improving the performance of convolutional neural networks via attention transfer. International Conference on Learning Representations (ICLR). https://arxiv.org/abs/1612.03928 [11] Park, W., Kim, D., Lu, Y., & Cho, M. (2019). Relational knowledge distillation. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 3967- 3976. https://arxiv.org/abs/1904.05068  
+
+3. Low energy consumption may one of the advantages of the developed architectures, it should be reasonable due to the less operations. However, it looks meaningless as they sacrifice the classification accuracy. In addition, they overlooked the energy consumption of light source. Can they
+
+<--- Page Split --->
+
+estimate the real energy consumption if the really mean "energy consumption"?  
+
+Response: We thank the reviewer for the valuable comment. With regards to the usefulness of the hybrid optical/digital architectures given that there is an accuracy drop, we agree that the reduction of accuracy will not be permitted for emergency situations (e.g., self- driving cars). However, computer vision tasks have been processed in many other circumstances, and we specifically point out a particular application field of statistical analysis in the Discussion- Applications section, where the ensemble average can compromise the inaccuracy from a single event.  
+
+In addition, we now include an ablation study comparing the hybrid optical/digital CNN result and the purely digital CNN (which consists only of a calibration layer and convolutional layers) in the last row of Table 1. The number of digital operations is identical in both cases, but the latter shows a significant drop (over \(20\%\) ) in testing accuracy due to the absence of the optical convolutional layer. This result demonstrates the benefits of our hybrid optical/digital architecture, beyond the straightforward relationship between fewer operations and reduced accuracy. The following sentences have been added to the manuscript:  
+
+"Additionally, compared to the backend- only results (Table 1), our hybrid optical/digital architecture achieves significantly higher classification accuracy (over \(20\%\) ), highlighting the crucial role of the optical convolutional encoder."  
+
+Regarding energy consumption, we can disregard the energy usage of light sources since we rely on ambient light (natural light), with no changes to the input image compared to the original CNN. However, in other works utilizing photonic integrated circuits, additional energy is required for light sources, as they depend on a coherent light source distinct from ambient light. The following sentences have been added to the manuscript Discussion- Energy consumption:  
+
+"Since we rely on ambient light - similar to real- world computer vision tasks - we do not require additional energy for the input light source."  
+
+With regards to energy consumption, we calculate both optical and digital energy consumption per image classification task. The color camera we used (Allied Vision Prosilica; GT 1930 C) has a total power consumption of \(3.4\mathrm{W}\) with 50.70 frames per second and 1,936×1,216 color pixels, which ends up with \(28\mathrm{nJ}\) per frame and pixel. Since we captured all \(\sim 2\) million pixels at the same time and cropped the region of interest, the energy consumption for one input image does not differ for the original CNN and hybrid CNN. However, if we can optimize the sensor configuration and number of pixels, we can calculate the minimum required number of pixels for both the original and hybrid CNN, and estimate the energy consumption for those. For the original CNN, we need 32×32 color pixels on camera. And for the hybrid CNN, we need 32×6×6 color pixels on the camera, where the 32 corresponds to number of multiplexed meta- optics and the 6×6 corresponds to number of pixels after the average pooling. We only need 6×6 pixels, not 32×32 pixels when the convolution is already done optically. Thus we estimate that the original CNN and hybrid CNN require an energy of about \(29.1\mu \mathrm{J}\) and \(32.8\mu \mathrm{J}\) , respectively, for the image capturing process per a single image. Next, the energy consumption for the computational backend is much larger for the original CNN compared to the hybrid CNN. For state- of- the- art computational systems, an energy consumption per a single MAC operation is \(1\mathrm{pJ}\) . Thus the energy consumption for a single object classification task for the hybrid CNN is about \(150\mathrm{nJ}\) , which is more than four orders of magnitude smaller than that of the original CNN, \(3.65\mathrm{mJ}\) . While the GPU we used (GeForce RTX 2080 Ti) has much larger energy consumption per a single MAC operation ( \(\sim 7.5\mathrm{pJ}\) ), making the energy consumption for a single object classification tasks for the hybrid CNN and original CNN \(1.13\mu \mathrm{J}\) and \(27.4\mathrm{mJ}\) . Considering the sensor power, the total system level energy consumption for a single object classification task dropped from \(3.68\mathrm{mJ}\) to \(0.03\mathrm{mJ}\) for the state- of- the- art digital processor, while \(27.4\mathrm{mJ}\) to \(34.0\mu \mathrm{J}\) for our GPU. We
+
+<--- Page Split --->
+
+emphasize that more than two orders of magnitude reduction in the system level computer vision tasks clearly provide strong benefits for practical implementation even with a partial accuracy drop for some application fields. We emphasize that to the best of our knowledge, our paper is the first paper which truly calculated the “energy consumption” by considering the sensor/ camera power.  
+
+We explain about the system energy consumption (with more details and an additional Table R1, highlighted in the manuscript) in the section “Energy consumption” in Discussion and add a sentence in Abstract as following:  
+
+“The proposed method can decrease total system-level energy more than two orders of magnitude per a single object classification.”  
+
+Table R1 (also Table 3 in the Revised Manuscript). System level energy consumption analysis per a single image classification task in each step of the computer vision depends on the network architecture.   
+
+<table><tr><td rowspan="2">Network architecture</td><td rowspan="2">Optical frontend</td><td colspan="2">Digital backend</td><td colspan="2">System</td></tr><tr><td>GeForce RTX 2080 Ti</td><td>State-of-the-art GPU</td><td>GeForce RTX 2080 Ti</td><td>State-of-the-art GPU</td></tr><tr><td>Original CNN with optimal camera pixels</td><td>2.91 × 10-5 J</td><td>2.74 × 10-2 J</td><td>3.65 × 10-3 J</td><td>2.74 × 10-2 J</td><td>3.68 × 10-3 J</td></tr><tr><td>Our hybrid optical/digital CNN with optimal camera pixels</td><td>3.28 × 10-5 J</td><td>1.13 × 10-6 J</td><td>1.50 × 10-7 J</td><td>3.40 × 10-5 J</td><td>3.30 × 10-5 J</td></tr></table>  
+
+4. Why the classification accuracy drops more significant for the CIFAR-10 dataset compared to the MNIST dataset? How about the classification accuracy for the MNIST dataset?  
+
+Response: Thank you for the valuable comment. In our previous work [12], the classification accuracy for MNIST dataset dropped from \(96.2\%\) to \(93.4\%\) ( \(= 2.8\%\) drop). In this work, the classification accuracy for CIFAR- 10 dataset drops from \(76.59\%\) to \(72.06\%\) ( \(73.17\%\) after retraining the backend, \(3.42\%\) drop). As the Reviewer mentioned, the classification accuracy drops slightly more in the case of CIFAR- 10 dataset, because it is more difficult and challenging to create multiple color PSFs from a single metasurface. The discrepancy between the ground- truth kernels and experimentally generated kernels for CIFAR- 10 dataset is the main reason why the CIFAR- 10 dataset has further accuracy drop. As we are working with more complex datasets, it is expected that the performance will decrease. MNIST is linearly separable and as such significantly simpler compared to CIFAR- 10. It is indeed expected that most neural networks will perform better when trained and evaluated on MNIST. As we mention in the paper, more complex scatters (not just a rectangular pillar) can adjust this discrepancy and reduce the accuracy drops while implementing the optical encoder experimentally.  
+
+If the Reviewer was referring to the accuracy drop during the knowledge distillation, it is correct that the classification accuracy dropped from \(98.4\%\) to \(96.2\%\) for MNIST dataset, while it drops from \(81.03\%\) to \(76.59\%\) for CIFAR- 10 dataset. This is simply due to the complexity of the dataset, where the accuracy drop is not severe for simple dataset such as monochromatic MNIST dataset.  
+
+[12] Wirth- Singh, A., Xiang, J., Choi, M., Fr'och, J.E., Huang, L., Colburn, S., Shlizerman, E., Majumdar, A.: Compressed meta- optical encoder for image classification. Advanced Photonics Nexus
+
+<--- Page Split --->
+
+## 5. Physical sizes about the metasurfaces should be provided for the testing samples.  
+
+Response: We thank the reviewer for carefully reviewing the paper. The physical size of the metasurface was described in Supplementary materials under "Fabrication of the Meta- optics" section. We agree that some description could be added to the main text, and we added a sentence in the main text where we describe the optical image of the metasurfaces in Figure 3a:  
+
+"Each convolutional meta-optic has a size of \(\sim 940 \times 940 \mu \mathrm{m}^2\) ."
+
+<--- Page Split --->
+
+Reviewer #2 (Remarks to the Author):  
+
+This work presents a novel approach for a hybrid optical/digital neural network, performing the computationally expensive convolution operation in optics. The manuscript introduces an innovative design that demonstrates significant computational efficiency while maintaining competitive classification accuracy. The work can be considered for publication; however, the following concerns should be addressed:  
+
+Response: We thank the reviewer for carefully reviewing the paper and providing valuable feedback. We have provided a point- by- point response and modified the paper accordingly.  
+
+1. Phase Coverage of Meta-Atoms: Based on Figure 2b, the phase coverage of the meta-atoms appears to be around 1 radian, even when assuming impractically small feature sizes. For designing an effective metasurface, a phase coverage is typically required. The authors should clarify how they achieve the necessary phase modulation with this limited range or discuss the implications of this constraint on device performance.  
+
+Response: We very much appreciate the comment by the reviewer. We realized that the phase coverage of the meta-atoms that we note 1 radian is actually \(2\pi\) radians, and corrected the axis of Figure 2b. Our data was normalized but that information was not reflected in the figure.  
+
+2. Kernel-Image Interaction: The interaction mechanism between the kernels and the image is unclear. The kernels are spatially distributed in different locations, yet they seem to interact with the entire image. The authors should elaborate on how this is physically implemented, particularly addressing whether there are optical multiplexing effects or specific alignment strategies that enable this global interaction.  
+
+Response: We appreciate the reviewer's question. The main reason we can interact with the whole image with spatially separated kernels is due to using incoherent light without any directionality (unlike a laser). As we state in the paper, we believe the optical neural network can be only beneficial when the information is already in the optical domain. The case we chose here is to emulate imaging under ambient radiation. Indeed, pictures of the same object can be taken even though the camera is not physically located exactly at the same location. Thus, being compatible with incoherent light, the alignment of single optics with the sensor becomes a non-trivial problem. This is also the reason for placing the kernels in different locations (more quantitative details and limitations are below) see the same image.  
+
+We provide more information on the physical implementation below. We have a total 32 convolutional meta- optics, corresponding to 16 positive and negative digital kernels. All the meta- optics are spatially distributed on a single chip, and either point spread functions (PSFs) or convolved images from all the meta- optics are captured at a color camera simultaneously. When the input light is a laser and a pinhole, we measure the PSFs. And when the input light is an image on an OLED display, we measure the convolved images. For optical convolutional imaging, we capture 32 different convolved colorful images on the camera, which then undergo digital post- processing. In our physical configuration, where the size of the input image is relatively small (32 by 32 pixels), the off- axis spatial variance in the display is not severe. And due to the large distance between the display and the metasurface ( \(\sim 105 \mathrm{mm}\) ) compared to the distance between the metasurface and the sensor ( \(\sim 2.46 \mathrm{mm}\) ), the off- axis spatial variance in the metasurface is also not severe. The relative intensity between each convolutional metasurfaces is adjusted by the calibration layer, which is scene- independent. We would like to emphasize that the spatial variance does not play an important role in our case of
+
+<--- Page Split --->
+
+classification tasks, as demonstrated by a small accuracy drop during the optical implementation, from \(76.59\%\) to \(72.06\%\) (73.17% after retraining the backend). However, those spatial distributions will be an issue for bigger input image size in a future work.  
+
+To clarify how we physically implement our optics, we add the following sentences in the manuscript:  
+
+"Then, we test the polychromatic optical encoder for the CIFAR- 10 dataset. By replacing the pinhole with an organic light- emitting diode (OLED) display, optical convolutional operations between the input image and the PSFs occur with spatially separated meta- optics on a single chip, then captured on a color camera (Fig. 3d). On the color camera, 32 different convolved images are captured and then subjected to digital backend of calibration and fully- connected layers."  
+
+"Because of small size of the input image as well as small distance between the metasurfaces and the camera ( \(\sim 2.46 \mathrm{mm}\) ) compared to the distance between the display and the metasurface ( \(\sim 105 \mathrm{mm}\) ), the spatial distribution of the metasurfaces does not affect much on the convolutional results."  
+
+3. Scope for Improvement in Hybrid CNN: The authors claim that "Our hybrid optical/digital CNN can be further improved by using complex meta-atoms to reproduce better PSFs optically." However, the current performance of the hybrid neural network is already close to that of the compressed CNN, suggesting limited room for further enhancement. It would be beneficial for the authors to expand on how complex meta-atoms could offer substantial performance gains and to discuss other potential strategies for improving compressed CNN performance, perhaps focusing on advanced knowledge distillation or transfer learning techniques.  
+
+Response: We sincerely appreciate the reviewer's insightful comments regarding the potential for further improvement in our hybrid optical/digital CNN. Below, we clarify how complex meta-atoms contribute to performance gains and propose additional enhancement strategies. We added the following sentences to the manuscript in the Discussion-Opportunities for improvement section.  
+
+"While the current hybrid CNN achieves competitive performance, a noticeable gap ( \(\sim 4.4\%\) ) remains compared to the compressed CNN. This discrepancy can be mitigated by employing a more sophisticated PSF design, which enhances optical processing capabilities and reduces information loss. By leveraging complex meta-atoms with improved phase and amplitude control, the optical system can more accurately approximate ideal convolutional operations, thereby closing the performance gap. Additionally, performance discrepancy exists between the original CNN and its compressed counterpart. To address this, we could leverage advanced knowledge distillation techniques to enhance the compressed model's learning efficiency [1- 3]. By integrating both improved PSF design and advanced knowledge distillation methods, our approach can effectively bridge these gaps."  
+
+[1] Tian, Y., Krishnan, D., & Isola, P. (2020). Contrastive representation distillation. International Conference on Learning Representations (ICLR). https://arxiv.org/abs/1910.10699[2] Zagoruyko, S., & Komodakis, N. (2017). Paying more attention to attention: Improving the performance of convolutional neural networks via attention transfer. International Conference on Learning Representations (ICLR). https://arxiv.org/abs/1612.03928[3] Park, W., Kim, D., Lu, Y., & Cho, M. (2019). Relational knowledge distillation. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 3967- 3976. https://arxiv.org/abs/1904.05068  
+
+4. Generalization Performance: While the hybrid system achieves reasonable transferability from
+
+<--- Page Split --->
+
+CIFAR- 10 to High- 10, the performance drop is notable. The authors could discuss potential design modifications or training strategies that might enhance the generalization capability of the optical encoder without the need for extensive backend retraining.  
+
+Response: We sincerely appreciate the reviewer's insightful question. Developing a generalized optical encoder is indeed challenging. Generally, an optical encoder optimized for a more complex and larger dataset becomes more capable of generalizing to simpler and smaller datasets. High- 10, while having the same number of classes as CIFAR- 10, contains fewer samples. The ablation study in Table 2 demonstrates that this transfer learning approach improves performance from \(40\%\) to \(66\%\) compared to end- to- end training, underscoring the effectiveness of transfer learning. It is important to emphasize that our physical optical encoder remains unchanged during the transfer process. Instead, we introduce a transfer learning layer (a fully connected layer) between the optical frontend and backend. This layer is fine- tuned exclusively during transfer learning to minimize extensive backend retraining.  
+
+We acknowledge the reviewer's observation that our current approach still underperforms compared to AlexNet with transfer learning. To enhance transfer learning quality, we propose designing a more accurate PSF and applying advanced knowledge distillation methods to better represent AlexNet, as elaborated in R2Q3.  
+
+Additionally, increasing the number of kernels can lead to better representation of AlexNet, thereby improving model generalization. For instance, while AlexNet has 256 kernels in its final convolutional layer, our system currently employs only 16 kernels. To enhance the single- layer optical encoder's ability to extract both shallow and deep features, and to better represent deep neural networks for higher accuracy, we refer the reviewer to R2Q7.  
+
+The following sentences have been added to the manuscript under Discussion- Transferability:  
+
+"The scalability of the hybrid approach to more complex, real- world datasets, such as ImageNet, remains a challenge due to large reduction in accuracy. The primary cause of reduced accuracy is network simplification, such as reducing the number of layers. For example, AlexNet has 256 kernels in its final convolutional layer, while we only employ 16 kernels.  
+
+Here, we transfer from CIFAR- 10 to the ImageNet subset (High- 10). High- 10 shares the same number of classes as CIFAR- 10 but contains fewer samples. Training High- 10 from scratch (with a simplified network) is already very challenging. The ablation study in Table 2 illustrates that this transfer learning approach improves performance from \(40\%\) to \(66\%\) compared to end- to- end training, reaffirming the efficacy of transfer learning.  
+
+MAC operations for the High- 10 dataset are the same as that of the CIFAR- 10 dataset, as the same CNN architecture is utilized. The train (test) accuracy of our hybrid CNN drops significantly by \(\sim 21.85\%\) ( \(\sim 25.22\%\) ) compared to the original CNN. Most of these losses occur during network compression, as the convolutional and fully connected layers are optimized for the CIFAR- 10 dataset. Nonetheless, our transfer learning and the optical encoder achieve a classification accuracy of \(\sim 60\%\) , outperforming other free- space optical neural networks. We emphasize that the optical frontend remains unchanged, and only the digital backend- comprising two fully connected layers and an additional transfer learning layer- is fine- tuned, showcasing the versatility of our hybrid CNN system.  
+
+Our current approach still lags behind AlexNet with transfer learning. To further enhance performance, we suggest potential design modifications and training strategies. From a design perspective, implementing additional optical kernels could be a solution. To achieve this within physical constraints, we could employ multiple cameras for different kernels or modify the
+
+<--- Page Split --->
+
+metasurfaces (e.g., by rotating them) while using a single camera. From a training perspective, adopting advanced knowledge distillation methods could better represent AlexNet with greater accuracy. Certain kernels could be aligned with the shallow- layer features of the electronic CNN, while others could focus on capturing deeper features."  
+
+5. Physical Realizability of PSFs: The paper acknowledges discrepancies between the ground-truth and experimentally measured PSFs (Figure 3c), partly due to fabrication imperfections and the inherent limitations of phase control across RGB channels. A more detailed analysis of how these imperfections affect system robustness and potential mitigation strategies would strengthen the paper.  
+
+Response: Thank you for the valuable comment. We use a 2 by 2 scatterer set to create the metasurface and make it more robust to fabrication imperfection (if one of four scatterers fall down, we still have three left). Still, we believe there are three major reasons for the discrepancies between the ground- truth and experimentally measured PSFs. We included the following in a new section Discussion-Opportunities for improvement.  
+
+(1) Imperfect fitting function for the phase over scatterer: We assumed the scatterers have constant transmission and do not have any resonant features in their relative phases, which is entirely accurate (Fig. 2b).  
+(2) Limited degree of freedom of the metasurface for multicolor PSFs: We optimized each of the metasurfaces targeting three different PSFs in red, green, and blue colors. The phase profiles of one rectangular scatterer for three different colors are not independent of each other. One solution is to use complex-shaped scatterers or super cells to have more degrees of freedom.  
+(3) Broadband light sources: We simulated the polychromatic metasurfaces at three discrete wavelengths (450, 532, and 635 nm). However, the OLED pixels have much broader wavelengths. We can reduce this discrepancy if we optimize the metasurface for more representative wavelengths.  
+
+However, the classification accuracy reported here for the CIFAR- 10 dataset is considerable compared to the other reported results in optical neural networks. There will be inevitable imperfection in optical implementation as the spectral information of the scene is always changing depending on the daylight, cloud, aerial, and many other conditions. Our results show that the digital backend can compensate for the non- ideal optical implementation and achieve a relatively high classification accuracy.  
+
+6. Energy Efficiency vs. Sensor Overhead: The authors highlight a significant reduction in computational energy consumption but mention that the optical encoder requires capturing more pixels than the original CNN, increasing the sensor's energy consumption. A deeper quantitative comparison of total system energy consumption, considering both sensor and backend, would provide a clearer picture of practical efficiency gains.  
+
+Response: Thank you for the valuable comment. The color camera we used (Allied Vision Prosilica; GT 1930 C) has a total power consumption of 3.4W with 50.70 frames per second and 1,936×1,216 color pixels, which ends up with 28 nJ per frame and pixel. Since we captured all \(\sim 2\) million pixels at the same time and cropped the region of interest, the energy consumption for one input image does not differ for the original CNN and hybrid CNN. However, if we can optimize the sensor configuration and number of pixels, we can calculate the minimum required number of pixels for both the original and hybrid CNN, and estimate the energy consumption for those. For the original CNN, we need 32×32 color pixels on camera. And for the hybrid CNN, we need 32×6×6 color pixels on
+
+<--- Page Split --->
+
+the camera, where the 32 corresponds to number of multiplexed meta- optics and the \(6 \times 6\) corresponds to number of pixels after the average pooling. We only need \(6 \times 6\) pixels, not \(32 \times 32\) pixels when the convolution is already done optically. Thus we estimate that the original CNN and hybrid CNN require an energy of about \(29.1 \mu \mathrm{J}\) and \(32.8 \mu \mathrm{J}\) , respectively, for the image capturing process per a single image. Next, the energy consumption for the computational backend is much larger for the original CNN compared to the hybrid CNN. For state- of- the- art computational systems, an energy consumption per a single MAC operation is \(1 \mathrm{pJ}\) . Thus the energy consumption for a single object classification task for the hybrid CNN is about \(150 \mathrm{nJ}\) , which is more than four orders of magnitude smaller than that of the original CNN, \(3.65 \mathrm{mJ}\) . While the GPU we used (GeForce RTX 2080 Ti) has much larger energy consumption per a single MAC operation ( \(\sim 7.5 \mathrm{pJ}\) ), making the energy consumption for a single object classification tasks for the hybrid CNN and original CNN \(1.13 \mu \mathrm{J}\) and \(27.4 \mathrm{mJ}\) . Considering the sensor power, the total system level energy consumption for a single object classification task dropped from \(3.68 \mathrm{mJ}\) to \(0.03 \mathrm{mJ}\) for the state- of- the- art digital processor, while \(27.4 \mathrm{mJ}\) to \(34.0 \mu \mathrm{J}\) for our GPU. We emphasize that more than two orders of magnitude reduction in the system level computer vision tasks clearly provide strong benefits for practical implementation even with a partial accuracy drop for some application fields. We emphasize that to the best of our knowledge, our paper is the first paper which truly calculated the “energy consumption” by considering the sensor/ camera power.  
+
+We explain about the system energy consumption (with more details and an additional Table R1, highlighted in the manuscript) in the section “Energy consumption” in Discussion and add a sentence in Abstract as following:  
+
+“The proposed method can decrease total system- level energy more than two orders of magnitude per a single object classification.”  
+
+Table R1 (also Table 3 in the Revised Manuscript). System level energy consumption analysis per a single image classification task in each step of the computer vision depends on the network architecture.   
+
+<table><tr><td rowspan="2">Network architecture</td><td>Optical frontend</td><td colspan="2">Digital backend</td><td colspan="2">System</td></tr><tr><td>GT 1930C</td><td>GeForce RTX 2080 Ti</td><td>State-of-the-art GPU</td><td>GeForce RTX 2080 Ti</td><td>State-of-the-art GPU</td></tr><tr><td>Original CNN with optimal camera pixels</td><td>2.91 × 10-5 J</td><td>2.74 × 10-2 J</td><td>3.65 × 10-3 J</td><td>2.74 × 10-2 J</td><td>3.68 × 10-3 J</td></tr><tr><td>Our hybrid optical/digital CNN with optimal camera pixels</td><td>3.28 × 10-5 J</td><td>1.13 × 10-6 J</td><td>1.50 × 10-7 J</td><td>3.40 × 10-5 J</td><td>3.30 × 10-5 J</td></tr></table>  
+
+7. Scalability and Practical Deployment: The current implementation focuses on relatively small datasets (CIFAR- 10, High- 10). The scalability of the system to more complex, real-world datasets (e.g., full ImageNet) remains unclear. It would be valuable if the authors discussed potential bottlenecks in scaling the optical encoder, such as fabrication complexity, alignment issues, or increased noise sensitivity.  
+
+Response: Thank you for the valuable comment. While we completely agree with the reviewer, we want to emphasize that the datasets used in our work are one of the most complex datasets reported in the optical neural network literature.
+
+<--- Page Split --->
+
+## Factors Causing Accuracy Drops:  
+
+Factors Causing Accuracy Drops:The accuracy drop arises from both network simplification (reducing layers) and physical implementation issues (e.g., fabrication misalignment). First, converting a deep neural network into a shallow single- layer CNN introduces an inherent accuracy drop. Second, the physical implementation of the hybrid approach, such as fabrication misalignment, further reduces performance. This issue becomes more pronounced when applied to more complex datasets, such as ImageNet. Using 16 kernels, we achieved top- 1 and top- 5 classification accuracies of \(35\%\) and \(58\%\) for ImageNet- 100, respectively, after compression, compared to the original AlexNet's top- 1 and top- 5 accuracies of \(56\%\) and \(78\%\) , respectively, before compression.  
+
+## Algorithmic and Structural Improvements:  
+
+Algorithmic and Structural Improvements:To address scalability for large datasets, we have demonstrated the potential to design a more accurate PSF and apply more advanced knowledge distillation methods to better represent AlexNet, as shown in R2Q3- 4. Beyond algorithmic improvements, structural modifications to the network can further enhance the performance of optical neural networks. To enable better representation of the empirical teacher network, we could convert a multi- layer CNN into a single- layer ONN. For example, certain kernels could be aligned with shallow- layer features from an electronic CNN, while others could be designated to capture deeper features. Additionally, employing residual blocks could mitigate information loss and improve feature representation.  
+
+## Optical Implementation:  
+
+First, as the number of kernels increases, it becomes challenging to measure all the convolved images simultaneously using a single camera. Multiple cameras may be required, which could introduce additional time delays in synchronizing readout events and significantly increase energy consumption for image readout. Alternatively, metasurfaces could be switched with temporal multiplexing (e.g., using a rotating wheel) while using a single camera; however, this would result in substantial time delays for capturing the complete set of image readouts.  
+
+We included the following in a new section "Transferability" and "Opportunities for improvement" in Discussion with highlighted notes..  
+
+8. Typos and Minor Errors: There are several typographical errors throughout the manuscript that need correction. A thorough proofreading is recommended to improve the manuscript's readability. Overall, this paper introduces an exciting advancement in hybrid optical/digital neural networks, but addressing the above concerns will significantly enhance its clarity, robustness, and impact.  
+
+Response: Thank you for the valuable comment. We reviewed the paper thoroughly and corrected all the typos in our best.
+
+<--- Page Split --->
+
+Reviewer #3 (Remarks to the Author):  
+
+This paper introduces a hybrid optical/digital neural network architecture that leverages polychromatic meta- optics to perform convolutional operations during image capture, reducing computational load. The system achieves \(73.17\%\) accuracy on CIFAR- 10 and demonstrates transferability to an ImageNet subset (High- 10) with moderate accuracy. The work addresses critical challenges in real- time, energy- efficient computer vision and offers a novel integration of optical encoding with digital backends. However, I feel the authors should clarify a few points below before being considered for publication in Nature Communications.  
+
+Response: We thank the reviewer for carefully reviewing the paper and providing valuable feedback. We have provided a point- by- point response and modified the paper accordingly.  
+
+1. Regarding the accuracy Trade-offs, the accuracy drop \((-8\%)\) from AlexNet on CIFAR-10, \(-25\%\) on High-10) may limit adoption in "safety-critical" applications. Could the authors describe how one could potentially achieve better accuracy?  
+
+Response: Thank you for the valuable comment. We agree that the reduction of accuracy will not be permitted for safety-critical situations (e.g., self- driving cars). Still, our accuracy on CIFAR- 10 is better than any other free- space optical implementation of neural networks. And there are usages of a moderate accuracy AI (but with high energy efficiency and low latency) for non- safety- critical situations as we describe in the Discussion- Applications section.  
+
+In order to reduce the accuracy drop, both optical and computational approaches can be applied to advance our current hybrid CNN. For an optical approach, we would like to answer the next question. For a computation approach, we could leverage advanced knowledge distillation techniques to enhance the compressed model's learning efficiency, such as those introduced in [1- 3]. We added the following sentences on the manuscript in the Discussion- Multichannel dataset section:  
+
+"While the current hybrid CNN achieves competitive performance, a noticeable gap \((- 4.4\%)\) remains compared to the compressed CNN. This discrepancy can be mitigated by employing a more sophisticated PSF design, which enhances optical processing capabilities and reduces information loss. By leveraging complex meta- atoms with improved phase and amplitude control, the optical system can more accurately approximate ideal convolutional operations, thereby closing the performance gap. Additionally, performance discrepancy exists between the original CNN and its compressed counterpart. To address this, we could leverage advanced knowledge distillation techniques to enhance the compressed model's learning efficiency [1- 3]. By integrating both improved PSF design and advanced knowledge distillation methods, our approach can effectively bridge these gaps."  
+
+[1] Tian, Y., Krishnan, D., & Isola, P. (2020). Contrastive representation distillation. International Conference on Learning Representations (ICLR). https://arxiv.org/abs/1910.10699[2] Zagoruyko, S., & Komodakis, N. (2017). Paying more attention to attention: Improving the performance of convolutional neural networks via attention transfer. International Conference on Learning Representations (ICLR). https://arxiv.org/abs/1612.03928[3] Park, W., Kim, D., Lu, Y., & Cho, M. (2019). Relational knowledge distillation. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 3967- 3976. https://arxiv.org/abs/1904.05068  
+
+2- 1. Could the author comment on fabrication Limitations: The impact of PSF discrepancies (e.g., \(\eta = 0.56\) for green) is under- analysed. Quantifying fabrication tolerances and robustness to
+
+<--- Page Split --->
+
+misalignment would strengthen the results.  
+
+2- 2. The proxy function(i.e., Eq. 1) ignores resonance effects. How does this approximation affect PSF fidelity, especially for green ( \(\eta = 0.56\) )? The authors are suggested to include an error analysis across all kernels.  
+
+Response: Thank you for the valuable comment. We used a 2 by 2 scatterer set to create a metasurface, making it more robust to fabrication imperfection (if one of four scatterers fall down, we still have three left). Still, we believe there are three major reasons for the discrepancies between the ground- truth and experimentally measured PSFs. We included the following in a new section "Opportunities for improvement" in Discussion.  
+
+(1) Imperfect fitting function for the phase over scatterer: We assumed the scatterers have constant transmission and do not have any resonant features in their relative phases, which is entirely accurate (Fig. 2b).  
+(2) Limited degree of freedom of the metasurface for multicolor PSFs: We optimized each of the metasurfaces targeting three different PSFs in red, green, and blue colors. The phase profiles of one rectangular scatterer for three different colors are not independent of each other. One solution is to use complex-shaped scatterers or super cells to have more degrees of freedom.  
+(3) Broadband light sources: We simulated the polychromatic metasurfaces at three discrete wavelengths (450, 532, and 635 nm). However, the OLED pixels have much broader wavelengths. We can reduce this discrepancy if we optimize the metasurface for more representative wavelengths.  
+
+However, the classification accuracy reported here for the CIFAR- 10 dataset is considerable compared to the other reported results in optical neural networks. There will be inevitable imperfection in optical implementation as the spectral information of the scene is always changing depending on the daylight, cloud, aerial, and many other conditions. Our results show that the digital backend can compensate for the non- ideal optical implementation and achieve a relatively high classification accuracy.  
+
+3. Scalability: The system uses 32 meta-optics for 16 kernels. How does this scale to deeper networks or larger datasets? A discussion on physical size constraints is missing.  
+
+Response: We thank the reviewer for carefully reviewing the paper. Scaling to deeper networks or larger datasets is challenging and requires careful consideration of both optical neural networks and knowledge distillation strategies.  
+
+## Factors Causing Accuracy Drops:  
+
+The accuracy drop arises from both network simplification (reducing layers) and physical implementation issues (e.g., fabrication misalignment). First, converting a deep neural network into a shallow single- layer CNN introduces an inherent accuracy drop. Second, the physical implementation of the hybrid approach, such as fabrication misalignment, further reduces performance. This issue becomes more pronounced when applied to more complex datasets, such as ImageNet. Using 16 kernels, we achieved top- 1 and top- 5 classification accuracies of \(35\%\) and \(58\%\) for ImageNet- 100, respectively, after compression, compared to the original AlexNet's top- 1 and top- 5 accuracies of \(56\%\) and \(78\%\) , respectively, before compression.  
+
+## Algorithmic and Structural Improvements:  
+
+To address scalability for large datasets, we have demonstrated the potential to design a more accurate PSF and apply more advanced knowledge distillation methods to better represent AlexNet, as shown in R2Q3- 4. Beyond algorithmic improvements, structural modifications to the network can further
+
+<--- Page Split --->
+
+enhance the performance of optical neural networks. To enable better representation of the empirical teacher network, we could convert a multi- layer CNN into a single- layer ONN. For example, certain kernels could be aligned with shallow- layer features from an electronic CNN, while others could be designated to capture deeper features. Additionally, employing residual blocks could mitigate information loss and improve feature representation.  
+
+## Optical Implementation:  
+
+First, as the number of kernels increases, it becomes challenging to measure all the convolved images simultaneously using a single camera. Multiple cameras may be required, which could introduce additional time delays in synchronizing readout events and significantly increase energy consumption for image readout. Alternatively, metasurfaces could be switched with temporal multiplexing (e.g., using a rotating wheel) while using a single camera; however, this would result in substantial time delays for capturing the complete set of image readouts.  
+
+The following sentences have been added to the manuscript Discussion- Transferability:  
+
+"The scalability of the hybrid approach to more complex, real- world datasets, such as ImageNet, remains a challenge due to large reduction in accuracy. The primary cause of reduced accuracy is network simplification, such as reducing the number of layers. For example, AlexNet has 256 kernels in its final convolutional layer, while we only employ 16 kernels.  
+
+Here, we transfer from CIFAR- 10 to the ImageNet subset (High- 10). High- 10 shares the same number of classes as CIFAR- 10 but contains fewer samples. Training High- 10 from scratch (with a simplified network) is already very challenging. The ablation study in Table 2 illustrates that this transfer learning approach improves performance from \(40\%\) to \(66\%\) compared to end- to- end training, reaffirming the efficacy of transfer learning.  
+
+MAC operations for the High- 10 dataset are the same as that of the CIFAR- 10 dataset, as the same CNN architecture is utilized. The train (test) accuracy of our hybrid CNN drops significantly by \(\sim 21.85\%\) ( \(\sim 25.22\%\) ) compared to the original CNN. Most of these losses occur during network compression, as the convolutional and fully connected layers are optimized for the CIFAR- 10 dataset. Nonetheless, our transfer learning and the optical encoder achieve a classification accuracy of \(\sim 60\%\) , outperforming other free- space optical neural networks. We emphasize that the optical frontend remains unchanged, and only the digital backend- comprising two fully connected layers and an additional transfer learning layer- is fine- tuned, showcasing the versatility of our hybrid CNN system.  
+
+Our current approach still lags behind AlexNet with transfer learning. To further enhance performance, we suggest potential design modifications and training strategies. From a design perspective, implementing additional optical kernels could be a solution. To achieve this within physical constraints, we could employ multiple cameras for different kernels or modify the metasurfaces (e.g., by rotating them) while using a single camera. From a training perspective, adopting advanced knowledge distillation methods could better represent AlexNet with greater accuracy. Certain kernels could be aligned with the shallow- layer features of the electronic CNN, while others could focus on capturing deeper features."  
+
+4. Calibration Layer Dependency: The reliance on a digital calibration layer to correct optical imperfections partially offsets the analog advantage. Clarify if this layer adds computational overhead.  
+
+Response: Thank you for the valuable comment. The digital calibration layer depends on fabrication
+
+<--- Page Split --->
+
+misalignment and other noise, and such misalignment can be removed theoretically. Assuming the original desired kernel parameter is \(\theta\) , and after fabrication, some misalignment occurs, the final parameter becomes \(\theta + \delta_{\theta}\) . Since this is a continuous layer, we can directly compute the impact of \(\delta_{\theta}\) on the output and subsequently remove the error \(\delta_{\theta}\) , ensuring minimal deviation from the intended values. we can express the desired output correction as:  
+
+\[y_{\text{desired}} = \theta *\text{Input}\]  
+
+\[y_{\text{optical}} = (\theta + \delta_{\theta}) * \text{Input} + \text{Noise}\]  
+
+\[y_{\text{desired}} = y_{\text{optical}} - \delta_{\theta} * \text{Input} - \text{Noise}\]  
+
+Where \(y_{\text{desired}}\) is the corrected output, \(y_{\text{optical}}\) is the raw optical network output. Noise accounts for additional sources of error such as system.  
+
+The calibration layer functions as a fully- connected layer and can be integrated into the backend without introducing additional computational overhead. Assuming the calibration parameters are \(\mathrm{W}_{1}\) and \(\mathrm{b}_{1}\) , and the final fully- connected layer in the backend has parameters \(\mathrm{W}_{2}\) and \(\mathrm{B}_{2}\) , we can merge them into a single layer. The new parameters would be:  
+
+\[W^{\prime} = W_{1}\times W_{2\] \[b^{\prime} = W_{2}b_{1} + b_{2\]  
+
+5. Could the authors comment on real-World applicability by testing under variable illumination/backgrounds (not just controlled lab conditions) as this way would better demonstrate practical utility?  
+
+Response: We really appreciate the reviewer for the question. We do agree with the Reviewer that implementing on a real- World scene means much more than the controlled laboratory conditions. The digital backend did well to compromise various errors and imperfections generated from the optical implementation even in the laboratory conditions. On the other hand, the uncontrolled situation in a natural scene (light intensity, color temperature, shadow, etc) will definitely cause much severe noise on our hybrid CNN. In fact, we are currently working on testing these optics under ambient illumination. However, creating a real- world dataset is not straight- forward. In order to do that, we may have to define constraints for the dataset, and put two cameras parallel and use a normal lens for one while use our metasurface for the other, a setup we recently reported [4].  
+
+[4] Fröch, J.E., Chakravarthula, P.K., Sun, J., Tseng, E., Colburn, S., Zhan, A., Miller, F., Wirth- Singh, A., Tanguy, Q.A., Han, Z., et al.: Beating bandwidth limits for large aperture broadband nano- optics. arXiv preprint arXiv:2402.06824 (2024)  
+
+6. The authors wrote "Thus the energy consumption for a single object classification task for the hybrid CNN is about 150nJ, which is more than four orders of magnitude smaller than that of the original CNN." Does the backend energy savings justify increased sensor costs?  
+
+Response: Thank you for the valuable comment. The color camera we used (Allied Vision Prosilica; GT 1930 C) has a total power consumption of 3.4W with 50.70 frames per second and 1,936×1,216 color pixels, which ends up with 28 nJ per frame and pixel. Since we captured all \(\sim 2\) million pixels at the same time and cropped the region of interest, the energy consumption for one input image does not differ for the original CNN and hybrid CNN. However, if we can optimize the sensor
+
+<--- Page Split --->
+
+configuration and number of pixels, we can calculate the minimum required number of pixels for both the original and hybrid CNN, and estimate the energy consumption for those. For the original CNN, we need \(32 \times 32\) color pixels on camera. And for the hybrid CNN, we need \(32 \times 6 \times 6\) color pixels on the camera, where the 32 corresponds to number of multiplexed meta- optics and the \(6 \times 6\) corresponds to number of pixels after the average pooling. We only need \(6 \times 6\) pixels, not \(32 \times 32\) pixels when the convolution is already done optically. Thus we estimate that the original CNN and hybrid CNN require an energy of about \(29.1 \mu \mathrm{J}\) and \(32.8 \mu \mathrm{J}\) , respectively, for the image capturing process per a single image. Next, the energy consumption for the computational backend is much larger for the original CNN compared to the hybrid CNN. For state- of- the- art computational systems, an energy consumption per a single MAC operation is \(1 \mathrm{pJ}\) . Thus the energy consumption for a single object classification task for the hybrid CNN is about \(150 \mathrm{nJ}\) , which is more than four orders of magnitude smaller than that of the original CNN, \(3.65 \mathrm{mJ}\) . While the GPU we used (GeForce RTX 2080 Ti) has much larger energy consumption per a single MAC operation ( \(\sim 7.5 \mathrm{pJ}\) ), making the energy consumption for a single object classification tasks for the hybrid CNN and original CNN \(1.13 \mu \mathrm{J}\) and \(27.4 \mathrm{mJ}\) . Considering the sensor power, the total system level energy consumption for a single object classification task dropped from \(3.68 \mathrm{mJ}\) to \(0.03 \mathrm{mJ}\) for the state- of- the- art digital processor, while \(27.4 \mathrm{mJ}\) to \(34.0 \mu \mathrm{J}\) for our GPU. We emphasize that more than two orders of magnitude reduction in the system level computer vision tasks clearly provide strong benefits for practical implementation even with a partial accuracy drop for some application fields. We emphasize that to the best of our knowledge, our paper is the first paper which truly calculated the “energy consumption” by considering the sensor/ camera power.  
+
+We explain about the system energy consumption (with more details and an additional Table R1, highlighted in the manuscript) in the section “Energy consumption” in Discussion and add a sentence in Abstract as following:  
+
+“The proposed method can decrease total system- level energy more than two orders of magnitude per a single object classification.”  
+
+Table R1 (also Table 3 in the Revised Manuscript). System level energy consumption analysis per a single image classification task in each step of the computer vision depends on the network architecture.  
+
+<table><tr><td rowspan="2">Network architecture</td><td>Optical frontend</td><td colspan="2">Digital backend</td><td colspan="2">System</td></tr><tr><td>GT 1930C</td><td>GeForce RTX 2080 Ti</td><td>State-of-the-art GPU</td><td>GeForce RTX 2080 Ti</td><td>State-of-the-art GPU</td></tr><tr><td>Original CNN with optimal camera pixels</td><td>2.91 × 10-5 J</td><td>2.74 × 10-2 J</td><td>3.65 × 10-3 J</td><td>2.74 × 10-2 J</td><td>3.68 × 10-3 J</td></tr><tr><td>Our hybrid optical/digital CNN with optimal camera pixels</td><td>3.28 × 10-5 J</td><td>1.13 × 10-6 J</td><td>1.50 × 10-7 J</td><td>3.40 × 10-5 J</td><td>3.30 × 10-5 J</td></tr></table>  
+
+7. The added “transfer learning layer” is not detailed in the main text. Could the authors provide architecture specifics and ablation studies?  
+
+Response: Thank you for the valuable comment. The transfer learning layer is a single fully connected layer used to reproject feature clustering. The ablation study is already included in Table 2, showing that this transfer learning layer improves performance from \(40\%\) to \(66\%\) .
+
+<--- Page Split --->
+
+8. By comparing accuracy/MAC reductions with recent works, could the author highlight how polychromatic encoding advances the field?  
+
+Response: Thank you for the valuable comment. For the CIFAR- 10 dataset, our hybrid optical/digital CNN reduced the number of MAC operations by a factor of \(\sim 24,000\) (Table A1). This reduction is more than two orders of magnitude higher than that of the MNIST hand-written dataset, where the meta- optical encoder reduced the number of MAC operations only by a factor of \(\sim 200\) [5]. The far increased reduction of digital operations for CIFAR- 10 dataset compared to MNIST dataset is an important benefit in both energy consumption and latency. This is mainly because of the complexity of the dataset (including the polychromatic nature of the dataset), where it requires a much larger number of operations for convolutional layers. We added this description at section "Multichannel dataset" in Discussion.  
+
+[5] Wirth- Singh, A., Xiang, J., Choi, M., Fr'och, J.E., Huang, L., Colburn, S., Shlizerman, E., Majumdar, A.: Compressed meta- optical encoder for image classification. Advanced Photonics Nexus 4(2), 026009- 026009 (2025)  
+
+9. Since the authors used the system to process static images, could they discuss feasibility for video streams and temporal feature extraction?  
+
+Response: We sincerely appreciate the reviewer's question. Real- time operation and latency remain critical challenges in AI. Our hybrid optical/digital architecture reduces the number of digital operations by more than four orders of magnitude, significantly lowering latency. Similar to the convolutional layers in AlexNet, the optical frontend has the potential to extract spatial features from individual frames. These spatial features can then be fed into RNNs, LSTMs, or Transformers to enhance temporal feature understanding. Thus, our approach could be highly relevant for video streaming applications, such as video conferencing, where real- time processing is essential but safety- critical constraints are minimal. We have added the following sentences to the manuscript in the Discussion- Applications section:  
+
+"On the other hand, real- time operation and latency remain among the most significant challenges in AI. Our hybrid optical/digital architecture reduced more than four orders of magnitude of digital operations, thereby substantially decreasing both power and latency. Consequently, our approach has the potential to create a significant impact on video conferencing which align well with non- safety- critical scenarios."
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+## Reviewer #1 (Remarks to the Author):  
+
+As the authors addressed all my concerns, I recommend it can be accepted for publication.  
+
+Response: We thank the reviewer for a careful review.  
+
+## Reviewer #3 (Remarks to the Author):  
+
+The authors have properly addressed all my comments from the previous round of review. I now recommend accepting this manuscript for publishing in Nature Communications.  
+
+Response: We thank the reviewer for a careful review.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Transparent Peer Review file__e8edc04180bfdf0eefaef4278be6e83a9bef2cad24998f6d65e53837953baf6f/supplementary_0_Transparent Peer Review file__e8edc04180bfdf0eefaef4278be6e83a9bef2cad24998f6d65e53837953baf6f_det.mmd

@@ -0,0 +1,639 @@
+<|ref|>title<|/ref|><|det|>[[72, 50, 295, 80]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[75, 96, 295, 118]]<|/det|>
+Peer Review File  
+
+<|ref|>title<|/ref|><|det|>[[73, 161, 820, 209]]<|/det|>
+# Transferable polychromatic optical encoder for neural networks  
+
+<|ref|>text<|/ref|><|det|>[[73, 224, 496, 241]]<|/det|>
+Corresponding Author: Professor Arka Majumdar  
+
+<|ref|>text<|/ref|><|det|>[[70, 274, 864, 289]]<|/det|>
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+<|ref|>text<|/ref|><|det|>[[73, 326, 144, 340]]<|/det|>
+Version 0:  
+
+<|ref|>text<|/ref|><|det|>[[73, 353, 219, 367]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[73, 379, 160, 393]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[73, 404, 238, 418]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 419, 920, 472]]<|/det|>
+This work demonstrates some results by developing an optical encoder that can perform convolution simultaneously in three color channels during the image capture, implementing several initial convolutional layers of a ANN. However, although the optical encoding can decrease computational operations, the classification accuracy ( \(\sim 73.2\%\) ) is too low for real applications. I donot think this manuscript reach the level of Nature Communications. Here are comments.  
+
+<|ref|>text<|/ref|><|det|>[[72, 472, 911, 511]]<|/det|>
+1. Many similar hybrid optical/digital architectures have been demonstrated. I donot find something new for the current one. Is there any new for the design? Just because they use metasurfaces? If fact, the testing accuracy drops significantly with their meta-optics.  
+
+<|ref|>text<|/ref|><|det|>[[72, 511, 910, 550]]<|/det|>
+2. It is strange that the authors said their architectures are integrated. But they compare the classification accuracy with the free space optical systems. It is possible to improve classification accuracy? How? At least, it should be comparable with integrated optical neural network architectures.  
+
+<|ref|>text<|/ref|><|det|>[[72, 550, 917, 602]]<|/det|>
+3. Low energy consumption may one of the advantages of the developed architectures, it should be reasonable due to the less operations. However, it looks meaningless as they sacrifice the classification accuracy. In addition, they overlooked the energy consumption of light source. Can they estimate the real energy consumption if the really mean "energy consumption"?  
+
+<|ref|>text<|/ref|><|det|>[[70, 601, 904, 629]]<|/det|>
+4. Why the classification accuracy drops more significant for the CIFAR-10 dataset compared to the MNIST dataset? How about the classification accuracy for the MNIST dataset?  
+
+<|ref|>text<|/ref|><|det|>[[72, 629, 650, 643]]<|/det|>
+5. Physical sizes about the metasurfaces should be provided for the testing samples.  
+
+<|ref|>text<|/ref|><|det|>[[73, 653, 161, 666]]<|/det|>
+Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[73, 679, 238, 692]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 693, 920, 745]]<|/det|>
+This work presents a novel approach for a hybrid optical/digital neural network, performing the computationally expensive convolution operation in optics. The manuscript introduces an innovative design that demonstrates significant computational efficiency while maintaining competitive classification accuracy. The work can be considered for publication; however, the following concerns should be addressed:  
+
+<|ref|>text<|/ref|><|det|>[[72, 756, 920, 810]]<|/det|>
+Phase Coverage of Meta- Atoms: Based on Figure 2b, the phase coverage of the meta- atoms appears to be around 1 radian, even when assuming impractically small feature sizes. For designing an effective metasurface, a phase coverage is typically required. The authors should clarify how they achieve the necessary phase modulation with this limited range or discuss the implications of this constraint on device performance.  
+
+<|ref|>text<|/ref|><|det|>[[72, 821, 902, 874]]<|/det|>
+Kernel- Image Interaction: The interaction mechanism between the kernels and the image is unclear. The kernels are spatially distributed in different locations, yet they seem to interact with the entire image. The authors should elaborate on how this is physically implemented, particularly addressing whether there are optical multiplexing effects or specific alignment strategies that enable this global interaction.  
+
+<|ref|>text<|/ref|><|det|>[[72, 885, 904, 939]]<|/det|>
+Scope for Improvement in Hybrid CNN: The authors claim that "Our hybrid optical/digital CNN can be further improved by using complex meta- atoms to reproduce better PSFs optically." However, the current performance of the hybrid neural network is already close to that of the compressed CNN, suggesting limited room for further enhancement. It would be beneficial for the authors to expand on how complex meta- atoms could offer substantial performance gains and to discuss
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 46, 852, 75]]<|/det|>
+other potential strategies for improving compressed CNN performance, perhaps focusing on advanced knowledge distillation or transfer learning techniques.  
+
+<|ref|>text<|/ref|><|det|>[[72, 86, 899, 127]]<|/det|>
+Generalization Performance: While the hybrid system achieves reasonable transferability from CIFAR- 10 to High- 10, the performance drop is notable. The authors could discuss potential design modifications or training strategies that might enhance the generalization capability of the optical encoder without the need for extensive backend retraining.  
+
+<|ref|>text<|/ref|><|det|>[[72, 138, 918, 192]]<|/det|>
+Physical Realizability of PSFs: The paper acknowledges discrepancies between the ground- truth and experimentally measured PSFs (Figure 3c), partly due to fabrication imperfections and the inherent limitations of phase control across RGB channels. A more detailed analysis of how these imperfections affect system robustness and potential mitigation strategies would strengthen the paper.  
+
+<|ref|>text<|/ref|><|det|>[[72, 202, 916, 257]]<|/det|>
+Energy Efficiency vs. Sensor Overhead: The authors highlight a significant reduction in computational energy consumption but mention that the optical encoder requires capturing more pixels than the original CNN, increasing the sensor's energy consumption. A deeper quantitative comparison of total system energy consumption, considering both sensor and backend, would provide a clearer picture of practical efficiency gains.  
+
+<|ref|>text<|/ref|><|det|>[[72, 268, 920, 321]]<|/det|>
+Scalability and Practical Deployment: The current implementation focuses on relatively small datasets (CIFAR- 10, High- 10). The scalability of the system to more complex, real- world datasets (e.g., full ImageNet) remains unclear. It would be valuable if the authors discussed potential bottlenecks in scaling the optical encoder, such as fabrication complexity, alignment issues, or increased noise sensitivity.  
+
+<|ref|>text<|/ref|><|det|>[[72, 332, 910, 361]]<|/det|>
+Typos and Minor Errors: There are several typographical errors throughout the manuscript that need correction. A thorough proofreading is recommended to improve the manuscript's readability.  
+
+<|ref|>text<|/ref|><|det|>[[72, 371, 904, 400]]<|/det|>
+Overall, this paper introduces an exciting advancement in hybrid optical/digital neural networks, but addressing the above concerns will significantly enhance its clarity, robustness, and impact.  
+
+<|ref|>sub_title<|/ref|><|det|>[[72, 411, 163, 425]]<|/det|>
+## Reviewer #3  
+
+<|ref|>text<|/ref|><|det|>[[72, 437, 922, 530]]<|/det|>
+(Remarks to the Author) This paper introduces a hybrid optical/digital neural network architecture that leverages polychromatic meta- optics to perform convolutional operations during image capture, reducing computational load. The system achieves \(73.17\%\) accuracy on CIFAR- 10 and demonstrates transferability to an ImageNet subset (High- 10) with moderate accuracy. The work addresses critical challenges in real- time, energy- efficient computer vision and offers a novel integration of optical encoding with digital backends. However, I feel the authors should clarify a few points below before being considered for publication in Nature Communications.  
+
+<|ref|>text<|/ref|><|det|>[[72, 541, 896, 570]]<|/det|>
+1. Regarding the accuracy Trade-offs, the accuracy drop ( \(\sim 8\%\) from AlexNet on CIFAR-10, \(\sim 25\%\) on High-10) may limit adoption in "safety-critical" applications. Could the authors describe how one could potentially achieve better accuracy?  
+
+<|ref|>text<|/ref|><|det|>[[70, 580, 905, 608]]<|/det|>
+2. Could the author comment on fabrication Limitations: The impact of PSF discrepancies (e.g., \(\eta = 0.56\) for green) is under- analysed. Quantifying fabrication tolerances and robustness to misalignment would strengthen the results.  
+
+<|ref|>text<|/ref|><|det|>[[70, 619, 901, 647]]<|/det|>
+Scalability: The system uses 32 meta- optics for 16 kernels. How does this scale to deeper networks or larger datasets? A discussion on physical size constraints is missing.  
+
+<|ref|>text<|/ref|><|det|>[[70, 658, 912, 686]]<|/det|>
+Calibration Layer Dependency: The reliance on a digital calibration layer to correct optical imperfections partially offsets the analog advantage. Clarify if this layer adds computational overhead.  
+
+<|ref|>text<|/ref|><|det|>[[70, 697, 875, 725]]<|/det|>
+2. Could the authors comment on real-World applicability by testing under variable illumination/backgrounds (not just controlled lab conditions) as this way would better demonstrate practical utility?  
+
+<|ref|>text<|/ref|><|det|>[[70, 736, 900, 764]]<|/det|>
+3. The proxy function(i.e., Eq. 1) ignores resonance effects. How does this approximation affect PSF fidelity, especially for green ( \(\eta = 0.56\) )? The authors are suggested to include an error analysis across all kernels.  
+
+<|ref|>text<|/ref|><|det|>[[70, 775, 881, 815]]<|/det|>
+4. The authors wrote "Thus the energy consumption for a single object classification task for the hybrid CNN is about 150nJ, which is more than four orders of magnitude smaller than that of the original CNN." Does the backend energy savings justify increased sensor costs?  
+
+<|ref|>text<|/ref|><|det|>[[70, 826, 900, 854]]<|/det|>
+5. The added "transfer learning layer" is not detailed in the main text. Could the authors provide architecture specifics and ablation studies?  
+
+<|ref|>text<|/ref|><|det|>[[70, 865, 875, 893]]<|/det|>
+6. By comparing accuracy/MAC reductions with recent works, could the author highlight how polychromatic encoding advances the field?  
+
+<|ref|>text<|/ref|><|det|>[[70, 904, 870, 932]]<|/det|>
+7. Since the authors used the system to processes static images, could they discuss feasibility for video streams and temporal feature extraction?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[73, 47, 145, 60]]<|/det|>
+Version 1:  
+
+<|ref|>text<|/ref|><|det|>[[73, 73, 220, 87]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[73, 100, 161, 113]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[73, 126, 692, 153]]<|/det|>
+(Remarks to the Author) As the authors addressed all my concerns, I recommend it can be accepted for publication.  
+
+<|ref|>text<|/ref|><|det|>[[73, 165, 163, 179]]<|/det|>
+Reviewer #3  
+
+<|ref|>text<|/ref|><|det|>[[73, 191, 240, 204]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[73, 204, 910, 231]]<|/det|>
+The authors have properly addressed all my comments from the previous round of review. I now recommend accepting this manuscript for publishing in Nature Communications.  
+
+<|ref|>text<|/ref|><|det|>[[72, 584, 916, 638]]<|/det|>
+Open Access This Peer Review 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.  
+
+<|ref|>text<|/ref|><|det|>[[72, 638, 797, 652]]<|/det|>
+In cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+<|ref|>text<|/ref|><|det|>[[72, 651, 911, 704]]<|/det|>
+The images or other third party material in this Peer Review 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.  
+
+<|ref|>text<|/ref|><|det|>[[72, 702, 618, 716]]<|/det|>
+To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 100, 325, 115]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 125, 422, 141]]<|/det|>
+## Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 150, 877, 232]]<|/det|>
+This work demonstrates some results by developing an optical encoder that can perform convolution simultaneously in three color channels during the image capture, implementing several initial convolutional layers of a ANN. However, although the optical encoding can decrease computational operations, the classification accuracy ( \(\sim 73.2\%\) ) is too low for real applications. I do not think this manuscript reaches the level of Nature Communications. Here are comments.  
+
+<|ref|>text<|/ref|><|det|>[[117, 240, 878, 437]]<|/det|>
+Response: We thank the reviewer for carefully reviewing the paper and providing valuable feedback. We have provided a point- by- point response and modified the paper accordingly. We respectfully disagree with the comments that our classification accuracy is too low. Our accuracy on CIFAR- 10 is better than any other free- space optical implementation of neural networks. And there are usages of a moderate accuracy AI (but with high energy efficiency and low latency) for non- safety- critical situations as we describe in the Discussion- Applications section. We emphasize that most optical neural networks work on MNIST or some toy problems [1- 5]. Majority of works cannot handle a colorful dataset like CIFAR- 10, while polychromatic information contains significant information in a real scenario. Moreover, we report a system level benefit in energy consumption (more than two orders of magnitude), while none of the other works describe the system level energy consumption including optoelectronic devices (e.g., laser, modulators, etc). In these regards, we believe that our work significantly extends the capability of existing optical neural networks.  
+
+<|ref|>text<|/ref|><|det|>[[115, 444, 878, 623]]<|/det|>
+[1] Chen, Y., Nazhamaiti, M., Xu, H., Meng, Y., Zhou, T., Li, G., Fan, J., Wei, Q., Wu, J., Qiao, F., et al.: All- analog photoelectronic chip for high- speed vision tasks. Nature 623(7985), 48- 57 (2023) [2] Lin, X., Rivenson, Y., Yardimci, N.T., Veli, M., Luo, Y., Jarrahi, M., Ozcan, A.: All- optical machine learning using diffractive deep neural networks. Science 361(6406), 1004- 1008 (2018) [3] Chen, Z., Sludds, A., Davis III, R., Christen, I., Bernstein, L., Ateshian, L., Heuser, T., Heermeier, N., Lott, J.A., Reitzenstein, S., et al.: Deep learning with coherent vcsel neural networks. Nature Photonics 17(8), 723- 730 (2023) [4] Xia, F., Kim, K., Eliezer, Y., Han, S., Shaughnessy, L., Gigan, S., Cao, H.: Non- linear optical encoding enabled by recurrent linear scattering. Nature Photonics, 1- 9 (2024) [5] Xue, Z., Zhou, T., Xu, Z., Yu, S., Dai, Q., Fang, L.: Fully forward mode training for optical neural networks. Nature 632(8024), 280- 286 (2024)  
+
+<|ref|>text<|/ref|><|det|>[[118, 654, 877, 705]]<|/det|>
+1. Many similar hybrid optical/digital architectures have been demonstrated. I do not find something new for the current one. Is there any new for the design? Just because they use metasurfaces? In fact, the testing accuracy drops significantly with their meta-optics.  
+
+<|ref|>text<|/ref|><|det|>[[117, 713, 878, 909]]<|/det|>
+Response: Thank you for the valuable comment. However, we respectfully disagree with the reviewer. First of all, we would like to emphasize that our work is the first optical/digital architecture which utilizes polychromatic meta-optics (a single metasurface performs different convolutions for three different colors: RGB), and this is an important feature to minimize the optical system as well as the system level energy consumption as we need less number of pixels to capture the image. This property is very difficult to realize using any other engineered optics, and critically relies on the complex dependence of the optical phase on the geometric parameters of the sub- wavelength meta- atoms. This is a completely new design, and indeed this can be achieved because we used sub- wavelength diffractive metasurface. Secondly, we will encourage the reviewers to provide references on hybrid architecture that has worked with complex datasets like CIFAR- 10. Optical structures have predominantly focused on MNIST, which is linearly separable and as such is extremely simple. Bringing optical structures to CIFAR- 10, and a subset of ImageNet (as we demonstrated) is
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 100, 201, 115]]<|/det|>
+non- trivial.  
+
+<|ref|>text<|/ref|><|det|>[[117, 124, 879, 368]]<|/det|>
+The Reviewer also mentioned the significant testing accuracy drop during the optical implementation. However, accuracy drop is relatively small, from \(76.59\%\) to \(72.06\%\) (73.17% after retraining the backend), compared to the digital compression variant. Furthermore, our result is state- of- the- art compared to other works published very recently [5- 8]. Again, we emphasize that demonstration with CIFAR- 10 and ImageNet is significantly more complex than with MNIST datasets, and to the best of our knowledge, our work is at par, if not better than most existing demonstrations of hybrid optical networks on CIFAR- 10. We explicitly describe the advantages of our system in energy consumption (more than two orders of magnitude), while none of the other works describe the system level energy consumption. Reported works that we are aware of have ignored the power of the camera, or intensifier or some other components. We will be happy to modify our claim if the reviewer can provide a reference that refutes it. In fact, the reviewer is correct to say that there is indeed a large body of work on the optical neural network, however, none of the works have shown system level energy consumption improvement (including both optoelectronic and digital energy consumption) as we show in our work. Majority of the work shows improvement in linear operations, but when the whole system is considered, it is not clear if the energy benefits are carried through.  
+
+<|ref|>text<|/ref|><|det|>[[116, 375, 878, 525]]<|/det|>
+[5] Xue, Z., Zhou, T., Xu, Z., Yu, S., Dai, Q., Fang, L.: Fully forward mode training for optical neural networks. Nature 632(8024), 280- 286 (2024) [6] Huo, Y., Bao, H., Peng, Y., Gao, C., Hua, W., Yang, Q., Li, H., Wang, R., Yoon, S.- E.: Optical neural network via loose neuron array and functional learning. Nature Communications 14(1), 2535 (2023) [7] Rahman, M.S.S., Ozcan, A.: Time- lapse image classification using a diffractive neural network. Advanced Intelligent Systems 5(5), 2200387 (2023) [8] Wei, K., Li, X., Froech, J., Chakravarthula, P., Whitehead, J., Tseng, E., Majumdar, A., Heide, F.: Spatially varying nanophotonic neural networks. Science Advances 10(45), 0391 (2024)  
+
+<|ref|>text<|/ref|><|det|>[[118, 548, 878, 613]]<|/det|>
+2. It is strange that the authors said their architectures are integrated. But they compare the classification accuracy with the free space optical systems. It is possible to improve classification accuracy? How? At least, it should be comparable with integrated optical neural network architectures.  
+
+<|ref|>text<|/ref|><|det|>[[118, 622, 878, 721]]<|/det|>
+Response: Thank you for the valuable comment. We claimed our system as a free- space optical encoder, not as integrated photonics. What might have been the source of confusion, is that we are using solid- state meta- optics, which can be manufactured using semiconductor manufacturing. They can also be integrated with a camera. Thus they are indeed an integrated system, but not a photonic integrated circuit (where the light flows via waveguide in the plane of the chip). In our system, the light travels perpendicular to the chip.  
+
+<|ref|>text<|/ref|><|det|>[[118, 729, 878, 811]]<|/det|>
+Free- space integrated systems can handle two- dimensional data and can directly interface with ambient incoherent light, unlike photonic integrated circuits. Our demonstrated free- space encoder can be easily integrated with conventional imaging systems with minimized modification of the system hardware. We did not compare our system with photonic integrated circuits since our performance in terms of space- bandwidth product is significantly better than those circuits.  
+
+<|ref|>text<|/ref|><|det|>[[118, 820, 878, 902]]<|/det|>
+For the other comment of the Reviewer about "improving classification accuracy", we can use a stronger knowledge distillation loss. Our current version is based on Geoffrey Hinton's 'Distilling the Knowledge in a Neural Network.' Recently, the community has introduced more advanced versions of knowledge distillation loss [9- 11]. Another possibility is that we can increase the number of kernels to enhance optical feature engineering; however it will increase the number of pixels required for image
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 99, 878, 164]]<|/det|>
+capturing, then increase the energy consumption for the optical side. Having said that, as explained above, our reported accuracy is better than current state- of- the- art on complex datasets like CIFAR- 10 using hybrid optical/digital neural networks. We added further information in the manuscript under Discussion- Opportunities for improvement addressing improvement in accuracy:  
+
+<|ref|>text<|/ref|><|det|>[[117, 172, 879, 353]]<|/det|>
+"While the current hybrid CNN achieves competitive performance, a noticeable gap ( \(\sim 4.4\%\) ) remains compared to the compressed CNN. This discrepancy can be mitigated by employing a more sophisticated PSF design, which enhances optical processing capabilities and reduces information loss. By leveraging complex meta- atoms with improved phase and amplitude control, the optical system can more accurately approximate ideal convolutional operations, thereby closing the performance gap. Additionally, performance discrepancy exists between the original CNN and its compressed counterpart. To address this, we could leverage advanced knowledge distillation techniques to enhance the compressed model's learning efficiency. By integrating both improved PSF design and advanced knowledge distillation methods, our approach can effectively bridge these gaps. We can also increase the number of kernels; however, this will increase the number of pixels required for image capture.  
+
+<|ref|>text<|/ref|><|det|>[[118, 361, 878, 411]]<|/det|>
+Figure 3c shows a clear discrepancy between the ground- truth convolutional kernels and the experimentally measured PSFs. Other than fabrication imperfections and optical misalignment, we identify three other reasons for the discrepancy.  
+
+<|ref|>text<|/ref|><|det|>[[118, 419, 878, 469]]<|/det|>
+1. Imperfect fitting function for the phase over scatterer: We assumed the scatterers have constant transmission and do not have any resonant features in their relative phases, which is entirely accurate (Fig. 2b).  
+
+<|ref|>text<|/ref|><|det|>[[118, 477, 878, 544]]<|/det|>
+2. Limited degree of freedom of the metasurface for multicolor PSFs: We optimized each of the metasurfaces targeting three different PSFs in red, green, and blue colors. The phase profiles of one rectangular scatterer for three different colors are not independent of each other. One solution is to use complex-shaped scatterers or super cells to have more degrees of freedom.  
+
+<|ref|>text<|/ref|><|det|>[[118, 552, 878, 602]]<|/det|>
+3. Broadband light sources: We simulated the polychromatic metasurfaces at three discrete wavelengths (450, 532, and 635 nm). However, the OLED pixels have much broader wavelengths. We can reduce this discrepancy if we optimize the metasurface for more representative wavelengths.  
+
+<|ref|>text<|/ref|><|det|>[[118, 609, 878, 692]]<|/det|>
+However, the classification accuracy reported here for the CIFAR- 10 dataset is considerable compared to the other reported results in optical neural networks. There will be inevitable imperfection in optical implementation as the spectral information of the scene is always changing depending on the daylight, cloud, aerial, and many other conditions. Our results show that the digital backend can compensate for the non- ideal optical implementation and achieve a relatively high classification accuracy."  
+
+<|ref|>text<|/ref|><|det|>[[117, 700, 879, 829]]<|/det|>
+[9] Tian, Y., Krishnan, D., & Isola, P. (2020). Contrastive representation distillation. International Conference on Learning Representations (ICLR). https://arxiv.org/abs/1910.10699 [10] Zagoruyko, S., & Komodakis, N. (2017). Paying more attention to attention: Improving the performance of convolutional neural networks via attention transfer. International Conference on Learning Representations (ICLR). https://arxiv.org/abs/1612.03928 [11] Park, W., Kim, D., Lu, Y., & Cho, M. (2019). Relational knowledge distillation. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 3967- 3976. https://arxiv.org/abs/1904.05068  
+
+<|ref|>text<|/ref|><|det|>[[118, 855, 878, 905]]<|/det|>
+3. Low energy consumption may one of the advantages of the developed architectures, it should be reasonable due to the less operations. However, it looks meaningless as they sacrifice the classification accuracy. In addition, they overlooked the energy consumption of light source. Can they
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 100, 707, 116]]<|/det|>
+estimate the real energy consumption if the really mean "energy consumption"?  
+
+<|ref|>text<|/ref|><|det|>[[118, 125, 878, 222]]<|/det|>
+Response: We thank the reviewer for the valuable comment. With regards to the usefulness of the hybrid optical/digital architectures given that there is an accuracy drop, we agree that the reduction of accuracy will not be permitted for emergency situations (e.g., self- driving cars). However, computer vision tasks have been processed in many other circumstances, and we specifically point out a particular application field of statistical analysis in the Discussion- Applications section, where the ensemble average can compromise the inaccuracy from a single event.  
+
+<|ref|>text<|/ref|><|det|>[[118, 231, 878, 346]]<|/det|>
+In addition, we now include an ablation study comparing the hybrid optical/digital CNN result and the purely digital CNN (which consists only of a calibration layer and convolutional layers) in the last row of Table 1. The number of digital operations is identical in both cases, but the latter shows a significant drop (over \(20\%\) ) in testing accuracy due to the absence of the optical convolutional layer. This result demonstrates the benefits of our hybrid optical/digital architecture, beyond the straightforward relationship between fewer operations and reduced accuracy. The following sentences have been added to the manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[118, 354, 877, 403]]<|/det|>
+"Additionally, compared to the backend- only results (Table 1), our hybrid optical/digital architecture achieves significantly higher classification accuracy (over \(20\%\) ), highlighting the crucial role of the optical convolutional encoder."  
+
+<|ref|>text<|/ref|><|det|>[[118, 412, 878, 494]]<|/det|>
+Regarding energy consumption, we can disregard the energy usage of light sources since we rely on ambient light (natural light), with no changes to the input image compared to the original CNN. However, in other works utilizing photonic integrated circuits, additional energy is required for light sources, as they depend on a coherent light source distinct from ambient light. The following sentences have been added to the manuscript Discussion- Energy consumption:  
+
+<|ref|>text<|/ref|><|det|>[[118, 502, 875, 536]]<|/det|>
+"Since we rely on ambient light - similar to real- world computer vision tasks - we do not require additional energy for the input light source."  
+
+<|ref|>text<|/ref|><|det|>[[117, 544, 878, 902]]<|/det|>
+With regards to energy consumption, we calculate both optical and digital energy consumption per image classification task. The color camera we used (Allied Vision Prosilica; GT 1930 C) has a total power consumption of \(3.4\mathrm{W}\) with 50.70 frames per second and 1,936×1,216 color pixels, which ends up with \(28\mathrm{nJ}\) per frame and pixel. Since we captured all \(\sim 2\) million pixels at the same time and cropped the region of interest, the energy consumption for one input image does not differ for the original CNN and hybrid CNN. However, if we can optimize the sensor configuration and number of pixels, we can calculate the minimum required number of pixels for both the original and hybrid CNN, and estimate the energy consumption for those. For the original CNN, we need 32×32 color pixels on camera. And for the hybrid CNN, we need 32×6×6 color pixels on the camera, where the 32 corresponds to number of multiplexed meta- optics and the 6×6 corresponds to number of pixels after the average pooling. We only need 6×6 pixels, not 32×32 pixels when the convolution is already done optically. Thus we estimate that the original CNN and hybrid CNN require an energy of about \(29.1\mu \mathrm{J}\) and \(32.8\mu \mathrm{J}\) , respectively, for the image capturing process per a single image. Next, the energy consumption for the computational backend is much larger for the original CNN compared to the hybrid CNN. For state- of- the- art computational systems, an energy consumption per a single MAC operation is \(1\mathrm{pJ}\) . Thus the energy consumption for a single object classification task for the hybrid CNN is about \(150\mathrm{nJ}\) , which is more than four orders of magnitude smaller than that of the original CNN, \(3.65\mathrm{mJ}\) . While the GPU we used (GeForce RTX 2080 Ti) has much larger energy consumption per a single MAC operation ( \(\sim 7.5\mathrm{pJ}\) ), making the energy consumption for a single object classification tasks for the hybrid CNN and original CNN \(1.13\mu \mathrm{J}\) and \(27.4\mathrm{mJ}\) . Considering the sensor power, the total system level energy consumption for a single object classification task dropped from \(3.68\mathrm{mJ}\) to \(0.03\mathrm{mJ}\) for the state- of- the- art digital processor, while \(27.4\mathrm{mJ}\) to \(34.0\mu \mathrm{J}\) for our GPU. We
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 99, 878, 166]]<|/det|>
+emphasize that more than two orders of magnitude reduction in the system level computer vision tasks clearly provide strong benefits for practical implementation even with a partial accuracy drop for some application fields. We emphasize that to the best of our knowledge, our paper is the first paper which truly calculated the “energy consumption” by considering the sensor/ camera power.  
+
+<|ref|>text<|/ref|><|det|>[[118, 173, 878, 223]]<|/det|>
+We explain about the system energy consumption (with more details and an additional Table R1, highlighted in the manuscript) in the section “Energy consumption” in Discussion and add a sentence in Abstract as following:  
+
+<|ref|>text<|/ref|><|det|>[[118, 231, 878, 265]]<|/det|>
+“The proposed method can decrease total system-level energy more than two orders of magnitude per a single object classification.”  
+
+<|ref|>table<|/ref|><|det|>[[118, 328, 878, 500]]<|/det|>
+<|ref|>table_caption<|/ref|><|det|>[[117, 273, 878, 323]]<|/det|>
+Table R1 (also Table 3 in the Revised Manuscript). System level energy consumption analysis per a single image classification task in each step of the computer vision depends on the network architecture.   
+
+<table><tr><td rowspan="2">Network architecture</td><td rowspan="2">Optical frontend</td><td colspan="2">Digital backend</td><td colspan="2">System</td></tr><tr><td>GeForce RTX 2080 Ti</td><td>State-of-the-art GPU</td><td>GeForce RTX 2080 Ti</td><td>State-of-the-art GPU</td></tr><tr><td>Original CNN with optimal camera pixels</td><td>2.91 × 10-5 J</td><td>2.74 × 10-2 J</td><td>3.65 × 10-3 J</td><td>2.74 × 10-2 J</td><td>3.68 × 10-3 J</td></tr><tr><td>Our hybrid optical/digital CNN with optimal camera pixels</td><td>3.28 × 10-5 J</td><td>1.13 × 10-6 J</td><td>1.50 × 10-7 J</td><td>3.40 × 10-5 J</td><td>3.30 × 10-5 J</td></tr></table>  
+
+<|ref|>text<|/ref|><|det|>[[118, 551, 878, 585]]<|/det|>
+4. Why the classification accuracy drops more significant for the CIFAR-10 dataset compared to the MNIST dataset? How about the classification accuracy for the MNIST dataset?  
+
+<|ref|>text<|/ref|><|det|>[[117, 592, 879, 791]]<|/det|>
+Response: Thank you for the valuable comment. In our previous work [12], the classification accuracy for MNIST dataset dropped from \(96.2\%\) to \(93.4\%\) ( \(= 2.8\%\) drop). In this work, the classification accuracy for CIFAR- 10 dataset drops from \(76.59\%\) to \(72.06\%\) ( \(73.17\%\) after retraining the backend, \(3.42\%\) drop). As the Reviewer mentioned, the classification accuracy drops slightly more in the case of CIFAR- 10 dataset, because it is more difficult and challenging to create multiple color PSFs from a single metasurface. The discrepancy between the ground- truth kernels and experimentally generated kernels for CIFAR- 10 dataset is the main reason why the CIFAR- 10 dataset has further accuracy drop. As we are working with more complex datasets, it is expected that the performance will decrease. MNIST is linearly separable and as such significantly simpler compared to CIFAR- 10. It is indeed expected that most neural networks will perform better when trained and evaluated on MNIST. As we mention in the paper, more complex scatters (not just a rectangular pillar) can adjust this discrepancy and reduce the accuracy drops while implementing the optical encoder experimentally.  
+
+<|ref|>text<|/ref|><|det|>[[118, 799, 878, 864]]<|/det|>
+If the Reviewer was referring to the accuracy drop during the knowledge distillation, it is correct that the classification accuracy dropped from \(98.4\%\) to \(96.2\%\) for MNIST dataset, while it drops from \(81.03\%\) to \(76.59\%\) for CIFAR- 10 dataset. This is simply due to the complexity of the dataset, where the accuracy drop is not severe for simple dataset such as monochromatic MNIST dataset.  
+
+<|ref|>text<|/ref|><|det|>[[118, 872, 878, 906]]<|/det|>
+[12] Wirth- Singh, A., Xiang, J., Choi, M., Fr'och, J.E., Huang, L., Colburn, S., Shlizerman, E., Majumdar, A.: Compressed meta- optical encoder for image classification. Advanced Photonics Nexus
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 140, 736, 157]]<|/det|>
+## 5. Physical sizes about the metasurfaces should be provided for the testing samples.  
+
+<|ref|>text<|/ref|><|det|>[[118, 167, 878, 232]]<|/det|>
+Response: We thank the reviewer for carefully reviewing the paper. The physical size of the metasurface was described in Supplementary materials under "Fabrication of the Meta- optics" section. We agree that some description could be added to the main text, and we added a sentence in the main text where we describe the optical image of the metasurfaces in Figure 3a:  
+
+<|ref|>text<|/ref|><|det|>[[118, 241, 580, 258]]<|/det|>
+"Each convolutional meta-optic has a size of \(\sim 940 \times 940 \mu \mathrm{m}^2\) ."
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 100, 404, 115]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 125, 878, 206]]<|/det|>
+This work presents a novel approach for a hybrid optical/digital neural network, performing the computationally expensive convolution operation in optics. The manuscript introduces an innovative design that demonstrates significant computational efficiency while maintaining competitive classification accuracy. The work can be considered for publication; however, the following concerns should be addressed:  
+
+<|ref|>text<|/ref|><|det|>[[118, 215, 875, 249]]<|/det|>
+Response: We thank the reviewer for carefully reviewing the paper and providing valuable feedback. We have provided a point- by- point response and modified the paper accordingly.  
+
+<|ref|>text<|/ref|><|det|>[[118, 273, 878, 355]]<|/det|>
+1. Phase Coverage of Meta-Atoms: Based on Figure 2b, the phase coverage of the meta-atoms appears to be around 1 radian, even when assuming impractically small feature sizes. For designing an effective metasurface, a phase coverage is typically required. The authors should clarify how they achieve the necessary phase modulation with this limited range or discuss the implications of this constraint on device performance.  
+
+<|ref|>text<|/ref|><|det|>[[118, 364, 878, 413]]<|/det|>
+Response: We very much appreciate the comment by the reviewer. We realized that the phase coverage of the meta-atoms that we note 1 radian is actually \(2\pi\) radians, and corrected the axis of Figure 2b. Our data was normalized but that information was not reflected in the figure.  
+
+<|ref|>text<|/ref|><|det|>[[118, 447, 878, 529]]<|/det|>
+2. Kernel-Image Interaction: The interaction mechanism between the kernels and the image is unclear. The kernels are spatially distributed in different locations, yet they seem to interact with the entire image. The authors should elaborate on how this is physically implemented, particularly addressing whether there are optical multiplexing effects or specific alignment strategies that enable this global interaction.  
+
+<|ref|>text<|/ref|><|det|>[[117, 538, 878, 686]]<|/det|>
+Response: We appreciate the reviewer's question. The main reason we can interact with the whole image with spatially separated kernels is due to using incoherent light without any directionality (unlike a laser). As we state in the paper, we believe the optical neural network can be only beneficial when the information is already in the optical domain. The case we chose here is to emulate imaging under ambient radiation. Indeed, pictures of the same object can be taken even though the camera is not physically located exactly at the same location. Thus, being compatible with incoherent light, the alignment of single optics with the sensor becomes a non-trivial problem. This is also the reason for placing the kernels in different locations (more quantitative details and limitations are below) see the same image.  
+
+<|ref|>text<|/ref|><|det|>[[117, 694, 878, 905]]<|/det|>
+We provide more information on the physical implementation below. We have a total 32 convolutional meta- optics, corresponding to 16 positive and negative digital kernels. All the meta- optics are spatially distributed on a single chip, and either point spread functions (PSFs) or convolved images from all the meta- optics are captured at a color camera simultaneously. When the input light is a laser and a pinhole, we measure the PSFs. And when the input light is an image on an OLED display, we measure the convolved images. For optical convolutional imaging, we capture 32 different convolved colorful images on the camera, which then undergo digital post- processing. In our physical configuration, where the size of the input image is relatively small (32 by 32 pixels), the off- axis spatial variance in the display is not severe. And due to the large distance between the display and the metasurface ( \(\sim 105 \mathrm{mm}\) ) compared to the distance between the metasurface and the sensor ( \(\sim 2.46 \mathrm{mm}\) ), the off- axis spatial variance in the metasurface is also not severe. The relative intensity between each convolutional metasurfaces is adjusted by the calibration layer, which is scene- independent. We would like to emphasize that the spatial variance does not play an important role in our case of
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 98, 877, 148]]<|/det|>
+classification tasks, as demonstrated by a small accuracy drop during the optical implementation, from \(76.59\%\) to \(72.06\%\) (73.17% after retraining the backend). However, those spatial distributions will be an issue for bigger input image size in a future work.  
+
+<|ref|>text<|/ref|><|det|>[[118, 157, 875, 175]]<|/det|>
+To clarify how we physically implement our optics, we add the following sentences in the manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[118, 183, 877, 265]]<|/det|>
+"Then, we test the polychromatic optical encoder for the CIFAR- 10 dataset. By replacing the pinhole with an organic light- emitting diode (OLED) display, optical convolutional operations between the input image and the PSFs occur with spatially separated meta- optics on a single chip, then captured on a color camera (Fig. 3d). On the color camera, 32 different convolved images are captured and then subjected to digital backend of calibration and fully- connected layers."  
+
+<|ref|>text<|/ref|><|det|>[[118, 273, 877, 323]]<|/det|>
+"Because of small size of the input image as well as small distance between the metasurfaces and the camera ( \(\sim 2.46 \mathrm{mm}\) ) compared to the distance between the display and the metasurface ( \(\sim 105 \mathrm{mm}\) ), the spatial distribution of the metasurfaces does not affect much on the convolutional results."  
+
+<|ref|>text<|/ref|><|det|>[[118, 356, 878, 472]]<|/det|>
+3. Scope for Improvement in Hybrid CNN: The authors claim that "Our hybrid optical/digital CNN can be further improved by using complex meta-atoms to reproduce better PSFs optically." However, the current performance of the hybrid neural network is already close to that of the compressed CNN, suggesting limited room for further enhancement. It would be beneficial for the authors to expand on how complex meta-atoms could offer substantial performance gains and to discuss other potential strategies for improving compressed CNN performance, perhaps focusing on advanced knowledge distillation or transfer learning techniques.  
+
+<|ref|>text<|/ref|><|det|>[[118, 480, 877, 547]]<|/det|>
+Response: We sincerely appreciate the reviewer's insightful comments regarding the potential for further improvement in our hybrid optical/digital CNN. Below, we clarify how complex meta-atoms contribute to performance gains and propose additional enhancement strategies. We added the following sentences to the manuscript in the Discussion-Opportunities for improvement section.  
+
+<|ref|>text<|/ref|><|det|>[[117, 554, 878, 719]]<|/det|>
+"While the current hybrid CNN achieves competitive performance, a noticeable gap ( \(\sim 4.4\%\) ) remains compared to the compressed CNN. This discrepancy can be mitigated by employing a more sophisticated PSF design, which enhances optical processing capabilities and reduces information loss. By leveraging complex meta-atoms with improved phase and amplitude control, the optical system can more accurately approximate ideal convolutional operations, thereby closing the performance gap. Additionally, performance discrepancy exists between the original CNN and its compressed counterpart. To address this, we could leverage advanced knowledge distillation techniques to enhance the compressed model's learning efficiency [1- 3]. By integrating both improved PSF design and advanced knowledge distillation methods, our approach can effectively bridge these gaps."  
+
+<|ref|>text<|/ref|><|det|>[[116, 725, 878, 857]]<|/det|>
+[1] Tian, Y., Krishnan, D., & Isola, P. (2020). Contrastive representation distillation. International Conference on Learning Representations (ICLR). https://arxiv.org/abs/1910.10699[2] Zagoruyko, S., & Komodakis, N. (2017). Paying more attention to attention: Improving the performance of convolutional neural networks via attention transfer. International Conference on Learning Representations (ICLR). https://arxiv.org/abs/1612.03928[3] Park, W., Kim, D., Lu, Y., & Cho, M. (2019). Relational knowledge distillation. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 3967- 3976. https://arxiv.org/abs/1904.05068  
+
+<|ref|>text<|/ref|><|det|>[[115, 881, 875, 899]]<|/det|>
+4. Generalization Performance: While the hybrid system achieves reasonable transferability from
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 99, 877, 148]]<|/det|>
+CIFAR- 10 to High- 10, the performance drop is notable. The authors could discuss potential design modifications or training strategies that might enhance the generalization capability of the optical encoder without the need for extensive backend retraining.  
+
+<|ref|>text<|/ref|><|det|>[[117, 157, 878, 320]]<|/det|>
+Response: We sincerely appreciate the reviewer's insightful question. Developing a generalized optical encoder is indeed challenging. Generally, an optical encoder optimized for a more complex and larger dataset becomes more capable of generalizing to simpler and smaller datasets. High- 10, while having the same number of classes as CIFAR- 10, contains fewer samples. The ablation study in Table 2 demonstrates that this transfer learning approach improves performance from \(40\%\) to \(66\%\) compared to end- to- end training, underscoring the effectiveness of transfer learning. It is important to emphasize that our physical optical encoder remains unchanged during the transfer process. Instead, we introduce a transfer learning layer (a fully connected layer) between the optical frontend and backend. This layer is fine- tuned exclusively during transfer learning to minimize extensive backend retraining.  
+
+<|ref|>text<|/ref|><|det|>[[118, 329, 877, 394]]<|/det|>
+We acknowledge the reviewer's observation that our current approach still underperforms compared to AlexNet with transfer learning. To enhance transfer learning quality, we propose designing a more accurate PSF and applying advanced knowledge distillation methods to better represent AlexNet, as elaborated in R2Q3.  
+
+<|ref|>text<|/ref|><|det|>[[118, 402, 877, 485]]<|/det|>
+Additionally, increasing the number of kernels can lead to better representation of AlexNet, thereby improving model generalization. For instance, while AlexNet has 256 kernels in its final convolutional layer, our system currently employs only 16 kernels. To enhance the single- layer optical encoder's ability to extract both shallow and deep features, and to better represent deep neural networks for higher accuracy, we refer the reviewer to R2Q7.  
+
+<|ref|>text<|/ref|><|det|>[[118, 493, 822, 510]]<|/det|>
+The following sentences have been added to the manuscript under Discussion- Transferability:  
+
+<|ref|>text<|/ref|><|det|>[[118, 524, 877, 589]]<|/det|>
+"The scalability of the hybrid approach to more complex, real- world datasets, such as ImageNet, remains a challenge due to large reduction in accuracy. The primary cause of reduced accuracy is network simplification, such as reducing the number of layers. For example, AlexNet has 256 kernels in its final convolutional layer, while we only employ 16 kernels.  
+
+<|ref|>text<|/ref|><|det|>[[118, 603, 877, 685]]<|/det|>
+Here, we transfer from CIFAR- 10 to the ImageNet subset (High- 10). High- 10 shares the same number of classes as CIFAR- 10 but contains fewer samples. Training High- 10 from scratch (with a simplified network) is already very challenging. The ablation study in Table 2 illustrates that this transfer learning approach improves performance from \(40\%\) to \(66\%\) compared to end- to- end training, reaffirming the efficacy of transfer learning.  
+
+<|ref|>text<|/ref|><|det|>[[118, 698, 877, 829]]<|/det|>
+MAC operations for the High- 10 dataset are the same as that of the CIFAR- 10 dataset, as the same CNN architecture is utilized. The train (test) accuracy of our hybrid CNN drops significantly by \(\sim 21.85\%\) ( \(\sim 25.22\%\) ) compared to the original CNN. Most of these losses occur during network compression, as the convolutional and fully connected layers are optimized for the CIFAR- 10 dataset. Nonetheless, our transfer learning and the optical encoder achieve a classification accuracy of \(\sim 60\%\) , outperforming other free- space optical neural networks. We emphasize that the optical frontend remains unchanged, and only the digital backend- comprising two fully connected layers and an additional transfer learning layer- is fine- tuned, showcasing the versatility of our hybrid CNN system.  
+
+<|ref|>text<|/ref|><|det|>[[118, 843, 877, 908]]<|/det|>
+Our current approach still lags behind AlexNet with transfer learning. To further enhance performance, we suggest potential design modifications and training strategies. From a design perspective, implementing additional optical kernels could be a solution. To achieve this within physical constraints, we could employ multiple cameras for different kernels or modify the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 99, 878, 165]]<|/det|>
+metasurfaces (e.g., by rotating them) while using a single camera. From a training perspective, adopting advanced knowledge distillation methods could better represent AlexNet with greater accuracy. Certain kernels could be aligned with the shallow- layer features of the electronic CNN, while others could focus on capturing deeper features."  
+
+<|ref|>text<|/ref|><|det|>[[118, 202, 878, 270]]<|/det|>
+5. Physical Realizability of PSFs: The paper acknowledges discrepancies between the ground-truth and experimentally measured PSFs (Figure 3c), partly due to fabrication imperfections and the inherent limitations of phase control across RGB channels. A more detailed analysis of how these imperfections affect system robustness and potential mitigation strategies would strengthen the paper.  
+
+<|ref|>text<|/ref|><|det|>[[118, 278, 878, 360]]<|/det|>
+Response: Thank you for the valuable comment. We use a 2 by 2 scatterer set to create the metasurface and make it more robust to fabrication imperfection (if one of four scatterers fall down, we still have three left). Still, we believe there are three major reasons for the discrepancies between the ground- truth and experimentally measured PSFs. We included the following in a new section Discussion-Opportunities for improvement.  
+
+<|ref|>text<|/ref|><|det|>[[123, 368, 880, 565]]<|/det|>
+(1) Imperfect fitting function for the phase over scatterer: We assumed the scatterers have constant transmission and do not have any resonant features in their relative phases, which is entirely accurate (Fig. 2b).  
+(2) Limited degree of freedom of the metasurface for multicolor PSFs: We optimized each of the metasurfaces targeting three different PSFs in red, green, and blue colors. The phase profiles of one rectangular scatterer for three different colors are not independent of each other. One solution is to use complex-shaped scatterers or super cells to have more degrees of freedom.  
+(3) Broadband light sources: We simulated the polychromatic metasurfaces at three discrete wavelengths (450, 532, and 635 nm). However, the OLED pixels have much broader wavelengths. We can reduce this discrepancy if we optimize the metasurface for more representative wavelengths.  
+
+<|ref|>text<|/ref|><|det|>[[118, 574, 878, 657]]<|/det|>
+However, the classification accuracy reported here for the CIFAR- 10 dataset is considerable compared to the other reported results in optical neural networks. There will be inevitable imperfection in optical implementation as the spectral information of the scene is always changing depending on the daylight, cloud, aerial, and many other conditions. Our results show that the digital backend can compensate for the non- ideal optical implementation and achieve a relatively high classification accuracy.  
+
+<|ref|>text<|/ref|><|det|>[[118, 692, 878, 774]]<|/det|>
+6. Energy Efficiency vs. Sensor Overhead: The authors highlight a significant reduction in computational energy consumption but mention that the optical encoder requires capturing more pixels than the original CNN, increasing the sensor's energy consumption. A deeper quantitative comparison of total system energy consumption, considering both sensor and backend, would provide a clearer picture of practical efficiency gains.  
+
+<|ref|>text<|/ref|><|det|>[[118, 782, 878, 912]]<|/det|>
+Response: Thank you for the valuable comment. The color camera we used (Allied Vision Prosilica; GT 1930 C) has a total power consumption of 3.4W with 50.70 frames per second and 1,936×1,216 color pixels, which ends up with 28 nJ per frame and pixel. Since we captured all \(\sim 2\) million pixels at the same time and cropped the region of interest, the energy consumption for one input image does not differ for the original CNN and hybrid CNN. However, if we can optimize the sensor configuration and number of pixels, we can calculate the minimum required number of pixels for both the original and hybrid CNN, and estimate the energy consumption for those. For the original CNN, we need 32×32 color pixels on camera. And for the hybrid CNN, we need 32×6×6 color pixels on
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 99, 879, 392]]<|/det|>
+the camera, where the 32 corresponds to number of multiplexed meta- optics and the \(6 \times 6\) corresponds to number of pixels after the average pooling. We only need \(6 \times 6\) pixels, not \(32 \times 32\) pixels when the convolution is already done optically. Thus we estimate that the original CNN and hybrid CNN require an energy of about \(29.1 \mu \mathrm{J}\) and \(32.8 \mu \mathrm{J}\) , respectively, for the image capturing process per a single image. Next, the energy consumption for the computational backend is much larger for the original CNN compared to the hybrid CNN. For state- of- the- art computational systems, an energy consumption per a single MAC operation is \(1 \mathrm{pJ}\) . Thus the energy consumption for a single object classification task for the hybrid CNN is about \(150 \mathrm{nJ}\) , which is more than four orders of magnitude smaller than that of the original CNN, \(3.65 \mathrm{mJ}\) . While the GPU we used (GeForce RTX 2080 Ti) has much larger energy consumption per a single MAC operation ( \(\sim 7.5 \mathrm{pJ}\) ), making the energy consumption for a single object classification tasks for the hybrid CNN and original CNN \(1.13 \mu \mathrm{J}\) and \(27.4 \mathrm{mJ}\) . Considering the sensor power, the total system level energy consumption for a single object classification task dropped from \(3.68 \mathrm{mJ}\) to \(0.03 \mathrm{mJ}\) for the state- of- the- art digital processor, while \(27.4 \mathrm{mJ}\) to \(34.0 \mu \mathrm{J}\) for our GPU. We emphasize that more than two orders of magnitude reduction in the system level computer vision tasks clearly provide strong benefits for practical implementation even with a partial accuracy drop for some application fields. We emphasize that to the best of our knowledge, our paper is the first paper which truly calculated the “energy consumption” by considering the sensor/ camera power.  
+
+<|ref|>text<|/ref|><|det|>[[118, 400, 878, 450]]<|/det|>
+We explain about the system energy consumption (with more details and an additional Table R1, highlighted in the manuscript) in the section “Energy consumption” in Discussion and add a sentence in Abstract as following:  
+
+<|ref|>text<|/ref|><|det|>[[118, 458, 878, 491]]<|/det|>
+“The proposed method can decrease total system- level energy more than two orders of magnitude per a single object classification.”  
+
+<|ref|>table<|/ref|><|det|>[[118, 556, 878, 730]]<|/det|>
+<|ref|>table_caption<|/ref|><|det|>[[117, 500, 878, 550]]<|/det|>
+Table R1 (also Table 3 in the Revised Manuscript). System level energy consumption analysis per a single image classification task in each step of the computer vision depends on the network architecture.   
+
+<table><tr><td rowspan="2">Network architecture</td><td>Optical frontend</td><td colspan="2">Digital backend</td><td colspan="2">System</td></tr><tr><td>GT 1930C</td><td>GeForce RTX 2080 Ti</td><td>State-of-the-art GPU</td><td>GeForce RTX 2080 Ti</td><td>State-of-the-art GPU</td></tr><tr><td>Original CNN with optimal camera pixels</td><td>2.91 × 10-5 J</td><td>2.74 × 10-2 J</td><td>3.65 × 10-3 J</td><td>2.74 × 10-2 J</td><td>3.68 × 10-3 J</td></tr><tr><td>Our hybrid optical/digital CNN with optimal camera pixels</td><td>3.28 × 10-5 J</td><td>1.13 × 10-6 J</td><td>1.50 × 10-7 J</td><td>3.40 × 10-5 J</td><td>3.30 × 10-5 J</td></tr></table>  
+
+<|ref|>text<|/ref|><|det|>[[118, 753, 878, 835]]<|/det|>
+7. Scalability and Practical Deployment: The current implementation focuses on relatively small datasets (CIFAR- 10, High- 10). The scalability of the system to more complex, real-world datasets (e.g., full ImageNet) remains unclear. It would be valuable if the authors discussed potential bottlenecks in scaling the optical encoder, such as fabrication complexity, alignment issues, or increased noise sensitivity.  
+
+<|ref|>text<|/ref|><|det|>[[118, 844, 878, 894]]<|/det|>
+Response: Thank you for the valuable comment. While we completely agree with the reviewer, we want to emphasize that the datasets used in our work are one of the most complex datasets reported in the optical neural network literature.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[119, 100, 386, 116]]<|/det|>
+## Factors Causing Accuracy Drops:  
+
+<|ref|>text<|/ref|><|det|>[[118, 116, 879, 246]]<|/det|>
+Factors Causing Accuracy Drops:The accuracy drop arises from both network simplification (reducing layers) and physical implementation issues (e.g., fabrication misalignment). First, converting a deep neural network into a shallow single- layer CNN introduces an inherent accuracy drop. Second, the physical implementation of the hybrid approach, such as fabrication misalignment, further reduces performance. This issue becomes more pronounced when applied to more complex datasets, such as ImageNet. Using 16 kernels, we achieved top- 1 and top- 5 classification accuracies of \(35\%\) and \(58\%\) for ImageNet- 100, respectively, after compression, compared to the original AlexNet's top- 1 and top- 5 accuracies of \(56\%\) and \(78\%\) , respectively, before compression.  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 260, 460, 276]]<|/det|>
+## Algorithmic and Structural Improvements:  
+
+<|ref|>text<|/ref|><|det|>[[118, 277, 875, 415]]<|/det|>
+Algorithmic and Structural Improvements:To address scalability for large datasets, we have demonstrated the potential to design a more accurate PSF and apply more advanced knowledge distillation methods to better represent AlexNet, as shown in R2Q3- 4. Beyond algorithmic improvements, structural modifications to the network can further enhance the performance of optical neural networks. To enable better representation of the empirical teacher network, we could convert a multi- layer CNN into a single- layer ONN. For example, certain kernels could be aligned with shallow- layer features from an electronic CNN, while others could be designated to capture deeper features. Additionally, employing residual blocks could mitigate information loss and improve feature representation.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 430, 313, 445]]<|/det|>
+## Optical Implementation:  
+
+<|ref|>text<|/ref|><|det|>[[118, 446, 878, 544]]<|/det|>
+First, as the number of kernels increases, it becomes challenging to measure all the convolved images simultaneously using a single camera. Multiple cameras may be required, which could introduce additional time delays in synchronizing readout events and significantly increase energy consumption for image readout. Alternatively, metasurfaces could be switched with temporal multiplexing (e.g., using a rotating wheel) while using a single camera; however, this would result in substantial time delays for capturing the complete set of image readouts.  
+
+<|ref|>text<|/ref|><|det|>[[118, 551, 876, 585]]<|/det|>
+We included the following in a new section "Transferability" and "Opportunities for improvement" in Discussion with highlighted notes..  
+
+<|ref|>text<|/ref|><|det|>[[118, 619, 878, 686]]<|/det|>
+8. Typos and Minor Errors: There are several typographical errors throughout the manuscript that need correction. A thorough proofreading is recommended to improve the manuscript's readability. Overall, this paper introduces an exciting advancement in hybrid optical/digital neural networks, but addressing the above concerns will significantly enhance its clarity, robustness, and impact.  
+
+<|ref|>text<|/ref|><|det|>[[118, 694, 875, 727]]<|/det|>
+Response: Thank you for the valuable comment. We reviewed the paper thoroughly and corrected all the typos in our best.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 100, 404, 115]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 125, 878, 239]]<|/det|>
+This paper introduces a hybrid optical/digital neural network architecture that leverages polychromatic meta- optics to perform convolutional operations during image capture, reducing computational load. The system achieves \(73.17\%\) accuracy on CIFAR- 10 and demonstrates transferability to an ImageNet subset (High- 10) with moderate accuracy. The work addresses critical challenges in real- time, energy- efficient computer vision and offers a novel integration of optical encoding with digital backends. However, I feel the authors should clarify a few points below before being considered for publication in Nature Communications.  
+
+<|ref|>text<|/ref|><|det|>[[118, 247, 875, 281]]<|/det|>
+Response: We thank the reviewer for carefully reviewing the paper and providing valuable feedback. We have provided a point- by- point response and modified the paper accordingly.  
+
+<|ref|>text<|/ref|><|det|>[[118, 305, 877, 355]]<|/det|>
+1. Regarding the accuracy Trade-offs, the accuracy drop \((-8\%)\) from AlexNet on CIFAR-10, \(-25\%\) on High-10) may limit adoption in "safety-critical" applications. Could the authors describe how one could potentially achieve better accuracy?  
+
+<|ref|>text<|/ref|><|det|>[[118, 363, 878, 445]]<|/det|>
+Response: Thank you for the valuable comment. We agree that the reduction of accuracy will not be permitted for safety-critical situations (e.g., self- driving cars). Still, our accuracy on CIFAR- 10 is better than any other free- space optical implementation of neural networks. And there are usages of a moderate accuracy AI (but with high energy efficiency and low latency) for non- safety- critical situations as we describe in the Discussion- Applications section.  
+
+<|ref|>text<|/ref|><|det|>[[118, 454, 878, 536]]<|/det|>
+In order to reduce the accuracy drop, both optical and computational approaches can be applied to advance our current hybrid CNN. For an optical approach, we would like to answer the next question. For a computation approach, we could leverage advanced knowledge distillation techniques to enhance the compressed model's learning efficiency, such as those introduced in [1- 3]. We added the following sentences on the manuscript in the Discussion- Multichannel dataset section:  
+
+<|ref|>text<|/ref|><|det|>[[117, 544, 878, 708]]<|/det|>
+"While the current hybrid CNN achieves competitive performance, a noticeable gap \((- 4.4\%)\) remains compared to the compressed CNN. This discrepancy can be mitigated by employing a more sophisticated PSF design, which enhances optical processing capabilities and reduces information loss. By leveraging complex meta- atoms with improved phase and amplitude control, the optical system can more accurately approximate ideal convolutional operations, thereby closing the performance gap. Additionally, performance discrepancy exists between the original CNN and its compressed counterpart. To address this, we could leverage advanced knowledge distillation techniques to enhance the compressed model's learning efficiency [1- 3]. By integrating both improved PSF design and advanced knowledge distillation methods, our approach can effectively bridge these gaps."  
+
+<|ref|>text<|/ref|><|det|>[[117, 716, 878, 845]]<|/det|>
+[1] Tian, Y., Krishnan, D., & Isola, P. (2020). Contrastive representation distillation. International Conference on Learning Representations (ICLR). https://arxiv.org/abs/1910.10699[2] Zagoruyko, S., & Komodakis, N. (2017). Paying more attention to attention: Improving the performance of convolutional neural networks via attention transfer. International Conference on Learning Representations (ICLR). https://arxiv.org/abs/1612.03928[3] Park, W., Kim, D., Lu, Y., & Cho, M. (2019). Relational knowledge distillation. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 3967- 3976. https://arxiv.org/abs/1904.05068  
+
+<|ref|>text<|/ref|><|det|>[[115, 871, 877, 905]]<|/det|>
+2- 1. Could the author comment on fabrication Limitations: The impact of PSF discrepancies (e.g., \(\eta = 0.56\) for green) is under- analysed. Quantifying fabrication tolerances and robustness to
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 100, 437, 116]]<|/det|>
+misalignment would strengthen the results.  
+
+<|ref|>text<|/ref|><|det|>[[118, 125, 878, 175]]<|/det|>
+2- 2. The proxy function(i.e., Eq. 1) ignores resonance effects. How does this approximation affect PSF fidelity, especially for green ( \(\eta = 0.56\) )? The authors are suggested to include an error analysis across all kernels.  
+
+<|ref|>text<|/ref|><|det|>[[118, 183, 878, 265]]<|/det|>
+Response: Thank you for the valuable comment. We used a 2 by 2 scatterer set to create a metasurface, making it more robust to fabrication imperfection (if one of four scatterers fall down, we still have three left). Still, we believe there are three major reasons for the discrepancies between the ground- truth and experimentally measured PSFs. We included the following in a new section "Opportunities for improvement" in Discussion.  
+
+<|ref|>text<|/ref|><|det|>[[147, 273, 878, 451]]<|/det|>
+(1) Imperfect fitting function for the phase over scatterer: We assumed the scatterers have constant transmission and do not have any resonant features in their relative phases, which is entirely accurate (Fig. 2b).  
+(2) Limited degree of freedom of the metasurface for multicolor PSFs: We optimized each of the metasurfaces targeting three different PSFs in red, green, and blue colors. The phase profiles of one rectangular scatterer for three different colors are not independent of each other. One solution is to use complex-shaped scatterers or super cells to have more degrees of freedom.  
+(3) Broadband light sources: We simulated the polychromatic metasurfaces at three discrete wavelengths (450, 532, and 635 nm). However, the OLED pixels have much broader wavelengths. We can reduce this discrepancy if we optimize the metasurface for more representative wavelengths.  
+
+<|ref|>text<|/ref|><|det|>[[118, 461, 878, 544]]<|/det|>
+However, the classification accuracy reported here for the CIFAR- 10 dataset is considerable compared to the other reported results in optical neural networks. There will be inevitable imperfection in optical implementation as the spectral information of the scene is always changing depending on the daylight, cloud, aerial, and many other conditions. Our results show that the digital backend can compensate for the non- ideal optical implementation and achieve a relatively high classification accuracy.  
+
+<|ref|>text<|/ref|><|det|>[[118, 577, 876, 611]]<|/det|>
+3. Scalability: The system uses 32 meta-optics for 16 kernels. How does this scale to deeper networks or larger datasets? A discussion on physical size constraints is missing.  
+
+<|ref|>text<|/ref|><|det|>[[118, 619, 877, 669]]<|/det|>
+Response: We thank the reviewer for carefully reviewing the paper. Scaling to deeper networks or larger datasets is challenging and requires careful consideration of both optical neural networks and knowledge distillation strategies.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 678, 386, 694]]<|/det|>
+## Factors Causing Accuracy Drops:  
+
+<|ref|>text<|/ref|><|det|>[[118, 695, 878, 824]]<|/det|>
+The accuracy drop arises from both network simplification (reducing layers) and physical implementation issues (e.g., fabrication misalignment). First, converting a deep neural network into a shallow single- layer CNN introduces an inherent accuracy drop. Second, the physical implementation of the hybrid approach, such as fabrication misalignment, further reduces performance. This issue becomes more pronounced when applied to more complex datasets, such as ImageNet. Using 16 kernels, we achieved top- 1 and top- 5 classification accuracies of \(35\%\) and \(58\%\) for ImageNet- 100, respectively, after compression, compared to the original AlexNet's top- 1 and top- 5 accuracies of \(56\%\) and \(78\%\) , respectively, before compression.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 839, 460, 855]]<|/det|>
+## Algorithmic and Structural Improvements:  
+
+<|ref|>text<|/ref|><|det|>[[118, 856, 875, 907]]<|/det|>
+To address scalability for large datasets, we have demonstrated the potential to design a more accurate PSF and apply more advanced knowledge distillation methods to better represent AlexNet, as shown in R2Q3- 4. Beyond algorithmic improvements, structural modifications to the network can further
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 99, 863, 185]]<|/det|>
+enhance the performance of optical neural networks. To enable better representation of the empirical teacher network, we could convert a multi- layer CNN into a single- layer ONN. For example, certain kernels could be aligned with shallow- layer features from an electronic CNN, while others could be designated to capture deeper features. Additionally, employing residual blocks could mitigate information loss and improve feature representation.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 201, 312, 216]]<|/det|>
+## Optical Implementation:  
+
+<|ref|>text<|/ref|><|det|>[[118, 216, 878, 314]]<|/det|>
+First, as the number of kernels increases, it becomes challenging to measure all the convolved images simultaneously using a single camera. Multiple cameras may be required, which could introduce additional time delays in synchronizing readout events and significantly increase energy consumption for image readout. Alternatively, metasurfaces could be switched with temporal multiplexing (e.g., using a rotating wheel) while using a single camera; however, this would result in substantial time delays for capturing the complete set of image readouts.  
+
+<|ref|>text<|/ref|><|det|>[[120, 322, 765, 339]]<|/det|>
+The following sentences have been added to the manuscript Discussion- Transferability:  
+
+<|ref|>text<|/ref|><|det|>[[118, 353, 878, 418]]<|/det|>
+"The scalability of the hybrid approach to more complex, real- world datasets, such as ImageNet, remains a challenge due to large reduction in accuracy. The primary cause of reduced accuracy is network simplification, such as reducing the number of layers. For example, AlexNet has 256 kernels in its final convolutional layer, while we only employ 16 kernels.  
+
+<|ref|>text<|/ref|><|det|>[[118, 432, 878, 514]]<|/det|>
+Here, we transfer from CIFAR- 10 to the ImageNet subset (High- 10). High- 10 shares the same number of classes as CIFAR- 10 but contains fewer samples. Training High- 10 from scratch (with a simplified network) is already very challenging. The ablation study in Table 2 illustrates that this transfer learning approach improves performance from \(40\%\) to \(66\%\) compared to end- to- end training, reaffirming the efficacy of transfer learning.  
+
+<|ref|>text<|/ref|><|det|>[[118, 527, 878, 658]]<|/det|>
+MAC operations for the High- 10 dataset are the same as that of the CIFAR- 10 dataset, as the same CNN architecture is utilized. The train (test) accuracy of our hybrid CNN drops significantly by \(\sim 21.85\%\) ( \(\sim 25.22\%\) ) compared to the original CNN. Most of these losses occur during network compression, as the convolutional and fully connected layers are optimized for the CIFAR- 10 dataset. Nonetheless, our transfer learning and the optical encoder achieve a classification accuracy of \(\sim 60\%\) , outperforming other free- space optical neural networks. We emphasize that the optical frontend remains unchanged, and only the digital backend- comprising two fully connected layers and an additional transfer learning layer- is fine- tuned, showcasing the versatility of our hybrid CNN system.  
+
+<|ref|>text<|/ref|><|det|>[[118, 671, 878, 802]]<|/det|>
+Our current approach still lags behind AlexNet with transfer learning. To further enhance performance, we suggest potential design modifications and training strategies. From a design perspective, implementing additional optical kernels could be a solution. To achieve this within physical constraints, we could employ multiple cameras for different kernels or modify the metasurfaces (e.g., by rotating them) while using a single camera. From a training perspective, adopting advanced knowledge distillation methods could better represent AlexNet with greater accuracy. Certain kernels could be aligned with the shallow- layer features of the electronic CNN, while others could focus on capturing deeper features."  
+
+<|ref|>text<|/ref|><|det|>[[118, 840, 878, 875]]<|/det|>
+4. Calibration Layer Dependency: The reliance on a digital calibration layer to correct optical imperfections partially offsets the analog advantage. Clarify if this layer adds computational overhead.  
+
+<|ref|>text<|/ref|><|det|>[[115, 882, 878, 900]]<|/det|>
+Response: Thank you for the valuable comment. The digital calibration layer depends on fabrication
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 98, 878, 181]]<|/det|>
+misalignment and other noise, and such misalignment can be removed theoretically. Assuming the original desired kernel parameter is \(\theta\) , and after fabrication, some misalignment occurs, the final parameter becomes \(\theta + \delta_{\theta}\) . Since this is a continuous layer, we can directly compute the impact of \(\delta_{\theta}\) on the output and subsequently remove the error \(\delta_{\theta}\) , ensuring minimal deviation from the intended values. we can express the desired output correction as:  
+
+<|ref|>equation<|/ref|><|det|>[[404, 190, 581, 212]]<|/det|>
+\[y_{\text{desired}} = \theta *\text{Input}\]  
+
+<|ref|>equation<|/ref|><|det|>[[339, 220, 650, 245]]<|/det|>
+\[y_{\text{optical}} = (\theta + \delta_{\theta}) * \text{Input} + \text{Noise}\]  
+
+<|ref|>equation<|/ref|><|det|>[[323, 253, 661, 278]]<|/det|>
+\[y_{\text{desired}} = y_{\text{optical}} - \delta_{\theta} * \text{Input} - \text{Noise}\]  
+
+<|ref|>text<|/ref|><|det|>[[118, 285, 877, 320]]<|/det|>
+Where \(y_{\text{desired}}\) is the corrected output, \(y_{\text{optical}}\) is the raw optical network output. Noise accounts for additional sources of error such as system.  
+
+<|ref|>text<|/ref|><|det|>[[118, 327, 878, 394]]<|/det|>
+The calibration layer functions as a fully- connected layer and can be integrated into the backend without introducing additional computational overhead. Assuming the calibration parameters are \(\mathrm{W}_{1}\) and \(\mathrm{b}_{1}\) , and the final fully- connected layer in the backend has parameters \(\mathrm{W}_{2}\) and \(\mathrm{B}_{2}\) , we can merge them into a single layer. The new parameters would be:  
+
+<|ref|>equation<|/ref|><|det|>[[428, 400, 564, 450]]<|/det|>
+\[W^{\prime} = W_{1}\times W_{2\] \[b^{\prime} = W_{2}b_{1} + b_{2\]  
+
+<|ref|>text<|/ref|><|det|>[[118, 471, 878, 522]]<|/det|>
+5. Could the authors comment on real-World applicability by testing under variable illumination/backgrounds (not just controlled lab conditions) as this way would better demonstrate practical utility?  
+
+<|ref|>text<|/ref|><|det|>[[117, 529, 879, 678]]<|/det|>
+Response: We really appreciate the reviewer for the question. We do agree with the Reviewer that implementing on a real- World scene means much more than the controlled laboratory conditions. The digital backend did well to compromise various errors and imperfections generated from the optical implementation even in the laboratory conditions. On the other hand, the uncontrolled situation in a natural scene (light intensity, color temperature, shadow, etc) will definitely cause much severe noise on our hybrid CNN. In fact, we are currently working on testing these optics under ambient illumination. However, creating a real- world dataset is not straight- forward. In order to do that, we may have to define constraints for the dataset, and put two cameras parallel and use a normal lens for one while use our metasurface for the other, a setup we recently reported [4].  
+
+<|ref|>text<|/ref|><|det|>[[118, 685, 877, 735]]<|/det|>
+[4] Fröch, J.E., Chakravarthula, P.K., Sun, J., Tseng, E., Colburn, S., Zhan, A., Miller, F., Wirth- Singh, A., Tanguy, Q.A., Han, Z., et al.: Beating bandwidth limits for large aperture broadband nano- optics. arXiv preprint arXiv:2402.06824 (2024)  
+
+<|ref|>text<|/ref|><|det|>[[117, 768, 877, 819]]<|/det|>
+6. The authors wrote "Thus the energy consumption for a single object classification task for the hybrid CNN is about 150nJ, which is more than four orders of magnitude smaller than that of the original CNN." Does the backend energy savings justify increased sensor costs?  
+
+<|ref|>text<|/ref|><|det|>[[117, 826, 878, 909]]<|/det|>
+Response: Thank you for the valuable comment. The color camera we used (Allied Vision Prosilica; GT 1930 C) has a total power consumption of 3.4W with 50.70 frames per second and 1,936×1,216 color pixels, which ends up with 28 nJ per frame and pixel. Since we captured all \(\sim 2\) million pixels at the same time and cropped the region of interest, the energy consumption for one input image does not differ for the original CNN and hybrid CNN. However, if we can optimize the sensor
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[116, 99, 879, 440]]<|/det|>
+configuration and number of pixels, we can calculate the minimum required number of pixels for both the original and hybrid CNN, and estimate the energy consumption for those. For the original CNN, we need \(32 \times 32\) color pixels on camera. And for the hybrid CNN, we need \(32 \times 6 \times 6\) color pixels on the camera, where the 32 corresponds to number of multiplexed meta- optics and the \(6 \times 6\) corresponds to number of pixels after the average pooling. We only need \(6 \times 6\) pixels, not \(32 \times 32\) pixels when the convolution is already done optically. Thus we estimate that the original CNN and hybrid CNN require an energy of about \(29.1 \mu \mathrm{J}\) and \(32.8 \mu \mathrm{J}\) , respectively, for the image capturing process per a single image. Next, the energy consumption for the computational backend is much larger for the original CNN compared to the hybrid CNN. For state- of- the- art computational systems, an energy consumption per a single MAC operation is \(1 \mathrm{pJ}\) . Thus the energy consumption for a single object classification task for the hybrid CNN is about \(150 \mathrm{nJ}\) , which is more than four orders of magnitude smaller than that of the original CNN, \(3.65 \mathrm{mJ}\) . While the GPU we used (GeForce RTX 2080 Ti) has much larger energy consumption per a single MAC operation ( \(\sim 7.5 \mathrm{pJ}\) ), making the energy consumption for a single object classification tasks for the hybrid CNN and original CNN \(1.13 \mu \mathrm{J}\) and \(27.4 \mathrm{mJ}\) . Considering the sensor power, the total system level energy consumption for a single object classification task dropped from \(3.68 \mathrm{mJ}\) to \(0.03 \mathrm{mJ}\) for the state- of- the- art digital processor, while \(27.4 \mathrm{mJ}\) to \(34.0 \mu \mathrm{J}\) for our GPU. We emphasize that more than two orders of magnitude reduction in the system level computer vision tasks clearly provide strong benefits for practical implementation even with a partial accuracy drop for some application fields. We emphasize that to the best of our knowledge, our paper is the first paper which truly calculated the “energy consumption” by considering the sensor/ camera power.  
+
+<|ref|>text<|/ref|><|det|>[[118, 448, 878, 499]]<|/det|>
+We explain about the system energy consumption (with more details and an additional Table R1, highlighted in the manuscript) in the section “Energy consumption�� in Discussion and add a sentence in Abstract as following:  
+
+<|ref|>text<|/ref|><|det|>[[118, 507, 877, 540]]<|/det|>
+“The proposed method can decrease total system- level energy more than two orders of magnitude per a single object classification.”  
+
+<|ref|>text<|/ref|><|det|>[[118, 548, 878, 597]]<|/det|>
+Table R1 (also Table 3 in the Revised Manuscript). System level energy consumption analysis per a single image classification task in each step of the computer vision depends on the network architecture.  
+
+<|ref|>table<|/ref|><|det|>[[118, 605, 877, 777]]<|/det|>
+
+<table><tr><td rowspan="2">Network architecture</td><td>Optical frontend</td><td colspan="2">Digital backend</td><td colspan="2">System</td></tr><tr><td>GT 1930C</td><td>GeForce RTX 2080 Ti</td><td>State-of-the-art GPU</td><td>GeForce RTX 2080 Ti</td><td>State-of-the-art GPU</td></tr><tr><td>Original CNN with optimal camera pixels</td><td>2.91 × 10-5 J</td><td>2.74 × 10-2 J</td><td>3.65 × 10-3 J</td><td>2.74 × 10-2 J</td><td>3.68 × 10-3 J</td></tr><tr><td>Our hybrid optical/digital CNN with optimal camera pixels</td><td>3.28 × 10-5 J</td><td>1.13 × 10-6 J</td><td>1.50 × 10-7 J</td><td>3.40 × 10-5 J</td><td>3.30 × 10-5 J</td></tr></table>  
+
+<|ref|>text<|/ref|><|det|>[[118, 801, 877, 835]]<|/det|>
+7. The added “transfer learning layer” is not detailed in the main text. Could the authors provide architecture specifics and ablation studies?  
+
+<|ref|>text<|/ref|><|det|>[[118, 844, 878, 894]]<|/det|>
+Response: Thank you for the valuable comment. The transfer learning layer is a single fully connected layer used to reproject feature clustering. The ablation study is already included in Table 2, showing that this transfer learning layer improves performance from \(40\%\) to \(66\%\) .
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 124, 875, 157]]<|/det|>
+8. By comparing accuracy/MAC reductions with recent works, could the author highlight how polychromatic encoding advances the field?  
+
+<|ref|>text<|/ref|><|det|>[[117, 166, 878, 312]]<|/det|>
+Response: Thank you for the valuable comment. For the CIFAR- 10 dataset, our hybrid optical/digital CNN reduced the number of MAC operations by a factor of \(\sim 24,000\) (Table A1). This reduction is more than two orders of magnitude higher than that of the MNIST hand-written dataset, where the meta- optical encoder reduced the number of MAC operations only by a factor of \(\sim 200\) [5]. The far increased reduction of digital operations for CIFAR- 10 dataset compared to MNIST dataset is an important benefit in both energy consumption and latency. This is mainly because of the complexity of the dataset (including the polychromatic nature of the dataset), where it requires a much larger number of operations for convolutional layers. We added this description at section "Multichannel dataset" in Discussion.  
+
+<|ref|>text<|/ref|><|det|>[[118, 321, 877, 371]]<|/det|>
+[5] Wirth- Singh, A., Xiang, J., Choi, M., Fr'och, J.E., Huang, L., Colburn, S., Shlizerman, E., Majumdar, A.: Compressed meta- optical encoder for image classification. Advanced Photonics Nexus 4(2), 026009- 026009 (2025)  
+
+<|ref|>text<|/ref|><|det|>[[118, 405, 877, 438]]<|/det|>
+9. Since the authors used the system to process static images, could they discuss feasibility for video streams and temporal feature extraction?  
+
+<|ref|>text<|/ref|><|det|>[[117, 447, 878, 592]]<|/det|>
+Response: We sincerely appreciate the reviewer's question. Real- time operation and latency remain critical challenges in AI. Our hybrid optical/digital architecture reduces the number of digital operations by more than four orders of magnitude, significantly lowering latency. Similar to the convolutional layers in AlexNet, the optical frontend has the potential to extract spatial features from individual frames. These spatial features can then be fed into RNNs, LSTMs, or Transformers to enhance temporal feature understanding. Thus, our approach could be highly relevant for video streaming applications, such as video conferencing, where real- time processing is essential but safety- critical constraints are minimal. We have added the following sentences to the manuscript in the Discussion- Applications section:  
+
+<|ref|>text<|/ref|><|det|>[[117, 602, 878, 684]]<|/det|>
+"On the other hand, real- time operation and latency remain among the most significant challenges in AI. Our hybrid optical/digital architecture reduced more than four orders of magnitude of digital operations, thereby substantially decreasing both power and latency. Consequently, our approach has the potential to create a significant impact on video conferencing which align well with non- safety- critical scenarios."
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 99, 325, 115]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 125, 422, 141]]<|/det|>
+## Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[117, 150, 787, 168]]<|/det|>
+As the authors addressed all my concerns, I recommend it can be accepted for publication.  
+
+<|ref|>text<|/ref|><|det|>[[118, 177, 520, 193]]<|/det|>
+Response: We thank the reviewer for a careful review.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 228, 422, 244]]<|/det|>
+## Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 253, 878, 287]]<|/det|>
+The authors have properly addressed all my comments from the previous round of review. I now recommend accepting this manuscript for publishing in Nature Communications.  
+
+<|ref|>text<|/ref|><|det|>[[118, 295, 520, 312]]<|/det|>
+Response: We thank the reviewer for a careful review.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Transparent Peer Review file__e953eb070ecccde9cf6628bea8dbe26c3e9c5cfaeccbb37878e543b2e142d42b/images_list.json

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

peer_reviews/supplementary_0_Transparent Peer Review file__e953eb070ecccde9cf6628bea8dbe26c3e9c5cfaeccbb37878e543b2e142d42b/supplementary_0_Transparent Peer Review file__e953eb070ecccde9cf6628bea8dbe26c3e9c5cfaeccbb37878e543b2e142d42b.mmd

@@ -0,0 +1,379 @@
+
+# Tumor immune dynamics and long-term clinical outcome of stage IIIA NSCLC patients treated with neoadjuvant chemoimmunotherapy  
+
+Corresponding Author: Dr Dominic Schmid  
+
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+Version 0:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+Thank you for the opportunity to review this interesting manuscript, wherein the authors report on multi- omic translational analysis of specimens obtained during the SAKK 16/14 trial. This study enrolled patients with Stage IIIA NSCLC (N2 involvement) treated with neoadjuvant platinum- based chemotherapy followed 3 cycles followed by 2 cycles of durvalumab and resection. Patients then underwent 1 year of adjuvant durvalumab treatment. Translational studies included tissue and blood interrogation at baseline (prior to chemoimmunotherapy; TP1), prior to neoadjuvant durvalumab (TP2), prior to surgery (TP3) and post 4 cycles of adjuvant durvalumab (TP4). This study reports the association of improved EFS with baseline tumor characteristics, including immune inflamed phenotype, tumor infiltrating CD8 T cell density and TLS size. The study also notes increased intra- tumoral T cell diversity and its association with improved EFS. Interrogation of the systemic immune system as measured through mass cytometry and multispectral flow cytometry on PMBC samples revealed post- treatment circulating Ki- 67- expressing CD39+ PD- 1+ CD8+ T cells that associated with improved EFS. Serum CCL15 elevations were noted in 3 patients with durable benefit to therapy for which CCL15 and its receptors CCR1 and CCR3 were explored in a pre- existing single- cell RNA dataset.  
+
+Major Comments/questions:  
+
+- Overall, the manuscript highlights previously known biomarkers of immune checkpoint blockade response with the potential to fundamentally understand the impact of chemotherapy on the modulation of key immune cells. The association of immune phenotype as it correlates to sequential chemotherapy and immunotherapy offers an incredibly opportunity to answer questions re: the role of these agents in neoadjuvant therapy for NSCLC. Additionally, this data set offers an opportunity to understand histology specific differences in immune profile and how that correlates with potential resistance mechanisms to neoadjuvant chemoimmunotherapy.  
+
+- In its current form, the manuscript does not take full advantage of this cohort's clinical impact. TP2 is a critically important time point, as it offers the opportunity to distinguish the impact of chemotherapy induction from that of immunotherapy. Recently published translational work in an LCMV model with sequential chemotherapy \(\rightarrow\) immunotherapy (https://doi.org/10.1158/1078-0432.CCR-23-1316) could be readily compared to these findings from human specimens.  
+
+- In addition, no mention is made of differential findings with respect to lung cancer histology, though it is well-recognized that the immune composition of lung cancer may vary significantly with respect to histology  
+
+- Do mutations described as having a potential deleterious impact on immunotherapy and chemotherapy response (i.e. STK11, KEAP1) confound the authors' findings at all?  
+
+- What is the significance of EFS relative to MPR/pCR and OS in this cohort? How does this differ from the other neoadjuvant IO +/- chemo studies?  
+
+- Clinical annotation of sites of recurrence could be of interest as well relative to tumor immune composition/response  
+
+Minor comments by page:  
+
+Pg 4, Line 80; multicentric \(\rightarrow\) multicenter  
+
+Pg 4, Line 88; does \(\rightarrow\) did or do not  
+
+Pg 5, Line 93; CM816 and other perioperative regimens have clearly demonstrated improved benefit for the inclusion of ICB for PD- L1 high expression. Only KN- 091 was inconsistent increasing levels of efficacy associated with higher level of expression of PD- L1. While NADIM did not reveal PD- L1 expression and TMB as associated with overall survival, this was a
+
+<--- Page Split --->
+
+correlation with response to treatment including MPR.  
+
+Pg 7, Figure 1; helpful to have median EFS and OS (NR) in the graph. Notable that there was a good amount of attrition for digital pathology analysis (only 21/67 patients) - why? Was NGS done on up- front tumor specimens? Change in mutations with treatment could be of interest (Ricciuti, JCO 2024)  
+
+Pg 8, Lines 144- 145; please provide classification criteria for inflamed and excluded as there is no consensus criteria for that determination. Seems this was arbitrarily determined by reading pathologists? How many fields of view were analyzed?  
+
+Pg 9, Figure 2; immune desert was not described in the results section and the distribution was not conveyed.  
+
+Pg 9, Figure 2; graphs c and d are very difficult to interpret statistically significant from non- significant data  
+
+Pg 9, Figure 2; immune inflamed and TC are similar (unless criteria is different?)  
+
+Pg 12, Figure 3; legend explanation of panel b and panel c are switched. Disambiguation of IB/RES should be in legend. Pg 12, Figure 3; C TLS size criteria? Versus above or below the median? What is the relationship between TLS size and PD- L1 score? Need to make sure this is not confounding.  
+
+PD- L1 score? Need to make sure this is not confounding.  
+
+Pg 13, Lines 221- 222; it would be interesting to evaluate the change in TCR diversity in the peripheral blood across various  
+
+time points from baseline through post treatment given the non- significant findings at the post- treatment time period. Numerous studies, including Yost (cited in this paper) report on the clonal expansion that is seen with ICB. It would be interesting to see the effect of TCR diversity from chemotherapy and immunotherapy, respectively, in the neoadjuvant setting.  
+
+Pg 13; what is the relationship between elevated intratumoral T cell clonal richness and TLS size? PD- L1?  
+
+Pg 14, Fig 4. The distribution of patients appears somewhat skewed with 15 patients within the EFS \(< 1\) year and 34 patients in the EFS \(>1\) year timeframe. Is there a sufficient sample size to create 3 cohorts of extremely poor responders, average responders and exceptional responders for further stratification of the data? Does TCR richness change with treatment?  
+
+Pg 15; was WES performed? Or just targeted NGS + RNAseq?  
+
+Pg 16, Lines 274 - 275 describes the expansion of PD1+ CD8+ T cells in the peripheral blood post- chemoiimmunotherapy that was primarily seen in responders. A recent paper by Marinello A et al. Clin Cancer Res 2024, describes the influence of platinum- based chemotherapy in reducing proliferative PD1+ CD8+ T cell expansion compared to anti- PD1 therapy alone in a LCMV mouse model. It would be interesting to see if the effects of chemotherapy are detrimental to the later use of immunotherapy based on the longitudinal assessment of proliferating PD1+ CD8+ T cells in this translational study across all the pre- surgical time points.  
+
+Pg 16, Lines 276 - 277, the presence of CD57+ CD4+ T cells has been described in several studies in the infectious disease field as marker of cell senescence with potential long- term memory components. It would be interesting to understand if the expansion of this cell subset was coupled with markers of memory (CD27, CD127, TCF1, CCR7/CD45RO/CD45RA) versus a more terminally differentiated T cell subset marked increases in checkpoint markers such as CTLA4, TIM- 3, LAG- 3, etc.  
+
+Pg 23; authors should comment more specifically on the impact of chemotherapy before immunotherapy - this is what is unique compared to the well- described translational datasets on neoadjuvant immunotherapy alone and combination chemoiimmunotherapy. For the digital path analysis, a significant limitation is that markers were identified on serial sections - much weaker than simultaneous measurement with a multiplex.  
+
+Pg 24, Line 444, ipilimumab is misspelled.  
+
+## Reviewer #2  
+
+(Remarks to the Author)  
+
+I co- reviewed this manuscript with one of the reviewers who provided the listed reports. This is part of the Nature Communications initiative to facilitate training in peer review and to provide appropriate recognition for Early Career Researchers who co- review manuscripts.  
+
+## Reviewer #4  
+
+(Remarks to the Author)  
+
+This manuscript profiles the immune response within a cohort of NSCLC patients that were treated as part of a phase II trial with neoadjuvant chemo and immunotherapy and a subsequent adjuvant immunotherapy regimen. These immune parameters, which are sourced from tissue, serum, and PBMCs at various time during the trial, are further correlated to 5- year survival outcomes. Major findings for patients with better outcomes include evidence of tumor inflammation, superior infiltration of CD8 T cells, TLS formation, TCR clonal diversity, and activated markers for infiltrating T cells, all of which have been accepted as signs of productive anti- tumor responses. Interestingly, tumor mutation burden did not correlate to outcomes. Novel findings include TIM- 3+ cDC1 among non- responders and signs of CCL15- CCR1 signaling among some responders. The dataset is extensive and impressive, and methods are sound. Enthusiasm is dampened somewhat in that the conclusions have already been highlighted in the field, and the patients are end up being responders seem to have more CD8 T cell infiltration on the onset. It is not clear if neoadjuvant therapy had any impact on turning more patients into responders. While the study seeks to elucidate the spatial dynamics of the tumor microenvironment, the study is rather descriptive. The Introduction states that the goal is to "evaluate whether emerging biomarkers are associated with sustained clinical benefit" but it is not clear if this had been accomplished by this retrospective analysis. In summary, the unique and deep dataset and its study is to be commended, but it is not clear if any hypotheses have been supported or even generated from the work. It is not clear whether this is suitable innovation to merit appearance in Nat Comm.  
+
+Version 1:  
+
+Reviewer comments:
+
+<--- Page Split --->
+
+Reviewer #1  
+
+(Remarks to the Author) The authors have satisfactorily addressed our comments.  
+
+Reviewer #2  
+
+(Remarks to the Author)  
+
+I co- reviewed this manuscript with one of the reviewers who provided the listed reports. This is part of the Nature Communications initiative to facilitate training in peer review and to provide appropriate recognition for Early Career Researchers who co- review manuscripts.  
+
+Reviewer #4  
+
+(Remarks to the Author)  
+
+Thank you for the thoughtful revision of the manuscript, which has incorporated my request for overall significance as well as the more detailed comments from other reviewers. I greatly appreciate the inclusion of a preliminary experiment in the rebuttal, which indicates that the authors have been considering fundamental aspects of the project and how this research can lead to innovation in tumor immunology. I am satisfied that the paper has addressed my concerns and recommend for acceptance.  
+
+Open Access This Peer Review 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 cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+The images or other third party material in this Peer Review 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 https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+
+# Tumor immune dynamics and long-term clinical outcome of stage IIIA NSCLC patients treated with neoadjuvant chemoimmunotherapySchmid et al.  
+
+## Point by point reply  
+
+Reviewer #1 (Remarks to the Author)  
+
+Thank you for the opportunity to review this interesting manuscript, wherein the authors report on multi- omic translational analysis of specimens obtained during the SAKK 16/14 trial. This study enrolled patients with Stage IIIA NSCLC (N2 involvement) treated with neoadjuvant platinum- based chemotherapy followed 3 cycles followed by 2 cycles of durvalumab and resection. Patients then underwent 1 year of adjuvant durvalumab treatment. Translational studies included tissue and blood interrogation at baseline (prior to chemoimmunotherapy; TP1), prior to neoadjuvant durvalumab (TP2), prior to surgery (TP3) and post 4 cycles of adjuvant durvalumab (TP4). This study reports the association of improved EFS with baseline tumor characteristics, including immune inflamed phenotype, tumor infiltrating CD8 T cell density and TLS size. The study also notes increased intra- tumoral T cell diversity and its association with improved EFS. Interrogation of the systemic immune system as measured through mass cytometry and multispectral flow cytometry on PMBC samples revealed post- treatment circulating Ki- 67- expressing CD39+ PD- 1+ CD8+ T cells that associated with improved EFS. Serum CCL15 elevations were noted in 3 patients with durable benefit to therapy for which CCL15 and its receptors CCR1 and CCR3 were explored in a pre- existing single- cell RNA dataset.  
+
+Major Comments/questions:  
+
+- Overall, the manuscript highlights previously known biomarkers of immune checkpoint blockade response with the potential to fundamentally understand the impact of chemotherapy on the modulation of key immune cells. The association of immune phenotype as it correlates to sequential chemotherapy and immunotherapy offers an incredibly opportunity to answer questions re: the role of these agents in neoadjuvant therapy for NSCLC. Additionally, this data set offers an opportunity to understand histology specific differences in immune profile and how that correlates with potential resistance mechanisms to neoadjuvant chemoimmunotherapy.  
+
+We sincerely appreciate your thoughtful and constructive feedback. Your insights into the potential of our dataset to investigate the effects of chemotherapy on immune cell modulation and the role of these agents in neoadjuvant therapy for NSCLC are highly valuable.  
+
+- In its current form, the manuscript does not take full advantage of this cohort's clinical impact. TP2 is a critically important time point, as it offers the opportunity to distinguish the impact of chemotherapy induction from that of immunotherapy. Recently published translational work in an LCMV model with sequential chemotherapy \(\rightarrow\) immunotherapy (https://doi.org/10.1158/1078-0432.CCR-23-1316) could be readily compared to these findings from human specimens.
+
+<--- Page Split --->
+
+We acknowledge that TP2 provides a valuable perspective on how chemotherapy and immunotherapy influence the host immune system. Notably, as highlighted in the referenced paper, the treatment sequence plays a critical role, as concurrent chemotherapy may dampen the immunostimulatory effects of checkpoint blockade.  
+
+In this revision, in response to the reviewer's suggestion, we have addressed this topic by conducting additional TCR sequencing on peripheral TP2 samples. Furthermore, we have expanded our discussion on the proliferative capacity of peripheral \(\mathrm{CD39^{+}}\) PD- 1 \(^+\) CD8 \(^+\) T cells during neoadjuvant chemoimmunotherapy. Please see our responses to the specific questions below for further details.  
+
+- In addition, no mention is made of differential findings with respect to lung cancer histology, though it is well-recognized that the immune composition of lung cancer may vary significantly with respect to histology  
+
+We appreciate this important comment concerning the separate analysis of different NSCLC histologies and the feasibility of further subtyping immune responses across different histological subtypes. Our dataset comprises a total of 37 cases of adenocarcinoma, 22 cases of squamous cell carcinoma, 1 case of large cell carcinoma, and 7 cases of NSCLC not otherwise specified (NOS). In line with clinical practice, we will categorize these cases in the revised manuscript as squamous cell carcinoma (SCC, \(n = 22\) ) and non- squamous cell carcinoma (non- SCC, \(n = 45\) ). Notably, our analysis reveals no significant differences in overall or event- free survival between patients with SCC and non- SCC (Extended Data Fig. 1a, b).  
+
+The histological response signature—including the inflamed immune phenotype, intra- tumoral \(\mathrm{CD8^{+}}\) T cell infiltrate infiltrates, and TLS size—is present across both subtypes and does not exhibit a statistically significant difference in frequency between SCC and non- SCC (Extended Data Fig. 2a- c). Furthermore, we observe comparable patterns in TCR indices (Extended Data Fig. 3a) and peripheral immune cell activation (Extended Data Fig. 8c) between SCC and non- SCC. However, the number of cases decreases when further subsetting the cohort, which may limit statistical power.  
+
+Despite these challenges, our analysis indicates that the observed immune phenotypes remain largely consistent across various histological subgroups of NSCLC. We recognize the need for further validation of these findings and a deeper investigation into distinct immune response subtypes within independent cohorts. Such validation would not only reinforce the broader applicability of our observations but also potentially identify histology- specific resistance mechanisms to neoadjuvant chemoimmunotherapy.  
+
+Further subtyping of immune responses within adenocarcinoma histological categories (lepidic, papillary, acinar, micropapillary and solid) presents considerable challenges in the neoadjuvant setting. First, such subtyping requires a comprehensive evaluation of resection specimens, whereas preoperative biopsies, though adequate for diagnosing lung adenocarcinoma, do not sufficiently capture the high degree of intratumoral heterogeneity. Second neoadjuvant treatment induces substantial histological alterations, making further subtyping or tumor grading unsuitable according to established WHO criteria in these samples. These limitations currently hinder the feasibility of this level of detailed analysis. We anticipate that future studies in larger, independent cohorts will build upon these observations.
+
+<--- Page Split --->
+
+- Do mutations described as having a potential deleterious impact on immunotherapy and chemotherapy response (i.e. STK11, KEAP1) confound the authors' findings at all?  
+
+In the SAKK 16/14 cohort, \(17\%\) of patients had a STK11 mutation \((n = 8)\) and \(15\%\) had a KEAP1 mutation \((n = 7)\) . All patients with STK11 mutations were diagnosed with adenocarcinoma and were current or former smokers while one patient with a KEAP1 mutation was a never smoker, and \(n = 2\) patients with KEAP1 mutations did not have adenocarcinoma \((n = 1\) with large cell carcinoma, \(n = 1\) NSCLC NOS).  
+
+Furthermore, when checking for KRAS mutation status, only \(n = 2\) patients with STK11 mutations had concurrent KRAS mutations, while \(n = 1\) patient had concurrent KRAS and KEAP1 mutations. This information is displayed in Extended Data Fig. 4f.  
+
+STK11 and KEAP1 have been identified as negative predictors of immunotherapy response in patients with KRAS- mutated adenocarcinoma who have a positive smoking history, but not in KRAS wild- type adenocarcinomas<sup>1</sup>. The limited number of patients in our cohorts prevents the calculation of odds ratios to accurately assess the impact of these two mutations. However, we note that both patients with KRAS/STK11 mutations exhibited a very low percentage of PD- L1<sup>+</sup> tumor cells (1%, data not shown), consistent with previous reports. Consequently, we cannot definitively determine whether the deleterious impact of these mutations may have influenced our results. We look forward to future translational trials with larger patient cohorts to further investigate these questions.  
+
+- What is the significance of EFS relative to MPR/pCR and OS in this cohort? How does this differ from the other neoadjuvant IO +/- chemo studies?  
+
+Since neoadjuvant treatment regimens are performed in earlier stage tumors, surrogate endpoints such as event- free survival and pathological response rates are commonly used for preliminary analysis. This is because assessing changes in overall survival requires a longer follow- up period compared to studies in more advanced tumor settings. Accordingly, EFS at 12 months was chosen as the primary endpoint of the SAKK 16/14 trial<sup>2</sup>.  
+
+We would like to reference a recent review that has confirmed the predictive value of MPR/pCR and EFS for OS in neoadjuvant checkpoint inhibition<sup>3</sup>, which notably includes the SAKK 16/14 trial. Furthermore, we now provide updated survival data in Extended Data Fig. 1c- f, illustrating the relationship between MPR/pCR, EFS and OS. Our analysis demonstrates that patients achieving MPR and pCR have significantly longer EFS \((p < 0.0001\) and \(p = 0.0858\) , respectively) and OS \((p < 0.0001\) and \(p = 0.0288\) , respectively).  
+
+- Clinical annotation of sites of recurrence could be of interest as well relative to tumor immune composition/response  
+
+Among the \(n = 67\) evaluable patients, \(n = 30\) experienced an EFS event, \(n = 25\) of which have a recorded site of progression. Of these, \(n = 14\) developed distant metastases, with the brain \((n = 6)\) and bone \((n = 4)\) being the most common metastatic sites. Meanwhile, \(n = 11\) exhibited purely locoregional recurrence, occurring either in the ipsi- or contralateral lung or in new lymph nodes. We observed no significant differences in EFS and OS between patients with locoregional vs. distant progression \((p = 0.4758\) and \(p = 0.7675\) , respectively). These data are presented in Extended Data Fig. 1g, h.
+
+<--- Page Split --->
+
+Unfortunately, digital pathology data were available for only a subset of patients with recorded progressions, limiting our ability to associate distinct tumor immune profiles with future sites of progression. We appreciate the reviewer's suggestion, as such an association would indeed be valuable to explore. We look forward to future studies with larger cohorts to further investigate this question.  
+
+Minor comments by page:  Pg 4, Line 80; multicentric -> multicenter  Pg 4, Line 88; does -> did or do not  
+
+Both of these mistakes were corrected.  
+
+Pg 5, Line 93; CM816 and other perioperative regimens have clearly demonstrated improved benefit for the inclusion of ICB for PD- L1 high expression. Only KN- 091 was inconsistent increasing levels of efficacy associated with higher level of expression of PD- L1. While NADIM did not reveal PD- L1 expression and TMB as associated with overall survival, this was a correlation with response to treatment including MPR.  
+
+We agree with the reviewer that in NSCLC patients being considered for perioperative immunotherapy, PD- L1 expression should be carefully assessed as a factor that may help guide the selection of the most appropriate treatment regimen. In this paragraph, our intention was to emphasize the relatively limited data on the predictive value of PD- L1 expression and TMB in the perioperative setting compared to their established role in purely systemic treatment for metastatic disease. We have revised this sentence accordingly.  
+
+Pg 7, Figure 1; helpful to have median EFS and OS (NR) in the graph. Notable that there was a good amount of attrition for digital pathology analysis (only 21/67 patients) - why?  
+
+We have added median EFS and OS to the survival curves in Fig. 1.  
+
+Regarding the attrition in digital pathology analysis, only a subset of the initial biopsies were tissue blocks suitable for digital pathology analysis. The remaining biopsies were cytology samples, primarily used for lung cancer diagnosis and PD- L1 assessment. The choice between these sample types was determined by the accessibility of the suspected lung tumor and local clinical protocols. Importantly, allowing both specimen types ensured that patient accrual was not hindered and enabled broad participation across multiple sites in Switzerland. We have clarified this point in the text accordingly.  
+
+Was NGS done on up- front tumor specimens? Change in mutations with treatment could be of interest (Ricciuti, JCO 2024)  
+
+Mutation profiling was conducted on mainly tumor resection samples, meaning the analysis was performed on tumors that had undergone neoadjuvant chemo- immunotherapy. The DNA
+
+<--- Page Split --->
+
+requirements for the assays used (50 ng for Foundation One CDx and 20 ng for Comprehensive Plus) were considered too high to be feasibly obtained from initial biopsies without compromising their use for other tissue- based analysis.  
+
+We agree with the reviewer that comparing mutational profiles before and after neoadjuvant immunotherapy could provide valuable mechanistic insights into the anti- tumor immune response. However, we note that in the referenced study<sup>4</sup>, samples were collected from patients who had demonstrated a confirmed response or stable disease for at least 3 months, with a median interval of 18.9 months between biopsies. In contrast, the shorter time frame inherent to the neoadjuvant setting must be considered when interpreting the dynamics of mutational profiles.  
+
+Additionally, we have clarified the sourcing of material for DNA mutation analysis in the Materials and Methods section.  
+
+Pg 8, Lines 144- 145; please provide classification criteria for inflamed and excluded as there is no consensus criteria for that determination. Seems this was arbitrarily determined by reading pathologists? How many fields of view were analyzed?  
+
+We appreciated the reviewer's attention to this missing information. Immune phenotyping was performed out by consensus of two board- certified pathologists (ABSB and VHK) following the recommendations of the International Immuno- Oncology Biomarker Working Group<sup>5,6</sup>. The complete tumor area was assessed to determine immune status, as previously described<sup>7</sup>. Specifically, the spatial distribution of tumor- infiltrating CD8<sup>+</sup> T cells was categorized into three immune phenotype: (1) immune desert – characterized by very rare and isolated CD8<sup>+</sup> T cells detected in any of the assessed tumor compartments, (2) immune excluded – defined by the presence of CD8<sup>+</sup> T cells at the tumor environment, primarily at the invasive margin or within the stroma, with only rare and isolated T cells found within the intratumoral compartment, and (3) inflamed – marked by CD8<sup>+</sup> T cells infiltrating the stromal compartment, directly contacting tumor cells, and penetrating the tumor parenchyma.  
+
+This information has been added to the Materials and Methods section.  
+
+Pg 9, Figure 2; immune desert was not described in the results section and the distribution was not conveyed.  
+
+This is correct. Fig. 2A presents generic examples of the three immune phenotypes in isolated tumor regions from SAKK 16/14 samples, intended to aid the reader's conceptual understanding. As noted, final immune classifications were determined by evaluating the entire tumor, rather than isolated regions. In the SAKK dataset, the immunologically "cold" subtype was entirely represented by immune- excluded tumors. While some excluded tumors can contain regions with very little immune infiltration, we did not identify any completely immune- desert tumors in the final dataset. To prevent any misunderstanding, we have clarified that these are generic examples from representative tumor regions.  
+
+Pg 9, Figure 2; graphs c and d are very difficult to interpret statistically significant from nonsignificant data
+
+<--- Page Split --->
+
+In Fig. 2c, d, our aim was to emphasize that infiltration densities of the indicated cell populations did not differ between immune- excluded and inflamed tumors, neither in the tumor compartment nor in the stroma. This observation holds true for all four analyzed populations (CD3+ T cells, CD8+ T cells, FoxP3+ T regulatory cells and CD20+ B cells). To ensure clarity, we have revised the text accordingly and have also added previously missing information on statistical testing in the figure legend.  
+
+By contrast, as expected, all immune populations tended to be more abundant in the stroma than in the tumor compartment. For improved readability, we have opted not to display test statistics for this comparison, as well as for cross- population comparisons (e.g. CD3+ T cells vs. CD20+ B cells).  
+
+Pg 9, Figure 2; immune inflamed and TC are similar (unless criteria is different?)  
+
+We apologize for any confusion. TC (tumor compartment) refers to the segmented area within the tissue block where tumor cells are located. This compartment is surrounded by stroma. Immune cell densities are quantified and expressed per \(\mathsf{mm}^2\) within either the TC or the stroma. In contrast, the immune phenotype (desert, excluded, inflamed) is determined based on the spatial distribution of immune cells across these compartments, as assessed by pathologist consensus (as detailed in our previous response).  
+
+Pg 12, Figure 3; legend explanation of panel b and panel c are switched. Disambiguation of IB/RES should be in legend.  
+
+We thank the reviewer for pointing out the missing information. Legend explanations for panels b and c have been corrected, IB/RES has been explained.  
+
+Pg 12, Figure 3; C TLS size criteria? Versus above or below the median? What is the relationship between TLS size and PD- L1 score? Need to make sure this is not confounding.  
+
+The average (i.e. arithmetic mean) area of TLS in each initial biopsy was calculated, and the median of these values was used as a cutoff to classify samples as either "TLS size high" or "TLS size low". This clarification has been added to the figure legend.  
+
+Furthermore, we have added Extended Data Fig. 2d to illustrate the relationship between PD- L1 score and average TLS size. Our analysis shows no significant difference in PD- L1 scores between samples with small and large TLS (p = 0.3502, Mann- Whitney test).  
+
+Pg 13, Lines 221- 222; it would be interesting to evaluate the change in TCR diversity in the peripheral blood across various time points from baseline through post treatment given the non- significant findings at the post- treatment time period. Numerous studies, including Yost (cited in this paper) report on the clonal expansion that is seen with ICB. It would be interesting
+
+<--- Page Split --->
+
+to see the effect of TCR diversity from chemotherapy and immunotherapy, respectively, in the neoadjuvant setting.  
+
+We agree that investigating clonal dynamics across different phases of neoadjuvant treatment is highly relevant. To complement our data, we have now performed TCR sequencing on timepoint 2 (TP2) peripheral samples (collected after neoadjuvant chemotherapy but before neoadjuvant durvalumab). Of note, no tissue samples were available at this timepoint.  
+
+Before interpreting the results, we must highlight the technical challenges associated with preparing sequencing libraries in different batches. Although RNA quality metrics were comparable, TP2 samples had fewer sequencing reads compared to TP1 and TP3 samples (Extended Data Fig. 3f). Since clonal richness (the number of unique TCR clones) correlates with sequencing depth, this prevented a direct comparison of clonal richness across timepoints in patients with different event- free survival (Extended Data Fig. 3g). However, other TCR repertoire metrics, such as evenness, were not affected (Extended Data Fig. 3h).  
+
+To address this issue, we performed random down- sampling of sequencing reads (10'000, 50'000, 250'000, 500'000, 750'000, 1'000'000, 1'500'000 and 2'000'000 reads) and analyzed the impact on clonal richness (number of TCR clones) and TCR evenness. This analysis, illustrated with data from four representative patients (Extended Data. Fig 3i, j) revealed a near- linear increase in the number of detected TCR clones with increasing sequencing depth. Conversely, TCR evenness plateaued as sequencing reads increased. Based on these observations, we concluded that down- sampling is an appropriate method to mitigate batch effects between TP1/TP3 and TP2 sequencing runs. Consequently, all further analyses for PBMC samples were conducted using 250'000 randomly selected sequencing reads. No down- sampling was performed for TCR sequencing data from tumor resections.  
+
+We observed a trend toward a higher number of TCR clones in PBMC samples from patients achieving \(\mathrm{EFS} \geq 12\) months both at baseline (TP1, \(\mathrm{p} = 0.1604\) ) and after neoadjuvant chemotherapy (TP2, \(\mathrm{p} = 0.0551\) , Fig. 4d). Similarly, we detected a trend toward higher TCR evenness in patients with \(\mathrm{EFS} \geq 12\) months at baseline (TP1, \(\mathrm{p} = 0.0853\) , Fig. 4e). Furthermore, Hill- Simpson diversity showed a marked decrease from baseline (TP1) to post- chemotherapy (TP2) in all patients, potentially reflecting the deleterious effect of chemotherapy on T cell expansion ( \(\mathrm{EFS} \geq 12\) months, \(\mathrm{p} = 0.0234\) ; \(\mathrm{EFS} < 12\) months, \(\mathrm{p} = 0.1079\) ). Notably, in patients with \(\mathrm{EFS} \geq 12\) months, Hill- Simpson diversity significantly increased again following neoadjuvant durvalumab ( \(\mathrm{p} = 0.0301\) , Fig. 4f), suggesting a potential recovery or expansion of T cell diversity after checkpoint blockade.  
+
+Pg 13; what is the relationship between elevated intratumoral T cell clonal richness and TLS size? PD- L1?  
+
+We found no association between intratumoral T cell clonal richness and TLS size in resections, as shown in Extended Data Fig. 3b. It is important to note that TCR sequencing was performed on resections, not on initial biopsies, due to sample limitations. In contrast, TLS were analyzed in both initial biopsies and resections. However, for survival analysis (Fig. 3 and Extended Data Fig. 2c, d), we only used initial biopsy data. This decision was made because resections included samples with complete pathological response, making it impossible to quantify intratumoral TLS in those cases.
+
+<--- Page Split --->
+
+Similarly, we found no association between T cell clonal richness and PD- L1 score, regardless of whether patients achieved \(\mathrm{EFS} \geq 12\) months or not. This is illustrated in **Extended Data Fig. 2e.**  
+
+Pg 14, Fig 4. The distribution of patients appears somewhat skewed with 15 patients within the EFS \(< 1\) year and 34 patients in the EFS \(>1\) year timeframe. Is there a sufficient sample size to create 3 cohorts of extremely poor responders, average responders and exceptional responders for further stratification of the data? Does TCR richness change with treatment?  
+
+As discussed in the paper, we selected a 12- month EFS cutoff because it has been the primary endpoint for evaluating the efficacy of this treatment regimen. However, as the reviewer correctly noted, this approach results in an imbalance in patient numbers between groups. To address this, we conducted an exploratory analysis by stratifying patients into three groups: EFS \(< 12\) months, EFS between 12 and 36 months, EFS \(>36\) months. We examined both T cell clonal richness (Fig. 4) and proliferation of tentatively tumor- reactive \(\mathrm{CD39^{+}}\) PD- \(1^{+}\) \(\mathrm{CD8^{+}}\) T cells (Fig. 5). While there appears to be no difference in clonal richness between the EFS \(\geq 12 < 36\) months and EFS \(>36\) months groups, we observed a difference in baseline proliferation of \(\mathrm{CD39^{+}}\) PD- \(1^{+}\) \(\mathrm{CD8^{+}}\) T T cells between these cohorts. This suggests that the proliferative capacity of tumor- reactive \(\mathrm{CD8^{+}}\) T cells at baseline may correlate with long- term event- free survival.  
+
+[editorial note: confidential, unpublished figures have been redacted]  
+
+However, the reduction in sample size limits the statistical power of the dataset, leading to a loss of statistical significance in several comparisons. We have addressed this limitation in the Discussion to acknowledge its potential impact on data interpretation.  
+
+Regarding TCR richness, we refer to our previous comment, where we discussed its association (or lack thereof) with event- free survival (EFS), TLS size, and PD- L1 score, as well as the impact of batch effects on the analysis.  
+
+Pg 15; was WES performed? Or just targeted NGS + RNAseq?  
+
+Mutational analysis was performed using one of two extensively validated targeted NGS assays on DNA isolated from FFPE- embedded tumor resection samples.
+
+<--- Page Split --->
+
+The Foundation One CDx assay investigates 324 cancer- specific genes for base substitutions, insertions/deletions, copy number alterations and recombinations. Furthermore, it provides tumor mutational burden. (https://www.foundationmedicine.de/de/our-services/cdx.html).  
+
+The Oncomine Comprehensive Assay Plus (Thermo Fisher Scientific) targets 517 genes and allows detection of base substitutions, insertions/deletions, copy number alterations, as well as tumor mutational burden estimation.  
+
+Gene expression profiling was run on FFPE- isolated RNA using the Oncomine Immune Response Research Assay (Thermo Fisher Scientific). Full RNAseq was not feasible due to poor RNA quality metrics, with a typical RNA integrity number (RIN) of \(\sim 2\) and low RNA input. Importantly, the Oncomine assay captures expression levels of 395 genes that are commonly involved in immunotherapy response, providing a targeted but robust alternative to full RNAseq.  
+
+Pg 16, Lines 274 - 275 describes the expansion of \(PD1 + CD8 + T\) cells in the peripheral blood post- chemoimmunotherapy that was primarily seen in responders. A recent paper by Marinello A et al. Clin Cancer Res 2024, describes the influence of platinum- based chemotherapy in reducing proliferative \(PD1 + CD8 + T\) cell expansion compared to anti- PD1 therapy alone in a LCMV mouse model. It would be interesting to see if the effects of chemotherapy are detrimental to the later use of immunotherapy based on the longitudinal assessment of proliferating \(PD1 + CD8 + T\) cells in this translational study across all the presurgical time points.  
+
+Interestingly, we observed increased proliferation in peripheral \(\mathrm{CD39^{+}}\) PD- \(1^{+}\) CD8 \(^+\) T cells after neoadjuvant chemotherapy at TP2 compared to baseline (TP1). Furthermore, only in patients with \(\mathrm{EFS} > 12\) months, proliferation continued to increase from TP2 to TP3 following neoadjuvant immunotherapy (Fig. 5g). While the proliferative response to chemotherapy alone may seem counterintuitive, we believe several clinical factors can explain this observation.  
+
+In the study by Marinello et al, mice received concomitant chemo- and immunotherapy every three days, with samples collected three days after the last administration. In this preclinical model, chemotherapy was shown to reduce expansion, proliferation, and cytokine secretion of LCMV- specific \(\mathrm{CD8 + }\) T cells. In contrast, in our clinical trial, patients received chemotherapy on the first day of a three- week cycle. Several key factors likely contribute to the observed T cell dynamics:  
+
+- Cytopenia and rebound hematopoiesis: Chemotherapy typically induces peripheral pancytopenia around one week after administration, followed by homeostatic hematopoiesis to restore immune cell counts.- G-CSF administration: Patients also received G-CSF to stimulate hematopoiesis and mitigate the risk of infection.- Timing of TP2 Sampling: TP2 samples were collected three weeks after the last chemotherapy dose, immediately before the first durvalumab dose. By this time, hematopoietic recovery
+
+<--- Page Split --->
+
+would likely be underway, contributing to the observed increase in \(\mathrm{CD8^{+}}\) T cell proliferation.  
+
+Thus, the dynamics of \(\mathrm{CD8^{+}}\) T cell proliferation appear to differ based on event- free survival (EFS). In patients with EFS \(\geq 12\) months, proliferation remains stable or even increases following neoadjuvant immunotherapy (TP2 to TP3), suggesting a sustained immune response. In contrast, in patients with EFS \(< 12\) months, proliferation decreases from TP2 to TP3, potentially indicating a lack of sustained T cell activation in response to durvalumab.  
+
+For further details on the SAKK 16/14 trial, including dosing schedules and treatment regimens, we refer to the study protocol available in the initial publication presenting the clinical findings<sup>2</sup>.  
+
+Pg 16, Lines 276 – 277, the presence of \(\mathrm{CD57 + CD4 + T}\) cells has been described in several studies in the infectious disease field as marker of cell senescence with potential long- term memory components. It would be interesting to understand if the expansion of this cell subset was coupled with markers of memory (CD27, CD127, TCF1, CCR7/CD45RO/CD45RA) versus a more terminally differentiated T cell subset marked increases in checkpoint markers such as CTLA4, TIM- 3, LAG- 3, etc.  
+
+We agree that \(\mathrm{CD4^{+}}\) \(\mathrm{CD57^{+}}\) cells are an intriguing subset, particularly given the growing recognition of immunosenescence as a potential resistance mechanism to cancer immunotherapy. To further characterize this population, we analyzed CyTOF data, and the results are presented in Extended Data. Fig. 7c- e.  
+
+Overall, \(\mathrm{CD4^{+}}\) \(\mathrm{CD57^{+}}\) T cells exhibit distinct phenotypic features:  
+
+- they express lower levels of CCR7, CD27, and CD127 compared to \(\mathrm{CD57^{-}}\) cells, suggesting a more differentiated state.- they predominantly express CD45RO rather than CD45RA, indicating a memory-like phenotype, further supported by the presence of TCF1, particularly in responders.  
+
+However, several markers also suggest terminal differentiation:  
+
+- \(\mathrm{CD57^{+}}\) cells express higher levels of PD-1 and TOX, both associated with exhaustion and terminal differentiation.- they exhibit lower expression of activation markers CD25 and CD28, potentially reflecting reduced activation capacity.  
+
+Interestingly, despite their differentiated phenotype, Ki67 expression was observed only in \(\mathrm{CD57^{+}}\) \(\mathrm{CD4^{+}}\) cells from responders, and proliferation further increased after neoadjuvant chemoimmunotherapy. This was unexpected, given that the overall abundance of \(\mathrm{CD4^{+}}\) \(\mathrm{CD57^{+}}\) cells was actually higher in non- responders.  
+
+These findings raise intriguing questions about the functional role of \(\mathrm{CD4^{+}}\) \(\mathrm{CD57^{+}}\) T cells in the context of immunotherapy, and we look forward to future studies that will further elucidate the dynamics and significance of this population over the course of treatment.
+
+<--- Page Split --->
+
+Pg 23; authors should comment more specifically on the impact of chemotherapy before immunotherapy - this is what is unique compared to the well- described translational datasets on neoadjuvant immunotherapy alone and combination chemoimmunotherapy. For the digital path analysis, a significant limitation is that markers were identified on serial sections - much weaker than simultaneous measurement with a multiplex.  
+
+We have expanded the Discussion on chemotherapy to further clarify the distinct effects of neoadjuvant chemotherapy versus durvalumab in the SAKK 16/14 trial.  
+
+We have also included a discussion on the limitations associated with using serial sections for digital pathology analysis. The primary reason for selecting single- marker IHC assays from routine diagnostics—rather than more complex multiplex approaches—was to maintain a strong translational focus, ensuring that our findings can be readily replicated in larger clinical trials. However, we acknowledge that this method does not allow for  
+
+- co-localization studies between different cell subsets (e.g. CD3+ T cells and CD20+ B cells)- Confirmation of cellular identity through co-staining (e.g. CD3+ CD8+ CD8 T cells or CD3+ FoxP3+ T reg cells).  
+
+Despite these limitations, the cellular densities presented in Fig. 2 were measured within specific tumor compartments, which were identified using H&E staining on the same slide as the IHC stain. This approach provides a spatially informed analysis while maintaining compatibility with clinical pathology workflows.  
+
+Pg 24, Line 444, ipilimumab is misspelled.  
+
+This has been corrected.  
+
+Reviewer #2 (Remarks to the Author):  
+
+I co- reviewed this manuscript with one of the reviewers who provided the listed reports. This is part of the Nature Communications initiative to facilitate training in peer review and to provide appropriate recognition for Early Career Researchers who co- review manuscripts.  
+
+We sincerely appreciate the time and effort that both you and your co- reviewer have dedicated to evaluating our manuscript. We fully support initiatives like this that facilitate training in peer review and provide well- deserved recognition for Early Career Researchers. Your constructive feedback has been invaluable in refining our work, and we are grateful for your thoughtful contributions.  
+
+Reviewer #4 (Remarks to the Author):  
+
+This manuscript profiles the immune response within a cohort of NSCLC patients that were treated as part of a phase II trial with neoadjuvant chemo and immunotherapy and a
+
+<--- Page Split --->
+
+subsequent adjuvant immunotherapy regiment. These immune parameters, which are sourced from tissue, serum, and PBMCs at various time during the trial, are further correlated to 5- year survival outcomes. Major findings for patients with better outcomes include evidence of tumor inflammation, superior infiltration of CD8 T cells, TLS formation, TCR clonal diversity, and activated markers for infiltrating T cells, all of which have been accepted as signs of productive anti- tumor responses. Interestingly, tumor mutation burden did not correlate to outcomes.  
+
+Novel findings include TIM- 3+ cDC1 among non- responders and signs of CCL15- CCR1 signaling among some responders. The dataset is extensive and impressive, and methods are sound. Enthusiasm is dampened somewhat in that the conclusions have already been highlighted in the field, and the patients are end up being responders seem to have more CD8 T cell infiltration on the onset. It is not clear if neoadjuvant therapy had any impact on turning more patients into responders.  
+
+While the study seeks to elucidate the spatial dynamics of the tumor microenvironment, the study is rather descriptive. The Introduction states that the goal is to "evaluate whether emerging biomarkers are associated with sustained clinical benefit" but it is not clear if this had been accomplished by this retrospective analysis.  
+
+In summary, the unique and deep dataset and its study is to be commended, but it is not clear if any hypotheses have been supported or even generated from the work. It is not clear whether this is suitable innovation to merit appearance in Nat Comm.  
+
+We thank the reviewer for raising these important conceptual questions.  
+
+Regarding the impact of neoadjuvant therapy on converting more patients into responders, we acknowledge that the single- arm design of the SAKK 16/14 study limits our ability to directly assess this question. However, previous studies have conclusively demonstrated that neoadjuvant chemoimmunotherapy increases the proportion of patients achieving major or complete pathological response and prolongs event- free and overall survival compared to neoadjuvant chemotherapy alone<sup>9- 14</sup>.  
+
+Unlike in melanoma<sup>15</sup>, a direct comparison between neoadjuvant and purely adjuvant (chemo- )immunotherapy regimens has not yet been performed in NSCLC. However, a recent review suggests that neoadjuvant immunotherapy may outperform purely adjuvant immunotherapy in NSCLC<sup>16</sup>.  
+
+We chose to primarily investigate pathology- based immune characteristics (such as \(\mathrm{CD8^{+}}\) T cell infiltration, TLS size, and immunophenotype) in initial biopsies because assessing these parameters in post- immunotherapy samples is technically more challenging. For example, in patients achieving pathological complete response (pCR), \(\mathrm{CD8^{+}}\) T cell densities become difficult to define due to the loss of clearly delineated tumor margins. Given that these patients have the longest EFS and OS (Extended Data Fig. 1e- f), these technical limitations could introduce bias when evaluating pathological correlates of response.  
+
+With regard to our stated goal in the Introduction, we aimed to highlight the challenges of biomarker discovery in the neoadjuvant NSCLC setting, where obtaining tumor specimens is inherently more complex than in other cancers such as melanoma. Developing and refining these methods is essential for future studies to stratify patients into treatment arms based on immune characteristics. We acknowledge that our original phrasing may have been misleading and have revised the introduction accordingly:
+
+<--- Page Split --->
+
+Finally, we would like to highlight a key hypothesis generated from our study. To our knowledge, the upregulation of TIM- 3 on peripheral cDCs in non- responders following anti- PD- L1 treatment has not been previously reported. This is particularly relevant as PD- L1 blockade on cDCs is a key mechanism driving response to treatment<sup>17</sup>.  
+
+To further investigate this mechanism, we are currently conducting reverse- translation studies in mice. In preliminary experiments, we observed that anti- PD- L1 treatment upregulates TIM- 3 on intratumoral cDCs, and this effect can be partially blocked by concurrent anti- TIM- 3 therapy. These experiments were performed in a murine 4T1 intramammary tumor model, designed to mimic the surgical resection and treatment scheme of the SAKK 16/14 trial.  
+
+[editorial note: confidential, unpublished figures have been redacted]  
+
+Given the preliminary nature of this experiment, we have chosen not to include it in the manuscript but to present it exclusively in this point- by- point response.  
+
+## References  
+
+1. Ricciuti, B. et al. Diminished Efficacy of Programmed Death-(Ligand)1 Inhibition in STK11- and KEAP1-Mutant Lung Adenocarcinoma Is Affected by KRAS Mutation Status. Journal of Thoracic Oncology 17, 399-410 (2022).  
+2. Rothschild, S. I. et al. SAKK 16/14: durvalumab in addition to neoadjuvant chemotherapy in patients with stage IIIA (N2) non-small-cell lung cancer—a multicenter single-arm phase II trial. Journal of clinical oncology 39, 2872-2880 (2021).  
+3. Nie, R. et al. Predictive value of radiological response, pathological response and relapse-free survival for overall survival in neoadjuvant immunotherapy trials: pooled analysis of 29 clinical trials. Eur J Cancer 186, 211-221 (2023).  
+4. Ricciuti, B. et al. Genomic and Immunophenotypic Landscape of Acquired Resistance to PD-(L)1 Blockade in Non-Small-Cell Lung Cancer. Journal of Clinical Oncology 42, 1311-1321 (2024).  
+5. Amgad, M. et al. Report on computational assessment of Tumor Infiltrating Lymphocytes from the International Immuno-Oncology Biomarker Working Group. npj Breast Cancer vol. 6 Preprint at https://doi.org/10.1038/s41523-020-0154-2 (2020).  
+6. Salgado, R. et al. The evaluation of tumor-infiltrating lymphocytes (TILS) in breast cancer: Recommendations by an International TILS Working Group 2014. Annals of Oncology vol. 26 259-271 Preprint at https://doi.org/10.1093/annonc/mdu450 (2015).
+
+<--- Page Split --->
+
+7. Sobottka, B. et al. Establishing standardized immune phenotyping of metastatic melanoma by digital pathology. Laboratory Investigation 101, 1561-1570 (2021).8. Marinello, A. et al. Platinum-Based Chemotherapy Attenuates the Effector Response of CD8 T Cells to Concomitant PD-1 Blockade. Clinical Cancer Research 30, 1833-1845 (2024).9. Provencio-Pulla, M. et al. Nivolumab+ chemotherapy versus chemotherapy as neoadjuvant treatment for resectable stage IIIA NSCLC: Primary endpoint results of pathological complete response (pCR) from phase II NADIM II trial. Preprint at (2022).10. Forde, P. M. et al. Neoadjuvant Nivolumab plus Chemotherapy in Resectable Lung Cancer. N Engl J Med 386, 1973-1985 (2022).11. Wakelee, H. et al. Perioperative Pembrolizumab for Early-Stage Non-Small-Cell Lung Cancer. New England Journal of Medicine 389, 491-503 (2023).12. Heymach, J. V. et al. Perioperative Durvalumab for Resectable Non-Small-Cell Lung Cancer. New England Journal of Medicine 389, 1672-1684 (2023).13. Lu, S. et al. Perioperative toripalimab + platinum-doublet chemotherapy vs chemotherapy in resectable stage II/III non-small cell lung cancer (NSCLC): Interim event-free survival (EFS) analysis of the phase III Neotorch study. Journal of Clinical Oncology 41, 425126-425126 (2023).14. Cascone, T. et al. LBA1 CheckMate 77T: Phase III study comparing neoadjuvant nivolumab (NIVO) plus chemotherapy (chemo) vs neoadjuvant placebo plus chemo followed by surgery and adjuvant NIVO or placebo for previously untreated, resectable stage II-IIIb NSCLC. Annals of Oncology 34, S1295 (2023).15. Patel, S. P. et al. Neoadjuvant-Adjuvant or Adjuvant-Only Pembrolizumab in Advanced Melanoma. New England Journal of Medicine 388, 813-823 (2023).16. Martins, R. S. et al. Neoadjuvant vs Adjuvant Chemoimmunotherapy for Stage II-IIIB Non-Small Cell Lung Cancer. in Annals of Thoracic Surgery vol. 118 672-681 (Elsevier Inc., 2024).17. Dammeijer, F. et al. The PD-1/PD-L1-Checkpoint Restrains T cell Immunity in Tumor- Draining Lymph Nodes. Cancer Cell 38, 685-700.e8 (2020).
+
+<--- Page Split --->

peer_reviews/supplementary_0_Transparent Peer Review file__e953eb070ecccde9cf6628bea8dbe26c3e9c5cfaeccbb37878e543b2e142d42b/supplementary_0_Transparent Peer Review file__e953eb070ecccde9cf6628bea8dbe26c3e9c5cfaeccbb37878e543b2e142d42b_det.mmd

@@ -0,0 +1,532 @@
+<|ref|>title<|/ref|><|det|>[[73, 163, 899, 237]]<|/det|>
+# Tumor immune dynamics and long-term clinical outcome of stage IIIA NSCLC patients treated with neoadjuvant chemoimmunotherapy  
+
+<|ref|>text<|/ref|><|det|>[[73, 249, 441, 267]]<|/det|>
+Corresponding Author: Dr Dominic Schmid  
+
+<|ref|>text<|/ref|><|det|>[[70, 299, 866, 314]]<|/det|>
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+<|ref|>text<|/ref|><|det|>[[73, 351, 144, 365]]<|/det|>
+Version 0:  
+
+<|ref|>text<|/ref|><|det|>[[73, 377, 220, 391]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[73, 403, 160, 417]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[73, 430, 237, 444]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 444, 922, 600]]<|/det|>
+Thank you for the opportunity to review this interesting manuscript, wherein the authors report on multi- omic translational analysis of specimens obtained during the SAKK 16/14 trial. This study enrolled patients with Stage IIIA NSCLC (N2 involvement) treated with neoadjuvant platinum- based chemotherapy followed 3 cycles followed by 2 cycles of durvalumab and resection. Patients then underwent 1 year of adjuvant durvalumab treatment. Translational studies included tissue and blood interrogation at baseline (prior to chemoimmunotherapy; TP1), prior to neoadjuvant durvalumab (TP2), prior to surgery (TP3) and post 4 cycles of adjuvant durvalumab (TP4). This study reports the association of improved EFS with baseline tumor characteristics, including immune inflamed phenotype, tumor infiltrating CD8 T cell density and TLS size. The study also notes increased intra- tumoral T cell diversity and its association with improved EFS. Interrogation of the systemic immune system as measured through mass cytometry and multispectral flow cytometry on PMBC samples revealed post- treatment circulating Ki- 67- expressing CD39+ PD- 1+ CD8+ T cells that associated with improved EFS. Serum CCL15 elevations were noted in 3 patients with durable benefit to therapy for which CCL15 and its receptors CCR1 and CCR3 were explored in a pre- existing single- cell RNA dataset.  
+
+<|ref|>text<|/ref|><|det|>[[73, 612, 264, 625]]<|/det|>
+Major Comments/questions:  
+
+<|ref|>text<|/ref|><|det|>[[72, 625, 911, 702]]<|/det|>
+- Overall, the manuscript highlights previously known biomarkers of immune checkpoint blockade response with the potential to fundamentally understand the impact of chemotherapy on the modulation of key immune cells. The association of immune phenotype as it correlates to sequential chemotherapy and immunotherapy offers an incredibly opportunity to answer questions re: the role of these agents in neoadjuvant therapy for NSCLC. Additionally, this data set offers an opportunity to understand histology specific differences in immune profile and how that correlates with potential resistance mechanisms to neoadjuvant chemoimmunotherapy.  
+
+<|ref|>text<|/ref|><|det|>[[72, 702, 904, 744]]<|/det|>
+- In its current form, the manuscript does not take full advantage of this cohort's clinical impact. TP2 is a critically important time point, as it offers the opportunity to distinguish the impact of chemotherapy induction from that of immunotherapy. Recently published translational work in an LCMV model with sequential chemotherapy \(\rightarrow\) immunotherapy (https://doi.org/10.1158/1078-0432.CCR-23-1316) could be readily compared to these findings from human specimens.  
+
+<|ref|>text<|/ref|><|det|>[[72, 744, 900, 771]]<|/det|>
+- In addition, no mention is made of differential findings with respect to lung cancer histology, though it is well-recognized that the immune composition of lung cancer may vary significantly with respect to histology  
+
+<|ref|>text<|/ref|><|det|>[[72, 771, 884, 798]]<|/det|>
+- Do mutations described as having a potential deleterious impact on immunotherapy and chemotherapy response (i.e. STK11, KEAP1) confound the authors' findings at all?  
+
+<|ref|>text<|/ref|><|det|>[[72, 798, 840, 825]]<|/det|>
+- What is the significance of EFS relative to MPR/pCR and OS in this cohort? How does this differ from the other neoadjuvant IO +/- chemo studies?  
+
+<|ref|>text<|/ref|><|det|>[[72, 825, 875, 839]]<|/det|>
+- Clinical annotation of sites of recurrence could be of interest as well relative to tumor immune composition/response  
+
+<|ref|>text<|/ref|><|det|>[[73, 852, 252, 866]]<|/det|>
+Minor comments by page:  
+
+<|ref|>text<|/ref|><|det|>[[72, 866, 350, 880]]<|/det|>
+Pg 4, Line 80; multicentric \(\rightarrow\) multicenter  
+
+<|ref|>text<|/ref|><|det|>[[72, 880, 315, 893]]<|/det|>
+Pg 4, Line 88; does \(\rightarrow\) did or do not  
+
+<|ref|>text<|/ref|><|det|>[[72, 894, 923, 936]]<|/det|>
+Pg 5, Line 93; CM816 and other perioperative regimens have clearly demonstrated improved benefit for the inclusion of ICB for PD- L1 high expression. Only KN- 091 was inconsistent increasing levels of efficacy associated with higher level of expression of PD- L1. While NADIM did not reveal PD- L1 expression and TMB as associated with overall survival, this was a
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 48, 440, 61]]<|/det|>
+correlation with response to treatment including MPR.  
+
+<|ref|>text<|/ref|><|det|>[[72, 60, 920, 101]]<|/det|>
+Pg 7, Figure 1; helpful to have median EFS and OS (NR) in the graph. Notable that there was a good amount of attrition for digital pathology analysis (only 21/67 patients) - why? Was NGS done on up- front tumor specimens? Change in mutations with treatment could be of interest (Ricciuti, JCO 2024)  
+
+<|ref|>text<|/ref|><|det|>[[72, 100, 925, 128]]<|/det|>
+Pg 8, Lines 144- 145; please provide classification criteria for inflamed and excluded as there is no consensus criteria for that determination. Seems this was arbitrarily determined by reading pathologists? How many fields of view were analyzed?  
+
+<|ref|>text<|/ref|><|det|>[[75, 128, 825, 141]]<|/det|>
+Pg 9, Figure 2; immune desert was not described in the results section and the distribution was not conveyed.  
+
+<|ref|>text<|/ref|><|det|>[[75, 141, 795, 155]]<|/det|>
+Pg 9, Figure 2; graphs c and d are very difficult to interpret statistically significant from non- significant data  
+
+<|ref|>text<|/ref|><|det|>[[75, 155, 620, 168]]<|/det|>
+Pg 9, Figure 2; immune inflamed and TC are similar (unless criteria is different?)  
+
+<|ref|>text<|/ref|><|det|>[[75, 168, 900, 195]]<|/det|>
+Pg 12, Figure 3; legend explanation of panel b and panel c are switched. Disambiguation of IB/RES should be in legend. Pg 12, Figure 3; C TLS size criteria? Versus above or below the median? What is the relationship between TLS size and PD- L1 score? Need to make sure this is not confounding.  
+
+<|ref|>text<|/ref|><|det|>[[75, 196, 920, 208]]<|/det|>
+PD- L1 score? Need to make sure this is not confounding.  
+
+<|ref|>text<|/ref|><|det|>[[75, 208, 916, 222]]<|/det|>
+Pg 13, Lines 221- 222; it would be interesting to evaluate the change in TCR diversity in the peripheral blood across various  
+
+<|ref|>text<|/ref|><|det|>[[75, 222, 884, 268]]<|/det|>
+time points from baseline through post treatment given the non- significant findings at the post- treatment time period. Numerous studies, including Yost (cited in this paper) report on the clonal expansion that is seen with ICB. It would be interesting to see the effect of TCR diversity from chemotherapy and immunotherapy, respectively, in the neoadjuvant setting.  
+
+<|ref|>text<|/ref|><|det|>[[75, 268, 803, 282]]<|/det|>
+Pg 13; what is the relationship between elevated intratumoral T cell clonal richness and TLS size? PD- L1?  
+
+<|ref|>text<|/ref|><|det|>[[75, 282, 920, 323]]<|/det|>
+Pg 14, Fig 4. The distribution of patients appears somewhat skewed with 15 patients within the EFS \(< 1\) year and 34 patients in the EFS \(>1\) year timeframe. Is there a sufficient sample size to create 3 cohorts of extremely poor responders, average responders and exceptional responders for further stratification of the data? Does TCR richness change with treatment?  
+
+<|ref|>text<|/ref|><|det|>[[75, 323, 500, 336]]<|/det|>
+Pg 15; was WES performed? Or just targeted NGS + RNAseq?  
+
+<|ref|>text<|/ref|><|det|>[[75, 336, 925, 412]]<|/det|>
+Pg 16, Lines 274 - 275 describes the expansion of PD1+ CD8+ T cells in the peripheral blood post- chemoiimmunotherapy that was primarily seen in responders. A recent paper by Marinello A et al. Clin Cancer Res 2024, describes the influence of platinum- based chemotherapy in reducing proliferative PD1+ CD8+ T cell expansion compared to anti- PD1 therapy alone in a LCMV mouse model. It would be interesting to see if the effects of chemotherapy are detrimental to the later use of immunotherapy based on the longitudinal assessment of proliferating PD1+ CD8+ T cells in this translational study across all the pre- surgical time points.  
+
+<|ref|>text<|/ref|><|det|>[[75, 412, 896, 478]]<|/det|>
+Pg 16, Lines 276 - 277, the presence of CD57+ CD4+ T cells has been described in several studies in the infectious disease field as marker of cell senescence with potential long- term memory components. It would be interesting to understand if the expansion of this cell subset was coupled with markers of memory (CD27, CD127, TCF1, CCR7/CD45RO/CD45RA) versus a more terminally differentiated T cell subset marked increases in checkpoint markers such as CTLA4, TIM- 3, LAG- 3, etc.  
+
+<|ref|>text<|/ref|><|det|>[[75, 478, 920, 532]]<|/det|>
+Pg 23; authors should comment more specifically on the impact of chemotherapy before immunotherapy - this is what is unique compared to the well- described translational datasets on neoadjuvant immunotherapy alone and combination chemoiimmunotherapy. For the digital path analysis, a significant limitation is that markers were identified on serial sections - much weaker than simultaneous measurement with a multiplex.  
+
+<|ref|>text<|/ref|><|det|>[[75, 532, 364, 545]]<|/det|>
+Pg 24, Line 444, ipilimumab is misspelled.  
+
+<|ref|>sub_title<|/ref|><|det|>[[75, 556, 162, 569]]<|/det|>
+## Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[75, 581, 238, 593]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[75, 593, 865, 633]]<|/det|>
+I co- reviewed this manuscript with one of the reviewers who provided the listed reports. This is part of the Nature Communications initiative to facilitate training in peer review and to provide appropriate recognition for Early Career Researchers who co- review manuscripts.  
+
+<|ref|>sub_title<|/ref|><|det|>[[75, 645, 162, 658]]<|/det|>
+## Reviewer #4  
+
+<|ref|>text<|/ref|><|det|>[[75, 671, 238, 684]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[75, 684, 920, 880]]<|/det|>
+This manuscript profiles the immune response within a cohort of NSCLC patients that were treated as part of a phase II trial with neoadjuvant chemo and immunotherapy and a subsequent adjuvant immunotherapy regimen. These immune parameters, which are sourced from tissue, serum, and PBMCs at various time during the trial, are further correlated to 5- year survival outcomes. Major findings for patients with better outcomes include evidence of tumor inflammation, superior infiltration of CD8 T cells, TLS formation, TCR clonal diversity, and activated markers for infiltrating T cells, all of which have been accepted as signs of productive anti- tumor responses. Interestingly, tumor mutation burden did not correlate to outcomes. Novel findings include TIM- 3+ cDC1 among non- responders and signs of CCL15- CCR1 signaling among some responders. The dataset is extensive and impressive, and methods are sound. Enthusiasm is dampened somewhat in that the conclusions have already been highlighted in the field, and the patients are end up being responders seem to have more CD8 T cell infiltration on the onset. It is not clear if neoadjuvant therapy had any impact on turning more patients into responders. While the study seeks to elucidate the spatial dynamics of the tumor microenvironment, the study is rather descriptive. The Introduction states that the goal is to "evaluate whether emerging biomarkers are associated with sustained clinical benefit" but it is not clear if this had been accomplished by this retrospective analysis. In summary, the unique and deep dataset and its study is to be commended, but it is not clear if any hypotheses have been supported or even generated from the work. It is not clear whether this is suitable innovation to merit appearance in Nat Comm.  
+
+<|ref|>text<|/ref|><|det|>[[75, 893, 145, 905]]<|/det|>
+Version 1:  
+
+<|ref|>text<|/ref|><|det|>[[75, 920, 219, 932]]<|/det|>
+Reviewer comments:
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[73, 47, 161, 60]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[73, 73, 465, 100]]<|/det|>
+(Remarks to the Author) The authors have satisfactorily addressed our comments.  
+
+<|ref|>text<|/ref|><|det|>[[73, 112, 162, 126]]<|/det|>
+Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[73, 139, 240, 152]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[73, 152, 865, 193]]<|/det|>
+I co- reviewed this manuscript with one of the reviewers who provided the listed reports. This is part of the Nature Communications initiative to facilitate training in peer review and to provide appropriate recognition for Early Career Researchers who co- review manuscripts.  
+
+<|ref|>text<|/ref|><|det|>[[73, 204, 161, 217]]<|/det|>
+Reviewer #4  
+
+<|ref|>text<|/ref|><|det|>[[73, 230, 240, 243]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[73, 243, 921, 310]]<|/det|>
+Thank you for the thoughtful revision of the manuscript, which has incorporated my request for overall significance as well as the more detailed comments from other reviewers. I greatly appreciate the inclusion of a preliminary experiment in the rebuttal, which indicates that the authors have been considering fundamental aspects of the project and how this research can lead to innovation in tumor immunology. I am satisfied that the paper has addressed my concerns and recommend for acceptance.  
+
+<|ref|>text<|/ref|><|det|>[[72, 661, 916, 715]]<|/det|>
+Open Access This Peer Review 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.  
+
+<|ref|>text<|/ref|><|det|>[[72, 715, 796, 728]]<|/det|>
+In cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+<|ref|>text<|/ref|><|det|>[[72, 728, 911, 780]]<|/det|>
+The images or other third party material in this Peer Review 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.  
+
+<|ref|>text<|/ref|><|det|>[[72, 780, 618, 793]]<|/det|>
+To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[118, 85, 878, 140]]<|/det|>
+# Tumor immune dynamics and long-term clinical outcome of stage IIIA NSCLC patients treated with neoadjuvant chemoimmunotherapySchmid et al.  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 195, 321, 214]]<|/det|>
+## Point by point reply  
+
+<|ref|>text<|/ref|><|det|>[[120, 244, 421, 260]]<|/det|>
+Reviewer #1 (Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[117, 275, 880, 527]]<|/det|>
+Thank you for the opportunity to review this interesting manuscript, wherein the authors report on multi- omic translational analysis of specimens obtained during the SAKK 16/14 trial. This study enrolled patients with Stage IIIA NSCLC (N2 involvement) treated with neoadjuvant platinum- based chemotherapy followed 3 cycles followed by 2 cycles of durvalumab and resection. Patients then underwent 1 year of adjuvant durvalumab treatment. Translational studies included tissue and blood interrogation at baseline (prior to chemoimmunotherapy; TP1), prior to neoadjuvant durvalumab (TP2), prior to surgery (TP3) and post 4 cycles of adjuvant durvalumab (TP4). This study reports the association of improved EFS with baseline tumor characteristics, including immune inflamed phenotype, tumor infiltrating CD8 T cell density and TLS size. The study also notes increased intra- tumoral T cell diversity and its association with improved EFS. Interrogation of the systemic immune system as measured through mass cytometry and multispectral flow cytometry on PMBC samples revealed post- treatment circulating Ki- 67- expressing CD39+ PD- 1+ CD8+ T cells that associated with improved EFS. Serum CCL15 elevations were noted in 3 patients with durable benefit to therapy for which CCL15 and its receptors CCR1 and CCR3 were explored in a pre- existing single- cell RNA dataset.  
+
+<|ref|>text<|/ref|><|det|>[[119, 544, 352, 558]]<|/det|>
+Major Comments/questions:  
+
+<|ref|>text<|/ref|><|det|>[[118, 559, 880, 684]]<|/det|>
+- Overall, the manuscript highlights previously known biomarkers of immune checkpoint blockade response with the potential to fundamentally understand the impact of chemotherapy on the modulation of key immune cells. The association of immune phenotype as it correlates to sequential chemotherapy and immunotherapy offers an incredibly opportunity to answer questions re: the role of these agents in neoadjuvant therapy for NSCLC. Additionally, this data set offers an opportunity to understand histology specific differences in immune profile and how that correlates with potential resistance mechanisms to neoadjuvant chemoimmunotherapy.  
+
+<|ref|>text<|/ref|><|det|>[[118, 699, 879, 746]]<|/det|>
+We sincerely appreciate your thoughtful and constructive feedback. Your insights into the potential of our dataset to investigate the effects of chemotherapy on immune cell modulation and the role of these agents in neoadjuvant therapy for NSCLC are highly valuable.  
+
+<|ref|>text<|/ref|><|det|>[[117, 792, 880, 886]]<|/det|>
+- In its current form, the manuscript does not take full advantage of this cohort's clinical impact. TP2 is a critically important time point, as it offers the opportunity to distinguish the impact of chemotherapy induction from that of immunotherapy. Recently published translational work in an LCMV model with sequential chemotherapy \(\rightarrow\) immunotherapy (https://doi.org/10.1158/1078-0432.CCR-23-1316) could be readily compared to these findings from human specimens.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 879, 147]]<|/det|>
+We acknowledge that TP2 provides a valuable perspective on how chemotherapy and immunotherapy influence the host immune system. Notably, as highlighted in the referenced paper, the treatment sequence plays a critical role, as concurrent chemotherapy may dampen the immunostimulatory effects of checkpoint blockade.  
+
+<|ref|>text<|/ref|><|det|>[[118, 162, 879, 241]]<|/det|>
+In this revision, in response to the reviewer's suggestion, we have addressed this topic by conducting additional TCR sequencing on peripheral TP2 samples. Furthermore, we have expanded our discussion on the proliferative capacity of peripheral \(\mathrm{CD39^{+}}\) PD- 1 \(^+\) CD8 \(^+\) T cells during neoadjuvant chemoimmunotherapy. Please see our responses to the specific questions below for further details.  
+
+<|ref|>text<|/ref|><|det|>[[118, 286, 879, 334]]<|/det|>
+- In addition, no mention is made of differential findings with respect to lung cancer histology, though it is well-recognized that the immune composition of lung cancer may vary significantly with respect to histology  
+
+<|ref|>text<|/ref|><|det|>[[118, 348, 880, 491]]<|/det|>
+We appreciate this important comment concerning the separate analysis of different NSCLC histologies and the feasibility of further subtyping immune responses across different histological subtypes. Our dataset comprises a total of 37 cases of adenocarcinoma, 22 cases of squamous cell carcinoma, 1 case of large cell carcinoma, and 7 cases of NSCLC not otherwise specified (NOS). In line with clinical practice, we will categorize these cases in the revised manuscript as squamous cell carcinoma (SCC, \(n = 22\) ) and non- squamous cell carcinoma (non- SCC, \(n = 45\) ). Notably, our analysis reveals no significant differences in overall or event- free survival between patients with SCC and non- SCC (Extended Data Fig. 1a, b).  
+
+<|ref|>text<|/ref|><|det|>[[118, 491, 880, 601]]<|/det|>
+The histological response signature—including the inflamed immune phenotype, intra- tumoral \(\mathrm{CD8^{+}}\) T cell infiltrate infiltrates, and TLS size—is present across both subtypes and does not exhibit a statistically significant difference in frequency between SCC and non- SCC (Extended Data Fig. 2a- c). Furthermore, we observe comparable patterns in TCR indices (Extended Data Fig. 3a) and peripheral immune cell activation (Extended Data Fig. 8c) between SCC and non- SCC. However, the number of cases decreases when further subsetting the cohort, which may limit statistical power.  
+
+<|ref|>text<|/ref|><|det|>[[118, 601, 879, 696]]<|/det|>
+Despite these challenges, our analysis indicates that the observed immune phenotypes remain largely consistent across various histological subgroups of NSCLC. We recognize the need for further validation of these findings and a deeper investigation into distinct immune response subtypes within independent cohorts. Such validation would not only reinforce the broader applicability of our observations but also potentially identify histology- specific resistance mechanisms to neoadjuvant chemoimmunotherapy.  
+
+<|ref|>text<|/ref|><|det|>[[118, 696, 880, 851]]<|/det|>
+Further subtyping of immune responses within adenocarcinoma histological categories (lepidic, papillary, acinar, micropapillary and solid) presents considerable challenges in the neoadjuvant setting. First, such subtyping requires a comprehensive evaluation of resection specimens, whereas preoperative biopsies, though adequate for diagnosing lung adenocarcinoma, do not sufficiently capture the high degree of intratumoral heterogeneity. Second neoadjuvant treatment induces substantial histological alterations, making further subtyping or tumor grading unsuitable according to established WHO criteria in these samples. These limitations currently hinder the feasibility of this level of detailed analysis. We anticipate that future studies in larger, independent cohorts will build upon these observations.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 880, 118]]<|/det|>
+- Do mutations described as having a potential deleterious impact on immunotherapy and chemotherapy response (i.e. STK11, KEAP1) confound the authors' findings at all?  
+
+<|ref|>text<|/ref|><|det|>[[118, 134, 880, 214]]<|/det|>
+In the SAKK 16/14 cohort, \(17\%\) of patients had a STK11 mutation \((n = 8)\) and \(15\%\) had a KEAP1 mutation \((n = 7)\) . All patients with STK11 mutations were diagnosed with adenocarcinoma and were current or former smokers while one patient with a KEAP1 mutation was a never smoker, and \(n = 2\) patients with KEAP1 mutations did not have adenocarcinoma \((n = 1\) with large cell carcinoma, \(n = 1\) NSCLC NOS).  
+
+<|ref|>text<|/ref|><|det|>[[118, 214, 880, 260]]<|/det|>
+Furthermore, when checking for KRAS mutation status, only \(n = 2\) patients with STK11 mutations had concurrent KRAS mutations, while \(n = 1\) patient had concurrent KRAS and KEAP1 mutations. This information is displayed in Extended Data Fig. 4f.  
+
+<|ref|>text<|/ref|><|det|>[[118, 260, 880, 400]]<|/det|>
+STK11 and KEAP1 have been identified as negative predictors of immunotherapy response in patients with KRAS- mutated adenocarcinoma who have a positive smoking history, but not in KRAS wild- type adenocarcinomas<sup>1</sup>. The limited number of patients in our cohorts prevents the calculation of odds ratios to accurately assess the impact of these two mutations. However, we note that both patients with KRAS/STK11 mutations exhibited a very low percentage of PD- L1<sup>+</sup> tumor cells (1%, data not shown), consistent with previous reports. Consequently, we cannot definitively determine whether the deleterious impact of these mutations may have influenced our results. We look forward to future translational trials with larger patient cohorts to further investigate these questions.  
+
+<|ref|>text<|/ref|><|det|>[[118, 447, 880, 481]]<|/det|>
+- What is the significance of EFS relative to MPR/pCR and OS in this cohort? How does this differ from the other neoadjuvant IO +/- chemo studies?  
+
+<|ref|>text<|/ref|><|det|>[[118, 499, 880, 577]]<|/det|>
+Since neoadjuvant treatment regimens are performed in earlier stage tumors, surrogate endpoints such as event- free survival and pathological response rates are commonly used for preliminary analysis. This is because assessing changes in overall survival requires a longer follow- up period compared to studies in more advanced tumor settings. Accordingly, EFS at 12 months was chosen as the primary endpoint of the SAKK 16/14 trial<sup>2</sup>.  
+
+<|ref|>text<|/ref|><|det|>[[118, 592, 880, 688]]<|/det|>
+We would like to reference a recent review that has confirmed the predictive value of MPR/pCR and EFS for OS in neoadjuvant checkpoint inhibition<sup>3</sup>, which notably includes the SAKK 16/14 trial. Furthermore, we now provide updated survival data in Extended Data Fig. 1c- f, illustrating the relationship between MPR/pCR, EFS and OS. Our analysis demonstrates that patients achieving MPR and pCR have significantly longer EFS \((p < 0.0001\) and \(p = 0.0858\) , respectively) and OS \((p < 0.0001\) and \(p = 0.0288\) , respectively).  
+
+<|ref|>text<|/ref|><|det|>[[118, 734, 880, 768]]<|/det|>
+- Clinical annotation of sites of recurrence could be of interest as well relative to tumor immune composition/response  
+
+<|ref|>text<|/ref|><|det|>[[118, 785, 880, 895]]<|/det|>
+Among the \(n = 67\) evaluable patients, \(n = 30\) experienced an EFS event, \(n = 25\) of which have a recorded site of progression. Of these, \(n = 14\) developed distant metastases, with the brain \((n = 6)\) and bone \((n = 4)\) being the most common metastatic sites. Meanwhile, \(n = 11\) exhibited purely locoregional recurrence, occurring either in the ipsi- or contralateral lung or in new lymph nodes. We observed no significant differences in EFS and OS between patients with locoregional vs. distant progression \((p = 0.4758\) and \(p = 0.7675\) , respectively). These data are presented in Extended Data Fig. 1g, h.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 880, 163]]<|/det|>
+Unfortunately, digital pathology data were available for only a subset of patients with recorded progressions, limiting our ability to associate distinct tumor immune profiles with future sites of progression. We appreciate the reviewer's suggestion, as such an association would indeed be valuable to explore. We look forward to future studies with larger cohorts to further investigate this question.  
+
+<|ref|>text<|/ref|><|det|>[[118, 209, 457, 261]]<|/det|>
+Minor comments by page:  Pg 4, Line 80; multicentric -> multicenter  Pg 4, Line 88; does -> did or do not  
+
+<|ref|>text<|/ref|><|det|>[[119, 278, 444, 293]]<|/det|>
+Both of these mistakes were corrected.  
+
+<|ref|>text<|/ref|><|det|>[[118, 340, 880, 429]]<|/det|>
+Pg 5, Line 93; CM816 and other perioperative regimens have clearly demonstrated improved benefit for the inclusion of ICB for PD- L1 high expression. Only KN- 091 was inconsistent increasing levels of efficacy associated with higher level of expression of PD- L1. While NADIM did not reveal PD- L1 expression and TMB as associated with overall survival, this was a correlation with response to treatment including MPR.  
+
+<|ref|>text<|/ref|><|det|>[[118, 445, 880, 541]]<|/det|>
+We agree with the reviewer that in NSCLC patients being considered for perioperative immunotherapy, PD- L1 expression should be carefully assessed as a factor that may help guide the selection of the most appropriate treatment regimen. In this paragraph, our intention was to emphasize the relatively limited data on the predictive value of PD- L1 expression and TMB in the perioperative setting compared to their established role in purely systemic treatment for metastatic disease. We have revised this sentence accordingly.  
+
+<|ref|>text<|/ref|><|det|>[[118, 586, 880, 621]]<|/det|>
+Pg 7, Figure 1; helpful to have median EFS and OS (NR) in the graph. Notable that there was a good amount of attrition for digital pathology analysis (only 21/67 patients) - why?  
+
+<|ref|>text<|/ref|><|det|>[[118, 638, 678, 654]]<|/det|>
+We have added median EFS and OS to the survival curves in Fig. 1.  
+
+<|ref|>text<|/ref|><|det|>[[118, 670, 880, 780]]<|/det|>
+Regarding the attrition in digital pathology analysis, only a subset of the initial biopsies were tissue blocks suitable for digital pathology analysis. The remaining biopsies were cytology samples, primarily used for lung cancer diagnosis and PD- L1 assessment. The choice between these sample types was determined by the accessibility of the suspected lung tumor and local clinical protocols. Importantly, allowing both specimen types ensured that patient accrual was not hindered and enabled broad participation across multiple sites in Switzerland. We have clarified this point in the text accordingly.  
+
+<|ref|>text<|/ref|><|det|>[[118, 825, 880, 859]]<|/det|>
+Was NGS done on up- front tumor specimens? Change in mutations with treatment could be of interest (Ricciuti, JCO 2024)  
+
+<|ref|>text<|/ref|><|det|>[[118, 876, 880, 908]]<|/det|>
+Mutation profiling was conducted on mainly tumor resection samples, meaning the analysis was performed on tumors that had undergone neoadjuvant chemo- immunotherapy. The DNA
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 879, 132]]<|/det|>
+requirements for the assays used (50 ng for Foundation One CDx and 20 ng for Comprehensive Plus) were considered too high to be feasibly obtained from initial biopsies without compromising their use for other tissue- based analysis.  
+
+<|ref|>text<|/ref|><|det|>[[118, 147, 880, 257]]<|/det|>
+We agree with the reviewer that comparing mutational profiles before and after neoadjuvant immunotherapy could provide valuable mechanistic insights into the anti- tumor immune response. However, we note that in the referenced study<sup>4</sup>, samples were collected from patients who had demonstrated a confirmed response or stable disease for at least 3 months, with a median interval of 18.9 months between biopsies. In contrast, the shorter time frame inherent to the neoadjuvant setting must be considered when interpreting the dynamics of mutational profiles.  
+
+<|ref|>text<|/ref|><|det|>[[118, 257, 880, 288]]<|/det|>
+Additionally, we have clarified the sourcing of material for DNA mutation analysis in the Materials and Methods section.  
+
+<|ref|>text<|/ref|><|det|>[[118, 334, 880, 386]]<|/det|>
+Pg 8, Lines 144- 145; please provide classification criteria for inflamed and excluded as there is no consensus criteria for that determination. Seems this was arbitrarily determined by reading pathologists? How many fields of view were analyzed?  
+
+<|ref|>text<|/ref|><|det|>[[118, 404, 880, 575]]<|/det|>
+We appreciated the reviewer's attention to this missing information. Immune phenotyping was performed out by consensus of two board- certified pathologists (ABSB and VHK) following the recommendations of the International Immuno- Oncology Biomarker Working Group<sup>5,6</sup>. The complete tumor area was assessed to determine immune status, as previously described<sup>7</sup>. Specifically, the spatial distribution of tumor- infiltrating CD8<sup>+</sup> T cells was categorized into three immune phenotype: (1) immune desert – characterized by very rare and isolated CD8<sup>+</sup> T cells detected in any of the assessed tumor compartments, (2) immune excluded – defined by the presence of CD8<sup>+</sup> T cells at the tumor environment, primarily at the invasive margin or within the stroma, with only rare and isolated T cells found within the intratumoral compartment, and (3) inflamed – marked by CD8<sup>+</sup> T cells infiltrating the stromal compartment, directly contacting tumor cells, and penetrating the tumor parenchyma.  
+
+<|ref|>text<|/ref|><|det|>[[120, 591, 707, 607]]<|/det|>
+This information has been added to the Materials and Methods section.  
+
+<|ref|>text<|/ref|><|det|>[[118, 653, 880, 686]]<|/det|>
+Pg 9, Figure 2; immune desert was not described in the results section and the distribution was not conveyed.  
+
+<|ref|>text<|/ref|><|det|>[[118, 705, 880, 830]]<|/det|>
+This is correct. Fig. 2A presents generic examples of the three immune phenotypes in isolated tumor regions from SAKK 16/14 samples, intended to aid the reader's conceptual understanding. As noted, final immune classifications were determined by evaluating the entire tumor, rather than isolated regions. In the SAKK dataset, the immunologically "cold" subtype was entirely represented by immune- excluded tumors. While some excluded tumors can contain regions with very little immune infiltration, we did not identify any completely immune- desert tumors in the final dataset. To prevent any misunderstanding, we have clarified that these are generic examples from representative tumor regions.  
+
+<|ref|>text<|/ref|><|det|>[[118, 876, 877, 909]]<|/det|>
+Pg 9, Figure 2; graphs c and d are very difficult to interpret statistically significant from nonsignificant data
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 100, 880, 195]]<|/det|>
+In Fig. 2c, d, our aim was to emphasize that infiltration densities of the indicated cell populations did not differ between immune- excluded and inflamed tumors, neither in the tumor compartment nor in the stroma. This observation holds true for all four analyzed populations (CD3+ T cells, CD8+ T cells, FoxP3+ T regulatory cells and CD20+ B cells). To ensure clarity, we have revised the text accordingly and have also added previously missing information on statistical testing in the figure legend.  
+
+<|ref|>text<|/ref|><|det|>[[118, 209, 880, 273]]<|/det|>
+By contrast, as expected, all immune populations tended to be more abundant in the stroma than in the tumor compartment. For improved readability, we have opted not to display test statistics for this comparison, as well as for cross- population comparisons (e.g. CD3+ T cells vs. CD20+ B cells).  
+
+<|ref|>text<|/ref|><|det|>[[118, 318, 771, 335]]<|/det|>
+Pg 9, Figure 2; immune inflamed and TC are similar (unless criteria is different?)  
+
+<|ref|>text<|/ref|><|det|>[[118, 352, 880, 446]]<|/det|>
+We apologize for any confusion. TC (tumor compartment) refers to the segmented area within the tissue block where tumor cells are located. This compartment is surrounded by stroma. Immune cell densities are quantified and expressed per \(\mathsf{mm}^2\) within either the TC or the stroma. In contrast, the immune phenotype (desert, excluded, inflamed) is determined based on the spatial distribution of immune cells across these compartments, as assessed by pathologist consensus (as detailed in our previous response).  
+
+<|ref|>text<|/ref|><|det|>[[118, 492, 880, 526]]<|/det|>
+Pg 12, Figure 3; legend explanation of panel b and panel c are switched. Disambiguation of IB/RES should be in legend.  
+
+<|ref|>text<|/ref|><|det|>[[118, 544, 880, 576]]<|/det|>
+We thank the reviewer for pointing out the missing information. Legend explanations for panels b and c have been corrected, IB/RES has been explained.  
+
+<|ref|>text<|/ref|><|det|>[[118, 621, 880, 656]]<|/det|>
+Pg 12, Figure 3; C TLS size criteria? Versus above or below the median? What is the relationship between TLS size and PD- L1 score? Need to make sure this is not confounding.  
+
+<|ref|>text<|/ref|><|det|>[[118, 673, 880, 721]]<|/det|>
+The average (i.e. arithmetic mean) area of TLS in each initial biopsy was calculated, and the median of these values was used as a cutoff to classify samples as either "TLS size high" or "TLS size low". This clarification has been added to the figure legend.  
+
+<|ref|>text<|/ref|><|det|>[[118, 737, 880, 784]]<|/det|>
+Furthermore, we have added Extended Data Fig. 2d to illustrate the relationship between PD- L1 score and average TLS size. Our analysis shows no significant difference in PD- L1 scores between samples with small and large TLS (p = 0.3502, Mann- Whitney test).  
+
+<|ref|>text<|/ref|><|det|>[[118, 829, 880, 899]]<|/det|>
+Pg 13, Lines 221- 222; it would be interesting to evaluate the change in TCR diversity in the peripheral blood across various time points from baseline through post treatment given the non- significant findings at the post- treatment time period. Numerous studies, including Yost (cited in this paper) report on the clonal expansion that is seen with ICB. It would be interesting
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 878, 118]]<|/det|>
+to see the effect of TCR diversity from chemotherapy and immunotherapy, respectively, in the neoadjuvant setting.  
+
+<|ref|>text<|/ref|><|det|>[[118, 135, 879, 214]]<|/det|>
+We agree that investigating clonal dynamics across different phases of neoadjuvant treatment is highly relevant. To complement our data, we have now performed TCR sequencing on timepoint 2 (TP2) peripheral samples (collected after neoadjuvant chemotherapy but before neoadjuvant durvalumab). Of note, no tissue samples were available at this timepoint.  
+
+<|ref|>text<|/ref|><|det|>[[118, 230, 879, 339]]<|/det|>
+Before interpreting the results, we must highlight the technical challenges associated with preparing sequencing libraries in different batches. Although RNA quality metrics were comparable, TP2 samples had fewer sequencing reads compared to TP1 and TP3 samples (Extended Data Fig. 3f). Since clonal richness (the number of unique TCR clones) correlates with sequencing depth, this prevented a direct comparison of clonal richness across timepoints in patients with different event- free survival (Extended Data Fig. 3g). However, other TCR repertoire metrics, such as evenness, were not affected (Extended Data Fig. 3h).  
+
+<|ref|>text<|/ref|><|det|>[[118, 355, 880, 511]]<|/det|>
+To address this issue, we performed random down- sampling of sequencing reads (10'000, 50'000, 250'000, 500'000, 750'000, 1'000'000, 1'500'000 and 2'000'000 reads) and analyzed the impact on clonal richness (number of TCR clones) and TCR evenness. This analysis, illustrated with data from four representative patients (Extended Data. Fig 3i, j) revealed a near- linear increase in the number of detected TCR clones with increasing sequencing depth. Conversely, TCR evenness plateaued as sequencing reads increased. Based on these observations, we concluded that down- sampling is an appropriate method to mitigate batch effects between TP1/TP3 and TP2 sequencing runs. Consequently, all further analyses for PBMC samples were conducted using 250'000 randomly selected sequencing reads. No down- sampling was performed for TCR sequencing data from tumor resections.  
+
+<|ref|>text<|/ref|><|det|>[[118, 526, 880, 684]]<|/det|>
+We observed a trend toward a higher number of TCR clones in PBMC samples from patients achieving \(\mathrm{EFS} \geq 12\) months both at baseline (TP1, \(\mathrm{p} = 0.1604\) ) and after neoadjuvant chemotherapy (TP2, \(\mathrm{p} = 0.0551\) , Fig. 4d). Similarly, we detected a trend toward higher TCR evenness in patients with \(\mathrm{EFS} \geq 12\) months at baseline (TP1, \(\mathrm{p} = 0.0853\) , Fig. 4e). Furthermore, Hill- Simpson diversity showed a marked decrease from baseline (TP1) to post- chemotherapy (TP2) in all patients, potentially reflecting the deleterious effect of chemotherapy on T cell expansion ( \(\mathrm{EFS} \geq 12\) months, \(\mathrm{p} = 0.0234\) ; \(\mathrm{EFS} < 12\) months, \(\mathrm{p} = 0.1079\) ). Notably, in patients with \(\mathrm{EFS} \geq 12\) months, Hill- Simpson diversity significantly increased again following neoadjuvant durvalumab ( \(\mathrm{p} = 0.0301\) , Fig. 4f), suggesting a potential recovery or expansion of T cell diversity after checkpoint blockade.  
+
+<|ref|>text<|/ref|><|det|>[[118, 729, 878, 762]]<|/det|>
+Pg 13; what is the relationship between elevated intratumoral T cell clonal richness and TLS size? PD- L1?  
+
+<|ref|>text<|/ref|><|det|>[[118, 781, 880, 891]]<|/det|>
+We found no association between intratumoral T cell clonal richness and TLS size in resections, as shown in Extended Data Fig. 3b. It is important to note that TCR sequencing was performed on resections, not on initial biopsies, due to sample limitations. In contrast, TLS were analyzed in both initial biopsies and resections. However, for survival analysis (Fig. 3 and Extended Data Fig. 2c, d), we only used initial biopsy data. This decision was made because resections included samples with complete pathological response, making it impossible to quantify intratumoral TLS in those cases.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 880, 131]]<|/det|>
+Similarly, we found no association between T cell clonal richness and PD- L1 score, regardless of whether patients achieved \(\mathrm{EFS} \geq 12\) months or not. This is illustrated in **Extended Data Fig. 2e.**  
+
+<|ref|>text<|/ref|><|det|>[[118, 178, 880, 248]]<|/det|>
+Pg 14, Fig 4. The distribution of patients appears somewhat skewed with 15 patients within the EFS \(< 1\) year and 34 patients in the EFS \(>1\) year timeframe. Is there a sufficient sample size to create 3 cohorts of extremely poor responders, average responders and exceptional responders for further stratification of the data? Does TCR richness change with treatment?  
+
+<|ref|>text<|/ref|><|det|>[[117, 265, 880, 438]]<|/det|>
+As discussed in the paper, we selected a 12- month EFS cutoff because it has been the primary endpoint for evaluating the efficacy of this treatment regimen. However, as the reviewer correctly noted, this approach results in an imbalance in patient numbers between groups. To address this, we conducted an exploratory analysis by stratifying patients into three groups: EFS \(< 12\) months, EFS between 12 and 36 months, EFS \(>36\) months. We examined both T cell clonal richness (Fig. 4) and proliferation of tentatively tumor- reactive \(\mathrm{CD39^{+}}\) PD- \(1^{+}\) \(\mathrm{CD8^{+}}\) T cells (Fig. 5). While there appears to be no difference in clonal richness between the EFS \(\geq 12 < 36\) months and EFS \(>36\) months groups, we observed a difference in baseline proliferation of \(\mathrm{CD39^{+}}\) PD- \(1^{+}\) \(\mathrm{CD8^{+}}\) T T cells between these cohorts. This suggests that the proliferative capacity of tumor- reactive \(\mathrm{CD8^{+}}\) T cells at baseline may correlate with long- term event- free survival.  
+
+<|ref|>text<|/ref|><|det|>[[208, 505, 787, 522]]<|/det|>
+[editorial note: confidential, unpublished figures have been redacted]  
+
+<|ref|>text<|/ref|><|det|>[[118, 677, 880, 724]]<|/det|>
+However, the reduction in sample size limits the statistical power of the dataset, leading to a loss of statistical significance in several comparisons. We have addressed this limitation in the Discussion to acknowledge its potential impact on data interpretation.  
+
+<|ref|>text<|/ref|><|det|>[[118, 739, 880, 786]]<|/det|>
+Regarding TCR richness, we refer to our previous comment, where we discussed its association (or lack thereof) with event- free survival (EFS), TLS size, and PD- L1 score, as well as the impact of batch effects on the analysis.  
+
+<|ref|>text<|/ref|><|det|>[[118, 832, 636, 849]]<|/det|>
+Pg 15; was WES performed? Or just targeted NGS + RNAseq?  
+
+<|ref|>text<|/ref|><|det|>[[118, 866, 880, 897]]<|/det|>
+Mutational analysis was performed using one of two extensively validated targeted NGS assays on DNA isolated from FFPE- embedded tumor resection samples.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 880, 147]]<|/det|>
+The Foundation One CDx assay investigates 324 cancer- specific genes for base substitutions, insertions/deletions, copy number alterations and recombinations. Furthermore, it provides tumor mutational burden. (https://www.foundationmedicine.de/de/our-services/cdx.html).  
+
+<|ref|>text<|/ref|><|det|>[[118, 162, 880, 209]]<|/det|>
+The Oncomine Comprehensive Assay Plus (Thermo Fisher Scientific) targets 517 genes and allows detection of base substitutions, insertions/deletions, copy number alterations, as well as tumor mutational burden estimation.  
+
+<|ref|>text<|/ref|><|det|>[[118, 225, 880, 320]]<|/det|>
+Gene expression profiling was run on FFPE- isolated RNA using the Oncomine Immune Response Research Assay (Thermo Fisher Scientific). Full RNAseq was not feasible due to poor RNA quality metrics, with a typical RNA integrity number (RIN) of \(\sim 2\) and low RNA input. Importantly, the Oncomine assay captures expression levels of 395 genes that are commonly involved in immunotherapy response, providing a targeted but robust alternative to full RNAseq.  
+
+<|ref|>text<|/ref|><|det|>[[117, 364, 881, 508]]<|/det|>
+Pg 16, Lines 274 - 275 describes the expansion of \(PD1 + CD8 + T\) cells in the peripheral blood post- chemoimmunotherapy that was primarily seen in responders. A recent paper by Marinello A et al. Clin Cancer Res 2024, describes the influence of platinum- based chemotherapy in reducing proliferative \(PD1 + CD8 + T\) cell expansion compared to anti- PD1 therapy alone in a LCMV mouse model. It would be interesting to see if the effects of chemotherapy are detrimental to the later use of immunotherapy based on the longitudinal assessment of proliferating \(PD1 + CD8 + T\) cells in this translational study across all the presurgical time points.  
+
+<|ref|>text<|/ref|><|det|>[[118, 524, 880, 618]]<|/det|>
+Interestingly, we observed increased proliferation in peripheral \(\mathrm{CD39^{+}}\) PD- \(1^{+}\) CD8 \(^+\) T cells after neoadjuvant chemotherapy at TP2 compared to baseline (TP1). Furthermore, only in patients with \(\mathrm{EFS} > 12\) months, proliferation continued to increase from TP2 to TP3 following neoadjuvant immunotherapy (Fig. 5g). While the proliferative response to chemotherapy alone may seem counterintuitive, we believe several clinical factors can explain this observation.  
+
+<|ref|>text<|/ref|><|det|>[[118, 633, 880, 728]]<|/det|>
+In the study by Marinello et al, mice received concomitant chemo- and immunotherapy every three days, with samples collected three days after the last administration. In this preclinical model, chemotherapy was shown to reduce expansion, proliferation, and cytokine secretion of LCMV- specific \(\mathrm{CD8 + }\) T cells. In contrast, in our clinical trial, patients received chemotherapy on the first day of a three- week cycle. Several key factors likely contribute to the observed T cell dynamics:  
+
+<|ref|>text<|/ref|><|det|>[[147, 730, 881, 890]]<|/det|>
+- Cytopenia and rebound hematopoiesis: Chemotherapy typically induces peripheral pancytopenia around one week after administration, followed by homeostatic hematopoiesis to restore immune cell counts.- G-CSF administration: Patients also received G-CSF to stimulate hematopoiesis and mitigate the risk of infection.- Timing of TP2 Sampling: TP2 samples were collected three weeks after the last chemotherapy dose, immediately before the first durvalumab dose. By this time, hematopoietic recovery
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[177, 84, 878, 116]]<|/det|>
+would likely be underway, contributing to the observed increase in \(\mathrm{CD8^{+}}\) T cell proliferation.  
+
+<|ref|>text<|/ref|><|det|>[[118, 130, 880, 225]]<|/det|>
+Thus, the dynamics of \(\mathrm{CD8^{+}}\) T cell proliferation appear to differ based on event- free survival (EFS). In patients with EFS \(\geq 12\) months, proliferation remains stable or even increases following neoadjuvant immunotherapy (TP2 to TP3), suggesting a sustained immune response. In contrast, in patients with EFS \(< 12\) months, proliferation decreases from TP2 to TP3, potentially indicating a lack of sustained T cell activation in response to durvalumab.  
+
+<|ref|>text<|/ref|><|det|>[[118, 240, 880, 288]]<|/det|>
+For further details on the SAKK 16/14 trial, including dosing schedules and treatment regimens, we refer to the study protocol available in the initial publication presenting the clinical findings<sup>2</sup>.  
+
+<|ref|>text<|/ref|><|det|>[[118, 333, 880, 440]]<|/det|>
+Pg 16, Lines 276 – 277, the presence of \(\mathrm{CD57 + CD4 + T}\) cells has been described in several studies in the infectious disease field as marker of cell senescence with potential long- term memory components. It would be interesting to understand if the expansion of this cell subset was coupled with markers of memory (CD27, CD127, TCF1, CCR7/CD45RO/CD45RA) versus a more terminally differentiated T cell subset marked increases in checkpoint markers such as CTLA4, TIM- 3, LAG- 3, etc.  
+
+<|ref|>text<|/ref|><|det|>[[118, 456, 880, 520]]<|/det|>
+We agree that \(\mathrm{CD4^{+}}\) \(\mathrm{CD57^{+}}\) cells are an intriguing subset, particularly given the growing recognition of immunosenescence as a potential resistance mechanism to cancer immunotherapy. To further characterize this population, we analyzed CyTOF data, and the results are presented in Extended Data. Fig. 7c- e.  
+
+<|ref|>text<|/ref|><|det|>[[118, 520, 650, 536]]<|/det|>
+Overall, \(\mathrm{CD4^{+}}\) \(\mathrm{CD57^{+}}\) T cells exhibit distinct phenotypic features:  
+
+<|ref|>text<|/ref|><|det|>[[147, 536, 880, 599]]<|/det|>
+- they express lower levels of CCR7, CD27, and CD127 compared to \(\mathrm{CD57^{-}}\) cells, suggesting a more differentiated state.- they predominantly express CD45RO rather than CD45RA, indicating a memory-like phenotype, further supported by the presence of TCF1, particularly in responders.  
+
+<|ref|>text<|/ref|><|det|>[[118, 615, 640, 631]]<|/det|>
+However, several markers also suggest terminal differentiation:  
+
+<|ref|>text<|/ref|><|det|>[[147, 632, 880, 695]]<|/det|>
+- \(\mathrm{CD57^{+}}\) cells express higher levels of PD-1 and TOX, both associated with exhaustion and terminal differentiation.- they exhibit lower expression of activation markers CD25 and CD28, potentially reflecting reduced activation capacity.  
+
+<|ref|>text<|/ref|><|det|>[[118, 710, 880, 775]]<|/det|>
+Interestingly, despite their differentiated phenotype, Ki67 expression was observed only in \(\mathrm{CD57^{+}}\) \(\mathrm{CD4^{+}}\) cells from responders, and proliferation further increased after neoadjuvant chemoimmunotherapy. This was unexpected, given that the overall abundance of \(\mathrm{CD4^{+}}\) \(\mathrm{CD57^{+}}\) cells was actually higher in non- responders.  
+
+<|ref|>text<|/ref|><|det|>[[118, 789, 880, 837]]<|/det|>
+These findings raise intriguing questions about the functional role of \(\mathrm{CD4^{+}}\) \(\mathrm{CD57^{+}}\) T cells in the context of immunotherapy, and we look forward to future studies that will further elucidate the dynamics and significance of this population over the course of treatment.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 880, 172]]<|/det|>
+Pg 23; authors should comment more specifically on the impact of chemotherapy before immunotherapy - this is what is unique compared to the well- described translational datasets on neoadjuvant immunotherapy alone and combination chemoimmunotherapy. For the digital path analysis, a significant limitation is that markers were identified on serial sections - much weaker than simultaneous measurement with a multiplex.  
+
+<|ref|>text<|/ref|><|det|>[[118, 189, 880, 221]]<|/det|>
+We have expanded the Discussion on chemotherapy to further clarify the distinct effects of neoadjuvant chemotherapy versus durvalumab in the SAKK 16/14 trial.  
+
+<|ref|>text<|/ref|><|det|>[[118, 237, 880, 315]]<|/det|>
+We have also included a discussion on the limitations associated with using serial sections for digital pathology analysis. The primary reason for selecting single- marker IHC assays from routine diagnostics—rather than more complex multiplex approaches—was to maintain a strong translational focus, ensuring that our findings can be readily replicated in larger clinical trials. However, we acknowledge that this method does not allow for  
+
+<|ref|>text<|/ref|><|det|>[[147, 316, 880, 379]]<|/det|>
+- co-localization studies between different cell subsets (e.g. CD3+ T cells and CD20+ B cells)- Confirmation of cellular identity through co-staining (e.g. CD3+ CD8+ CD8 T cells or CD3+ FoxP3+ T reg cells).  
+
+<|ref|>text<|/ref|><|det|>[[118, 396, 880, 459]]<|/det|>
+Despite these limitations, the cellular densities presented in Fig. 2 were measured within specific tumor compartments, which were identified using H&E staining on the same slide as the IHC stain. This approach provides a spatially informed analysis while maintaining compatibility with clinical pathology workflows.  
+
+<|ref|>text<|/ref|><|det|>[[118, 475, 463, 491]]<|/det|>
+Pg 24, Line 444, ipilimumab is misspelled.  
+
+<|ref|>text<|/ref|><|det|>[[118, 509, 323, 524]]<|/det|>
+This has been corrected.  
+
+<|ref|>text<|/ref|><|det|>[[118, 586, 426, 602]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 618, 880, 666]]<|/det|>
+I co- reviewed this manuscript with one of the reviewers who provided the listed reports. This is part of the Nature Communications initiative to facilitate training in peer review and to provide appropriate recognition for Early Career Researchers who co- review manuscripts.  
+
+<|ref|>text<|/ref|><|det|>[[118, 681, 880, 760]]<|/det|>
+We sincerely appreciate the time and effort that both you and your co- reviewer have dedicated to evaluating our manuscript. We fully support initiatives like this that facilitate training in peer review and provide well- deserved recognition for Early Career Researchers. Your constructive feedback has been invaluable in refining our work, and we are grateful for your thoughtful contributions.  
+
+<|ref|>text<|/ref|><|det|>[[118, 836, 426, 852]]<|/det|>
+Reviewer #4 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 872, 880, 906]]<|/det|>
+This manuscript profiles the immune response within a cohort of NSCLC patients that were treated as part of a phase II trial with neoadjuvant chemo and immunotherapy and a
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 84, 880, 209]]<|/det|>
+subsequent adjuvant immunotherapy regiment. These immune parameters, which are sourced from tissue, serum, and PBMCs at various time during the trial, are further correlated to 5- year survival outcomes. Major findings for patients with better outcomes include evidence of tumor inflammation, superior infiltration of CD8 T cells, TLS formation, TCR clonal diversity, and activated markers for infiltrating T cells, all of which have been accepted as signs of productive anti- tumor responses. Interestingly, tumor mutation burden did not correlate to outcomes.  
+
+<|ref|>text<|/ref|><|det|>[[117, 211, 880, 316]]<|/det|>
+Novel findings include TIM- 3+ cDC1 among non- responders and signs of CCL15- CCR1 signaling among some responders. The dataset is extensive and impressive, and methods are sound. Enthusiasm is dampened somewhat in that the conclusions have already been highlighted in the field, and the patients are end up being responders seem to have more CD8 T cell infiltration on the onset. It is not clear if neoadjuvant therapy had any impact on turning more patients into responders.  
+
+<|ref|>text<|/ref|><|det|>[[117, 318, 880, 388]]<|/det|>
+While the study seeks to elucidate the spatial dynamics of the tumor microenvironment, the study is rather descriptive. The Introduction states that the goal is to "evaluate whether emerging biomarkers are associated with sustained clinical benefit" but it is not clear if this had been accomplished by this retrospective analysis.  
+
+<|ref|>text<|/ref|><|det|>[[117, 389, 880, 442]]<|/det|>
+In summary, the unique and deep dataset and its study is to be commended, but it is not clear if any hypotheses have been supported or even generated from the work. It is not clear whether this is suitable innovation to merit appearance in Nat Comm.  
+
+<|ref|>text<|/ref|><|det|>[[118, 459, 712, 476]]<|/det|>
+We thank the reviewer for raising these important conceptual questions.  
+
+<|ref|>text<|/ref|><|det|>[[118, 491, 880, 586]]<|/det|>
+Regarding the impact of neoadjuvant therapy on converting more patients into responders, we acknowledge that the single- arm design of the SAKK 16/14 study limits our ability to directly assess this question. However, previous studies have conclusively demonstrated that neoadjuvant chemoimmunotherapy increases the proportion of patients achieving major or complete pathological response and prolongs event- free and overall survival compared to neoadjuvant chemotherapy alone<sup>9- 14</sup>.  
+
+<|ref|>text<|/ref|><|det|>[[118, 600, 880, 664]]<|/det|>
+Unlike in melanoma<sup>15</sup>, a direct comparison between neoadjuvant and purely adjuvant (chemo- )immunotherapy regimens has not yet been performed in NSCLC. However, a recent review suggests that neoadjuvant immunotherapy may outperform purely adjuvant immunotherapy in NSCLC<sup>16</sup>.  
+
+<|ref|>text<|/ref|><|det|>[[117, 678, 880, 789]]<|/det|>
+We chose to primarily investigate pathology- based immune characteristics (such as \(\mathrm{CD8^{+}}\) T cell infiltration, TLS size, and immunophenotype) in initial biopsies because assessing these parameters in post- immunotherapy samples is technically more challenging. For example, in patients achieving pathological complete response (pCR), \(\mathrm{CD8^{+}}\) T cell densities become difficult to define due to the loss of clearly delineated tumor margins. Given that these patients have the longest EFS and OS (Extended Data Fig. 1e- f), these technical limitations could introduce bias when evaluating pathological correlates of response.  
+
+<|ref|>text<|/ref|><|det|>[[117, 789, 880, 883]]<|/det|>
+With regard to our stated goal in the Introduction, we aimed to highlight the challenges of biomarker discovery in the neoadjuvant NSCLC setting, where obtaining tumor specimens is inherently more complex than in other cancers such as melanoma. Developing and refining these methods is essential for future studies to stratify patients into treatment arms based on immune characteristics. We acknowledge that our original phrasing may have been misleading and have revised the introduction accordingly:
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 880, 148]]<|/det|>
+Finally, we would like to highlight a key hypothesis generated from our study. To our knowledge, the upregulation of TIM- 3 on peripheral cDCs in non- responders following anti- PD- L1 treatment has not been previously reported. This is particularly relevant as PD- L1 blockade on cDCs is a key mechanism driving response to treatment<sup>17</sup>.  
+
+<|ref|>text<|/ref|><|det|>[[118, 148, 880, 243]]<|/det|>
+To further investigate this mechanism, we are currently conducting reverse- translation studies in mice. In preliminary experiments, we observed that anti- PD- L1 treatment upregulates TIM- 3 on intratumoral cDCs, and this effect can be partially blocked by concurrent anti- TIM- 3 therapy. These experiments were performed in a murine 4T1 intramammary tumor model, designed to mimic the surgical resection and treatment scheme of the SAKK 16/14 trial.  
+
+<|ref|>text<|/ref|><|det|>[[174, 325, 755, 343]]<|/det|>
+[editorial note: confidential, unpublished figures have been redacted]  
+
+<|ref|>text<|/ref|><|det|>[[118, 447, 880, 480]]<|/det|>
+Given the preliminary nature of this experiment, we have chosen not to include it in the manuscript but to present it exclusively in this point- by- point response.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 510, 239, 529]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[115, 544, 870, 904]]<|/det|>
+1. Ricciuti, B. et al. Diminished Efficacy of Programmed Death-(Ligand)1 Inhibition in STK11- and KEAP1-Mutant Lung Adenocarcinoma Is Affected by KRAS Mutation Status. Journal of Thoracic Oncology 17, 399-410 (2022).  
+2. Rothschild, S. I. et al. SAKK 16/14: durvalumab in addition to neoadjuvant chemotherapy in patients with stage IIIA (N2) non-small-cell lung cancer—a multicenter single-arm phase II trial. Journal of clinical oncology 39, 2872-2880 (2021).  
+3. Nie, R. et al. Predictive value of radiological response, pathological response and relapse-free survival for overall survival in neoadjuvant immunotherapy trials: pooled analysis of 29 clinical trials. Eur J Cancer 186, 211-221 (2023).  
+4. Ricciuti, B. et al. Genomic and Immunophenotypic Landscape of Acquired Resistance to PD-(L)1 Blockade in Non-Small-Cell Lung Cancer. Journal of Clinical Oncology 42, 1311-1321 (2024).  
+5. Amgad, M. et al. Report on computational assessment of Tumor Infiltrating Lymphocytes from the International Immuno-Oncology Biomarker Working Group. npj Breast Cancer vol. 6 Preprint at https://doi.org/10.1038/s41523-020-0154-2 (2020).  
+6. Salgado, R. et al. The evaluation of tumor-infiltrating lymphocytes (TILS) in breast cancer: Recommendations by an International TILS Working Group 2014. Annals of Oncology vol. 26 259-271 Preprint at https://doi.org/10.1093/annonc/mdu450 (2015).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 82, 875, 610]]<|/det|>
+7. Sobottka, B. et al. Establishing standardized immune phenotyping of metastatic melanoma by digital pathology. Laboratory Investigation 101, 1561-1570 (2021).8. Marinello, A. et al. Platinum-Based Chemotherapy Attenuates the Effector Response of CD8 T Cells to Concomitant PD-1 Blockade. Clinical Cancer Research 30, 1833-1845 (2024).9. Provencio-Pulla, M. et al. Nivolumab+ chemotherapy versus chemotherapy as neoadjuvant treatment for resectable stage IIIA NSCLC: Primary endpoint results of pathological complete response (pCR) from phase II NADIM II trial. Preprint at (2022).10. Forde, P. M. et al. Neoadjuvant Nivolumab plus Chemotherapy in Resectable Lung Cancer. N Engl J Med 386, 1973-1985 (2022).11. Wakelee, H. et al. Perioperative Pembrolizumab for Early-Stage Non-Small-Cell Lung Cancer. New England Journal of Medicine 389, 491-503 (2023).12. Heymach, J. V. et al. Perioperative Durvalumab for Resectable Non-Small-Cell Lung Cancer. New England Journal of Medicine 389, 1672-1684 (2023).13. Lu, S. et al. Perioperative toripalimab + platinum-doublet chemotherapy vs chemotherapy in resectable stage II/III non-small cell lung cancer (NSCLC): Interim event-free survival (EFS) analysis of the phase III Neotorch study. Journal of Clinical Oncology 41, 425126-425126 (2023).14. Cascone, T. et al. LBA1 CheckMate 77T: Phase III study comparing neoadjuvant nivolumab (NIVO) plus chemotherapy (chemo) vs neoadjuvant placebo plus chemo followed by surgery and adjuvant NIVO or placebo for previously untreated, resectable stage II-IIIb NSCLC. Annals of Oncology 34, S1295 (2023).15. Patel, S. P. et al. Neoadjuvant-Adjuvant or Adjuvant-Only Pembrolizumab in Advanced Melanoma. New England Journal of Medicine 388, 813-823 (2023).16. Martins, R. S. et al. Neoadjuvant vs Adjuvant Chemoimmunotherapy for Stage II-IIIB Non-Small Cell Lung Cancer. in Annals of Thoracic Surgery vol. 118 672-681 (Elsevier Inc., 2024).17. Dammeijer, F. et al. The PD-1/PD-L1-Checkpoint Restrains T cell Immunity in Tumor- Draining Lymph Nodes. Cancer Cell 38, 685-700.e8 (2020).
+
+<--- Page Split --->

peer_reviews/supplementary_0_Transparent Peer Review file__e9f305359274f364160106c26e3b423ede3cbec231bbbc95ee32f2afdd3ba013/images_list.json

@@ -0,0 +1,78 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Fig. S2.1. Temperature-dependent extraordinary and ordinary refractive indices of the NJU001 material at \\(1300 \\mathrm{nm}\\) and \\(650 \\mathrm{nm}\\) . The refractive indices were determined by analyzing the fringe spacing extracted from interference patterns measured at \\(5^{\\circ}\\mathrm{C}\\) intervals.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 8
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_1.jpg",
+    "caption": "Fig. S2.2. Temperature-dependent variation of \\(d_{33}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        252,
+        106,
+        718,
+        255
+      ]
+    ],
+    "page_idx": 9
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_2.jpg",
+    "caption": "Fig. S15a. Target orientation distribution \\(\\alpha\\) of polar LCs.",
+    "footnote": [],
+    "bbox": [
+      [
+        278,
+        75,
+        736,
+        380
+      ]
+    ],
+    "page_idx": 10
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Revised Fig. 5:",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 11
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_3.jpg",
+    "caption": "Fig. S21. Photoalignment conditions for the LC structures in our manuscript",
+    "footnote": [],
+    "bbox": [
+      [
+        270,
+        199,
+        733,
+        572
+      ]
+    ],
+    "page_idx": 16
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_4.jpg",
+    "caption": "Fig. S16. A schematic of a cross-sections of the LC cell.",
+    "footnote": [],
+    "bbox": [
+      [
+        260,
+        95,
+        700,
+        216
+      ]
+    ],
+    "page_idx": 17
+  }
+]

peer_reviews/supplementary_0_Transparent Peer Review file__e9f305359274f364160106c26e3b423ede3cbec231bbbc95ee32f2afdd3ba013/supplementary_0_Transparent Peer Review file__e9f305359274f364160106c26e3b423ede3cbec231bbbc95ee32f2afdd3ba013.mmd

@@ -0,0 +1,703 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+# Periodically-modulated unipolar and bipolar orders in nematic fluids towards miniaturized nonlinear vectorial optics  
+
+Corresponding Author: Professor Yan-qing Lu  
+
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+Version 0:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+Review: Periodically-modulated unipolar and bipolar orders in nematic fluids: towards 1 miniaturized microscopic nonlinear photonics  
+
+The article presents the experimental results related to a new nematic liquid crystal mesophase, designated as the Nx phase, which was observed to exist between the nematic and ferroelectric nematic liquid crystal phases. The distinctive characteristics of this mesophase manifest as unipolar and bipolar orders. The authors employed these properties in the fabrication of three- dimensional polar topological structures. In the presented work, the authors demonstrate a strong understanding of liquid crystal photonics, as evidenced by their description of the preparation of a liquid crystal mixture, its properties, and the presentation of detailed information about the experimental setup. The experimental verification of the observed Nx mesophase included differential scanning calorimetry and second harmonic generation. The results obtained from these methods are detailed in the following sections. All results present a comprehensive study and are convincing. The authors successfully prepared multidomain NLC structures characterized by micron- size domain resolution. The article's structure and organization of ideas are appropriate, with a clear introduction, methodology, results, and discussion. The references are carefully chosen and are appropriate to the scope of the article. The article effectively highlights advances in the generation of topological states of polar matter and demonstrates the potential for microscopic nonlinear optical microdevices.  
+
+I would recommend this manuscript for publication in Nature Communications.  
+
+It would be beneficial to clarify the authors' precise intentions in writing about a "meter- scale vectorial beam generator." Line 107: "the nonlinear vectorial beam generator from meter to sub- millimeter scale".  
+
+It is probable that a typographical error has been made in the following text:  
+
+Line 448: "N,N'- Bis(2,5- di- tert- butylphenyl)- 3,3,9,10- 448 perylenedicarboximide" instead of: "Fluorescent dye N,N'- Bis(2,5- di- tert- butylphenyl)- 3,4,9,10- 448 perylenedicarboximide"  
+
+## Reviewer #2  
+
+(Remarks to the Author)  
+
+The authors describe a novel device based on LCs that directly generates perfect vector beams in the second harmonic. The paper is novel and the devices demonstrated have potential. However, the results are, in many places, very poorly described. Quite a few things are unclear to me, so I cannot assess whether they are wrong or only poorly written. The paper contains many fancy words but fails to convey its meaning in simple terms. This is already evident in the abstract and continues throughout the text. There are also no conclusions to summarize the findings of the paper. In this view also the novelty is not clear. I believe the paper can be published, but it should be significantly rewritten.  
+
+In the introduction, there are many non- optics related references, which are not well related to the paper's topic. Conversely, the generation of optical vortices and vector beams within LCs is insufficiently referenced. There are many papers about generating various beams via transmission through the structure, laser light emission, and SHG (10.1038/s41467- 024-
+
+<--- Page Split --->
+
+53040- 8). While the last paper (by the same authors) is cited, it is not discussed in enough detail, especially since the work of this paper is very similar to the former one. The novelty is, therefore, not obvious.  
+
+The authors give a factor of \(10^{\wedge} - 6\) in volume reduction but do not explain how they got this number. Anyway, I think it is not useful to talk about volume. Specifically, the lateral dimensions of an optical device are designed for the desired size of an optical beam. In many instances, it is not desirable to make a device small in lateral size; on the opposite, to be useful, it should be large enough. But it is useful to make them thinner. Although, in the case of this paper, the device is very thin, it is also very inefficient, compared, for example, to solid- state phase- matched crystals. So, there is no point in comparing their sizes. I urge the authors to delete this number and rather just emphasize that a single device (thin layer) can do the same task as a few regular optical components.  
+
+In Fig 1a- c there are some checkmarks, X and some crossed checkmarks. It is not clear what it means. Also, RM734 and DIO have an X for SHG activity in N and Nx phases, but for NJU001, from these graphics, someone could understand that it has SHG activity across all phases. Also, what does "promising platform" mean? I suggest to remove all of these descriptions from the fig 1.  
+
+"SHG efficiency is defined as the SHG intensity ratio of the NJU001 material to that of a Y- cut quartz plate." Is this meant for the same thickness in both cases? Giving an absolute number for the nonlinear coefficient in pm/V would be nice.  
+
+Lines 249- 273: This paragraph is unnecessary and also partially wrong as it is written now. The authors derive two obvious equations. This could be fine, but they also invented two new names, "nonlinear Malus's law" and "polar polarizers". They further write as if this is some new concept: "Therefore, the nonlinear Malus's law provides a new concept of "polar polarizers". But this is, of course, known for many decades. It is just an obvious property of nonlinear optical processes. Why invent new names and write that this is a new concept? They further write: "This relationship describes a new scenario in which the manipulation of light in the nonlinear optical regime is now vectorized, i.e., the vector field of the SH wave is head- to- tail inequivalent (Fig. S13). The vector field of the SH wave is determined by the orientation of local dipoles in the range of [0, 2π]. In contrast, the traditional Malus's law disables the head- to- tail inequivalence of polarization." This is, of course, wrong. In the nonlinear regime, we still have head- to- tail equivalence of polarization. In other words, the vector field of the SH wave is head- to- tail equivalent. The only difference is the phase, which is a completely separate property from the polarization. Anyway, the authors do not mention phase in this paragraph. They say the relationship (line 262) is vectorized and head- to- tail inequivalent. Which is wrong.  
+
+The section about generating perfect vector beams is extremely short (lines 275- 295), providing almost no details. All of the following topics should be discussed in more detail:  
+
+I. What are perfect vector beams? These are much less known to readers than, for example, vortex beams. So they must be described.  
+
+II. What are the structures in Fig4, and how were they made? Provide details on the surface anchoring patterns. Also polar director field must be shown, like it is shown in Fig. S17.  
+
+III. How were the two phases achieved (Fig4bi and ii). It is not obvious how the light's phase is modulated. Specifically, how is the spiral phase modulation achieved? With LCs this is usually achieved by illuminating a q-plate (e.g. a radial configuration) with a circularly polarized beam. But here, the input beam is linearly polarized. Also, to achieve a spiral phase, the left-right symmetry has to be broken. For example, in vortex generation by a q-plate, the symmetry is broken by the circularly polarized input beam. But in this paper, as far as I can tell, the symmetry is not broken by the input beam or the LC structure. Anyway, why a spiral phase is needed anyway? This would be used for vortex beams, but vortex beams are not mentioned in the paper. As far as I know, the perfect vector beams do not need a spiral phase. The authors are confusing vector and vortex beams. They also use the term topological charge, but polarization topological charge is relevant for vector beams.  
+
+IV. Also, how was the axicon phase modulation achieved? Again this is not mentioned at all. The phase is alternating between 0 and pi/2. However, the phase of SHG can only be either 0 or pi (Fig S13b).  
+
+V. It is not clear at all what the electric field does, how it is applied, how the director changes, and how the output exactly changes.  
+
+In connection to III., in fig4 it is not clear what the LC structure is, the same is also not clear from the beginning of the paper. Photoalignment is only mentioned once in the introduction, but is not clear where and how was it used for the rest of the results. For example, are the patterns in fig2 self- formed or due to photoalignment?  
+
+Not clear also what is the surface anchoring on the other non- patterned surface or are both surfaces patterned? Some cross- sections of the LC cells should help.  
+
+Another thing that is not clear to me is how they were able to generate a perfect circle as drawn in Fig4a and seen in 4e. No light should be generated when the director is perpendicular to the incoming polarization. Therefore there should be no SH light perpendicular to the incoming polarization. For this reason, all the rings should have gaps. From some images, this is evident; from some, others is not. In any case, Fig4a is wrong. Also, since there are intensity gaps in the ring, then polarization topological charge cannot be defined. In other words, we cannot define how many times the polarization rotates when going around one full circle since, in some positions, the phase is not defined due to the intensity being zero.  
+
+The structure of the paper is quite strange. The authors first talk about LC materials, then structures in LCs, then optics, and the final section is named "discussion." But the discussion does not discuss the results, but it is a section by itself, discussing some new things but not the optics at all. Therefore, there is no actual discussion/conclusion. The "discussion" chapter
+
+<--- Page Split --->
+
+would fit before the optics part. In general, the paper also feels like two or even three disconnected parts.  
+
+Version 1:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+In my opinion, the corrected version of the manuscript should include: "Periodically- Modulated Unipolar and Bipolar Orders in Nematic Fluids: Towards Miniaturized Nonlinear Vectorial Optics," the authors successfully addressed majority of the concerns raised by the reviewers. The details of the experiments are much clearer, including the improved abstract, additional discussion, and conclusion, as well as more readable figures.  
+
+One typographical error needs to be corrected:  
+
+Line 507: "polar nad orientational orders" instead of: "polar and orientational orders"  
+
+Reviewer #2  
+
+(Remarks to the Author)  
+
+The authors have addressed most of my comments, except for two critical points (below). When these two points are addressed, I suggest publishing the article.  
+
+The novelty compared to their previous work is still not appropriately addressed. The authors should state this more clearly.  
+
+The authors removed the earlier assertive framing about "Nonlinear Malus's law" and "polar polarizers", rewrote the section, and dropped the headline claim—but they still describe the medium analogously as a "specialized 'nonlinear polar polarizer'," and state the transmission axis shows head- to- tail asymmetry via a Jones- style description. So the term specialized "nonlinear polar polarizer" should be removed. The head- to- tail asymmetry should be removed because it is incorrect.  
+
+Version 2:  
+
+Reviewer comments:  
+
+Reviewer #2  
+
+(Remarks to the Author)  
+
+All comments have been addressed. I recommend publication.
+
+<--- Page Split --->
+
+Open Access This Peer Review 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 cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+The images or other third party material in this Peer Review 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 https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+
+## Re: manuscript NCOMMS-25-04493-T  
+
+We sincerely thank the respected reviewers for their valuable time and helpful comments. We have carefully revised our manuscript and Supplementary Information by taking the respected reviewers' comments as appropriate into account.  
+
+Our point- by- point response to the reviewers' comments and the changes made in our revised manuscript with yellow color highlighted track change as follows.  
+
+## Reviewer 1  
+
+The article presents the experimental results related to a new nematic liquid crystal mesophase, designated as the Nx phase, which was observed to exist between the nematic and ferroelectric nematic liquid crystal phases. The distinctive characteristics of this mesophase manifest as unipolar and bipolar orders. The authors employed these properties in the fabrication of three- dimensional polar topological structures. In the presented work, the authors demonstrate a strong understanding of liquid crystal photonics, as evidenced by their description of the preparation of a liquid crystal mixture, its properties, and the presentation of detailed information about the experimental setup. The experimental verification of the observed Nx mesophase included differential scanning calorimetry and second harmonic generation. The results obtained from these methods are detailed in the following sections. All results present a comprehensive study and are convincing. The authors successfully prepared multidomain NLC structures characterized by micron- size domain resolution. The article's structure and organization of ideas are appropriate, with a clear introduction, methodology, results, and discussion. The references are carefully chosen and are appropriate to the scope of the article. The article effectively highlights advances in the generation of topological states of polar matter and demonstrates the potential for microscopic nonlinear optical microdevices. I would recommend this manuscript for publication in Nature Communications.  
+
+## Response:  
+
+We appreciate Reviewer 1's encouraging comments and recommendation on our work.  
+
+Q1.1 \\*It would be beneficial to clarify the authors' precise intentions in writing about a "meter- scale vectorial beam generator." Line 107: "the nonlinear vectorial beam generator from meter to sub- millimeter scale".  
+
+## A1.1 Response:  
+
+We appreciate this insightful suggestion. We aimed to demonstrate the miniaturization of our nonlinear vectorial photonic platform from meter- scale systems to sub- millimeter scale. Traditional nonlinear optical systems often necessitate large- scale setups, which typically involve beam splitting (the first step is to separate the two polarization components into two propagating beams by using a polarizing beam splitter) and beam regathering with a separated SHG process and a linear light field manipulation process. Thus, the optical paths are complex, requiring many optical components [Photonics Res. 7, 1340 (2019). Chin. Phys. Lett. 39, 034201 (2022). Appl. Phys. Lett. 119, 011104 (2021)]. By leveraging the engineerable molecular orientation of polar nematics, we demonstrate the integration of vectorial field generation and nonlinear optical frequency conversion into a single, miniaturized device—achieving a significantly reduced volume. This paves the way for portable and scalable photonic technologies with broad application potential, especially integrated photonics.  
+
+To better clarify this point in our manuscript, we added a paragraph in our "Discussion and Conclusion" section.  
+
+## Added Text in Revised Manuscript:  
+
+Line 447- 458, page 15: "As photonics technology advances from scalar to vectorial regimes \(^{30,51 - 54}\) , the construction of nonlinear vectorial light fields has become increasingly essential for applications including super- resolution imaging \(^{55}\) , high- capacity optical communications \(^{51}\) , and high- precision laser micromachining \(^{56}\) . Note that creating such vectorial nonlinear modes would require several passes through a spatial light modulator (lossy) or mixing beams with interferometry (difficult to align), demanding complex optical systems with cascaded linear and nonlinear light field manipulation. Here, the
+
+<--- Page Split --->
+
+generation of SH PVBs is achieved directly from a single, miniaturized polar LC device under a scalar Gaussian FW incidence, thanks to the flexible tailoring of tunable polar topologies. This is a feat that has not yet been accomplished with other ferroelectric materials. Such intriguing light- matter interaction could lead to the next generation of portable, scalable, and reconfigurable photonic technologies with wide- ranging applications, e.g., all- optical communications, quantum computing, soft intelligent robotics, and the biomedical industry."  
+
+Q1.2 \*It is probable that a typographical error has been made in the following text:  
+
+Line 448: "N,N'- Bis(2,5- di- tert- butylphenyl)- 3,3,9,10- 448 perylenedicarboximide" instead of: "Fluorescent dye N,N'- Bis(2,5- di- tert- butylphenyl)- 3,4,9,10- 448 perylenedicarboximide"  
+
+## A1.2 Response:  
+
+We apologize for the misstate. We have corrected this accordingly.  
+
+## Original Text in Manuscript:  
+
+Line 448- 449, page 17: "Fluorescent dye N,N'- Bis(2,5- di- tert- butylphenyl)- 3,3,9,10- 448 perylenedicarboximide."  
+
+Revised Text in Revised Manuscript:  
+
+Line 495- 496, page 16: "Fluorescent dye N,N'- Bis(2,5- di- tert- butylphenyl)- 3,4,9,10- 448 perylenedicarboximide."  
+
+## Reviewer #2  
+
+The authors describe a novel device based on LCs that directly generates perfect vector beams in the second harmonic. The paper is novel and the devices demonstrated have potential. However, the results are, in many places, very poorly described. Quite a few things are unclear to me, so I cannot assess whether they are wrong or only poorly written. The paper contains many fancy words but fails to convey its meaning in simple terms. This is already evident in the abstract and continues throughout the text. There are also no conclusions to summarize the findings of the paper. In this view also the novelty is not clear. I believe the paper can be published, but it should be significantly rewritten.  
+
+## Response:  
+
+We acknowledge the review's concerns and thank the suggestions. In response to the review's feedback, we have conducted experiments, made substantial revisions, and rewritten sections to improve the clarity, conciseness, and overall presentation of our findings, ensuring that our research is conveyed more effectively.  
+
+Q2.1 \*In the introduction, there are many non optics related references, which are not well related to the paper's topic. Conversely, the generation of optical vortices and vector beams within LCs is insufficiently referenced. There are many papers about generating various beams via transmission through the structure, laser light emission, and SHG (10.1038/s41467- 024- 53040- 8). While the last paper (by the same authors) is cited, it is not discussed in enough detail, especially since the work of this paper is very similar to the former one. The novelty is, therefore, not obvious.  
+
+## A2.1 Response:  
+
+We thank the reviewer for his/her constructive suggestions. We have removed some non- optics related references and added some new references related to optical vortices and vector beams with LCs to enrich the introduction.  
+
+## (1) References about linear optical vortices and vector beams: structure  
+
+[New J. Phys. 9, 78 (2007)] (LC- POV); [Opt. Lett. 41, 2205- 2208 (2016)] (LC- SLM based PVVB); [Appl. Phys. Lett. 113, 121101 (2018)] (LC- POV/PVB);
+
+<--- Page Split --->
+
+[Opt. Commun. 469, 125807 (2020)] (LC- SLM based PVVB); laser light emission [Opt. Express. 18, 212 (2020).] (LC- SLM based VB) [Proc. Natl. Acad. Sci. 118, 2110839118 (2021)] (Topological liquid crystal superstructures as structured light lasers);  
+
+(2) References about nonlinear optical vortices and vector beams:  
+
+SHG/nonlinear crystal  
+
+[Opt. Lett. 43, 5981- 5984 (2018)]; (LC- SLM based Nonlinear VB)  
+
+Added Optics- Related References in Revised Manuscript:  
+
+Line 73- 74, page 2: "... have opened up new vistas to explore topological counterparts in optics<sup>22- 30</sup>."  
+
+24 Maurer, C., Jesacher, A., Fürhapter, S., Bernet, S. & Ritsch- Marte, M. Tailoring of arbitrary optical vector beams. New J. Phys. 9, 78- 78 (2007).  
+
+25 Li, P. et al. Generation of perfect vectorial vortex beams. Opt. Lett. 41, 2205 (2016).  
+
+26 Wei, B.- Y. et al. Vortex Airy beams directly generated via liquid crystal q- Airy- plates. Appl. Phys. Lett. 112, 121101 (2018).  
+
+27 Mandal, A., Maji, S. & Brundavanam, M. M. Common- path generation of stable cylindrical perfect vector vortex beams of arbitrary order. Optics Communications 469, 125807 (2020).  
+
+28 Bashkansky, M., Park, D. & Fatemi, F. K. Azimuthally and radially polarized light with a nematic SLM. Opt. Express 18, 212 (2020).  
+
+29 Papic, M. et al. Topological liquid crystal superstructures as structured light lasers. Proc. Natl. Acad. Sci. U.S.A. 118, 2110839118 (2021).  
+
+30 Liu, H., Li, H., Zheng, Y. & Chen, X. Nonlinear frequency conversion and manipulation of vector beams. Opt. Lett. 43, 5981 (2018).  
+
+We appreciate the reviewer's insightful observation regarding the similarity between our current manuscript and the previously published work (Nat. Commun.10.1038/s41467- 024- 53040- 8). While the two studies share methodological foundations, they differ fundamentally in their scientific focus and contributions, as explained below:  
+
+## (1) Distinct Focus on Phases with Unique Physical Properties:  
+
+The previous work (Nat. Commun.10.1038/s41467- 024- 53040- 8) primarily investigated the ferroelectric nematic phase, which exhibits spontaneous polarization and robust ferroelectricity. In contrast, this manuscript explores an intermediate mesophase between the ferroelectric nematic phase and the conventional nematic phase. This mesophase demonstrates unipolar and bipolar polar orderings, with physical properties (e.g., polarization magnitude, orientational order) intermediate between ferroelectric nematic and nematic phases. A new structural model is established that elucidates the formation of the unique unipolar orderings in this mesophase—a key advance beyond the scope of our prior work. Crucially, it enables programmable and defect- free domain engineering due to the mesophase's unique polar order, a capability not demonstrated in the ferroelectric nematic phase.  
+
+## (2) Structures:  
+
+We fabricate an azimuthally- variant q- plate structure in the previous work (Fig. 3 in Nat. Commun.10.1038/s41467- 024- 53040- 8). On the other hand, this work demonstrates more complex topological hierarchical superstructures (revised Fig. 5) with a high performance in nonlinear vectorial optics.  
+
+## (3) Distinct Nonlinear Optical Phenomena:  
+
+Our prior work focused on achieving scalar nonlinear optical vortices in both left- handed and right- handed circular polarization channels under circular polarization incidence, in order to demonstrate the nonlinear geometric phase principle in LCs. On the other hand, this manuscript reports the generation of perfect vectorial nonlinear beams (e.g., vector beams
+
+<--- Page Split --->
+
+with spatially varying polarization states) under linear polarization incidence. This distinction expands the toolkit for polarization- controlled nonlinear optics, leveraging the mesophase's flexible polar domain engineering.  
+
+In summary, the current work represents a conceptual and experimental progression by targeting an unexplored polar mesophase structure with emergent properties, developing a structural model, and demonstrating novel nonlinear vectorial optical functionalities. We have clarified these distinctions in the revised manuscript (see Introduction and Discussion sections) to better highlight the novelty of this work for readers.  
+
+## Added Text in Revised Manuscript:  
+
+Line 85- 91, page 3: "By photopatterning an azimuthally- variant q- plate structure, our group recently demonstrated the generation of second- harmonic scalar optical vortices with spin- locked topological charges by leveraging cascaded linear and nonlinear optical spin- orbit interactions<sup>34</sup>. However, developing more complex topological polar LC electrooptic devices remains a critical challenge due to the intricate energy competition among elastic energy, Landau energy, polarization gradients, flexoelectricity, and electrostatics, which limits the fabrication of defect- free and high- performance polar structures<sup>36,38,39</sup>."  
+
+Q2.2 \*The authors give a factor of \(10^{\wedge} - 6\) in volume reduction but do not explain how they got this number. Anyway, I think it is not useful to talk about volume. Specifically, the lateral dimensions of an optical device are designed for the desired size of an optical beam. In many instances, it is not desirable to make a device small in lateral size; on the opposite, to be useful, it should be large enough. But it is useful to make them thinner. Although, in the case of this paper, the device is very thin, it is also very inefficient, compared, for example, to solid- state phase- matched crystals. So, there is no point in comparing their sizes. I urge the authors to delete this number and rather just emphasize that a single device (thin layer) can do the same task as a few regular optical components.  
+
+## A2.2 Response:  
+
+We agree with the reviewer that different application scenarios necessitate varying device dimensions. Accordingly, we have deleted this number and emphasized that we use a single- component device in a thin layer form to achieve the same task that usually should employ several bulky optical components.  
+
+## Original Text in Manuscript:  
+
+Line 35- 37, page 1: "We succeed in scaling down the volume of the nonlinear vectorial beam generation devices by a factor over \(10^{- 6}\) , providing a foundation for the microscopic nonlinear optical apparatuses."  
+
+## Revised Text in Revised Manuscript:  
+
+Line 35- 39, page 1: "A case study demonstrates the facile generation of second- harmonic perfect vector beams through photopatterned topological polar liquid crystal superstructures. Remarkably, this single- layer, micrometer- thin- film device attains functionalities comparable to multiple conventional optical components, establishing a foundation for advanced nonlinear vectorial optics at the microscopic scale."  
+
+## Original Text in Manuscript:  
+
+Line 106- 108, page 3: "We succeed in miniaturizing the nonlinear vectorial beam generator from meter to sub- millimeter scale, scaling down the volume of the devices by a factor over \(10^{- 6}\) ."  
+
+## Revised Text in Revised Manuscript:  
+
+Line 98- 100, page 3: "Additionally, we successfully generate second- harmonic perfect vector beams using a single device featuring a micrometer- thin LC layer, which performs the same functions as a large- scale optical system made up of multiple optical components."  
+
+Q2.3 \*In Fig 1a- c there are some checkmarks, X and some crossed checkmarks. It is not clear what it means. Also, RM734 and DIO have an X for SHG activity in N and Nx phases, but for NJU001, from these graphics, someone could understand
+
+<--- Page Split --->
+
+that it has SHG activity across all phases. Also, what does "promising platform" mean? I suggest to remove all of these descriptions from the fig 1.  
+
+## A2.3 Response:  
+
+We sincerely thank the reviewer for the great suggestions. We have removed these descriptions from the Fig. 1 accordingly.  
+
+## Original Fig. 1:  
+
+![](images/Figure_unknown_0.jpg)
+
+
+<--- Page Split --->
+
+Revised Fig. 1:  
+
+![](images/Figure_unknown_1.jpg)
+  
+
+Q2.4 \*\\*SHG efficiency is defined as the SHG intensity ratio of the NJU001 material to that of a Y- cut quartz plate." Is this meant for the same thickness in both cases? Giving an absolute number for the nonlinear coefficient in pm/V would be nice.  
+
+## A2.4 Response:  
+
+In our experiment, the thickness of the Y- cut quartz plate was \(2\mathrm{mm}\) and the thickness of the NJU001 sample was \(5.2\mu \mathrm{m}\) . We have supplemented this information in the revised manuscript. Also, we agree with the reviewer that it would be better to provide an absolute number for the nonlinear coefficient \(d_{33}\) in pm/V. Relevant experimental result have been added to the revised manuscript.  
+
+The nonlinear optical coefficient \(d_{33}\) can be determined using the following equation for second- harmonic generation (SHG):  
+
+\[I_{2} = \frac{2\omega_{2}^{2}\sin^{2}\left(\frac{\Delta kd}{2}\right)}{\epsilon_{0}n_{2}n_{1}^{2}c^{3}\left(\Delta k\right)^{2}} d_{33}^{2}I_{1}^{2} = \mathrm{C}\frac{\sin^{2}\left(\frac{\Delta kd}{2}\right)}{n_{2}n_{1}^{2}\left(\Delta k\right)^{2}} d_{33}^{2}I_{1}^{2}\]  
+
+where:  
+
+\(\Delta k = \frac{2\pi\omega_{1}(n_{1} - n_{2})}{c}\) is the wavevector mismatch;  
+
+\(\epsilon_{0}\) is the vacuum permittivity;  
+
+\(\omega_{1}\) and \(\omega_{2}\) are the angular frequencies of the fundamental wave (FW) and SH wave; \(c\) is the speed of light;  
+
+\(n_{1}\) and \(n_{2}\) represent the refractive indices of the FW and SH wave, respectively;  
+
+\(C\) is a constant related to the measuring system;  
+
+\(d_{33}\) is the nonlinear optical coefficient,  
+
+\(d\) is the sample thickness;
+
+<--- Page Split --->
+
+\(I_{1}\) and \(I_{2}\) represent the intensities of the FW and the SH wave.  
+
+Once C is determined, and \(n_{1}\) and \(n_{2}\) are measured, we can determine the \(d_{33}\) of the NJU001 material by fitting the intensity correlation between FW \(I_{1}\) and the SH wave \(I_{2}\) .  
+
+## (1) Determination of the System Constant C  
+
+The constant C was calibrated using a \(z\) - cut \(\mathrm{LiNbO_3}\) crystal ( \(d = 1 \mathrm{mm}\) in our experiment), where the nonlinear interaction is governed by the \(d_{22}\) coefficient under the experimental configuration: optic axis aligned along the \(x\) - axis, and incident FW polarization along the \(y\) - axis. Given that \(d_{22} = 6.3 \mathrm{pm / V}\) [G. D. Boyd et al. Appl. Phys. Lett. 5, 234- 236 (1964)], the refractive indices \(n_{1} = 2.218\) and \(n_{2} = 2.279\) at \(650 \mathrm{nm}\) and \(1300 \mathrm{nm}\) [O. Gayer et al. Appl. Phys. B. 91, 343- 348 (2008)], and \(\Delta k\) calculated from the dispersion relation, C was determined by fitting the \(I_{1} - I_{2}\) intensity correlation.  
+
+## (2) Measurement of the Refractive Indices in NJU001  
+
+The extraordinary \((n_{\mathrm{e}})\) and ordinary \((n_{\mathrm{o}})\) refractive indices of NJU001 were measured using an interference method [Priyanka Kumari et al. Science. 383, 1364 (2024)].  
+
+Sample Preparation:  
+
+A commercial wedge cell (EHC KCRK- 05, KERUN EXPERIMENTAL EQUIPMENT CO., LTD) was filled with NJU001 at \(145^{\circ}\mathrm{C}\) via capillary. The rubbing direction was along the thickness gradient.  
+
+Interference Fringe Analysis:  
+
+Polarized light microscopy with red/near-infrared filters ( \(\lambda = 650 / 1300 \mathrm{nm}\) , bandwidth \(10 \mathrm{nm}\) ).  
+
+Dihedral angle \(\alpha\) determined from air fringes:  
+
+\[\alpha = \frac{\lambda}{2n_{\mathrm{air}}u_{\mathrm{air}}} (\lambda = 650 \mathrm{nm}, n_{\mathrm{air}} = 1).\]  
+
+Refractive indices calculated via:  
+
+\[n_{\mathrm{o},\mathrm{e}} = \frac{\lambda}{2\alpha u_{\mathrm{LC}}},\]  
+
+where \(u_{\mathrm{LC}}\) is the fringe spacing in NJU001.  
+
+Measurement Protocol:  
+
+\(n_{\mathrm{e}}\) measured when the director was parallel to the light polarization,  
+
+\(n_{\mathrm{o}}\) measured when the director was perpendicular.  
+
+The temperature- dependent refractive indices at \(650 \mathrm{nm}\) and \(1300 \mathrm{nm}\) are shown in the following figure.  
+
+![](images/Figure_unknown_2.jpg)
+
+<center>Fig. S2.1. Temperature-dependent extraordinary and ordinary refractive indices of the NJU001 material at \(1300 \mathrm{nm}\) and \(650 \mathrm{nm}\) . The refractive indices were determined by analyzing the fringe spacing extracted from interference patterns measured at \(5^{\circ}\mathrm{C}\) intervals. </center>  
+
+After we have obtained the material parameters of \(n_{2}\) , \(n_{1}\) , and C, \(\Delta k\) was calculated. Subsequently, the nonlinear optical coefficient \(d_{33}\) was extracted by fitting the intensity correlation between FW \(I_{1}\) and the SH wave \(I_{2}\) . The temperature- dependent variation of \(d_{33}\) is presented in the following figure. As seen from the figure, the laser power/energy meter begins
+
+<--- Page Split --->
+
+to detect the SH signal at approximately \(64^{\circ}\mathrm{C}\) , which shows a slight discrepancy compared to that ( \(\sim 70^{\circ}\mathrm{C}\) ) in Fig. 1e, which is derived from the light intensity processed by the CMOS camera in the main text. This difference is attributed to the sensitivity of the detection instruments; specifically, the CMOS camera is capable of detecting weaker SHG signals.  
+
+![](images/Figure_5.jpg)
+
+<center>Fig. S2.2. Temperature-dependent variation of \(d_{33}\) . </center>  
+
+The NJU001 material exhibits \(C_{\infty \nu}\) symmetry and \(P_{5}\) in the direction of the director (i.e., longitudinal ferroelectricity), analogous to the behavior observed in RM734. Usually, the coefficient \(d_{31}\) can be measured by illuminating the sample with ordinary light (polarization perpendicular to the director). However, the accurate determination of this coefficient was difficult because of the high \(d_{33} / d_{31}\) ratio. As a consequence, the signal remained undetected due to the limited sensitivity of the photodetector (Laser Power and Energy Meter, SKU2256258, Coherent). Thus, we just characterized the \(d_{33}\) of the NJU001 material.  
+
+## Added a Section 2 in Revised Supplementary Information  
+
+## Added Text in Revised Manuscript:  
+
+Line 130- 132, page 4: "The nonlinear coefficient \(d_{33}\) is experimentally determined to be approximately 5.7 pm/V and 6.9 pm/V at \(61.0^{\circ}\mathrm{C}\) (Nx) and \(45.7^{\circ}\mathrm{C}\) (NF), respectively (Supplementary Section 2)."  
+
+## Added Text in Revised Manuscript:  
+
+Line 139- 140, page 4: "The thickness of the Y-cut and NJU001 sample was 2 mm and 5.2 \(\mu \mathrm{m}\) , respectively."  
+
+Q2.5 \\*Lines 249- 273: This paragraph is unnecessary and also partially wrong as it is written now. The authors derive two obvious equations. This could be fine, but they also invented two new names, "nonlinear Malus's law" and "polar polarizers". They further write as if this is some new concept: "Therefore, the nonlinear Malus's law provides a new concept of "polar polarizers"." But this is, of course, known for many decades. It is just an obvious property of nonlinear optical processes. Why invent new names and write that this is a new concept? They further write: "This relationship describes a new scenario in which the manipulation of light in the nonlinear optical regime is now vectorized, i.e., the vector field of the SH wave is head- to tail inequivalent (Fig. S13). The vector field of the SH wave is determined by the orientation of local dipoles in the range of [0, \(2\pi\) ]. In contrast, the traditional Malus's law disables the head- to- tail inequivalence of polarization." This is, of course, wrong. In the nonlinear regime, we still have head- to- tail equivalence of polarization. In other words, the vector field of the SH wave is head- to- tail equivalent. The only difference is the phase, which is a completely separate property from the polarization. Anyway, the authors do not mention phase in this paragraph. They say the relationship (line 262) is vectorized and head- to- tail inequivalent. Which is wrong.  
+
+## A2.5 Response:  
+
+We sincerely thank the reviewer for their valuable and insightful comment. Upon reconsideration, we fully agree that certain conceptual discussions in the original manuscript were excessive for the present study's scope. Accordingly, we have significantly revised this paragraph to focus on explaining the physical meaning of the derived equation, and we also added Supplementary section 5 and revised Fig. S14 to make it clearer.
+
+<--- Page Split --->
+
+Original Text in Manuscript:  
+
+Line 249- 283, page 10: "When a linearly- polarized light at an angle \(\phi\) is input and passes through a polarizer with its transmission axis at \(\theta \in (0, \pi)\) , the output electric field in the linear optical regime is expressed as \(\mathbf{E} = \left(\frac{E_x}{E_y}\right) = \mathbf{A}_0 \left(\cos^2 \theta \sin \theta \sin^2 \theta\right) \left(\cos \phi \sin \phi\right) = \mathbf{A}_0 \cos (\phi - \theta) \hat{\mathbf{e}}_\theta\) , where \(\hat{\mathbf{e}}_\theta\) represents the unit vector of \(\theta\) - axis. This teaches us the intensity of the output beam as \(I = \mathrm{A}_0^2 \cos^2 (\phi - \theta)\) , known as Malus's law. Systems with inversion symmetry breaking exhibit a non- zero second- order nonlinear coefficient, leading to additional second- order nonlinear optical responses like SHG. When an input polarized FW excites a nonlinear medium belonging to the point group \(C_{\infty \nu}\) , e.g., polar nematic phase composed of orientated dipoles, an SH nonlinear polarization arises as (Supplementary Section 3):  
+
+\[\begin{array}{rl} & {\mathbf{P}_{\alpha}^{2\omega} = 2\mathrm{A}_0^2\epsilon_0d_{33}\cos^2 (\phi -\alpha)\hat{\mathbf{e}}_\alpha}\\ & {\qquad = 2\mathrm{A}_0^2\epsilon_0d_{33}\cos (\phi -\alpha)\left(\cos^2\alpha \cos \alpha \sin \alpha \sin^2\alpha\right)\left(\cos \phi \right).} \end{array} \quad (1)\]  
+
+\(\mathrm{A}_0\) is the amplitude of FW, \(\omega\) the angular frequency, \(\phi\) the polarization angle relative to the \(x\) - axis, \(\epsilon_0\) the vacuum permittivity, \(\alpha\) the azimuthal angle of the dipole, and \(\hat{\mathbf{e}}_\alpha\) the unit vector along the \(\alpha\) direction. The intensity of the emitted SH wave is proportional to the quartic amplitude of the electric field of FW, \(I^{2\omega}(\alpha) \propto \mathrm{A}_0^4 \cos^4 (\phi - \alpha)\) , corresponding to an extended nonlinear version of Malus's law (Fig. 4a). This relationship describes a new scenario in which the manipulation of light in the nonlinear optical regime is now vectorized, i.e., the vector field of the SH wave is head- to- tail inequivalent (Fig. S13). The vector field of the SH wave is determined by the orientation of local dipoles in the range of \([0, 2\pi ]\) . In contrast, the traditional Malus's law disables the head- to- tail inequivalence of polarization. Therefore, the nonlinear Malus's law provides a new concept of "polar polarizers". This immediately generates interest in pixelating and engineering the so- called polar polarizers with arbitrary polarization angles to develop unprecedented nonlinear vectorial optical devices. However, as a longstanding unsolved problem, the traditional solid- state polar systems barely own the flexibility to realize spatially dependent dipole orientations. The polar nematics, such as the Nx state, obviously break this limitation, and their designability of polar fields offers us a revolutionary platform.  
+
+As a proof- of- concept demonstration, we show the manipulation of vectorial nonlinear photonic functionalities by leveraging the direct interaction between light and a minimalist single- layer nonlinear perfect vector beam (PVB) generator with in- plane toroidal polar topological superstructure (Fig. 4a). Fig. 4b shows the design of the fabricated phase profile based on the nonlinear Malus principle, which is superposed of a vortex phase and an axicon phase function. This design highlights the unique hierarchical feature of our device; it integrates both the continuous and discontinuous variations of dipole orientations, which is confirmed by the visible Maltese cross and the periodic circular domain walls (Fig. 4c and Fig. S14). The polar ordering is characterized by SHG- I imaging (Fig. S15)."  
+
+## Revised Text in Revised Manuscript:  
+
+Line 347- 393, page 12- 13: "When an input polarized FW excites a nonlinear medium belonging to the point group \(C_{\infty \nu}\)  
+
+(e.g., polar LCs composed of orientated dipoles), an nonlinear polarization field arises as (Supplementary Section 4):  
+
+\[\begin{array}{rl} & {\mathbf{P}_{\alpha}^{2\omega} = 2\mathrm{A}_0^2\epsilon_0d_{33}\cos^2 (\phi -\alpha)\hat{\mathbf{e}}_\alpha}\\ & {\qquad = 2\mathrm{A}_0^2\epsilon_0d_{33}\cos (\phi -\alpha)\left(\cos^2\alpha \cos \alpha \sin \alpha \sin^2\alpha\right)\left(\cos \phi \right).} \end{array} \quad (3)\]  
+
+\(\mathrm{A}_0\) is the amplitude of FW, \(\epsilon_0\) the vacuum permittivity, \(d_{33}\) the nonlinear coefficient, \(\omega\) the angular frequency, \(\phi\) the polarization angle relative to the \(x\) - axis, \(\alpha\) the azimuthal angle of the dipole, and \(\hat{\mathbf{e}}_\alpha\) the unit vector along the \(\alpha\) direction, \(\left(\frac{\cos\alpha}{\sin\alpha}\right)\) . The intensity is proportional to the quartic amplitude of the electric field of FW,  
+
+\[I^{2\omega}(\alpha) \propto \mathrm{A}_0^4 \cos^4 (\phi - \alpha) \quad (4)\]  
+
+These relationships bear a striking resemblance to the light propagation through a linear polarizer and the well- established Malus's law (Supplementary Section 5). Analogously, the oriented polar LCs function as a specialized "nonlinear polar
+
+<--- Page Split --->
+
+polarizer", characterized by a Jones matrix formalism [Jones vector: \(\cos (\phi - \alpha)\left(\frac{\cos^2\alpha}{\cos\alpha}\sin \alpha\right)\) ], where the transmission axis exhibits head- to- tail asymmetry. This configuration endows the system with two distinct capabilities: 1) the nonlinear optical response governed by the material's nonlinearity; 2) the selective excitation of instantaneous electric field vector, which follows the polar LC's orientation \(\alpha\) across the full angular range \([0, 2\pi ]\) (Fig. S14).  
+
+As a proof- of- concept demonstration, we demonstrate the capability of generating both linear and nonlinear perfect vector beams (PVBs) using a single- layer patterned polar LC device (Fig. 5a). As a new type of vector beam \(^{45}\) with cylindrical symmetry in polarization, PVB has sparked considerable interest because its radius and intensity profile are independent of the polarization topological charge \(l\) , demonstrating superior capabilities in optical manipulation, microscopy imaging, and laser micromachining. Although PVBs have been widely studied in the framework of linear optics \(^{46,47}\) , the direct generation of SH PVBs from a single nonlinear optical element has not been realized due to previous challenges in manipulating in- plane polar ordering. Here, we design the orientation distribution \(\alpha\) of polar LCs based on Eq. (3) and (4), taking into account the near- field polarization, phase, and intensity distributions. Fig. S15a depicts the designed orientation distribution of polar LCs, \(\alpha = \frac{1}{2} l\phi + \frac{1}{2}\arg \left[2\sum_{n = 1}^{\infty}a_{n}\sin \left(2\pi \frac{n}{r} r\right)\right]\) , \(r \geq 0\) . \(a_{n}\) represents the Fourier series coefficients, \(T = 100 \mu \mathrm{m}\) is the period of concentric rings, \(r\) and \(\phi\) denote the radial distance and azimuthal angle in the polar coordinate, and \(l\) is the polarization topological charge. In this case, the simulated near- field instantaneous electric field vector, when Fourier- transformed to the far field, enables an intensity profile and polarization distribution that meet the criteria of a nonlinear perfect vector beam. Accordingly, we show the surface anchoring pattern for both substrates of the device in Fig. 5a. The alignment direction continuously changes in the azimuthal angle, and the orientations in adjacent rings maintain a consistent perpendicular relationship in the radial direction. By employing photoalignment technology in conjunction with a self- developed digital micromirror device- based microlithography system (Methods) \(^{34,48}\) , we achieved high- quality in- plane polar LC domain engineering, accomplished through the effective utilization of surface anchoring energy to dictate the LC orientation. The resulting polar LC superstructure is characterized under a crossed- PLM, which highlights the unique hierarchical and topological features of our device, incorporating both continuous and discontinuous variations of dipole orientations. The visible Maltese cross and the periodic circular domain walls shown in Fig. 5b and Fig. S15b indicate that the LC director orientations are faithfully imprinted according to our design. A schematic of a cross- section of the LC cell is provided in Fig. S16. The polar ordering is further confirmed by SHG- I imaging (Fig. S17), with the local polar orientation distribution at the center shown in Fig. 5c.  
+
+## Original Text in Manuscript:  
+
+Line 298- 299, page 12: "Nonlinear Malus's law and vectorial nonlinear photonic functionality enabled by flexible tailoring of the polar ordering in the miniaturized PVB generation device."  
+
+Revised Text in Revised Manuscript:  
+
+Line 417- 419, page 14: "Schematic of LC- based nonlinear vectorial optics. This photonic functionality is enabled by flexible tailoring of the polar ordering in the miniaturized PVB generation device."  
+
+Q2.6 \*The section about generating perfect vector beams is extremely short (lines 275- 295), providing almost no details. All of the following topics should be discussed in more detail:  
+
+I. What are perfect vector beams? These are much less known to readers than, for example, vortex beams. So they must be described.  
+
+II. What are the structures in Fig4, and how were they made? Provide details on the surface anchoring patterns. Also polar director field must be shown, like it is shown in Fig. S17.  
+
+III. How were the two phases achieved (Fig4bi and ii). It is not obvious how the light's phase is modulated. Specifically, how is the spiral phase modulation achieved? With LCs this is usually achieved by illuminating a q-plate (e.g. a radial configuration) with a circularly polarized beam. But here, the input beam is linearly polarized. Also, to achieve a spiral phase, the left-right symmetry has to be broken. For example, in vortex generation by a q-plate, the symmetry is broken by the circularly polarized
+
+<--- Page Split --->
+
+input beam. But in this paper, as far as I can tell, the symmetry is not broken by the input beam or the LC structure. Anyway, why a spiral phase is needed anyway? This would be used for vortex beams, but vortex beams are not mentioned in the paper. As far as I know, the perfect vector beams do not need a spiral phase. The authors are confusing vector and vortex beams. They also use the term topological charge, but polarization topological charge is relevant for vector beams.  
+
+IV. Also, how was the axicon phase modulation achieved? Again this is not mentioned at all. The phase is alternating between 0 and pi/2. However, the phase of SHG can only be either 0 or pi (Fig S13b).  
+
+V. It is not clear at all what the electric field does, how it is applied, how the director changes, and how the output exactly changes.  
+
+## A2.6 Response:  
+
+We are deeply grateful to the reviewer for his/her valuable and constructive comments, which have significantly helped improve the quality of our manuscript.  
+
+I: We agree with the reviewer and we have added descriptions about the perfect vector beams and related references.  
+
+## Added Text in Revised Manuscript:  
+
+Line 365- 372, page 12: "As a proof- of- concept demonstration, we demonstrate the capability of generating both linear and nonlinear perfect vector beams (PVBs) using a single- layer patterned polar LC device (Fig. 5a). As a new type of vector beam \(^{45}\) with cylindrical symmetry in polarization, PVB has sparked considerable interest because its radius and intensity profile are independent of the polarization topological charge \(I\) , demonstrating superior capabilities in optical manipulation, microscopy imaging, and laser micromachining. Although PVBs have been widely studied in the framework of linear optics \(^{46,47}\) , the direct generation of SH PVBs from a single nonlinear optical element has not been realized due to previous challenges in manipulating in- plane polar ordering."  
+
+II: We sincerely apologize for not providing sufficient details about the structure and fabrication in our original manuscript. We have revised this part accordingly.  
+
+## Original Text in Manuscript:  
+
+Line 278- 283, page 10: "Fig. 4b shows the design of the fabricated phase profile based on the nonlinear Malus principle, which is superposed of a vortex phase and an axicon phase function. This design highlights the unique hierarchical feature of our device; it integrates both the continuous and discontinuous variations of dipole orientations, which is confirmed by the visible Maltese cross and the periodic circular domain walls (Fig. 4c and Fig. S14). The polar ordering is characterized by SHG- I imaging (Fig. S15)."  
+
+## Revised Text in Revised Manuscript:  
+
+Line 380- 393, page 12- 13: "Accordingly, we show the surface anchoring pattern for both substrates of the device in Fig. 5a. The alignment direction continuously changes in the azimuthal direction, and the orientations in adjacent rings maintain a consistent perpendicular relationship in the radial direction. By employing photoalignment technology in conjunction with a self- developed digital micromirror device- based microlithography system (Methods) \(^{34,48}\) , we achieved high- quality in- plane polar LC domain engineering, accomplished through the effective utilization of surface anchoring energy to dictate the LC orientation. The resulting polar LC superstructure is characterized under a crossed- PLM, which highlights the unique hierarchical and topological features of our device, incorporating both continuous and discontinuous variations of dipole orientations. The visible Maltese cross and the periodic circular domain walls shown in Fig. 5b and Fig. S15b indicate that the LC director orientations are faithfully imprinted according to our design. A schematic of a cross- section of the LC cell is provided in Fig. S16. The polar ordering is further confirmed by SHG- I imaging (Fig. S17), with the local polar orientation distribution at the center shown in Fig. 5c."  
+
+III and IV: We sincerely apologize for any confusion caused by the brevity of our previous writing. We recognize that our initial presentation was insufficiently detailed and partially inaccurate, and we appreciate the reviewer's patience in this matter. The two subfigures illustrated in Fig. 4b- i and ii (original version) are not used to independently generate a vortex beam and implement axicon focusing, respectively. Instead, these two are combined via linear superposition to create a composite
+
+<--- Page Split --->
+
+configuration to obtain the designed optical axis distribution of the toroidal polar topological superstructure (Fig. 4b- iii, original version), thus for producing PVBs. As evident from Fig. 4b- iii, the alignment directions in adjacent rings exhibit a consistent perpendicular relationship in the radial direction. This spatial arrangement corresponds to “the phase alternating between 0 and \(\pi /2\) ” as the reviewer said. This is from the structure aspect regarding the orientation angle distribution.  
+
+However, it is important to distinguish this from the SHG phase characteristic shown in Fig. S14f. Here, the phase refers specifically to phase distribution of the near- field vector light field, which enforces a binary phase constraint - the phase can only assume discrete values of either 0 or \(\pi\) for the resulting linearly polarized light. This is from the optics aspect regarding the optical phase of the near- field polarization distribution.  
+
+To enhance clarity, we have revised our presentation by: 1) directly providing the mathematical formulation of the final \(\alpha ; 2\) showing the illustration of target polar LC orientation distribution (Fig. S15a); 3) presenting the corresponding surface anchoring pattern in revised Fig. 5a, which is implemented through photopatterning to control the LC director orientation. Accordingly, we have now provided a more comprehensive explanation to address this issue. Also, we have changed the term topological charge to polarization topological charge in the revised manuscript.  
+
+## Original Text in Manuscript:  
+
+Line 278- 279, page 10: “Fig. 4b shows the design of the fabricated phase profile based on the nonlinear Malus principle, which is superposed of a vortex phase and an axicon phase function.”  
+
+## Revised Text in Revised Manuscript:  
+
+Line 372- 380, page 12: “Here, we design the orientation distribution \(\alpha\) of polar LCs based on Eq. (3) and (4), taking into account the near- field polarization, phase, and intensity distributions. Fig. S15a depicts the designed orientation distribution of polar LCs, \(\alpha = \frac{1}{2} l\phi +\frac{1}{2}\arg \left[2\sum_{n = 1}^{n}a_{n}\sin \left(2\pi \frac{n}{r}\tau\right)\right], r \geq 0\) . \(a_{n}\) represents the Fourier series coefficients, \(T = 100 \mu \mathrm{m}\) is the period of concentric rings, \(r\) and \(\phi\) denote the radial distance and azimuthal angle in the polar coordinate, and \(l\) is the polarization topological charge. In this case, the simulated near- field instantaneous electric field vector, when Fourier- transformed to the far field, enables an intensity profile and polarization distribution that meet the criteria of a nonlinear perfect vector beam.”  
+
+## Added Text in Revised Manuscript:  
+
+Line 367- 370, page 12: “PVB has sparked considerable interest because its radius and intensity profile are independent of the polarization topological charge \(l\) , demonstrating superior capabilities in optical manipulation, microscopy imaging, and laser micromachining.”  
+
+## Added Text in Revised Manuscript:  
+
+Line 377, page 12: “\(l\) is the polarization topological charge.”  
+
+## Original Text in Manuscript:  
+
+Line 284, page 10: “topological charge \(|l| = 2\) .”  
+
+Revised Text in Revised Manuscript:  
+
+Line 395- 396, page 13: “polarization topological charge \(|l| = 2\) .”  
+
+## Original Text in Manuscript:  
+
+Line 290- 291, page 11: “characterized by the invariant radius and intensity profiles under changing topological charges.”  
+
+Revised Text in Revised Manuscript:  
+
+Line 406, page 13: “characterized by the invariant radius and intensity profiles under changing polarization topological charges.”  
+
+## Original Text in Manuscript:  
+
+Line 308, page 13: “Dependences of the ring diameter on the topological charge \((|l| = 1 \sim 6)\) .”  
+
+Revised Text in Revised Manuscript:  
+
+Line 427- 428, page 14- 15: “Dependences of the ring diameter on the polarization topological charge \((|l| = 1 \sim 6)\) .”
+
+<--- Page Split --->
+
+Newly added Fig. S15a:  
+
+![](images/Figure_unknown_3.jpg)
+
+<center>Fig. S15a. Target orientation distribution \(\alpha\) of polar LCs. </center>
+
+<--- Page Split --->
+![](images/Figure_unknown_4.jpg)
+
+
+<--- Page Split --->
+![PLACEHOLDER_18_0]
+
+<center>Revised Fig. 5: </center>  
+
+V: Due to space limitations in the main text, the electrical switching characteristics of our nonlinear optical LC device are presented in Supplementary Section 7 (revised version). But to enhance reader comprehension, we have added a brief paragraph that provides additional context about this particular aspect of our work.  
+
+## Original Text in Manuscript:  
+
+Line 285- 287, page 10- 11: "an SH vector beam with a ring- shaped intensity profile is also obtained using electric tunability (Fig. 4e,h-j; Supplementary Section 5)."  
+
+## Revised Text in Revised Manuscript:  
+
+Line 410- 415, page 13: "Furthermore, by leveraging the dynamic tunability of LCs, we demonstrate active and reversible switching of SH-PVBs under an ultra- low electric field. When applying a triangular electric field (peak- to- peak voltage \(\mathrm{V_{pp}} = 0.12 \mathrm{V / \mu m}\) , frequency \(f = 0.5 \mathrm{Hz}\) ) across the device, the topological polar LC superstructure exhibits periodic switching behavior (Fig. 5h and 5i). This results in a dynamically tunable nonlinear vector optical field with periodic on- off features (Fig. 5j). More details are provided in Supplementary Section 7."  
+
+Q2.7 \*In connection to III., in fig4 it is not clear what the LC structure is, the same is also not clear from the beginning of the
+
+<--- Page Split --->
+
+paper. Photoalignment is only mentioned once in the introduction, but is not clear where and how was it used for the rest of the results. For example, are the patterns in fig2 self-formed or due to photoalignment?  
+
+## A2.7 Response:  
+
+We have gone through the whole manuscript and supplement the details about the structures and the fabrications. As stated in Methods (Line 490- 491, Page 17) that if not specifically denoted, the tests in our work employed LC cells that were photoaligned. We have now included a list for all photoalignment patterns as below, which is added in our revised Supplementary Information. The patterns in Fig. 2 are self- formed under homogeneous photoalignment.  
+
+![PLACEHOLDER_19_0]
+
+<center>Fig. S21. Photoalignment conditions for the LC structures in our manuscript </center>  
+
+Added Text in Revised Manuscript:  
+
+Line 482, page 16: "We summarize the alignment pattern in Fig. S21."  
+
+Original Text in Manuscript:  
+
+Line 233, page 8: "By encoding the brightness of the artwork into the spatially- variant polar field."  
+
+Revised Text in Manuscript:  
+
+Line 330, page 10: "By encoding the brightness of the artwork (from 0 to 255) into the spatially- variant polar field."  
+
+Added Text in Manuscript:  
+
+Line 341- 342, page 11: "The grayscale values from 0 to 255 are proportionally mapped to the distribution of \(\alpha\) ."  
+
+Q2.8 \*Not clear also what is the surface anchoring on the other non- patterned surface or are both surfaces patterned? Some cross- sections of the LC cells should help.  
+
+## A2.8 Response:  
+
+The two substrates are both surfaces photopatterned and the patterns are the same (revised Fig. 5a, added description of "photoalignment for both substrates"). We have added a schematic of a cross- sections of the LC cell in Fig. S16 for
+
+<--- Page Split --->
+
+clarification.  
+
+![PLACEHOLDER_20_0]
+
+<center>Fig. S16. A schematic of a cross-sections of the LC cell. </center>  
+
+Added Text in Revised Manuscript:  
+
+Line 481- 482, page 16: "The SD1 alignment layers on both substrates were simultaneously photoaligned."  
+
+Line 391, page 13: "A schematic of a cross-sections of the LC cell is provided in Fig. S16."  
+
+Line 419- 420, page 14: "The photoalignment patterns for the two substrates of the LC device are the same shown in the right panel."  
+
+Q2.9 \*Another thing that is not clear to me is how they were able to generate a perfect circle as drawn in Fig4a and seen in 4e. No light should be generated when the director is perpendicular to the incoming polarization. Therefore there should be no SH light perpendicular to the incoming polarization. For this reason, all the rings should have gaps. From some images, this is evident; from some, others is not. In any case, Fig4a is wrong. Also, since there are intensity gaps in the ring, then polarization topological charge cannot be defined. In other words, we cannot define how many times the polarization rotates when going around one full circle since, in some positions, the phase is not defined due to the intensity being zero.  
+
+## A2.9 Response:  
+
+We apologize for the lack of clarity in our previous manuscript. The reviewer is right in stating that no light should be emitted when the director is perpendicular to the incoming polarization. However, it is important to note that the light patterns in revised Fig. 5e correspond to a specific far- field nonlinear PVB with the polarization topological charge of \(l = 2\) as well as its polarization characterization. The nonlinear PVB itself does not have gaps (the first panel in revised Fig. 5e). The SH light fields in the second and third panels in revised Fig. 5e are analyzed by an analyzer (polarizer before the image plane). Extinction regions (i.e., gaps) occur where the polarization is perpendicular to the analyzer. Thus, by analyzing these extinction locations, the nonlinear PVB with a specific space- variant polarization distribution is confirmed, which is provided in the upper right corner of the first panel of revised Fig. 5e.  
+
+Another thing should be noted that all the rings of the polar LC superstructure would generate ring- shaped SH signals in the near- field with gaps. Each ring has two gaps that are aligned in a line. However, the lines' orientations for odd and even rings are perpendicular to each other due to the designed polar LC superstructure, where the orientations of the LC dipoles within adjacent rings always differs by pi/2. Then, the light pattern observed at the far- field focal point, after Fourier transformation, displays a perfect circular shape, as shown in revised Fig. 5a.  
+
+To make it clearer, we have revised the corresponding descriptions accordingly.  
+
+## Added Text in Revised Manuscript:  
+
+Line 399- 404, page 13: "Extinction occurs when the polarization is perpendicular to the analyzer. Therefore, by examining
+
+<--- Page Split --->
+
+the locations of extinction, the nonlinear PVB with a specific space- variant polarization distribution is confirmed, as shown in the upper right corner of the first panel of Fig. 5e. The polarization topological charge is determined to be \(l = 2\) . These results align well with the expected intensities and polarization distributions."  
+
+## Added Text in Revised Manuscript:  
+
+Line 426- 427, page 14: "The polarization distributions of the PVBs are provided in the upper right corners of the first panels of (d) and (e)."  
+
+## Original Text in Manuscript:  
+
+Line 285- 286, page 10- 11: "Simultaneously, an SH vector beam with a ring- shaped intensity profile is also obtained using electric tunability"  
+
+## Revised Text in Manuscript:  
+
+Line 398, page 13: "Simultaneously, an SH vector beam with a ring- shaped intensity profile is also obtained (the first panel in Fig. 5e)."  
+
+Q2.10 \*The structure of the paper is quite strange. The authors first talk about LC materials, then structures in LCs, then optics, and the final section is named "discussion." But the discussion does not discuss the results, but it is a section by itself, discussing some new things but not the optics at all. Therefore, there is no actual discussion/conclusion. The "discussion" chapter would fit before the optics part. In general, the paper also feels like two or even three disconnected parts.  
+
+## A2.10 Response:  
+
+We are sincerely grateful for the reviewer's valuable advice. We have restructured our manuscript to follow a logical flow, from materials to structures and finally to optics. The former "Discussion" section (about polar ordering simulations) has been relocated prior to the optics part. Additionally, we have added a new paragraph in the revised "discussion and conclusion" section to succinctly summarize our findings and the significance.  
+
+## Revised "discussion and conclusion" section  
+
+Line 436- 460, page 15: "Systems that exhibit phase competition, order parameter coexistence, and emergent order parameter topologies constitute a crucial part of modern condensed- matter physics<sup>50</sup>. In this work, we establish a key link between paraelectric (apolar) and ferroelectric (polar) states, based on a distinct liquid of ferroelectrics with nematicity, which acts as an intermediate state between the recently identified ferroelectric nematic and traditional apolar nematic phases. A new structural model is established with our extended Oseen- Frank free- energy functional, which elucidates the formation of the unique unipolar orderings in this mesophase. Unlike conventional ferroelectric materials, the Nx phase exhibits an exceptional capability of in- plane polar domain engineering, enabling the creation of complex hierarchical architectures with customizable, precise, and defect- free LC ordering on a large scale. This may provide an exciting platform for developing new ferroelectric materials with tailored properties and for exploring emergent phenomena in condensed- matter physics and nonlinear photonics.  
+
+As photonics technology advances from scalar to vectorial regimes<sup>30,51- 54</sup>, the construction of nonlinear vectorial light fields has become increasingly essential for applications including super- resolution imaging<sup>55</sup>, high- capacity optical communications<sup>51</sup>, and high- precision laser micromachining<sup>56</sup>. Note that creating such vectorial nonlinear modes would require several passes through a spatial light modulator (lossy) or mixing beams with interferometry (difficult to align), demanding complex optical systems with cascaded linear and nonlinear light field manipulation. Here, the generation of SH PVBs is achieved directly from a single, miniaturized polar LC device under a scalar Gaussian FW incidence, thanks to the flexible tailoring of tunable polar topologies. This is a feat that has not yet been accomplished with other ferroelectric materials. Such intriguing light- matter interaction could lead to the next generation of portable, scalable, and reconfigurable photonic technologies with wide- ranging applications, e.g., all- optical communications, quantum computing, soft intelligent robotics, and the biomedical industry."
+
+<--- Page Split --->
+
+## Re: manuscript NCOMMS-25-04493A  
+
+We sincerely thank the respected reviewers for their valuable time and helpful comments. We have carefully revised our manuscript and Supplementary Information by taking the respected reviewers' comments as appropriate into account.  
+
+Our point- by- point response to the reviewers' comments and the changes made in our revised manuscript with yellow color highlighted track change as follows.  
+
+## Reviewer #1  
+
+Q1.1 \*One typographical error needs to be corrected: Line 507: "polar nad orientational orders" instead of: "polar and orientational orders".  
+
+## A1.1 Response:  
+
+We apologize for the error. We have corrected it accordingly.  
+
+Original Text in Manuscript:  
+
+Line 507, page 16: "... polar nad orientational orders ..."  
+
+Revised Text in Revised Manuscript:  
+
+Line 516, page 16: "... polar and orientational orders ..."  
+
+## Reviewer #2  
+
+Q2.1 \*The novelty compared to their previous work is still not appropriately addressed. The authors should state this more clearly.  
+
+We apologize that, while we explained the core differences between our previous work and the present one in our last reply to the reviewer (as repeatedly explained below), we did not make enough revisions in our manuscript. In this revision, we made revisions to make readers understand these points well.  
+
+## The core differences between our previous work and the present one  
+
+## (1) New polar orders:  
+
+The previous work (Nat. Commun.10.1038/s41467- 024- 53040- 8) primarily investigated the ferroelectric nematic phase, which exhibits spontaneous polarization and robust ferroelectricity. In contrast, this work explores an intermediate mesophase between the ferroelectric nematic phase and the conventional nematic phase. This mesophase demonstrates unipolar and bipolar polar orderings, with its physical properties (e.g., polarization magnitude, orientational order) intermediate between ferroelectric nematic and nematic phases. A new structural model is established that elucidates the formation of the unique unipolar orderings in this mesophase—a key advance beyond the scope of our prior work.  
+
+## (2) Domain engineering capability:  
+
+The Nx phase exhibits greater potential than the NF phase for in- plane domain engineering of nonlinear optical elements. Although prior study (Nat. Commun.10.1038/s41467- 024- 53040- 8) have demonstrated that NF ordering can generate pure splay orientational patterns with good quality, it remains fundamentally limited in achieving complex hierarchical architectures with customizable deformations (e.g., intricate bending geometries), precise alignment, and defect- free polar ordering at scale. This limitation is clearly demonstrated in Fig. 5, where photopatterning of a concentric topological polar
+
+<--- Page Split --->
+
+LC superstructure in the NF phase results in extensive defects and domain walls (Fig. 5b_iv and Fig. S15). Conversely, the Nx phase enables defect- free hierarchical topological superstructures (Fig. 5b_ii and Fig. S15), thus yielding high- performance nonlinear perfect vector beams. Such complex hierarchical superstructures surpass the azimuthal complexity of polar orientational field designs like \(q\) - plate in our previous work (Nat. Commun.10.1038/s41467- 024- 53040- 8) realized in the NF phase.  
+
+## (3) From scalar vortices to vectorial beams:  
+
+Our prior work demonstrates the parallel generations of scalar nonlinear optical vortices with nonlinear geometric phase encoded NF \(q\) - plate structure. On the other hand, this manuscript reports the generation of vectorial nonlinear perfect vortex beams. This distinction expands the toolkit for nonlinear vectorial optics, leveraging the mesophase's flexible polar domain engineering.  
+
+## Revised Manuscript  
+
+## Original Text in Manuscript:  
+
+Line 436- 447, page 15: "Systems that exhibit phase competition, order parameter coexistence, and emergent order parameter topologies constitute a crucial part of modern condensed- matter physics<sup>50</sup>. In this work, we establish a key link between paraelectric (apolar) and ferroelectric (polar) states, based on a distinct liquid of ferroelectrics with nematicity, which acts as an intermediate state between the recently identified ferroelectric nematic and traditional apolar nematic phases. A new structural model is established with our extended Oseen- Frank free- energy functional, which elucidates the formation of the unique unipolar orderings in this mesophase. Unlike conventional ferroelectric materials, the Nx phase exhibits an exceptional capability of in- plane polar domain engineering, enabling the creation of complex hierarchical architectures with customizable, precise, and defect- free LC ordering on a large scale. This may provide an exciting platform for developing new ferroelectric materials with tailored properties and for exploring emergent phenomena in condensed- matter physics and nonlinear photonics."  
+
+## Revised Text in Revised Manuscript:  
+
+Line 433- 456, page 15: "Systems that exhibit phase competition, order parameter coexistence, and emergent order parameter topologies constitute a crucial part of modern condensed- matter physics<sup>50</sup>. In this work, we establish a key link between paraelectric (apolar) and ferroelectric (polar) states, based on a distinct liquid of ferroelectrics with nematicity, which acts as an intermediate state between the recently identified ferroelectric nematic and traditional apolar nematic phases. A new structural model is established with our extended Oseen- Frank free- energy functional, which elucidates the formation of the unique unipolar and bipolar orderings in this mesophase. Importantly, the discovered polar orders make a sharp contrast to the existing NF order in the structure, and demonstrate a better potential in domain engineering for developing nonlinear optical elements. For the former, this study reveals a novel energy competition scenario among the Landau energies (\(\Delta F_{\mathrm{L},\mathrm{A}}\) and \(\Delta F_{\mathrm{L},\mathrm{P}}\) for apolar and polar Landau energy penalties) and flexoelectricity (\(\Delta F_{\mathrm{flex}}\)) with decoupled \(S\) and \(S_{\mathrm{P}}\) . It means that carefully adjusting multiple energy landscapes with either coupling or decoupling polar and LC nematic orders would trigger a broad range of unknown polar orders. For the latter, the Nx phase exhibits an exceptional capability of in- plane polar domain engineering. Though preceding works have shown that NF order can be employed to generate pure splay orientational patterns with good quality<sup>34,36,51</sup>, complex hierarchical architectures with customizable (e.g., with more complicated deformations with bending), precise, and defect- free polar ordering on a large scale is difficult (see defect- mediated NF structure in Fig. 5b (iv), Fig. S11, and Fig. S15). The Nx order has shown the capability to overcome these drawbacks, enabling hierarchical topological superstructures that surpass the azimuthal complexity of polar orientational field designs like prior \(q\) - plate<sup>34</sup> and periodic splay patterns<sup>36,51</sup>, yielding high- performance nonlinear perfect vector beams — a capability neither previously conceptualized nor realized in the NF phase. This may provide an exciting platform for developing new ferroelectric materials with tailored properties and for exploring emergent phenomena in condensed- matter physics and nonlinear photonics."  
+
+Q2.2 \*The authors removed the earlier assertive framing about “Nonlinear Malus’s law” and “polar polarizers”, rewrote the
+
+<--- Page Split --->
+
+section, and dropped the headline claim—but they still describe the medium analogously as a “specialized ‘nonlinear polar polarizer’,” and state the transmission axis shows head- to- tail asymmetry via a Jones- style description. So the term specialized “nonlinear polar polarizer” should be removed. The head- to- tail asymmetry should be removed because it is incorrect.  
+
+## A2.2 Response:  
+
+We have remove both phrases "specialized ‘nonlinear polar polarizer' and "head- to- tail asymmetry" accordingly to eliminate any possible ambiguity.  
+
+## Our Text in Manuscript:  
+
+Line 357- 360, page 12: "Analogously, the oriented polar LCs function as a specialized ‘nonlinear polar polarizer”, characterized by a Jones matrix formalism [Jones vector: \(\cos (\phi - \alpha)\left(\begin{array}{cc}{\cos^{2}\alpha}&{\cos\alpha\sin\alpha}\\{\cos\alpha\sin\alpha}&{\sin^{2}\alpha}\end{array}\right)]\) , where the transmission axis exhibits head- to- tail asymmetry. This configuration endows the system with two distinct capabilities..."  
+
+## Revised Text in Revised Manuscript:  
+
+Line 356- 358, page 12: "The oriented polar LCs is characterized by a Jones matrix formalism: \(\cos (\phi - \alpha)\left(\begin{array}{cc}{\cos^{2}\alpha}&{\cos\alpha\sin\alpha}\\{\cos\alpha\sin\alpha}&{\sin^{2}\alpha}\end{array}\right)\) , endowing the system with two distinct capabilities..."
+
+<--- Page Split --->

peer_reviews/supplementary_0_Transparent Peer Review file__e9f305359274f364160106c26e3b423ede3cbec231bbbc95ee32f2afdd3ba013/supplementary_0_Transparent Peer Review file__e9f305359274f364160106c26e3b423ede3cbec231bbbc95ee32f2afdd3ba013_det.mmd

@@ -0,0 +1,988 @@
+<|ref|>title<|/ref|><|det|>[[73, 53, 295, 80]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[75, 96, 296, 118]]<|/det|>
+Peer Review File  
+
+<|ref|>title<|/ref|><|det|>[[73, 161, 833, 236]]<|/det|>
+# Periodically-modulated unipolar and bipolar orders in nematic fluids towards miniaturized nonlinear vectorial optics  
+
+<|ref|>text<|/ref|><|det|>[[73, 249, 468, 266]]<|/det|>
+Corresponding Author: Professor Yan-qing Lu  
+
+<|ref|>text<|/ref|><|det|>[[72, 298, 864, 313]]<|/det|>
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+<|ref|>text<|/ref|><|det|>[[73, 351, 144, 365]]<|/det|>
+Version 0:  
+
+<|ref|>text<|/ref|><|det|>[[73, 377, 219, 391]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[73, 403, 160, 416]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[73, 430, 238, 443]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 443, 914, 470]]<|/det|>
+Review: Periodically-modulated unipolar and bipolar orders in nematic fluids: towards 1 miniaturized microscopic nonlinear photonics  
+
+<|ref|>text<|/ref|><|det|>[[72, 470, 912, 640]]<|/det|>
+The article presents the experimental results related to a new nematic liquid crystal mesophase, designated as the Nx phase, which was observed to exist between the nematic and ferroelectric nematic liquid crystal phases. The distinctive characteristics of this mesophase manifest as unipolar and bipolar orders. The authors employed these properties in the fabrication of three- dimensional polar topological structures. In the presented work, the authors demonstrate a strong understanding of liquid crystal photonics, as evidenced by their description of the preparation of a liquid crystal mixture, its properties, and the presentation of detailed information about the experimental setup. The experimental verification of the observed Nx mesophase included differential scanning calorimetry and second harmonic generation. The results obtained from these methods are detailed in the following sections. All results present a comprehensive study and are convincing. The authors successfully prepared multidomain NLC structures characterized by micron- size domain resolution. The article's structure and organization of ideas are appropriate, with a clear introduction, methodology, results, and discussion. The references are carefully chosen and are appropriate to the scope of the article. The article effectively highlights advances in the generation of topological states of polar matter and demonstrates the potential for microscopic nonlinear optical microdevices.  
+
+<|ref|>text<|/ref|><|det|>[[73, 639, 607, 653]]<|/det|>
+I would recommend this manuscript for publication in Nature Communications.  
+
+<|ref|>text<|/ref|><|det|>[[72, 664, 885, 690]]<|/det|>
+It would be beneficial to clarify the authors' precise intentions in writing about a "meter- scale vectorial beam generator." Line 107: "the nonlinear vectorial beam generator from meter to sub- millimeter scale".  
+
+<|ref|>text<|/ref|><|det|>[[72, 702, 590, 716]]<|/det|>
+It is probable that a typographical error has been made in the following text:  
+
+<|ref|>text<|/ref|><|det|>[[70, 716, 907, 742]]<|/det|>
+Line 448: "N,N'- Bis(2,5- di- tert- butylphenyl)- 3,3,9,10- 448 perylenedicarboximide" instead of: "Fluorescent dye N,N'- Bis(2,5- di- tert- butylphenyl)- 3,4,9,10- 448 perylenedicarboximide"  
+
+<|ref|>sub_title<|/ref|><|det|>[[73, 768, 161, 781]]<|/det|>
+## Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[73, 794, 238, 807]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 807, 923, 885]]<|/det|>
+The authors describe a novel device based on LCs that directly generates perfect vector beams in the second harmonic. The paper is novel and the devices demonstrated have potential. However, the results are, in many places, very poorly described. Quite a few things are unclear to me, so I cannot assess whether they are wrong or only poorly written. The paper contains many fancy words but fails to convey its meaning in simple terms. This is already evident in the abstract and continues throughout the text. There are also no conclusions to summarize the findings of the paper. In this view also the novelty is not clear. I believe the paper can be published, but it should be significantly rewritten.  
+
+<|ref|>text<|/ref|><|det|>[[72, 897, 916, 938]]<|/det|>
+In the introduction, there are many non- optics related references, which are not well related to the paper's topic. Conversely, the generation of optical vortices and vector beams within LCs is insufficiently referenced. There are many papers about generating various beams via transmission through the structure, laser light emission, and SHG (10.1038/s41467- 024-
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[70, 46, 911, 75]]<|/det|>
+53040- 8). While the last paper (by the same authors) is cited, it is not discussed in enough detail, especially since the work of this paper is very similar to the former one. The novelty is, therefore, not obvious.  
+
+<|ref|>text<|/ref|><|det|>[[72, 85, 916, 180]]<|/det|>
+The authors give a factor of \(10^{\wedge} - 6\) in volume reduction but do not explain how they got this number. Anyway, I think it is not useful to talk about volume. Specifically, the lateral dimensions of an optical device are designed for the desired size of an optical beam. In many instances, it is not desirable to make a device small in lateral size; on the opposite, to be useful, it should be large enough. But it is useful to make them thinner. Although, in the case of this paper, the device is very thin, it is also very inefficient, compared, for example, to solid- state phase- matched crystals. So, there is no point in comparing their sizes. I urge the authors to delete this number and rather just emphasize that a single device (thin layer) can do the same task as a few regular optical components.  
+
+<|ref|>text<|/ref|><|det|>[[72, 190, 916, 244]]<|/det|>
+In Fig 1a- c there are some checkmarks, X and some crossed checkmarks. It is not clear what it means. Also, RM734 and DIO have an X for SHG activity in N and Nx phases, but for NJU001, from these graphics, someone could understand that it has SHG activity across all phases. Also, what does "promising platform" mean? I suggest to remove all of these descriptions from the fig 1.  
+
+<|ref|>text<|/ref|><|det|>[[72, 254, 911, 283]]<|/det|>
+"SHG efficiency is defined as the SHG intensity ratio of the NJU001 material to that of a Y- cut quartz plate." Is this meant for the same thickness in both cases? Giving an absolute number for the nonlinear coefficient in pm/V would be nice.  
+
+<|ref|>text<|/ref|><|det|>[[72, 293, 923, 450]]<|/det|>
+Lines 249- 273: This paragraph is unnecessary and also partially wrong as it is written now. The authors derive two obvious equations. This could be fine, but they also invented two new names, "nonlinear Malus's law" and "polar polarizers". They further write as if this is some new concept: "Therefore, the nonlinear Malus's law provides a new concept of "polar polarizers". But this is, of course, known for many decades. It is just an obvious property of nonlinear optical processes. Why invent new names and write that this is a new concept? They further write: "This relationship describes a new scenario in which the manipulation of light in the nonlinear optical regime is now vectorized, i.e., the vector field of the SH wave is head- to- tail inequivalent (Fig. S13). The vector field of the SH wave is determined by the orientation of local dipoles in the range of [0, 2π]. In contrast, the traditional Malus's law disables the head- to- tail inequivalence of polarization." This is, of course, wrong. In the nonlinear regime, we still have head- to- tail equivalence of polarization. In other words, the vector field of the SH wave is head- to- tail equivalent. The only difference is the phase, which is a completely separate property from the polarization. Anyway, the authors do not mention phase in this paragraph. They say the relationship (line 262) is vectorized and head- to- tail inequivalent. Which is wrong.  
+
+<|ref|>text<|/ref|><|det|>[[72, 461, 911, 490]]<|/det|>
+The section about generating perfect vector beams is extremely short (lines 275- 295), providing almost no details. All of the following topics should be discussed in more detail:  
+
+<|ref|>text<|/ref|><|det|>[[72, 489, 911, 515]]<|/det|>
+I. What are perfect vector beams? These are much less known to readers than, for example, vortex beams. So they must be described.  
+
+<|ref|>text<|/ref|><|det|>[[72, 515, 911, 542]]<|/det|>
+II. What are the structures in Fig4, and how were they made? Provide details on the surface anchoring patterns. Also polar director field must be shown, like it is shown in Fig. S17.  
+
+<|ref|>text<|/ref|><|det|>[[72, 541, 920, 660]]<|/det|>
+III. How were the two phases achieved (Fig4bi and ii). It is not obvious how the light's phase is modulated. Specifically, how is the spiral phase modulation achieved? With LCs this is usually achieved by illuminating a q-plate (e.g. a radial configuration) with a circularly polarized beam. But here, the input beam is linearly polarized. Also, to achieve a spiral phase, the left-right symmetry has to be broken. For example, in vortex generation by a q-plate, the symmetry is broken by the circularly polarized input beam. But in this paper, as far as I can tell, the symmetry is not broken by the input beam or the LC structure. Anyway, why a spiral phase is needed anyway? This would be used for vortex beams, but vortex beams are not mentioned in the paper. As far as I know, the perfect vector beams do not need a spiral phase. The authors are confusing vector and vortex beams. They also use the term topological charge, but polarization topological charge is relevant for vector beams.  
+
+<|ref|>text<|/ref|><|det|>[[72, 660, 891, 688]]<|/det|>
+IV. Also, how was the axicon phase modulation achieved? Again this is not mentioned at all. The phase is alternating between 0 and pi/2. However, the phase of SHG can only be either 0 or pi (Fig S13b).  
+
+<|ref|>text<|/ref|><|det|>[[72, 688, 896, 714]]<|/det|>
+V. It is not clear at all what the electric field does, how it is applied, how the director changes, and how the output exactly changes.  
+
+<|ref|>text<|/ref|><|det|>[[72, 725, 911, 764]]<|/det|>
+In connection to III., in fig4 it is not clear what the LC structure is, the same is also not clear from the beginning of the paper. Photoalignment is only mentioned once in the introduction, but is not clear where and how was it used for the rest of the results. For example, are the patterns in fig2 self- formed or due to photoalignment?  
+
+<|ref|>text<|/ref|><|det|>[[72, 775, 920, 803]]<|/det|>
+Not clear also what is the surface anchoring on the other non- patterned surface or are both surfaces patterned? Some cross- sections of the LC cells should help.  
+
+<|ref|>text<|/ref|><|det|>[[72, 814, 920, 894]]<|/det|>
+Another thing that is not clear to me is how they were able to generate a perfect circle as drawn in Fig4a and seen in 4e. No light should be generated when the director is perpendicular to the incoming polarization. Therefore there should be no SH light perpendicular to the incoming polarization. For this reason, all the rings should have gaps. From some images, this is evident; from some, others is not. In any case, Fig4a is wrong. Also, since there are intensity gaps in the ring, then polarization topological charge cannot be defined. In other words, we cannot define how many times the polarization rotates when going around one full circle since, in some positions, the phase is not defined due to the intensity being zero.  
+
+<|ref|>text<|/ref|><|det|>[[72, 905, 923, 946]]<|/det|>
+The structure of the paper is quite strange. The authors first talk about LC materials, then structures in LCs, then optics, and the final section is named "discussion." But the discussion does not discuss the results, but it is a section by itself, discussing some new things but not the optics at all. Therefore, there is no actual discussion/conclusion. The "discussion" chapter
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 47, 785, 61]]<|/det|>
+would fit before the optics part. In general, the paper also feels like two or even three disconnected parts.  
+
+<|ref|>text<|/ref|><|det|>[[73, 99, 145, 112]]<|/det|>
+Version 1:  
+
+<|ref|>text<|/ref|><|det|>[[73, 125, 220, 138]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[73, 151, 160, 164]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[73, 177, 238, 190]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[73, 190, 913, 242]]<|/det|>
+In my opinion, the corrected version of the manuscript should include: "Periodically- Modulated Unipolar and Bipolar Orders in Nematic Fluids: Towards Miniaturized Nonlinear Vectorial Optics," the authors successfully addressed majority of the concerns raised by the reviewers. The details of the experiments are much clearer, including the improved abstract, additional discussion, and conclusion, as well as more readable figures.  
+
+<|ref|>text<|/ref|><|det|>[[73, 255, 393, 268]]<|/det|>
+One typographical error needs to be corrected:  
+
+<|ref|>text<|/ref|><|det|>[[73, 268, 646, 282]]<|/det|>
+Line 507: "polar nad orientational orders" instead of: "polar and orientational orders"  
+
+<|ref|>text<|/ref|><|det|>[[73, 294, 161, 307]]<|/det|>
+Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[73, 320, 238, 334]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[73, 346, 867, 373]]<|/det|>
+The authors have addressed most of my comments, except for two critical points (below). When these two points are addressed, I suggest publishing the article.  
+
+<|ref|>text<|/ref|><|det|>[[70, 384, 911, 399]]<|/det|>
+The novelty compared to their previous work is still not appropriately addressed. The authors should state this more clearly.  
+
+<|ref|>text<|/ref|><|det|>[[72, 410, 915, 476]]<|/det|>
+The authors removed the earlier assertive framing about "Nonlinear Malus's law" and "polar polarizers", rewrote the section, and dropped the headline claim—but they still describe the medium analogously as a "specialized 'nonlinear polar polarizer'," and state the transmission axis shows head- to- tail asymmetry via a Jones- style description. So the term specialized "nonlinear polar polarizer" should be removed. The head- to- tail asymmetry should be removed because it is incorrect.  
+
+<|ref|>text<|/ref|><|det|>[[73, 501, 145, 514]]<|/det|>
+Version 2:  
+
+<|ref|>text<|/ref|><|det|>[[73, 528, 220, 541]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[73, 554, 161, 567]]<|/det|>
+Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[73, 580, 238, 593]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[73, 593, 500, 606]]<|/det|>
+All comments have been addressed. I recommend publication.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 45, 916, 99]]<|/det|>
+Open Access This Peer Review 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.  
+
+<|ref|>text<|/ref|><|det|>[[72, 100, 797, 113]]<|/det|>
+In cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+<|ref|>text<|/ref|><|det|>[[72, 113, 911, 166]]<|/det|>
+The images or other third party material in this Peer Review 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.  
+
+<|ref|>text<|/ref|><|det|>[[72, 166, 618, 180]]<|/det|>
+To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[59, 52, 355, 68]]<|/det|>
+## Re: manuscript NCOMMS-25-04493-T  
+
+<|ref|>text<|/ref|><|det|>[[58, 79, 940, 115]]<|/det|>
+We sincerely thank the respected reviewers for their valuable time and helpful comments. We have carefully revised our manuscript and Supplementary Information by taking the respected reviewers' comments as appropriate into account.  
+
+<|ref|>text<|/ref|><|det|>[[58, 135, 940, 171]]<|/det|>
+Our point- by- point response to the reviewers' comments and the changes made in our revised manuscript with yellow color highlighted track change as follows.  
+
+<|ref|>sub_title<|/ref|><|det|>[[59, 199, 152, 214]]<|/det|>
+## Reviewer 1  
+
+<|ref|>text<|/ref|><|det|>[[57, 228, 941, 465]]<|/det|>
+The article presents the experimental results related to a new nematic liquid crystal mesophase, designated as the Nx phase, which was observed to exist between the nematic and ferroelectric nematic liquid crystal phases. The distinctive characteristics of this mesophase manifest as unipolar and bipolar orders. The authors employed these properties in the fabrication of three- dimensional polar topological structures. In the presented work, the authors demonstrate a strong understanding of liquid crystal photonics, as evidenced by their description of the preparation of a liquid crystal mixture, its properties, and the presentation of detailed information about the experimental setup. The experimental verification of the observed Nx mesophase included differential scanning calorimetry and second harmonic generation. The results obtained from these methods are detailed in the following sections. All results present a comprehensive study and are convincing. The authors successfully prepared multidomain NLC structures characterized by micron- size domain resolution. The article's structure and organization of ideas are appropriate, with a clear introduction, methodology, results, and discussion. The references are carefully chosen and are appropriate to the scope of the article. The article effectively highlights advances in the generation of topological states of polar matter and demonstrates the potential for microscopic nonlinear optical microdevices. I would recommend this manuscript for publication in Nature Communications.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 488, 135, 502]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[60, 505, 677, 522]]<|/det|>
+We appreciate Reviewer 1's encouraging comments and recommendation on our work.  
+
+<|ref|>text<|/ref|><|det|>[[58, 542, 940, 578]]<|/det|>
+Q1.1 \\*It would be beneficial to clarify the authors' precise intentions in writing about a "meter- scale vectorial beam generator." Line 107: "the nonlinear vectorial beam generator from meter to sub- millimeter scale".  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 599, 175, 614]]<|/det|>
+## A1.1 Response:  
+
+<|ref|>text<|/ref|><|det|>[[57, 617, 941, 782]]<|/det|>
+We appreciate this insightful suggestion. We aimed to demonstrate the miniaturization of our nonlinear vectorial photonic platform from meter- scale systems to sub- millimeter scale. Traditional nonlinear optical systems often necessitate large- scale setups, which typically involve beam splitting (the first step is to separate the two polarization components into two propagating beams by using a polarizing beam splitter) and beam regathering with a separated SHG process and a linear light field manipulation process. Thus, the optical paths are complex, requiring many optical components [Photonics Res. 7, 1340 (2019). Chin. Phys. Lett. 39, 034201 (2022). Appl. Phys. Lett. 119, 011104 (2021)]. By leveraging the engineerable molecular orientation of polar nematics, we demonstrate the integration of vectorial field generation and nonlinear optical frequency conversion into a single, miniaturized device—achieving a significantly reduced volume. This paves the way for portable and scalable photonic technologies with broad application potential, especially integrated photonics.  
+
+<|ref|>text<|/ref|><|det|>[[60, 802, 855, 819]]<|/det|>
+To better clarify this point in our manuscript, we added a paragraph in our "Discussion and Conclusion" section.  
+
+<|ref|>sub_title<|/ref|><|det|>[[59, 840, 329, 856]]<|/det|>
+## Added Text in Revised Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 858, 941, 949]]<|/det|>
+Line 447- 458, page 15: "As photonics technology advances from scalar to vectorial regimes \(^{30,51 - 54}\) , the construction of nonlinear vectorial light fields has become increasingly essential for applications including super- resolution imaging \(^{55}\) , high- capacity optical communications \(^{51}\) , and high- precision laser micromachining \(^{56}\) . Note that creating such vectorial nonlinear modes would require several passes through a spatial light modulator (lossy) or mixing beams with interferometry (difficult to align), demanding complex optical systems with cascaded linear and nonlinear light field manipulation. Here, the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 42, 940, 133]]<|/det|>
+generation of SH PVBs is achieved directly from a single, miniaturized polar LC device under a scalar Gaussian FW incidence, thanks to the flexible tailoring of tunable polar topologies. This is a feat that has not yet been accomplished with other ferroelectric materials. Such intriguing light- matter interaction could lead to the next generation of portable, scalable, and reconfigurable photonic technologies with wide- ranging applications, e.g., all- optical communications, quantum computing, soft intelligent robotics, and the biomedical industry."  
+
+<|ref|>text<|/ref|><|det|>[[57, 172, 647, 189]]<|/det|>
+Q1.2 \*It is probable that a typographical error has been made in the following text:  
+
+<|ref|>text<|/ref|><|det|>[[57, 191, 940, 225]]<|/det|>
+Line 448: "N,N'- Bis(2,5- di- tert- butylphenyl)- 3,3,9,10- 448 perylenedicarboximide" instead of: "Fluorescent dye N,N'- Bis(2,5- di- tert- butylphenyl)- 3,4,9,10- 448 perylenedicarboximide"  
+
+<|ref|>sub_title<|/ref|><|det|>[[57, 248, 174, 262]]<|/det|>
+## A1.2 Response:  
+
+<|ref|>text<|/ref|><|det|>[[57, 265, 530, 281]]<|/det|>
+We apologize for the misstate. We have corrected this accordingly.  
+
+<|ref|>sub_title<|/ref|><|det|>[[57, 303, 278, 318]]<|/det|>
+## Original Text in Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 321, 872, 337]]<|/det|>
+Line 448- 449, page 17: "Fluorescent dye N,N'- Bis(2,5- di- tert- butylphenyl)- 3,3,9,10- 448 perylenedicarboximide."  
+
+<|ref|>text<|/ref|><|det|>[[57, 340, 336, 355]]<|/det|>
+Revised Text in Revised Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 358, 870, 374]]<|/det|>
+Line 495- 496, page 16: "Fluorescent dye N,N'- Bis(2,5- di- tert- butylphenyl)- 3,4,9,10- 448 perylenedicarboximide."  
+
+<|ref|>sub_title<|/ref|><|det|>[[57, 420, 152, 436]]<|/det|>
+## Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[57, 448, 940, 559]]<|/det|>
+The authors describe a novel device based on LCs that directly generates perfect vector beams in the second harmonic. The paper is novel and the devices demonstrated have potential. However, the results are, in many places, very poorly described. Quite a few things are unclear to me, so I cannot assess whether they are wrong or only poorly written. The paper contains many fancy words but fails to convey its meaning in simple terms. This is already evident in the abstract and continues throughout the text. There are also no conclusions to summarize the findings of the paper. In this view also the novelty is not clear. I believe the paper can be published, but it should be significantly rewritten.  
+
+<|ref|>sub_title<|/ref|><|det|>[[57, 581, 137, 595]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[57, 598, 940, 652]]<|/det|>
+We acknowledge the review's concerns and thank the suggestions. In response to the review's feedback, we have conducted experiments, made substantial revisions, and rewritten sections to improve the clarity, conciseness, and overall presentation of our findings, ensuring that our research is conveyed more effectively.  
+
+<|ref|>text<|/ref|><|det|>[[57, 690, 940, 781]]<|/det|>
+Q2.1 \*In the introduction, there are many non optics related references, which are not well related to the paper's topic. Conversely, the generation of optical vortices and vector beams within LCs is insufficiently referenced. There are many papers about generating various beams via transmission through the structure, laser light emission, and SHG (10.1038/s41467- 024- 53040- 8). While the last paper (by the same authors) is cited, it is not discussed in enough detail, especially since the work of this paper is very similar to the former one. The novelty is, therefore, not obvious.  
+
+<|ref|>sub_title<|/ref|><|det|>[[57, 804, 176, 818]]<|/det|>
+## A2.1 Response:  
+
+<|ref|>text<|/ref|><|det|>[[57, 821, 940, 855]]<|/det|>
+We thank the reviewer for his/her constructive suggestions. We have removed some non- optics related references and added some new references related to optical vortices and vector beams with LCs to enrich the introduction.  
+
+<|ref|>sub_title<|/ref|><|det|>[[57, 858, 528, 889]]<|/det|>
+## (1) References about linear optical vortices and vector beams: structure  
+
+<|ref|>text<|/ref|><|det|>[[57, 894, 480, 947]]<|/det|>
+[New J. Phys. 9, 78 (2007)] (LC- POV); [Opt. Lett. 41, 2205- 2208 (2016)] (LC- SLM based PVVB); [Appl. Phys. Lett. 113, 121101 (2018)] (LC- POV/PVB);
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 43, 901, 115]]<|/det|>
+[Opt. Commun. 469, 125807 (2020)] (LC- SLM based PVVB); laser light emission [Opt. Express. 18, 212 (2020).] (LC- SLM based VB) [Proc. Natl. Acad. Sci. 118, 2110839118 (2021)] (Topological liquid crystal superstructures as structured light lasers);  
+
+<|ref|>text<|/ref|><|det|>[[57, 136, 558, 152]]<|/det|>
+(2) References about nonlinear optical vortices and vector beams:  
+
+<|ref|>text<|/ref|><|det|>[[57, 156, 229, 170]]<|/det|>
+SHG/nonlinear crystal  
+
+<|ref|>text<|/ref|><|det|>[[57, 174, 533, 189]]<|/det|>
+[Opt. Lett. 43, 5981- 5984 (2018)]; (LC- SLM based Nonlinear VB)  
+
+<|ref|>text<|/ref|><|det|>[[57, 210, 494, 225]]<|/det|>
+Added Optics- Related References in Revised Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 228, 777, 245]]<|/det|>
+Line 73- 74, page 2: "... have opened up new vistas to explore topological counterparts in optics<sup>22- 30</sup>."  
+
+<|ref|>text<|/ref|><|det|>[[57, 265, 940, 300]]<|/det|>
+24 Maurer, C., Jesacher, A., Fürhapter, S., Bernet, S. & Ritsch- Marte, M. Tailoring of arbitrary optical vector beams. New J. Phys. 9, 78- 78 (2007).  
+
+<|ref|>text<|/ref|><|det|>[[57, 303, 688, 319]]<|/det|>
+25 Li, P. et al. Generation of perfect vectorial vortex beams. Opt. Lett. 41, 2205 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[57, 321, 940, 338]]<|/det|>
+26 Wei, B.- Y. et al. Vortex Airy beams directly generated via liquid crystal q- Airy- plates. Appl. Phys. Lett. 112, 121101 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[57, 340, 940, 375]]<|/det|>
+27 Mandal, A., Maji, S. & Brundavanam, M. M. Common- path generation of stable cylindrical perfect vector vortex beams of arbitrary order. Optics Communications 469, 125807 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[57, 377, 940, 412]]<|/det|>
+28 Bashkansky, M., Park, D. & Fatemi, F. K. Azimuthally and radially polarized light with a nematic SLM. Opt. Express 18, 212 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[57, 415, 940, 450]]<|/det|>
+29 Papic, M. et al. Topological liquid crystal superstructures as structured light lasers. Proc. Natl. Acad. Sci. U.S.A. 118, 2110839118 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[57, 452, 940, 487]]<|/det|>
+30 Liu, H., Li, H., Zheng, Y. & Chen, X. Nonlinear frequency conversion and manipulation of vector beams. Opt. Lett. 43, 5981 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[57, 524, 940, 578]]<|/det|>
+We appreciate the reviewer's insightful observation regarding the similarity between our current manuscript and the previously published work (Nat. Commun.10.1038/s41467- 024- 53040- 8). While the two studies share methodological foundations, they differ fundamentally in their scientific focus and contributions, as explained below:  
+
+<|ref|>sub_title<|/ref|><|det|>[[57, 599, 523, 615]]<|/det|>
+## (1) Distinct Focus on Phases with Unique Physical Properties:  
+
+<|ref|>text<|/ref|><|det|>[[57, 617, 941, 764]]<|/det|>
+The previous work (Nat. Commun.10.1038/s41467- 024- 53040- 8) primarily investigated the ferroelectric nematic phase, which exhibits spontaneous polarization and robust ferroelectricity. In contrast, this manuscript explores an intermediate mesophase between the ferroelectric nematic phase and the conventional nematic phase. This mesophase demonstrates unipolar and bipolar polar orderings, with physical properties (e.g., polarization magnitude, orientational order) intermediate between ferroelectric nematic and nematic phases. A new structural model is established that elucidates the formation of the unique unipolar orderings in this mesophase—a key advance beyond the scope of our prior work. Crucially, it enables programmable and defect- free domain engineering due to the mesophase's unique polar order, a capability not demonstrated in the ferroelectric nematic phase.  
+
+<|ref|>sub_title<|/ref|><|det|>[[57, 785, 170, 800]]<|/det|>
+## (2) Structures:  
+
+<|ref|>text<|/ref|><|det|>[[57, 803, 940, 856]]<|/det|>
+We fabricate an azimuthally- variant q- plate structure in the previous work (Fig. 3 in Nat. Commun.10.1038/s41467- 024- 53040- 8). On the other hand, this work demonstrates more complex topological hierarchical superstructures (revised Fig. 5) with a high performance in nonlinear vectorial optics.  
+
+<|ref|>sub_title<|/ref|><|det|>[[57, 877, 381, 892]]<|/det|>
+## (3) Distinct Nonlinear Optical Phenomena:  
+
+<|ref|>text<|/ref|><|det|>[[57, 895, 940, 949]]<|/det|>
+Our prior work focused on achieving scalar nonlinear optical vortices in both left- handed and right- handed circular polarization channels under circular polarization incidence, in order to demonstrate the nonlinear geometric phase principle in LCs. On the other hand, this manuscript reports the generation of perfect vectorial nonlinear beams (e.g., vector beams
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 42, 940, 78]]<|/det|>
+with spatially varying polarization states) under linear polarization incidence. This distinction expands the toolkit for polarization- controlled nonlinear optics, leveraging the mesophase's flexible polar domain engineering.  
+
+<|ref|>text<|/ref|><|det|>[[57, 98, 944, 170]]<|/det|>
+In summary, the current work represents a conceptual and experimental progression by targeting an unexplored polar mesophase structure with emergent properties, developing a structural model, and demonstrating novel nonlinear vectorial optical functionalities. We have clarified these distinctions in the revised manuscript (see Introduction and Discussion sections) to better highlight the novelty of this work for readers.  
+
+<|ref|>sub_title<|/ref|><|det|>[[59, 192, 329, 207]]<|/det|>
+## Added Text in Revised Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 210, 941, 318]]<|/det|>
+Line 85- 91, page 3: "By photopatterning an azimuthally- variant q- plate structure, our group recently demonstrated the generation of second- harmonic scalar optical vortices with spin- locked topological charges by leveraging cascaded linear and nonlinear optical spin- orbit interactions<sup>34</sup>. However, developing more complex topological polar LC electrooptic devices remains a critical challenge due to the intricate energy competition among elastic energy, Landau energy, polarization gradients, flexoelectricity, and electrostatics, which limits the fabrication of defect- free and high- performance polar structures<sup>36,38,39</sup>."  
+
+<|ref|>text<|/ref|><|det|>[[57, 339, 941, 467]]<|/det|>
+Q2.2 \*The authors give a factor of \(10^{\wedge} - 6\) in volume reduction but do not explain how they got this number. Anyway, I think it is not useful to talk about volume. Specifically, the lateral dimensions of an optical device are designed for the desired size of an optical beam. In many instances, it is not desirable to make a device small in lateral size; on the opposite, to be useful, it should be large enough. But it is useful to make them thinner. Although, in the case of this paper, the device is very thin, it is also very inefficient, compared, for example, to solid- state phase- matched crystals. So, there is no point in comparing their sizes. I urge the authors to delete this number and rather just emphasize that a single device (thin layer) can do the same task as a few regular optical components.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 489, 175, 503]]<|/det|>
+## A2.2 Response:  
+
+<|ref|>text<|/ref|><|det|>[[58, 505, 940, 559]]<|/det|>
+We agree with the reviewer that different application scenarios necessitate varying device dimensions. Accordingly, we have deleted this number and emphasized that we use a single- component device in a thin layer form to achieve the same task that usually should employ several bulky optical components.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 580, 278, 595]]<|/det|>
+## Original Text in Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[58, 598, 940, 632]]<|/det|>
+Line 35- 37, page 1: "We succeed in scaling down the volume of the nonlinear vectorial beam generation devices by a factor over \(10^{- 6}\) , providing a foundation for the microscopic nonlinear optical apparatuses."  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 636, 337, 651]]<|/det|>
+## Revised Text in Revised Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[58, 654, 940, 726]]<|/det|>
+Line 35- 39, page 1: "A case study demonstrates the facile generation of second- harmonic perfect vector beams through photopatterned topological polar liquid crystal superstructures. Remarkably, this single- layer, micrometer- thin- film device attains functionalities comparable to multiple conventional optical components, establishing a foundation for advanced nonlinear vectorial optics at the microscopic scale."  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 747, 278, 762]]<|/det|>
+## Original Text in Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[58, 765, 940, 800]]<|/det|>
+Line 106- 108, page 3: "We succeed in miniaturizing the nonlinear vectorial beam generator from meter to sub- millimeter scale, scaling down the volume of the devices by a factor over \(10^{- 6}\) ."  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 804, 337, 818]]<|/det|>
+## Revised Text in Revised Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[58, 821, 940, 874]]<|/det|>
+Line 98- 100, page 3: "Additionally, we successfully generate second- harmonic perfect vector beams using a single device featuring a micrometer- thin LC layer, which performs the same functions as a large- scale optical system made up of multiple optical components."  
+
+<|ref|>text<|/ref|><|det|>[[57, 913, 940, 948]]<|/det|>
+Q2.3 \*In Fig 1a- c there are some checkmarks, X and some crossed checkmarks. It is not clear what it means. Also, RM734 and DIO have an X for SHG activity in N and Nx phases, but for NJU001, from these graphics, someone could understand
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 43, 940, 77]]<|/det|>
+that it has SHG activity across all phases. Also, what does "promising platform" mean? I suggest to remove all of these descriptions from the fig 1.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 100, 176, 114]]<|/det|>
+## A2.3 Response:  
+
+<|ref|>text<|/ref|><|det|>[[60, 117, 930, 133]]<|/det|>
+We sincerely thank the reviewer for the great suggestions. We have removed these descriptions from the Fig. 1 accordingly.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 153, 176, 167]]<|/det|>
+## Original Fig. 1:  
+
+<|ref|>image<|/ref|><|det|>[[156, 170, 828, 655]]<|/det|>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 40, 171, 55]]<|/det|>
+Revised Fig. 1:  
+
+<|ref|>image<|/ref|><|det|>[[156, 60, 833, 510]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[57, 549, 940, 585]]<|/det|>
+Q2.4 \*\\*SHG efficiency is defined as the SHG intensity ratio of the NJU001 material to that of a Y- cut quartz plate." Is this meant for the same thickness in both cases? Giving an absolute number for the nonlinear coefficient in pm/V would be nice.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 606, 177, 621]]<|/det|>
+## A2.4 Response:  
+
+<|ref|>text<|/ref|><|det|>[[57, 624, 941, 696]]<|/det|>
+In our experiment, the thickness of the Y- cut quartz plate was \(2\mathrm{mm}\) and the thickness of the NJU001 sample was \(5.2\mu \mathrm{m}\) . We have supplemented this information in the revised manuscript. Also, we agree with the reviewer that it would be better to provide an absolute number for the nonlinear coefficient \(d_{33}\) in pm/V. Relevant experimental result have been added to the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[60, 715, 939, 733]]<|/det|>
+The nonlinear optical coefficient \(d_{33}\) can be determined using the following equation for second- harmonic generation (SHG):  
+
+<|ref|>equation<|/ref|><|det|>[[343, 735, 655, 770]]<|/det|>
+\[I_{2} = \frac{2\omega_{2}^{2}\sin^{2}\left(\frac{\Delta kd}{2}\right)}{\epsilon_{0}n_{2}n_{1}^{2}c^{3}\left(\Delta k\right)^{2}} d_{33}^{2}I_{1}^{2} = \mathrm{C}\frac{\sin^{2}\left(\frac{\Delta kd}{2}\right)}{n_{2}n_{1}^{2}\left(\Delta k\right)^{2}} d_{33}^{2}I_{1}^{2}\]  
+
+<|ref|>text<|/ref|><|det|>[[58, 774, 108, 787]]<|/det|>
+where:  
+
+<|ref|>text<|/ref|><|det|>[[57, 796, 397, 824]]<|/det|>
+\(\Delta k = \frac{2\pi\omega_{1}(n_{1} - n_{2})}{c}\) is the wavevector mismatch;  
+
+<|ref|>text<|/ref|><|det|>[[58, 828, 275, 844]]<|/det|>
+\(\epsilon_{0}\) is the vacuum permittivity;  
+
+<|ref|>text<|/ref|><|det|>[[57, 846, 668, 880]]<|/det|>
+\(\omega_{1}\) and \(\omega_{2}\) are the angular frequencies of the fundamental wave (FW) and SH wave; \(c\) is the speed of light;  
+
+<|ref|>text<|/ref|><|det|>[[57, 883, 639, 899]]<|/det|>
+\(n_{1}\) and \(n_{2}\) represent the refractive indices of the FW and SH wave, respectively;  
+
+<|ref|>text<|/ref|><|det|>[[58, 902, 397, 917]]<|/det|>
+\(C\) is a constant related to the measuring system;  
+
+<|ref|>text<|/ref|><|det|>[[58, 920, 331, 935]]<|/det|>
+\(d_{33}\) is the nonlinear optical coefficient,  
+
+<|ref|>text<|/ref|><|det|>[[58, 939, 242, 954]]<|/det|>
+\(d\) is the sample thickness;
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 43, 511, 59]]<|/det|>
+\(I_{1}\) and \(I_{2}\) represent the intensities of the FW and the SH wave.  
+
+<|ref|>text<|/ref|><|det|>[[57, 61, 940, 98]]<|/det|>
+Once C is determined, and \(n_{1}\) and \(n_{2}\) are measured, we can determine the \(d_{33}\) of the NJU001 material by fitting the intensity correlation between FW \(I_{1}\) and the SH wave \(I_{2}\) .  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 118, 395, 134]]<|/det|>
+## (1) Determination of the System Constant C  
+
+<|ref|>text<|/ref|><|det|>[[57, 136, 941, 227]]<|/det|>
+The constant C was calibrated using a \(z\) - cut \(\mathrm{LiNbO_3}\) crystal ( \(d = 1 \mathrm{mm}\) in our experiment), where the nonlinear interaction is governed by the \(d_{22}\) coefficient under the experimental configuration: optic axis aligned along the \(x\) - axis, and incident FW polarization along the \(y\) - axis. Given that \(d_{22} = 6.3 \mathrm{pm / V}\) [G. D. Boyd et al. Appl. Phys. Lett. 5, 234- 236 (1964)], the refractive indices \(n_{1} = 2.218\) and \(n_{2} = 2.279\) at \(650 \mathrm{nm}\) and \(1300 \mathrm{nm}\) [O. Gayer et al. Appl. Phys. B. 91, 343- 348 (2008)], and \(\Delta k\) calculated from the dispersion relation, C was determined by fitting the \(I_{1} - I_{2}\) intensity correlation.  
+
+<|ref|>sub_title<|/ref|><|det|>[[57, 247, 465, 263]]<|/det|>
+## (2) Measurement of the Refractive Indices in NJU001  
+
+<|ref|>text<|/ref|><|det|>[[57, 265, 940, 301]]<|/det|>
+The extraordinary \((n_{\mathrm{e}})\) and ordinary \((n_{\mathrm{o}})\) refractive indices of NJU001 were measured using an interference method [Priyanka Kumari et al. Science. 383, 1364 (2024)].  
+
+<|ref|>text<|/ref|><|det|>[[58, 304, 204, 319]]<|/det|>
+Sample Preparation:  
+
+<|ref|>text<|/ref|><|det|>[[57, 321, 940, 357]]<|/det|>
+A commercial wedge cell (EHC KCRK- 05, KERUN EXPERIMENTAL EQUIPMENT CO., LTD) was filled with NJU001 at \(145^{\circ}\mathrm{C}\) via capillary. The rubbing direction was along the thickness gradient.  
+
+<|ref|>text<|/ref|><|det|>[[58, 359, 266, 374]]<|/det|>
+Interference Fringe Analysis:  
+
+<|ref|>text<|/ref|><|det|>[[57, 377, 720, 393]]<|/det|>
+Polarized light microscopy with red/near-infrared filters ( \(\lambda = 650 / 1300 \mathrm{nm}\) , bandwidth \(10 \mathrm{nm}\) ).  
+
+<|ref|>text<|/ref|><|det|>[[58, 396, 382, 411]]<|/det|>
+Dihedral angle \(\alpha\) determined from air fringes:  
+
+<|ref|>equation<|/ref|><|det|>[[375, 417, 621, 444]]<|/det|>
+\[\alpha = \frac{\lambda}{2n_{\mathrm{air}}u_{\mathrm{air}}} (\lambda = 650 \mathrm{nm}, n_{\mathrm{air}} = 1).\]  
+
+<|ref|>text<|/ref|><|det|>[[58, 450, 295, 465]]<|/det|>
+Refractive indices calculated via:  
+
+<|ref|>equation<|/ref|><|det|>[[448, 472, 548, 499]]<|/det|>
+\[n_{\mathrm{o},\mathrm{e}} = \frac{\lambda}{2\alpha u_{\mathrm{LC}}},\]  
+
+<|ref|>text<|/ref|><|det|>[[58, 506, 363, 522]]<|/det|>
+where \(u_{\mathrm{LC}}\) is the fringe spacing in NJU001.  
+
+<|ref|>text<|/ref|><|det|>[[58, 525, 225, 540]]<|/det|>
+Measurement Protocol:  
+
+<|ref|>text<|/ref|><|det|>[[57, 544, 537, 559]]<|/det|>
+\(n_{\mathrm{e}}\) measured when the director was parallel to the light polarization,  
+
+<|ref|>text<|/ref|><|det|>[[58, 563, 410, 577]]<|/det|>
+\(n_{\mathrm{o}}\) measured when the director was perpendicular.  
+
+<|ref|>text<|/ref|><|det|>[[58, 580, 795, 596]]<|/det|>
+The temperature- dependent refractive indices at \(650 \mathrm{nm}\) and \(1300 \mathrm{nm}\) are shown in the following figure.  
+
+<|ref|>image<|/ref|><|det|>[[306, 608, 780, 810]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[56, 821, 941, 875]]<|/det|>
+<center>Fig. S2.1. Temperature-dependent extraordinary and ordinary refractive indices of the NJU001 material at \(1300 \mathrm{nm}\) and \(650 \mathrm{nm}\) . The refractive indices were determined by analyzing the fringe spacing extracted from interference patterns measured at \(5^{\circ}\mathrm{C}\) intervals. </center>  
+
+<|ref|>text<|/ref|><|det|>[[57, 895, 941, 950]]<|/det|>
+After we have obtained the material parameters of \(n_{2}\) , \(n_{1}\) , and C, \(\Delta k\) was calculated. Subsequently, the nonlinear optical coefficient \(d_{33}\) was extracted by fitting the intensity correlation between FW \(I_{1}\) and the SH wave \(I_{2}\) . The temperature- dependent variation of \(d_{33}\) is presented in the following figure. As seen from the figure, the laser power/energy meter begins
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 42, 941, 95]]<|/det|>
+to detect the SH signal at approximately \(64^{\circ}\mathrm{C}\) , which shows a slight discrepancy compared to that ( \(\sim 70^{\circ}\mathrm{C}\) ) in Fig. 1e, which is derived from the light intensity processed by the CMOS camera in the main text. This difference is attributed to the sensitivity of the detection instruments; specifically, the CMOS camera is capable of detecting weaker SHG signals.  
+
+<|ref|>image<|/ref|><|det|>[[252, 106, 718, 255]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[316, 266, 679, 283]]<|/det|>
+<center>Fig. S2.2. Temperature-dependent variation of \(d_{33}\) . </center>  
+
+<|ref|>text<|/ref|><|det|>[[57, 302, 941, 411]]<|/det|>
+The NJU001 material exhibits \(C_{\infty \nu}\) symmetry and \(P_{5}\) in the direction of the director (i.e., longitudinal ferroelectricity), analogous to the behavior observed in RM734. Usually, the coefficient \(d_{31}\) can be measured by illuminating the sample with ordinary light (polarization perpendicular to the director). However, the accurate determination of this coefficient was difficult because of the high \(d_{33} / d_{31}\) ratio. As a consequence, the signal remained undetected due to the limited sensitivity of the photodetector (Laser Power and Energy Meter, SKU2256258, Coherent). Thus, we just characterized the \(d_{33}\) of the NJU001 material.  
+
+<|ref|>sub_title<|/ref|><|det|>[[57, 433, 496, 449]]<|/det|>
+## Added a Section 2 in Revised Supplementary Information  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 470, 328, 485]]<|/det|>
+## Added Text in Revised Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 488, 940, 523]]<|/det|>
+Line 130- 132, page 4: "The nonlinear coefficient \(d_{33}\) is experimentally determined to be approximately 5.7 pm/V and 6.9 pm/V at \(61.0^{\circ}\mathrm{C}\) (Nx) and \(45.7^{\circ}\mathrm{C}\) (NF), respectively (Supplementary Section 2)."  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 544, 328, 559]]<|/det|>
+## Added Text in Revised Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[58, 562, 830, 578]]<|/det|>
+Line 139- 140, page 4: "The thickness of the Y-cut and NJU001 sample was 2 mm and 5.2 \(\mu \mathrm{m}\) , respectively."  
+
+<|ref|>text<|/ref|><|det|>[[57, 616, 942, 838]]<|/det|>
+Q2.5 \\*Lines 249- 273: This paragraph is unnecessary and also partially wrong as it is written now. The authors derive two obvious equations. This could be fine, but they also invented two new names, "nonlinear Malus's law" and "polar polarizers". They further write as if this is some new concept: "Therefore, the nonlinear Malus's law provides a new concept of "polar polarizers"." But this is, of course, known for many decades. It is just an obvious property of nonlinear optical processes. Why invent new names and write that this is a new concept? They further write: "This relationship describes a new scenario in which the manipulation of light in the nonlinear optical regime is now vectorized, i.e., the vector field of the SH wave is head- to tail inequivalent (Fig. S13). The vector field of the SH wave is determined by the orientation of local dipoles in the range of [0, \(2\pi\) ]. In contrast, the traditional Malus's law disables the head- to- tail inequivalence of polarization." This is, of course, wrong. In the nonlinear regime, we still have head- to- tail equivalence of polarization. In other words, the vector field of the SH wave is head- to- tail equivalent. The only difference is the phase, which is a completely separate property from the polarization. Anyway, the authors do not mention phase in this paragraph. They say the relationship (line 262) is vectorized and head- to- tail inequivalent. Which is wrong.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 860, 176, 874]]<|/det|>
+## A2.5 Response:  
+
+<|ref|>text<|/ref|><|det|>[[58, 877, 941, 949]]<|/det|>
+We sincerely thank the reviewer for their valuable and insightful comment. Upon reconsideration, we fully agree that certain conceptual discussions in the original manuscript were excessive for the present study's scope. Accordingly, we have significantly revised this paragraph to focus on explaining the physical meaning of the derived equation, and we also added Supplementary section 5 and revised Fig. S14 to make it clearer.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[58, 62, 277, 77]]<|/det|>
+Original Text in Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 80, 941, 245]]<|/det|>
+Line 249- 283, page 10: "When a linearly- polarized light at an angle \(\phi\) is input and passes through a polarizer with its transmission axis at \(\theta \in (0, \pi)\) , the output electric field in the linear optical regime is expressed as \(\mathbf{E} = \left(\frac{E_x}{E_y}\right) = \mathbf{A}_0 \left(\cos^2 \theta \sin \theta \sin^2 \theta\right) \left(\cos \phi \sin \phi\right) = \mathbf{A}_0 \cos (\phi - \theta) \hat{\mathbf{e}}_\theta\) , where \(\hat{\mathbf{e}}_\theta\) represents the unit vector of \(\theta\) - axis. This teaches us the intensity of the output beam as \(I = \mathrm{A}_0^2 \cos^2 (\phi - \theta)\) , known as Malus's law. Systems with inversion symmetry breaking exhibit a non- zero second- order nonlinear coefficient, leading to additional second- order nonlinear optical responses like SHG. When an input polarized FW excites a nonlinear medium belonging to the point group \(C_{\infty \nu}\) , e.g., polar nematic phase composed of orientated dipoles, an SH nonlinear polarization arises as (Supplementary Section 3):  
+
+<|ref|>equation<|/ref|><|det|>[[264, 247, 937, 300]]<|/det|>
+\[\begin{array}{rl} & {\mathbf{P}_{\alpha}^{2\omega} = 2\mathrm{A}_0^2\epsilon_0d_{33}\cos^2 (\phi -\alpha)\hat{\mathbf{e}}_\alpha}\\ & {\qquad = 2\mathrm{A}_0^2\epsilon_0d_{33}\cos (\phi -\alpha)\left(\cos^2\alpha \cos \alpha \sin \alpha \sin^2\alpha\right)\left(\cos \phi \right).} \end{array} \quad (1)\]  
+
+<|ref|>text<|/ref|><|det|>[[57, 302, 941, 523]]<|/det|>
+\(\mathrm{A}_0\) is the amplitude of FW, \(\omega\) the angular frequency, \(\phi\) the polarization angle relative to the \(x\) - axis, \(\epsilon_0\) the vacuum permittivity, \(\alpha\) the azimuthal angle of the dipole, and \(\hat{\mathbf{e}}_\alpha\) the unit vector along the \(\alpha\) direction. The intensity of the emitted SH wave is proportional to the quartic amplitude of the electric field of FW, \(I^{2\omega}(\alpha) \propto \mathrm{A}_0^4 \cos^4 (\phi - \alpha)\) , corresponding to an extended nonlinear version of Malus's law (Fig. 4a). This relationship describes a new scenario in which the manipulation of light in the nonlinear optical regime is now vectorized, i.e., the vector field of the SH wave is head- to- tail inequivalent (Fig. S13). The vector field of the SH wave is determined by the orientation of local dipoles in the range of \([0, 2\pi ]\) . In contrast, the traditional Malus's law disables the head- to- tail inequivalence of polarization. Therefore, the nonlinear Malus's law provides a new concept of "polar polarizers". This immediately generates interest in pixelating and engineering the so- called polar polarizers with arbitrary polarization angles to develop unprecedented nonlinear vectorial optical devices. However, as a longstanding unsolved problem, the traditional solid- state polar systems barely own the flexibility to realize spatially dependent dipole orientations. The polar nematics, such as the Nx state, obviously break this limitation, and their designability of polar fields offers us a revolutionary platform.  
+
+<|ref|>text<|/ref|><|det|>[[57, 542, 941, 671]]<|/det|>
+As a proof- of- concept demonstration, we show the manipulation of vectorial nonlinear photonic functionalities by leveraging the direct interaction between light and a minimalist single- layer nonlinear perfect vector beam (PVB) generator with in- plane toroidal polar topological superstructure (Fig. 4a). Fig. 4b shows the design of the fabricated phase profile based on the nonlinear Malus principle, which is superposed of a vortex phase and an axicon phase function. This design highlights the unique hierarchical feature of our device; it integrates both the continuous and discontinuous variations of dipole orientations, which is confirmed by the visible Maltese cross and the periodic circular domain walls (Fig. 4c and Fig. S14). The polar ordering is characterized by SHG- I imaging (Fig. S15)."  
+
+<|ref|>sub_title<|/ref|><|det|>[[59, 692, 336, 707]]<|/det|>
+## Revised Text in Revised Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[59, 716, 940, 735]]<|/det|>
+Line 347- 393, page 12- 13: "When an input polarized FW excites a nonlinear medium belonging to the point group \(C_{\infty \nu}\)  
+
+<|ref|>text<|/ref|><|det|>[[59, 746, 899, 764]]<|/det|>
+(e.g., polar LCs composed of orientated dipoles), an nonlinear polarization field arises as (Supplementary Section 4):  
+
+<|ref|>equation<|/ref|><|det|>[[264, 766, 937, 816]]<|/det|>
+\[\begin{array}{rl} & {\mathbf{P}_{\alpha}^{2\omega} = 2\mathrm{A}_0^2\epsilon_0d_{33}\cos^2 (\phi -\alpha)\hat{\mathbf{e}}_\alpha}\\ & {\qquad = 2\mathrm{A}_0^2\epsilon_0d_{33}\cos (\phi -\alpha)\left(\cos^2\alpha \cos \alpha \sin \alpha \sin^2\alpha\right)\left(\cos \phi \right).} \end{array} \quad (3)\]  
+
+<|ref|>text<|/ref|><|det|>[[57, 820, 941, 888]]<|/det|>
+\(\mathrm{A}_0\) is the amplitude of FW, \(\epsilon_0\) the vacuum permittivity, \(d_{33}\) the nonlinear coefficient, \(\omega\) the angular frequency, \(\phi\) the polarization angle relative to the \(x\) - axis, \(\alpha\) the azimuthal angle of the dipole, and \(\hat{\mathbf{e}}_\alpha\) the unit vector along the \(\alpha\) direction, \(\left(\frac{\cos\alpha}{\sin\alpha}\right)\) . The intensity is proportional to the quartic amplitude of the electric field of FW,  
+
+<|ref|>equation<|/ref|><|det|>[[401, 894, 937, 911]]<|/det|>
+\[I^{2\omega}(\alpha) \propto \mathrm{A}_0^4 \cos^4 (\phi - \alpha) \quad (4)\]  
+
+<|ref|>text<|/ref|><|det|>[[57, 914, 941, 950]]<|/det|>
+These relationships bear a striking resemblance to the light propagation through a linear polarizer and the well- established Malus's law (Supplementary Section 5). Analogously, the oriented polar LCs function as a specialized "nonlinear polar
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 45, 940, 134]]<|/det|>
+polarizer", characterized by a Jones matrix formalism [Jones vector: \(\cos (\phi - \alpha)\left(\frac{\cos^2\alpha}{\cos\alpha}\sin \alpha\right)\) ], where the transmission axis exhibits head- to- tail asymmetry. This configuration endows the system with two distinct capabilities: 1) the nonlinear optical response governed by the material's nonlinearity; 2) the selective excitation of instantaneous electric field vector, which follows the polar LC's orientation \(\alpha\) across the full angular range \([0, 2\pi ]\) (Fig. S14).  
+
+<|ref|>text<|/ref|><|det|>[[56, 153, 941, 598]]<|/det|>
+As a proof- of- concept demonstration, we demonstrate the capability of generating both linear and nonlinear perfect vector beams (PVBs) using a single- layer patterned polar LC device (Fig. 5a). As a new type of vector beam \(^{45}\) with cylindrical symmetry in polarization, PVB has sparked considerable interest because its radius and intensity profile are independent of the polarization topological charge \(l\) , demonstrating superior capabilities in optical manipulation, microscopy imaging, and laser micromachining. Although PVBs have been widely studied in the framework of linear optics \(^{46,47}\) , the direct generation of SH PVBs from a single nonlinear optical element has not been realized due to previous challenges in manipulating in- plane polar ordering. Here, we design the orientation distribution \(\alpha\) of polar LCs based on Eq. (3) and (4), taking into account the near- field polarization, phase, and intensity distributions. Fig. S15a depicts the designed orientation distribution of polar LCs, \(\alpha = \frac{1}{2} l\phi + \frac{1}{2}\arg \left[2\sum_{n = 1}^{\infty}a_{n}\sin \left(2\pi \frac{n}{r} r\right)\right]\) , \(r \geq 0\) . \(a_{n}\) represents the Fourier series coefficients, \(T = 100 \mu \mathrm{m}\) is the period of concentric rings, \(r\) and \(\phi\) denote the radial distance and azimuthal angle in the polar coordinate, and \(l\) is the polarization topological charge. In this case, the simulated near- field instantaneous electric field vector, when Fourier- transformed to the far field, enables an intensity profile and polarization distribution that meet the criteria of a nonlinear perfect vector beam. Accordingly, we show the surface anchoring pattern for both substrates of the device in Fig. 5a. The alignment direction continuously changes in the azimuthal angle, and the orientations in adjacent rings maintain a consistent perpendicular relationship in the radial direction. By employing photoalignment technology in conjunction with a self- developed digital micromirror device- based microlithography system (Methods) \(^{34,48}\) , we achieved high- quality in- plane polar LC domain engineering, accomplished through the effective utilization of surface anchoring energy to dictate the LC orientation. The resulting polar LC superstructure is characterized under a crossed- PLM, which highlights the unique hierarchical and topological features of our device, incorporating both continuous and discontinuous variations of dipole orientations. The visible Maltese cross and the periodic circular domain walls shown in Fig. 5b and Fig. S15b indicate that the LC director orientations are faithfully imprinted according to our design. A schematic of a cross- section of the LC cell is provided in Fig. S16. The polar ordering is further confirmed by SHG- I imaging (Fig. S17), with the local polar orientation distribution at the center shown in Fig. 5c.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 616, 277, 632]]<|/det|>
+## Original Text in Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 635, 940, 670]]<|/det|>
+Line 298- 299, page 12: "Nonlinear Malus's law and vectorial nonlinear photonic functionality enabled by flexible tailoring of the polar ordering in the miniaturized PVB generation device."  
+
+<|ref|>text<|/ref|><|det|>[[58, 672, 336, 688]]<|/det|>
+Revised Text in Revised Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 690, 940, 726]]<|/det|>
+Line 417- 419, page 14: "Schematic of LC- based nonlinear vectorial optics. This photonic functionality is enabled by flexible tailoring of the polar ordering in the miniaturized PVB generation device."  
+
+<|ref|>text<|/ref|><|det|>[[57, 764, 915, 800]]<|/det|>
+Q2.6 \*The section about generating perfect vector beams is extremely short (lines 275- 295), providing almost no details. All of the following topics should be discussed in more detail:  
+
+<|ref|>text<|/ref|><|det|>[[57, 802, 940, 837]]<|/det|>
+I. What are perfect vector beams? These are much less known to readers than, for example, vortex beams. So they must be described.  
+
+<|ref|>text<|/ref|><|det|>[[57, 839, 940, 874]]<|/det|>
+II. What are the structures in Fig4, and how were they made? Provide details on the surface anchoring patterns. Also polar director field must be shown, like it is shown in Fig. S17.  
+
+<|ref|>text<|/ref|><|det|>[[57, 876, 944, 949]]<|/det|>
+III. How were the two phases achieved (Fig4bi and ii). It is not obvious how the light's phase is modulated. Specifically, how is the spiral phase modulation achieved? With LCs this is usually achieved by illuminating a q-plate (e.g. a radial configuration) with a circularly polarized beam. But here, the input beam is linearly polarized. Also, to achieve a spiral phase, the left-right symmetry has to be broken. For example, in vortex generation by a q-plate, the symmetry is broken by the circularly polarized
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 42, 940, 112]]<|/det|>
+input beam. But in this paper, as far as I can tell, the symmetry is not broken by the input beam or the LC structure. Anyway, why a spiral phase is needed anyway? This would be used for vortex beams, but vortex beams are not mentioned in the paper. As far as I know, the perfect vector beams do not need a spiral phase. The authors are confusing vector and vortex beams. They also use the term topological charge, but polarization topological charge is relevant for vector beams.  
+
+<|ref|>text<|/ref|><|det|>[[57, 115, 940, 150]]<|/det|>
+IV. Also, how was the axicon phase modulation achieved? Again this is not mentioned at all. The phase is alternating between 0 and pi/2. However, the phase of SHG can only be either 0 or pi (Fig S13b).  
+
+<|ref|>text<|/ref|><|det|>[[57, 152, 940, 188]]<|/det|>
+V. It is not clear at all what the electric field does, how it is applied, how the director changes, and how the output exactly changes.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 211, 176, 225]]<|/det|>
+## A2.6 Response:  
+
+<|ref|>text<|/ref|><|det|>[[57, 229, 940, 263]]<|/det|>
+We are deeply grateful to the reviewer for his/her valuable and constructive comments, which have significantly helped improve the quality of our manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[58, 280, 884, 296]]<|/det|>
+I: We agree with the reviewer and we have added descriptions about the perfect vector beams and related references.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 315, 327, 330]]<|/det|>
+## Added Text in Revised Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 333, 940, 461]]<|/det|>
+Line 365- 372, page 12: "As a proof- of- concept demonstration, we demonstrate the capability of generating both linear and nonlinear perfect vector beams (PVBs) using a single- layer patterned polar LC device (Fig. 5a). As a new type of vector beam \(^{45}\) with cylindrical symmetry in polarization, PVB has sparked considerable interest because its radius and intensity profile are independent of the polarization topological charge \(I\) , demonstrating superior capabilities in optical manipulation, microscopy imaging, and laser micromachining. Although PVBs have been widely studied in the framework of linear optics \(^{46,47}\) , the direct generation of SH PVBs from a single nonlinear optical element has not been realized due to previous challenges in manipulating in- plane polar ordering."  
+
+<|ref|>text<|/ref|><|det|>[[57, 479, 936, 512]]<|/det|>
+II: We sincerely apologize for not providing sufficient details about the structure and fabrication in our original manuscript. We have revised this part accordingly.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 532, 277, 547]]<|/det|>
+## Original Text in Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 550, 940, 640]]<|/det|>
+Line 278- 283, page 10: "Fig. 4b shows the design of the fabricated phase profile based on the nonlinear Malus principle, which is superposed of a vortex phase and an axicon phase function. This design highlights the unique hierarchical feature of our device; it integrates both the continuous and discontinuous variations of dipole orientations, which is confirmed by the visible Maltese cross and the periodic circular domain walls (Fig. 4c and Fig. S14). The polar ordering is characterized by SHG- I imaging (Fig. S15)."  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 643, 336, 657]]<|/det|>
+## Revised Text in Revised Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 660, 940, 861]]<|/det|>
+Line 380- 393, page 12- 13: "Accordingly, we show the surface anchoring pattern for both substrates of the device in Fig. 5a. The alignment direction continuously changes in the azimuthal direction, and the orientations in adjacent rings maintain a consistent perpendicular relationship in the radial direction. By employing photoalignment technology in conjunction with a self- developed digital micromirror device- based microlithography system (Methods) \(^{34,48}\) , we achieved high- quality in- plane polar LC domain engineering, accomplished through the effective utilization of surface anchoring energy to dictate the LC orientation. The resulting polar LC superstructure is characterized under a crossed- PLM, which highlights the unique hierarchical and topological features of our device, incorporating both continuous and discontinuous variations of dipole orientations. The visible Maltese cross and the periodic circular domain walls shown in Fig. 5b and Fig. S15b indicate that the LC director orientations are faithfully imprinted according to our design. A schematic of a cross- section of the LC cell is provided in Fig. S16. The polar ordering is further confirmed by SHG- I imaging (Fig. S17), with the local polar orientation distribution at the center shown in Fig. 5c."  
+
+<|ref|>text<|/ref|><|det|>[[57, 882, 940, 955]]<|/det|>
+III and IV: We sincerely apologize for any confusion caused by the brevity of our previous writing. We recognize that our initial presentation was insufficiently detailed and partially inaccurate, and we appreciate the reviewer's patience in this matter. The two subfigures illustrated in Fig. 4b- i and ii (original version) are not used to independently generate a vortex beam and implement axicon focusing, respectively. Instead, these two are combined via linear superposition to create a composite
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 42, 941, 115]]<|/det|>
+configuration to obtain the designed optical axis distribution of the toroidal polar topological superstructure (Fig. 4b- iii, original version), thus for producing PVBs. As evident from Fig. 4b- iii, the alignment directions in adjacent rings exhibit a consistent perpendicular relationship in the radial direction. This spatial arrangement corresponds to “the phase alternating between 0 and \(\pi /2\) ” as the reviewer said. This is from the structure aspect regarding the orientation angle distribution.  
+
+<|ref|>text<|/ref|><|det|>[[57, 135, 941, 208]]<|/det|>
+However, it is important to distinguish this from the SHG phase characteristic shown in Fig. S14f. Here, the phase refers specifically to phase distribution of the near- field vector light field, which enforces a binary phase constraint - the phase can only assume discrete values of either 0 or \(\pi\) for the resulting linearly polarized light. This is from the optics aspect regarding the optical phase of the near- field polarization distribution.  
+
+<|ref|>text<|/ref|><|det|>[[57, 227, 941, 319]]<|/det|>
+To enhance clarity, we have revised our presentation by: 1) directly providing the mathematical formulation of the final \(\alpha ; 2\) showing the illustration of target polar LC orientation distribution (Fig. S15a); 3) presenting the corresponding surface anchoring pattern in revised Fig. 5a, which is implemented through photopatterning to control the LC director orientation. Accordingly, we have now provided a more comprehensive explanation to address this issue. Also, we have changed the term topological charge to polarization topological charge in the revised manuscript.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 336, 278, 351]]<|/det|>
+## Original Text in Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 352, 940, 386]]<|/det|>
+Line 278- 279, page 10: “Fig. 4b shows the design of the fabricated phase profile based on the nonlinear Malus principle, which is superposed of a vortex phase and an axicon phase function.”  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 386, 338, 401]]<|/det|>
+## Revised Text in Revised Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 401, 941, 530]]<|/det|>
+Line 372- 380, page 12: “Here, we design the orientation distribution \(\alpha\) of polar LCs based on Eq. (3) and (4), taking into account the near- field polarization, phase, and intensity distributions. Fig. S15a depicts the designed orientation distribution of polar LCs, \(\alpha = \frac{1}{2} l\phi +\frac{1}{2}\arg \left[2\sum_{n = 1}^{n}a_{n}\sin \left(2\pi \frac{n}{r}\tau\right)\right], r \geq 0\) . \(a_{n}\) represents the Fourier series coefficients, \(T = 100 \mu \mathrm{m}\) is the period of concentric rings, \(r\) and \(\phi\) denote the radial distance and azimuthal angle in the polar coordinate, and \(l\) is the polarization topological charge. In this case, the simulated near- field instantaneous electric field vector, when Fourier- transformed to the far field, enables an intensity profile and polarization distribution that meet the criteria of a nonlinear perfect vector beam.”  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 547, 328, 562]]<|/det|>
+## Added Text in Revised Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 562, 941, 612]]<|/det|>
+Line 367- 370, page 12: “PVB has sparked considerable interest because its radius and intensity profile are independent of the polarization topological charge \(l\) , demonstrating superior capabilities in optical manipulation, microscopy imaging, and laser micromachining.”  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 630, 328, 644]]<|/det|>
+## Added Text in Revised Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 646, 494, 662]]<|/det|>
+Line 377, page 12: “\(l\) is the polarization topological charge.”  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 679, 278, 694]]<|/det|>
+## Original Text in Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 696, 395, 711]]<|/det|>
+Line 284, page 10: “topological charge \(|l| = 2\) .”  
+
+<|ref|>text<|/ref|><|det|>[[58, 713, 338, 727]]<|/det|>
+Revised Text in Revised Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 728, 518, 744]]<|/det|>
+Line 395- 396, page 13: “polarization topological charge \(|l| = 2\) .”  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 761, 278, 776]]<|/det|>
+## Original Text in Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 778, 920, 793]]<|/det|>
+Line 290- 291, page 11: “characterized by the invariant radius and intensity profiles under changing topological charges.”  
+
+<|ref|>text<|/ref|><|det|>[[58, 794, 338, 808]]<|/det|>
+Revised Text in Revised Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 810, 940, 842]]<|/det|>
+Line 406, page 13: “characterized by the invariant radius and intensity profiles under changing polarization topological charges.”  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 860, 278, 874]]<|/det|>
+## Original Text in Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 876, 725, 891]]<|/det|>
+Line 308, page 13: “Dependences of the ring diameter on the topological charge \((|l| = 1 \sim 6)\) .”  
+
+<|ref|>text<|/ref|><|det|>[[58, 893, 338, 907]]<|/det|>
+Revised Text in Revised Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 909, 870, 925]]<|/det|>
+Line 427- 428, page 14- 15: “Dependences of the ring diameter on the polarization topological charge \((|l| = 1 \sim 6)\) .”
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 42, 238, 56]]<|/det|>
+Newly added Fig. S15a:  
+
+<|ref|>image<|/ref|><|det|>[[278, 75, 736, 380]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[295, 394, 700, 410]]<|/det|>
+<center>Fig. S15a. Target orientation distribution \(\alpha\) of polar LCs. </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[156, 71, 825, 720]]<|/det|>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[155, 65, 825, 650]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[56, 42, 171, 56]]<|/det|>
+<center>Revised Fig. 5: </center>  
+
+<|ref|>text<|/ref|><|det|>[[56, 653, 941, 707]]<|/det|>
+V: Due to space limitations in the main text, the electrical switching characteristics of our nonlinear optical LC device are presented in Supplementary Section 7 (revised version). But to enhance reader comprehension, we have added a brief paragraph that provides additional context about this particular aspect of our work.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 728, 277, 743]]<|/det|>
+## Original Text in Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[56, 746, 940, 780]]<|/det|>
+Line 285- 287, page 10- 11: "an SH vector beam with a ring- shaped intensity profile is also obtained using electric tunability (Fig. 4e,h-j; Supplementary Section 5)."  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 783, 336, 798]]<|/det|>
+## Revised Text in Revised Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[56, 801, 941, 892]]<|/det|>
+Line 410- 415, page 13: "Furthermore, by leveraging the dynamic tunability of LCs, we demonstrate active and reversible switching of SH-PVBs under an ultra- low electric field. When applying a triangular electric field (peak- to- peak voltage \(\mathrm{V_{pp}} = 0.12 \mathrm{V / \mu m}\) , frequency \(f = 0.5 \mathrm{Hz}\) ) across the device, the topological polar LC superstructure exhibits periodic switching behavior (Fig. 5h and 5i). This results in a dynamically tunable nonlinear vector optical field with periodic on- off features (Fig. 5j). More details are provided in Supplementary Section 7."  
+
+<|ref|>text<|/ref|><|det|>[[60, 930, 940, 947]]<|/det|>
+Q2.7 \*In connection to III., in fig4 it is not clear what the LC structure is, the same is also not clear from the beginning of the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 43, 940, 77]]<|/det|>
+paper. Photoalignment is only mentioned once in the introduction, but is not clear where and how was it used for the rest of the results. For example, are the patterns in fig2 self-formed or due to photoalignment?  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 100, 176, 114]]<|/det|>
+## A2.7 Response:  
+
+<|ref|>text<|/ref|><|det|>[[57, 117, 941, 189]]<|/det|>
+We have gone through the whole manuscript and supplement the details about the structures and the fabrications. As stated in Methods (Line 490- 491, Page 17) that if not specifically denoted, the tests in our work employed LC cells that were photoaligned. We have now included a list for all photoalignment patterns as below, which is added in our revised Supplementary Information. The patterns in Fig. 2 are self- formed under homogeneous photoalignment.  
+
+<|ref|>image<|/ref|><|det|>[[270, 199, 733, 572]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[225, 580, 770, 596]]<|/det|>
+<center>Fig. S21. Photoalignment conditions for the LC structures in our manuscript </center>  
+
+<|ref|>text<|/ref|><|det|>[[58, 618, 328, 633]]<|/det|>
+Added Text in Revised Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 636, 562, 652]]<|/det|>
+Line 482, page 16: "We summarize the alignment pattern in Fig. S21."  
+
+<|ref|>text<|/ref|><|det|>[[58, 673, 279, 688]]<|/det|>
+Original Text in Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[58, 691, 764, 707]]<|/det|>
+Line 233, page 8: "By encoding the brightness of the artwork into the spatially- variant polar field."  
+
+<|ref|>text<|/ref|><|det|>[[58, 710, 273, 725]]<|/det|>
+Revised Text in Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[58, 728, 886, 745]]<|/det|>
+Line 330, page 10: "By encoding the brightness of the artwork (from 0 to 255) into the spatially- variant polar field."  
+
+<|ref|>text<|/ref|><|det|>[[58, 766, 264, 781]]<|/det|>
+Added Text in Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[58, 784, 857, 801]]<|/det|>
+Line 341- 342, page 11: "The grayscale values from 0 to 255 are proportionally mapped to the distribution of \(\alpha\) ."  
+
+<|ref|>text<|/ref|><|det|>[[57, 839, 940, 873]]<|/det|>
+Q2.8 \*Not clear also what is the surface anchoring on the other non- patterned surface or are both surfaces patterned? Some cross- sections of the LC cells should help.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 896, 176, 910]]<|/det|>
+## A2.8 Response:  
+
+<|ref|>text<|/ref|><|det|>[[57, 914, 941, 948]]<|/det|>
+The two substrates are both surfaces photopatterned and the patterns are the same (revised Fig. 5a, added description of "photoalignment for both substrates"). We have added a schematic of a cross- sections of the LC cell in Fig. S16 for
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 44, 150, 58]]<|/det|>
+clarification.  
+
+<|ref|>image<|/ref|><|det|>[[260, 95, 700, 216]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[298, 229, 699, 245]]<|/det|>
+<center>Fig. S16. A schematic of a cross-sections of the LC cell. </center>  
+
+<|ref|>text<|/ref|><|det|>[[58, 266, 328, 282]]<|/det|>
+Added Text in Revised Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[60, 284, 816, 301]]<|/det|>
+Line 481- 482, page 16: "The SD1 alignment layers on both substrates were simultaneously photoaligned."  
+
+<|ref|>text<|/ref|><|det|>[[58, 320, 710, 337]]<|/det|>
+Line 391, page 13: "A schematic of a cross-sections of the LC cell is provided in Fig. S16."  
+
+<|ref|>text<|/ref|><|det|>[[57, 357, 940, 393]]<|/det|>
+Line 419- 420, page 14: "The photoalignment patterns for the two substrates of the LC device are the same shown in the right panel."  
+
+<|ref|>text<|/ref|><|det|>[[57, 450, 941, 561]]<|/det|>
+Q2.9 \*Another thing that is not clear to me is how they were able to generate a perfect circle as drawn in Fig4a and seen in 4e. No light should be generated when the director is perpendicular to the incoming polarization. Therefore there should be no SH light perpendicular to the incoming polarization. For this reason, all the rings should have gaps. From some images, this is evident; from some, others is not. In any case, Fig4a is wrong. Also, since there are intensity gaps in the ring, then polarization topological charge cannot be defined. In other words, we cannot define how many times the polarization rotates when going around one full circle since, in some positions, the phase is not defined due to the intensity being zero.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 581, 176, 596]]<|/det|>
+## A2.9 Response:  
+
+<|ref|>text<|/ref|><|det|>[[57, 598, 941, 745]]<|/det|>
+We apologize for the lack of clarity in our previous manuscript. The reviewer is right in stating that no light should be emitted when the director is perpendicular to the incoming polarization. However, it is important to note that the light patterns in revised Fig. 5e correspond to a specific far- field nonlinear PVB with the polarization topological charge of \(l = 2\) as well as its polarization characterization. The nonlinear PVB itself does not have gaps (the first panel in revised Fig. 5e). The SH light fields in the second and third panels in revised Fig. 5e are analyzed by an analyzer (polarizer before the image plane). Extinction regions (i.e., gaps) occur where the polarization is perpendicular to the analyzer. Thus, by analyzing these extinction locations, the nonlinear PVB with a specific space- variant polarization distribution is confirmed, which is provided in the upper right corner of the first panel of revised Fig. 5e.  
+
+<|ref|>text<|/ref|><|det|>[[57, 764, 941, 856]]<|/det|>
+Another thing should be noted that all the rings of the polar LC superstructure would generate ring- shaped SH signals in the near- field with gaps. Each ring has two gaps that are aligned in a line. However, the lines' orientations for odd and even rings are perpendicular to each other due to the designed polar LC superstructure, where the orientations of the LC dipoles within adjacent rings always differs by pi/2. Then, the light pattern observed at the far- field focal point, after Fourier transformation, displays a perfect circular shape, as shown in revised Fig. 5a.  
+
+<|ref|>text<|/ref|><|det|>[[57, 876, 620, 892]]<|/det|>
+To make it clearer, we have revised the corresponding descriptions accordingly.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 914, 328, 929]]<|/det|>
+## Added Text in Revised Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 932, 940, 949]]<|/det|>
+Line 399- 404, page 13: "Extinction occurs when the polarization is perpendicular to the analyzer. Therefore, by examining
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[58, 41, 940, 96]]<|/det|>
+the locations of extinction, the nonlinear PVB with a specific space- variant polarization distribution is confirmed, as shown in the upper right corner of the first panel of Fig. 5e. The polarization topological charge is determined to be \(l = 2\) . These results align well with the expected intensities and polarization distributions."  
+
+<|ref|>sub_title<|/ref|><|det|>[[59, 118, 327, 133]]<|/det|>
+## Added Text in Revised Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[58, 136, 940, 172]]<|/det|>
+Line 426- 427, page 14: "The polarization distributions of the PVBs are provided in the upper right corners of the first panels of (d) and (e)."  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 192, 279, 207]]<|/det|>
+## Original Text in Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[58, 210, 940, 245]]<|/det|>
+Line 285- 286, page 10- 11: "Simultaneously, an SH vector beam with a ring- shaped intensity profile is also obtained using electric tunability"  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 248, 273, 263]]<|/det|>
+## Revised Text in Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[58, 266, 940, 301]]<|/det|>
+Line 398, page 13: "Simultaneously, an SH vector beam with a ring- shaped intensity profile is also obtained (the first panel in Fig. 5e)."  
+
+<|ref|>text<|/ref|><|det|>[[58, 339, 940, 412]]<|/det|>
+Q2.10 \*The structure of the paper is quite strange. The authors first talk about LC materials, then structures in LCs, then optics, and the final section is named "discussion." But the discussion does not discuss the results, but it is a section by itself, discussing some new things but not the optics at all. Therefore, there is no actual discussion/conclusion. The "discussion" chapter would fit before the optics part. In general, the paper also feels like two or even three disconnected parts.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 433, 184, 448]]<|/det|>
+## A2.10 Response:  
+
+<|ref|>text<|/ref|><|det|>[[58, 451, 940, 523]]<|/det|>
+We are sincerely grateful for the reviewer's valuable advice. We have restructured our manuscript to follow a logical flow, from materials to structures and finally to optics. The former "Discussion" section (about polar ordering simulations) has been relocated prior to the optics part. Additionally, we have added a new paragraph in the revised "discussion and conclusion" section to succinctly summarize our findings and the significance.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 544, 390, 559]]<|/det|>
+## Revised "discussion and conclusion" section  
+
+<|ref|>text<|/ref|><|det|>[[57, 561, 941, 745]]<|/det|>
+Line 436- 460, page 15: "Systems that exhibit phase competition, order parameter coexistence, and emergent order parameter topologies constitute a crucial part of modern condensed- matter physics<sup>50</sup>. In this work, we establish a key link between paraelectric (apolar) and ferroelectric (polar) states, based on a distinct liquid of ferroelectrics with nematicity, which acts as an intermediate state between the recently identified ferroelectric nematic and traditional apolar nematic phases. A new structural model is established with our extended Oseen- Frank free- energy functional, which elucidates the formation of the unique unipolar orderings in this mesophase. Unlike conventional ferroelectric materials, the Nx phase exhibits an exceptional capability of in- plane polar domain engineering, enabling the creation of complex hierarchical architectures with customizable, precise, and defect- free LC ordering on a large scale. This may provide an exciting platform for developing new ferroelectric materials with tailored properties and for exploring emergent phenomena in condensed- matter physics and nonlinear photonics.  
+
+<|ref|>text<|/ref|><|det|>[[57, 764, 941, 950]]<|/det|>
+As photonics technology advances from scalar to vectorial regimes<sup>30,51- 54</sup>, the construction of nonlinear vectorial light fields has become increasingly essential for applications including super- resolution imaging<sup>55</sup>, high- capacity optical communications<sup>51</sup>, and high- precision laser micromachining<sup>56</sup>. Note that creating such vectorial nonlinear modes would require several passes through a spatial light modulator (lossy) or mixing beams with interferometry (difficult to align), demanding complex optical systems with cascaded linear and nonlinear light field manipulation. Here, the generation of SH PVBs is achieved directly from a single, miniaturized polar LC device under a scalar Gaussian FW incidence, thanks to the flexible tailoring of tunable polar topologies. This is a feat that has not yet been accomplished with other ferroelectric materials. Such intriguing light- matter interaction could lead to the next generation of portable, scalable, and reconfigurable photonic technologies with wide- ranging applications, e.g., all- optical communications, quantum computing, soft intelligent robotics, and the biomedical industry."
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[58, 53, 348, 68]]<|/det|>
+## Re: manuscript NCOMMS-25-04493A  
+
+<|ref|>text<|/ref|><|det|>[[57, 80, 940, 115]]<|/det|>
+We sincerely thank the respected reviewers for their valuable time and helpful comments. We have carefully revised our manuscript and Supplementary Information by taking the respected reviewers' comments as appropriate into account.  
+
+<|ref|>text<|/ref|><|det|>[[57, 135, 940, 171]]<|/det|>
+Our point- by- point response to the reviewers' comments and the changes made in our revised manuscript with yellow color highlighted track change as follows.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 201, 152, 215]]<|/det|>
+## Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[57, 248, 940, 282]]<|/det|>
+Q1.1 \*One typographical error needs to be corrected: Line 507: "polar nad orientational orders" instead of: "polar and orientational orders".  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 304, 174, 318]]<|/det|>
+## A1.1 Response:  
+
+<|ref|>text<|/ref|><|det|>[[58, 322, 490, 337]]<|/det|>
+We apologize for the error. We have corrected it accordingly.  
+
+<|ref|>text<|/ref|><|det|>[[58, 358, 280, 373]]<|/det|>
+Original Text in Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[58, 377, 470, 392]]<|/det|>
+Line 507, page 16: "... polar nad orientational orders ..."  
+
+<|ref|>text<|/ref|><|det|>[[58, 396, 336, 410]]<|/det|>
+Revised Text in Revised Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[58, 415, 470, 429]]<|/det|>
+Line 516, page 16: "... polar and orientational orders ..."  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 479, 152, 492]]<|/det|>
+## Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[57, 544, 940, 579]]<|/det|>
+Q2.1 \*The novelty compared to their previous work is still not appropriately addressed. The authors should state this more clearly.  
+
+<|ref|>text<|/ref|><|det|>[[58, 599, 940, 652]]<|/det|>
+We apologize that, while we explained the core differences between our previous work and the present one in our last reply to the reviewer (as repeatedly explained below), we did not make enough revisions in our manuscript. In this revision, we made revisions to make readers understand these points well.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 674, 610, 690]]<|/det|>
+## The core differences between our previous work and the present one  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 693, 222, 707]]<|/det|>
+## (1) New polar orders:  
+
+<|ref|>text<|/ref|><|det|>[[58, 710, 940, 819]]<|/det|>
+The previous work (Nat. Commun.10.1038/s41467- 024- 53040- 8) primarily investigated the ferroelectric nematic phase, which exhibits spontaneous polarization and robust ferroelectricity. In contrast, this work explores an intermediate mesophase between the ferroelectric nematic phase and the conventional nematic phase. This mesophase demonstrates unipolar and bipolar polar orderings, with its physical properties (e.g., polarization magnitude, orientational order) intermediate between ferroelectric nematic and nematic phases. A new structural model is established that elucidates the formation of the unique unipolar orderings in this mesophase—a key advance beyond the scope of our prior work.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 841, 320, 856]]<|/det|>
+## (2) Domain engineering capability:  
+
+<|ref|>text<|/ref|><|det|>[[58, 859, 940, 949]]<|/det|>
+The Nx phase exhibits greater potential than the NF phase for in- plane domain engineering of nonlinear optical elements. Although prior study (Nat. Commun.10.1038/s41467- 024- 53040- 8) have demonstrated that NF ordering can generate pure splay orientational patterns with good quality, it remains fundamentally limited in achieving complex hierarchical architectures with customizable deformations (e.g., intricate bending geometries), precise alignment, and defect- free polar ordering at scale. This limitation is clearly demonstrated in Fig. 5, where photopatterning of a concentric topological polar
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 42, 942, 133]]<|/det|>
+LC superstructure in the NF phase results in extensive defects and domain walls (Fig. 5b_iv and Fig. S15). Conversely, the Nx phase enables defect- free hierarchical topological superstructures (Fig. 5b_ii and Fig. S15), thus yielding high- performance nonlinear perfect vector beams. Such complex hierarchical superstructures surpass the azimuthal complexity of polar orientational field designs like \(q\) - plate in our previous work (Nat. Commun.10.1038/s41467- 024- 53040- 8) realized in the NF phase.  
+
+<|ref|>sub_title<|/ref|><|det|>[[59, 155, 386, 170]]<|/det|>
+## (3) From scalar vortices to vectorial beams:  
+
+<|ref|>text<|/ref|><|det|>[[57, 173, 941, 245]]<|/det|>
+Our prior work demonstrates the parallel generations of scalar nonlinear optical vortices with nonlinear geometric phase encoded NF \(q\) - plate structure. On the other hand, this manuscript reports the generation of vectorial nonlinear perfect vortex beams. This distinction expands the toolkit for nonlinear vectorial optics, leveraging the mesophase's flexible polar domain engineering.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 266, 222, 281]]<|/det|>
+## Revised Manuscript  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 285, 277, 300]]<|/det|>
+## Original Text in Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 302, 943, 486]]<|/det|>
+Line 436- 447, page 15: "Systems that exhibit phase competition, order parameter coexistence, and emergent order parameter topologies constitute a crucial part of modern condensed- matter physics<sup>50</sup>. In this work, we establish a key link between paraelectric (apolar) and ferroelectric (polar) states, based on a distinct liquid of ferroelectrics with nematicity, which acts as an intermediate state between the recently identified ferroelectric nematic and traditional apolar nematic phases. A new structural model is established with our extended Oseen- Frank free- energy functional, which elucidates the formation of the unique unipolar orderings in this mesophase. Unlike conventional ferroelectric materials, the Nx phase exhibits an exceptional capability of in- plane polar domain engineering, enabling the creation of complex hierarchical architectures with customizable, precise, and defect- free LC ordering on a large scale. This may provide an exciting platform for developing new ferroelectric materials with tailored properties and for exploring emergent phenomena in condensed- matter physics and nonlinear photonics."  
+
+<|ref|>sub_title<|/ref|><|det|>[[59, 507, 335, 522]]<|/det|>
+## Revised Text in Revised Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 525, 942, 893]]<|/det|>
+Line 433- 456, page 15: "Systems that exhibit phase competition, order parameter coexistence, and emergent order parameter topologies constitute a crucial part of modern condensed- matter physics<sup>50</sup>. In this work, we establish a key link between paraelectric (apolar) and ferroelectric (polar) states, based on a distinct liquid of ferroelectrics with nematicity, which acts as an intermediate state between the recently identified ferroelectric nematic and traditional apolar nematic phases. A new structural model is established with our extended Oseen- Frank free- energy functional, which elucidates the formation of the unique unipolar and bipolar orderings in this mesophase. Importantly, the discovered polar orders make a sharp contrast to the existing NF order in the structure, and demonstrate a better potential in domain engineering for developing nonlinear optical elements. For the former, this study reveals a novel energy competition scenario among the Landau energies (\(\Delta F_{\mathrm{L},\mathrm{A}}\) and \(\Delta F_{\mathrm{L},\mathrm{P}}\) for apolar and polar Landau energy penalties) and flexoelectricity (\(\Delta F_{\mathrm{flex}}\)) with decoupled \(S\) and \(S_{\mathrm{P}}\) . It means that carefully adjusting multiple energy landscapes with either coupling or decoupling polar and LC nematic orders would trigger a broad range of unknown polar orders. For the latter, the Nx phase exhibits an exceptional capability of in- plane polar domain engineering. Though preceding works have shown that NF order can be employed to generate pure splay orientational patterns with good quality<sup>34,36,51</sup>, complex hierarchical architectures with customizable (e.g., with more complicated deformations with bending), precise, and defect- free polar ordering on a large scale is difficult (see defect- mediated NF structure in Fig. 5b (iv), Fig. S11, and Fig. S15). The Nx order has shown the capability to overcome these drawbacks, enabling hierarchical topological superstructures that surpass the azimuthal complexity of polar orientational field designs like prior \(q\) - plate<sup>34</sup> and periodic splay patterns<sup>36,51</sup>, yielding high- performance nonlinear perfect vector beams — a capability neither previously conceptualized nor realized in the NF phase. This may provide an exciting platform for developing new ferroelectric materials with tailored properties and for exploring emergent phenomena in condensed- matter physics and nonlinear photonics."  
+
+<|ref|>text<|/ref|><|det|>[[58, 931, 940, 949]]<|/det|>
+Q2.2 \*The authors removed the earlier assertive framing about “Nonlinear Malus’s law” and “polar polarizers”, rewrote the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 42, 941, 97]]<|/det|>
+section, and dropped the headline claim—but they still describe the medium analogously as a “specialized ‘nonlinear polar polarizer’,” and state the transmission axis shows head- to- tail asymmetry via a Jones- style description. So the term specialized “nonlinear polar polarizer” should be removed. The head- to- tail asymmetry should be removed because it is incorrect.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 118, 176, 133]]<|/det|>
+## A2.2 Response:  
+
+<|ref|>text<|/ref|><|det|>[[57, 136, 940, 172]]<|/det|>
+We have remove both phrases "specialized ‘nonlinear polar polarizer' and "head- to- tail asymmetry" accordingly to eliminate any possible ambiguity.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 210, 244, 225]]<|/det|>
+## Our Text in Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 228, 941, 301]]<|/det|>
+Line 357- 360, page 12: "Analogously, the oriented polar LCs function as a specialized ‘nonlinear polar polarizer”, characterized by a Jones matrix formalism [Jones vector: \(\cos (\phi - \alpha)\left(\begin{array}{cc}{\cos^{2}\alpha}&{\cos\alpha\sin\alpha}\\{\cos\alpha\sin\alpha}&{\sin^{2}\alpha}\end{array}\right)]\) , where the transmission axis exhibits head- to- tail asymmetry. This configuration endows the system with two distinct capabilities..."  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 303, 336, 318]]<|/det|>
+## Revised Text in Revised Manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[57, 321, 940, 375]]<|/det|>
+Line 356- 358, page 12: "The oriented polar LCs is characterized by a Jones matrix formalism: \(\cos (\phi - \alpha)\left(\begin{array}{cc}{\cos^{2}\alpha}&{\cos\alpha\sin\alpha}\\{\cos\alpha\sin\alpha}&{\sin^{2}\alpha}\end{array}\right)\) , endowing the system with two distinct capabilities..."
+
+<--- Page Split --->

peer_reviews/supplementary_0_Transparent Peer Review file__ea1ec4a4902aa61e5b36f33c79e40e7d7bfc3e82b5a0ebffabff1a0fa26242a4/images_list.json

@@ -0,0 +1,34 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Mass spectra of ubiquitination modifications at 16 lysine sites of CAD",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 8
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_1.jpg",
+    "caption": "Mass spectra of ubiquitination modifications at 16 lysine sites of CAD",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 8
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_2.jpg",
+    "caption": "Supplementary Table 2. Structures of small molecule compounds.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 9
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_3.jpg",
+    "caption": "Case#9",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 9
+  }
+]

peer_reviews/supplementary_0_Transparent Peer Review file__ea3baab8c78d31e3169d6916ccba0f903a7ba7d2977a3fd23353f046862522cf/images_list.json

@@ -0,0 +1,10 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Rebuttal Figure 2. U2OS-ACE2 cells were infected with SARS-CoV-2 WT MOI 0.5 and cells were fixed and stained with the indicated antibody combinations at 6 hours post infection.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 9
+  }
+]

peer_reviews/supplementary_0_Transparent Peer Review file__ea3baab8c78d31e3169d6916ccba0f903a7ba7d2977a3fd23353f046862522cf/supplementary_0_Transparent Peer Review file__ea3baab8c78d31e3169d6916ccba0f903a7ba7d2977a3fd23353f046862522cf.mmd

@@ -0,0 +1,592 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+# SARS-CoV-2 N protein recruits G3BP to double membrane vesicles to promote translation of viral mRNAs  
+
+Corresponding Author: Professor Gerald Mclnerney  
+
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+Version 0:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author) Summary/Key Results  
+
+The manuscript Long et al. investigates the pro- viral role of G3BP1/2 during SARS- CoV- 2 infection. Previous studies have established that the SARS- CoV- 2 Nucleocapsid (N) protein binds with G3BP1/2, inhibiting stress granule formation (which is believed to be antiviral). However, whether G3BP1/2 is itself pro- or anti- viral is a matter of some debate. To investigate the importance of N:G3BP1/2 binding, the authors use reverse genetics to generate a recombinant virus with a two amino acid substitution in the N protein (N: 115A; F17A). Compared to infection with wild type SARS- CoV- 2, the authors mutant exhibits reduced replication, N:G3BP1/2 binding, and increased stress granule formation.  
+
+The most interesting findings of this article are mechanistic. The authors clearly demonstrate that G3BP1 co- localizes with NSP3 on double membrane vesicles and that this interaction requires N:G3BP1 binding. Furthermore, they provide evidence that this results in enhanced translation at RTCs, evidenced by the co- localization of 40S ribosomal subunits, ribopuromycinylation experiments, and the number of ribosomes localized to DMVs by TEM.  
+
+Overall, this manuscript is both interesting and scientifically sound. In particular, the last 1/3 (Fig. 5 and 6) of the paper linking N:G3BP1 binding with enhanced translation at DMVs is novel. Pro- viral functions of G3BP1 during SARS- CoV- 2 infection have been proposed, but the data presented here are the first to elucidate a concrete mechanism.  
+
+Despite this article's strengths, there are some major weaknesses. First, the overall novelty of the paper is lessened by the fact that a similar article, Yang et al. 2024, has already been published. In Yang et al., the authors produced a single substitution mutant (F17A) SARS- CoV- 2, and described the effects on in vitro replication, G3BP1/2 binding, phase separation, stress granule formation, and in vivo infection. Given the similarity in the mutants (F17A vs I15A/F17A) and that the results and conclusions are almost identical (they differ only in magnitude), most of the data presented in Figures 2 – 4 are merely a confirmation of prior work. At minimum, the authors need to cite this past work.  
+
+In conclusion, the mechanistic work present in the figures 5 and 6 of this manuscript are novel, interesting, and important. However, this novelty is harmed by the fact that the data in Figures 2 – 4 is largely a confirmation of a manuscript of another group, which the authors do not cite. The authors can improve their manuscript by a) emphasizing and expanding upon those findings which are novel and b) citing Yang et al. and addressing the minor points of difference between the two studies.  
+
+Specific Critiques:  
+
+Major:  
+
+1. Novelty: Yang et al. have already published on a recombinant SARS-CoV-2 mutant that disrupts N:G3BP1/2 binding. Their mutant contains 1 of the 2 mutations used by the authors (F17A) and show similar results. Thus, Yang et al. has essentially already described the following data included in this manuscript:  
+
+a. structure of the N:G3BP interface within the NFT2L domain (Fig. 2A)
+
+<--- Page Split --->
+
+b. disruption of N:G3BP1 binding through the F17A mutation (Fig. 2C)  
+c. increased stress granules in cells infected with the F17A mutant compared to WT (Fig. 2D)  
+d. decreased viral replication in vitro in the F17A mutation (Fig. 2G)  
+e. decreased pathogenesis and replication in vivo using a small animal model (Fig. 3B, 3C)  
+f. That G3BP1 facilitates LLPS, which is disrupted by the F17A mutation (Fig. 4)  
+
+While the mutants are not identical (F17A alone in Yang, vs I15A and F17A in the present manuscript) the same conclusions are reached. At minimum, Yang et. al. should be cited. Overall, this reduces the novelty of the research significantly.  
+
+2. Specificity of Mutant: While the underlying conclusions of Yang et. al are concordant with the authors work, there is a major difference in the magnitude of the effects seen. For example, Yang et. al. reports \(\sim 1\) -log reduction in virus replication with their mutant (F17A) but the present study reports \(\sim 2\) -log reduction with theirs (I15A, F17A). The Yang et al. manuscript spends a great deal of time demonstrating that the F17A mutation is highly specific, disrupting SARS2 N:G3BP1/2 binding while not affecting SARS2 N's affinity to any other cellular proteins according to mass spectrometry analysis. Thus, one possible explanation for the differences in the magnitude of phenotypes between the two studies is that the addition of I15A results in off target effects. Control experiments addressing the specificity of the double mutant are needed to alleviate these concerns.  
+
+3. Viral replication in K18-hACE2 mice. Fig. 3 compares the pathogenesis of SARS2 WT vs the I15A, F17A mutant, demonstrating that the I15A, F17A mutation reduces weight loss, H&E staining, and antigen staining relative to WT. The absence of viral lung titer is a striking omission. Viral lung titers between SARS2 WT and the I15A, F17A mutation should be examined to confirm replication differences in vivo.  
+
+4. Translation of viral proteins: As a mentioned in the summary, the authors finding that N re-localizes G3BP1 from stress granules to replication transcription complexes to enhance translation is a major strength and a real step forward for the field. However, when digging into the details of their model, the authors propose that the ribopuromycinylation staining co-localizing with NSP3 indicates enhanced translation of viral genes as the emerge from DMVs. While plausible, this is not adequately supported. Looking at Fig. 2C, the protein levels of N itself are not affected much by the I15A, F17A mutation. This should be explained, or the levels of other viral proteins should be examined. Given the normal abundance of N within infected cells, viral factors translated from longer subgenomic transcripts may be affected more drastically and validate their model.  
+
+Minor:  
+
+1. While the microscopy data presented in Fig.5 and Fig 6 (IF and TEM) is convincing, it would be improved by any sort of quantification that the authors would be able to provide. For instance, the number of ribosomes per DMV in Fig. 6C, etc.  
+
+2. The scale bars on all the IF images throughout the manuscript are too small to read.  
+
+## Citations  
+
+Yang Z, Johnson BA, Meliopoulos VA, Ju X, Zhang P, Hughes MP, Wu J, Koreski KP, Clary JE, Chang TC, Wu G, Hixon J, Duffner J, Wong K, Lemieux R, Lokugamage KG, Alvarado RE, Crocqet-Valdes PA, Walker DH, Plante KS, Plante JA, Weaver SC, Kim HJ, Meyers R, Schultz-Cherry S, Ding Q, Menachery VD, Taylor JP. Interaction between host G3BP and viral nucleocapsid protein regulates SARS-CoV-2 replication and pathogenicity. Cell Rep. 2024 Mar 26;43(3):113965. doi: 10.1016/j.celrep.2024.113965. Epub 2024 Mar 15. PMID: 38492217.  
+
+## Reviewer #2  
+
+(Remarks to the Author) This manuscript addresses the mechanisms by G3BP proteins affect the ability of SARS2 to infect human cells. The work concludes that the G3BP proteins interact with the SARS2 N protein to promote the translation of viral RNAs that are emerging from the membrane bound replication organelles. Although this is potentially an interesting manuscript, as detailed below, significant additional analyses would be required to support the major conclusions.  
+
+This review is from Roy Parker and I would be happy to clarify these comments for the authors if needed.  
+
+## Specific Comments:  
+
+1) The novel conclusion of this work is that a complex of G3BP-N is targeted to the pore complex of double membrane replication organelles to recruit ribosomes to the emerging viral mRNAs. I am not yet convinced of this conclusion given the current data and make the following comments/suggestions. a) The first argument for this function is that viruses with an N protein mutation that blocking interact with G3BP proteins showed reduced replication (similar to what was observed in Yang et al., 2024, Cell Reports). However, it remains possible that G3BP inhibits SARS replication and in the absence of N proteins inhibiting G3BP, there is an inhibitory effect on SARS (as suggested at least in part in Burke et al., 2024, Science Advances). I realize the field is split on this issue, but the authors could clarify this issue by examining WT and RATA SARS replication in WT and g3bpΔ cells. This would clarify whether G3BP is a host factor, or just limits SARS infection, and how those putative roles are affected by interaction with the N protein. Without clearing demonstrating G3BP is a required host factor it is difficult to argue the N-G3BP protein interaction is required for viral growth.  
+
+b) The second argument for this function is that G3BP overlaps with dsRNA and the N protein earlier in infection. This
+
+<--- Page Split --->
+
+experiment could be improved by i) showing the individual channels (at least in supplement) since it is not possible to assess how significant these overlaps are with just one small zoom showing individual channels, and ii) some type of quantification at all the time points. I find it notable that at 12 hours it looks like G3BP is excluded from the area of dsRNA, which is similar to what we observed (Burke et al., 2024, Science Advances). How would exclusion of G3BP from replication areas at this time point fit with the proposed model?  
+
+c) The third argument for this function of GFP-G3BP is the overlaps of dsRNA, N, and the Nsp3 protein (as a marker of replication organelles). I make two comments on this experiment. i) I am concerned this staining pattern could be affected by the GFP tag on G3BP1. I suggest this possibility because we observed a different distribution of GFP-G3BP and untagged G3BP in SAR infected cells (see Figure below). Because of this difference, we avoided the use of GFP tagged G3BP for SARS experiments. At a minimum, the authors need to show this subcellular distribution of proteins is not affected by the GFP tag. ii) In addition, however this experiment is performed (e.g. GFP or IF), the experiment could be improved by showing the individual channels (at least in supplement) and quantifying the extent of co-localization.  
+
+d) The fourth argument for G3BP promoting local translation is that puromycin labeling overlaps with G3BP IF. It was my understanding that puromycin labeling can no longer be relied upon to identify sites of local translation due to rapid diffusion of released peptides, even when using translation elongation inhibitors in conjunction (Enam et al., 2020, eLife). Given this caveat, additional observations would be needed to make a robust conclusion for G3BP marking the sites of translation. (Can you see ribosome clusters overlapping with G3BP by CLEM or EM with gold labeled antibodies?)  
+
+e) The final argument for this function of G3BP is EM imaging showing differences between WT and RATA SARS infection. The key observations are that in RATA mutants and g3bpΔΔ cell lines the ribosomes are more diffuse and less concentrated around the double membrane viral organelles. While a single EM image can be interesting, to make these points robustly, those differences need to be quantified in some manner (beyond just the size of the LVCVs).  
+
+## Additional Issues:  
+
+2) How do the authors identify the infected cells in Figure 1a/b since the N protein IF is dim at 6 hours. Is the cell making stress granules at 6 hours infected? This should be clarified.  
+
+3) The specific model the authors put forth predicts that in the RATA virus (or in G3BPΔΔ cell lines), the translation efficiency of the viral RNAs would be low early in infection. This could be directly tested by measuring the rate of viral protein production and the levels of viral RNAs at the same time point and comparing WT to RATA.  
+
+Figure for authors showing differences in endogenous G3BP1 distribution and the distribution of GFP-G3BP1:  
+
+(Since the website will not let me insert the Figure, I will email it directly to the authors with a copy of my review. Note: I also uploaded to the review site as well.)  
+
+## Reviewer #3  
+
+(Remarks to the Author)  
+
+In this article, Long and colleagues generate a SARS- CoV- 2 mutant (RATA) that lacks the ability to interact with G3BP1/2 proteins, which are RNA- binding proteins that promote stress granule assembly. The RATA mutant virus is attenuated in cell culture models and is less pathogenic in K18- hACE2 transgenic mice. In comparison to WT SARS- CoV- 2, G3BP displays reduced co- localization with dsRNA and N at DMV replication factories in the RATA mutant virus infection. The authors also observe less incorporation of puromycin at DMV in the RATA mutant virus infection, suggesting a reduction in local translation of viral RNA at the DMV. Based on these observations, the authors claim that G3BP1/2 proteins are host factors that are required for SARS- CoV- 2 replication by enhancing localized translation at viral DMV.  
+
+In general, the authors describe interesting cellular biology relating to how SARS- CoV- 2 N modulates G3BP functions during infection, which adds to a growing body of literature. The paper is well- written and the data is of high quality.  
+
+However, their data do not support their primary conclusion that G3BP is a pro- viral protein required for SARS- CoV- 2 replication. Specifically, the observation that SARS- CoV- 2 replicates to similar titers in parental and G3BP- KO cells (data not shown), which is consistent with other studies (Burke et al., RNA 2021 PMID: 34315815), indicates that G3BP is not a pro- viral protein that is required for SARS- CoV- 2 replication. These data strongly argue against their primary model. An alternative explanation for the attenuation of the RATA virus is that the reduction in interaction between G3BP1 and N during RATA mutant virus infection leads to enhanced G3BP1- mediated antiviral activity (i.e., type I IFN response, as suggested in their model) and thus reduce RATA mutant virus replication. However, the authors do not examine if and how the RATA mutant alters antiviral responses, nor do they test if the replication of the RATA mutant would be rescued in G3BP- KO cells as would be expected by this function of N. Because the RATA mutations of N could disrupt a number of putative N functions, it is unclear if altered G3BP interactions contribute to RATA virus attenuation.  
+
+Overall, their data do not support their primary conclusion that G3BP- N interaction is required for optimal viral replication via localization of translation at viral DMV, which weakens the impact of this article.  
+
+Specific comments:  
+
+1. Extended Figure 3D. Most infected cells do not contain G3BP1 complexes in cells infected with either RATA or WT virus. Most cells with G3BP1 granules do not appear to be infected. Are these SGs generated through paracrine signaling from
+
+<--- Page Split --->
+
+infected cells? Also, eIF4G staining is only observed in SARS- CoV- 2 infected cells. Does SARS- CoV- 2 infection lead to an increase in eIF4G, or is this signal spectral crossover from viral N staining?  
+
+2. Burke et al. 2024 (PMID: 38295168) showed that an N-resistant G3BP1 could cause G3BP1 aggregates to form in SARS-CoV-2 infected cells, and that inhibition of eIF4A but not sodium arsenite-induced phosphorylation of eIF2-alpha increased G3BP1 interactions with viral RNA in large aggregates containing viral RNA and dsRNA. Because these findings are similar to those made by the authors in this study, the authors should cite accordingly.  
+
+3. Fig. 1A. It is not clear that the cells with SGs are infected since they lack N protein.  
+
+4. Fig. 2G. The authors show that the SARS-CoV-2-RATA mutant is attenuated, as it replicates to lower titers in several cell types. While the authors claim that this is the result of disrupting G3BP1-N interactions required for maximal viral replication capacity, it could also be that this mutation disrupts normal N functions or disrupts interactions with other host proteins. If the attenuation of the RATA mutant is due to disruption of G3BP1, then knockout of G3BP1 would be expected to reduce SARS-CoV-2-WT virus replication capacity similarly. The authors should test if SARS-CoV-2 replicates to lower titers in G3BP-KO cells in comparison to parental cells, and if so, show that rescue of G3BP1 rescues viral replication. Notably, Knockout of G3BP1/2-KO did not reduce SARS-CoV-2 replication in A549 cells (Burke et al., 2024).  
+
+5. Fig. 2G. If the RATA mutant leads to enhanced antiviral activities of G3BP, then it should replicate to equal titers as WT virus in G3BP-KO cells. The authors should consider testing this.  
+
+6. Lines 182-183. "As expected, N-RATA colocalized neither with G3BP1 nor dsRNA (Fig. 5b), demonstrating that SARS-CoV-2 N recruits G3BP1 to RTC". This statement is misleading based on the data in the figure, as N-RATA does co-localize with dsRNA but not with G3BP1.  
+
+7. Fig. 6C. It is not obvious that ribosomes are reduced near DMV in the RATA mutant compared to WT virus. The authors should quantify this result.  
+
+Version 1:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+The manuscript Long et al. investigates the pro-viral role of G3BP1/2 during SARS-CoV- 2 infection. Previous studies have established that the SARS-CoV- 2 Nucleocapsid (N) protein binds with G3BP1/2, inhibiting stress granule formation (which is believed to be antiviral). However, whether G3BP1/2 is itself pro- or anti-viral is a matter of some debate. To investigate the importance of N:G3BP1/2 binding, the authors use reverse genetics to generate a recombinant virus with a two amino acid substitution in the N protein (N: 115A; F17A). Compared to infection with wild type SARS-CoV- 2, the authors mutant exhibits reduced replication, N:G3BP1/2 binding, and increased stress granule formation.  
+
+Compared to the initial submission, this manuscript has been great improved. All of my major concerns were addressed, specifically 1) my concerns regarding novelty, 2) the specificity of the 115A, F17A mutant compared to F17A alone, 3) the lack of replication data in vivo, and 4) data strengthening their claim regarding an effect on translation of viral proteins. In addition, they have adequately addressed all of my minor critiques.  
+
+As such, I recommend this article is suitable for publication as is.  
+
+Reviewer #2  
+
+(Remarks to the Author)  
+
+In this revised manuscript, the authors have improved the work. I am supportive of publication but recommend some final alterations be made to the manuscript to strengthen the work and remove ambiguities.  
+
+1) An important experiment is how the WT and RATA virus replicate in WT and G3BPΔ cells since this addresses whether G3BP proteins are primarily antiviral or proviral and the nature of the alteration in the RATA mutant. I thank the authors for adding these experiments. The work would be strengthened by clarifying the specific differences observed in g3bpΔ cells. As I look at the figure I observe:  
+
+a) The RATA mutant is 10-50 times less replicative than WT virus in WT cell lines, but only \(\sim 3X\) worse that WT in g3bpΔ cells, consistent with the authors interpretation that the major effects of this mutation are due to G3BP proteins. 
+b) It looks like WT virus is hindered in g3bpΔ cells suggesting G3BP proteins can promote replication, but this effect is not rescued by the reintroduction of G3BP1 (assuming I am interpreting the figure correctly). 
+I suggest: i) The authors clarify what the differences are for WT and RATA virus in the different cell lines and then ii) describe how they interpret those differences for the functional consequences of the N-G3BP interaction.
+
+<--- Page Split --->
+
+2) The authors interpretation that N-G3BP promotes viral translation could be strengthened by quantifying the puromycin labeling experiments on a single cell level (using data they already have). Since most of the translation at this stage will be viral mRNAs, it would be predicted that either RATA in WT cells, or WT virus in g3bpΔΔ cells, should show overall reduced translation rates.  
+
+3) It would be appropriate to at least discuss the alternative model wherein G3BP plays a role in promoting virion packaging and release from the cells (Murigneux et al., Nature Communications, 2024). Could both models be true, or might there be a simpler resolution?  
+
+4) A terrific experiment would be to show immuno-gold localization of G3BP to the DMV with ribosomes on them. I would not require that for publication, but it would really strengthen the work.  
+
+## Reviewer #3  
+
+(Remarks to the Author)  
+
+In this revised manuscript, the authors adequately addressed many reviewer concerns. However, their data in Fig. 3G do not support that G3BP1 is pro- viral during SARS- CoV- 2 infection. Specifically, the authors show that the RATA virus displays a reduced ability to generate plaque forming units in several cell lines, including Vero cells, MA- 104, and U2OS cells. This indicates that the RATA virus is attenuated. However, whether this is due to the inability of N to interact with G3BP1 during RATA infection was unknown.  
+
+To address this, the authors examined growth kinetics of WT and RATA via plaque assays in parental (U2OS- ACE2), G3BP1- KO (GFP), and rescue (GFP- G1- WT) U2OS cell lines. Several observations do not support that G3BP1 is a pro- viral host factor required for SARS- CoV- 2 replication based on data in Fig. 3G:  
+
+1. Knockout of G3BP1 in U2OS cells resulted in higher titers of WT virus by 36 hrs. p.i. Thus, G3BP1 is not required for SARS-CoV-2 replication, but in fact could reduce SARS-CoV-2 replication. Moreover, rescue of GFP-G3BP1 in the G3BP-KO cells (GFP-G1-WT) did not enhance WT SARS-CoV-2 replication kinetics or in increase final PFU titers. This indicates that G3BP1 is not a host factor that enhances SARS-CoV-2 replication.  
+
+2. Knockout of G3BP1 in U2OS cells increased RATA virus replication, suggesting that G3BP1 perturbs RATA virus replication. Notably, RATA virus replication is reduced by GFP-G3BP1 in (GFP-G1-WT) U2OS cell lines.  
+
+3. The RATA virus is attenuated in G3BP1-KO (GFP) cells in comparison to WT cells. This indicates that the mutations in N that lead to attenuation of the RATA virus is G3BP1-independent.  
+
+4. Despite the overall higher replication of WT vs. RATA in all the cell lines, the growth kinetics between WT and RATA viruses is similar between parental (U2OS-ACE2) and G3BP1-KO (GFP) cells.  
+
+In summary, if the interaction between SARS-CoV- 2 N and G3BP1 were required to enhance viral replication, then the authors would have observed a reduction in WT virus replication in G3BP1/2-KO cells to titers equivalent to RATA. Moreover, the reduction in WT virus fitness would be rescued in (GFP-G1-WT) U2OS cell lines. However, the authors do not observe these effects. Instead, they observe that knockout of G3BP1/2 leads to higher replication of both WT and RATA, and that rescuing G3BP1 expression only reduces RATA. Combined, these data indicate that the RATA mutations attenuate SARS-CoV-2 replication independently of G3BP1/2 interactions, and that the slower replicating RATA virus might be more sensitive to the general antiviral effects of G3BP1 (interferon-independent).  
+
+## Additional comments  
+
+Considering the observation that their viral N staining is capable of contaminating other channels, the authors should consider repeating key results with no- primary controls to confirm that any co- localization results are not due to spectral crossover.  
+
+Version 2:  
+
+Reviewer comments:  
+
+Reviewer #2  
+
+(Remarks to the Author) The authors have adequately addressed my comments.  
+
+Reviewer #3
+
+<--- Page Split --->
+
+(Remarks to the Author)  
+
+The manuscript is improved from initial submission, and the authors have addressed many reviewer concerns. Overall, this is a thorough study and thus publication is recommended.  
+
+Open Access This Peer Review 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 cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+The images or other third party material in this Peer Review 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 https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+## Reviewer #1 (Remarks to the Author):  
+
+Summary/Key Results  
+
+The manuscript Long et al. investigates the pro- viral role of G3BP1/2 during SARS- CoV- 2 infection. Previous studies have established that the SARS- CoV- 2 Nucleocapsid (N) protein binds with G3BP1/2, inhibiting stress granule formation (which is believed to be antiviral). However, whether G3BP1/2 is itself pro- or anti- viral is a matter of some debate. To investigate the importance of N:G3BP1/2 binding, the authors use reverse genetics to generate a recombinant virus with a two amino acid substitution in the N protein (N: 115A; F17A). Compared to infection with wild type SARS- CoV- 2, the authors mutant exhibits reduced replication, N:G3BP1/2 binding, and increased stress granule formation.  
+
+The most interesting findings of this article are mechanistic. The authors clearly demonstrate that G3BP1 co- localizes with NSP3 on double membrane vesicles and that this interaction requires N:G3BP1 binding. Furthermore, they provide evidence that this results in enhanced translation at RTCs, evidenced by the co- localization of 40S ribosomal subunits, ribopuromycincylation experiments, and the number of ribosomes localized to DMVs by TEM.  
+
+Overall, this manuscript is both interesting and scientifically sound. In particular, the last 1/3 (Fig. 5 and 6) of the paper linking N:G3BP1 binding with enhanced translation at DMVs is novel. Pro- viral functions of G3BP1 during SARS- CoV- 2 infection have been proposed, but the data presented here are the first to elucidate a concrete mechanism.  
+
+Despite this article's strengths, there are some major weaknesses. First, the overall novelty of the paper is lessened by the fact that a similar article, Yang et al. 2024, has already been published. In Yang et al., the authors produced a single substitution mutant (F17A) SARS- CoV- 2, and described the effects on in vitro replication, G3BP1/2 binding, phase separation, stress granule formation, and in vivo infection. Given the similarity in the mutants (F17A vs I15A/F17A) and that the results and conclusions are almost identical (they differ only in magnitude), most of the data presented in Figures 2 – 4 are merely a confirmation of prior work. At minimum, the authors need to cite this past work.  
+
+In conclusion, the mechanistic work present in the figures 5 and 6 of this manuscript are novel, interesting, and important. However, this novelty is harmed by the fact that the data in Figures 2 – 4 is largely a confirmation of a manuscript of another group, which the authors do not cite. The authors can improve their manuscript by a) emphasizing and expanding upon those findings which are novel and b) citing Yang et al. and addressing the minor points of difference between the two studies.  
+
+We thank the reviewer for the positive comments and valuable suggestions.  
+
+Specific Critiques:  
+
+Major:  
+
+1. Novelty: Yang et al. have already published on a recombinant SARS-CoV-2 mutant that disrupts N:G3BP1/2 binding. Their mutant contains 1 of the 2 mutations used by the authors (F17A) and show similar results. Thus, Yang et al. has essentially already described the following
+
+<--- Page Split --->
+
+data included in this manuscript:  
+
+a. structure of the N:G3BP interface within the NFT2L domain (Fig. 2A)  
+b. disruption of N:G3BP1 binding through the F17A mutation (Fig. 2C)  
+c. increased stress granules in cells infected with the F17A mutant compared to WT (Fig. 2D)  
+d. decreased viral replication in vitro in the F17A mutation (Fig. 2G)  
+e. decreased pathogenesis and replication in vivo using a small animal model (Fig. 3B, 3C)  
+f. That G3BP1 facilitates LLPS, which is disrupted by the F17A mutation (Fig. 4)  
+
+While the mutants are not identical (F17A alone in Yang, vs I15A and F17A in the present manuscript) the same conclusions are reached. At minimum, Yang et. al. should be cited. Overall, this reduces the novelty of the research significantly.  
+
+The points about the decreased novelty of parts of our paper after the publication of the Yang paper are well taken. It is an area of great interest and indeed, some of those discoveries listed had been made before Yang, including (a) by Biswal et al., J Mol Biol 2022, PMID: 35240128 and (b) by Kruse et al., Nature Communications 2021; PMID 34799561; Huang et al., 2021, Cell Discovery, PMID: 34400613: Biswal et al., J Mol Biol 2022). The Yang paper was cited in the original manuscript (reference 23) and we have now cited it more prominently and discussed their data on lines 130- 132 in the revised manuscript.  
+
+A major difference between the studies is our inclusion of the I15A mutation. Our strategy, based on our previous work on viral FG- based motifs binding to G3BP, was to mutate more than one critical residue to confidently predict that no binding would take place in infected cells.  
+
+The NTF2- like domain of G3BP1 features a long binding groove formed by two \(\alpha\) - helices and two \(\beta\) - sheets, comprising a \(5.6 \AA\) wide groove and a \(3.5 \AA\) narrow groove. In our previous structural work, a dual groove- insertion mode was observed in complexes involving alphavirus nsP3/G3BP1- NTF2L (Schulte et al., Open Biology 2016, PMID: 27383630) and the host protein Caprin1/G3BP1- NTF2L (Schulte et al., Open Biology 2023, PMID: 37161291). Taking N for example in this binding mode, the N- F17 aromatic ring inserts into the aromatic cage at the centre of NTF2L binding groove, stabilised by multiple \(\pi\) - stacking interactions, while the bulky hydrophobic side chain of N- I15 inserts into the small groove, coordinated by G3BP residues L10, V11, and P6. Notably, the Caprin1- Y370A mutation (equivalent to N- I15) significantly reduced Caprin1's binding to G3BP- NTF2L (Schulte et al., Open Biology 2016, PMID: 27383630). Therefore, we postulated that both I15 and F17 in N were crucial for its interaction with G3BP- NTF2L and mutating them to alanine could maximally disrupt this interaction.  
+
+In the Biswal and Yang papers, the F17A single mutation abolished binding to N in in vitro assays, but a low but detectable level of binding was detected in a smaller study by Huang et al., 2021, Cell Discovery, PMID: 34400613, see their Fig S1C). We were concerned that in infected cells, where there will be other mechanisms for grouping and concentrating viral and cellular components, weak binding affinities might be compensated by high local concentrations of the relevant proteins.  
+
+Indeed, the data presented by Yang et al, in Figure 6E and F, show that there is weak, but readily detectable colocalisation of G3BP and N- F17A at 24 hours of infection in VeroE6- TMPRSS2 cells, (a detail from their Figure 6F reproduced below in Rebuttal Fig 1). Compare those data with the equivalent experiments from our study with RATA (Figs 6a, c, e, 7a, S3a, S5) where there is negligible colocalisation of N- RATA and G3BP1.
+
+<--- Page Split --->
+
+Rebuttal Figure 1: Data reproduced from Yang et al. Cell Reports, figure 6F with red lines and box added to indicate sites of colocalisation of G3BP1 (yellow), viral genomic RNA (magenta) and N- 17A (cyan).  
+
+This minimal level of recruitment of G3BP by N- F17A is probably insufficient to counteract the antiviral effect, and the conclusions of Yang and colleagues are unaffected. However, it remains possible that a low level of interaction of G3BP and N- F17A, earlier in infection might be able provide some of the proviral effects that might explain the difference in magnitude of the attenuation between F17A (Yang) and I15A+F17A (our work).  
+
+These data, combined with the specificity of the I15A mutation for G3BP (see our response to comment 2, below), reinforces our decision to mutate both residues to create viral mutant that is truly defective for G3BP interaction. We believe this adds to the novelty of our work and hope the reviewer agrees.  
+
+In a further change to strengthen the novelty of our work relative to that of Yang and colleagues, we have moved data using virus variants of concern, from supplement to the main figures (new Fig 2). We show, similarly to Yang et al., that the P13L mutation in the N protein of the omicron lineage causes a slight decrease in the binding to G3BP. However, we extend that work with infectious WA- 1 (P13) and XBB.1.5 (L13) viruses, showing that, despite the slightly decreased binding, no difference in SG inhibition is observed, suggesting it unlikely that this mutation might contribute to reduced virulence of omicron variants, as suggested by Yang et al.  
+
+But, as the reviewer points out, the real novelty in our work lies in the mechanistic work which is "the first to elucidate a concrete mechanism" for G3BP's proviral effects. In response to other comments (below), we have further strengthened those experiments and now believe the work represents an advance to the field. We hope the reviewers and editor agree that the paper is now worthy of publication.  
+
+2. Specificity of Mutant: While the underlying conclusions of Yang et. al are concordant with the authors work, there is a major difference in the magnitude of the effects seen. For example, Yang et. al. reports \(\sim 1\) -log reduction in virus replication with their mutant (F17A) but the present study reports \(\sim 2\) -log reduction with theirs (I15A, F17A). The Yang et al. manuscript spends a great deal of time demonstrating that the F17A mutation is highly specific, disrupting SARS2 N:G3BP1/2 binding while not affecting SARS2 N's affinity to any other cellular proteins according to mass spectrometry analysis. Thus, one possible explanation for the differences in the magnitude of phenotypes between the two studies is that the addition of I15A results in off target effects. Control experiments addressing the specificity of the double mutant are needed to alleviate these concerns.
+
+<--- Page Split --->
+
+We do not believe that the I15A mutation results in off target effects. Our main evidence for this is an experiment performed in our collaborator Jakob Nilsson's lab in Copenhagen, and presented in Kruse et al., Nature Communications 2021; PMID 34799561), figure 2i (reproduced here as Rebuttal Fig 2). The results show that the R14A and I15A double mutant is very specific for G3BP1 and G3BP2. This is in agreement with our observations (mentioned above) that both I15 and F17 are critical for G3BP interaction. Mutation of either residue abolishes G3BP interaction in most experiments, but we chose to mutate both to be sure to exclude any weak interactions.  
+
+for G3BP1 and G3BP2. This is in agreement with our observations (mentioned above) that both I15 and F17 are critical for G3BP interaction. Mutation of either residue abolishes G3BP interaction in most experiments, but we chose to mutate both to be sure to exclude any weak interactions.  
+
+Rebuttal Figure 2: Reproduced from Kruse et al., 2021, figure 2i legend - "Quantitative mass spectrometry analysis of YFP- tagged SARS- CoV- 2 N WT or 2A purified from HeLa cells (n = 4 technical replicates)."  
+
+3. Viral replication in K18-hACE2 mice. Fig. 3 compares the pathogenesis of SARS2 WT vs the I15A, F17A mutant, demonstrating that the I15A, F17A mutation reduces weight loss, H&E staining, and antigen staining relative to WT. The absence of viral lung titer is a striking omission. Viral lung titers between SARS2 WT and the I15A, F17A mutation should be examined to confirm replication differences in vivo.  
+
+We present RT- qPCR data from fixed lung tissue for viral RNA quantification as a measure of viral load in animals from the primary challenge and from the re- challenge. The results, now included in the revised manuscript as Fig 4c and 4f respectively (and reproduced here in Rebuttal Fig 3), clearly show that the RATA mutant is attenuated for replication in vivo (primary challenge) and that mice previously infected with RATA are better able to control WT virus infection (rechallenge)  
+
+We did not save fresh- frozen lung tissue from which to quantify infectious viral lung titres, but we believe repeating the experiment to generate live viral titres would be unnecessary and not be in accordance with the "3R" guiding principles for more ethical use of animals (Reduction).  
+
+![](images/Figure_2.jpg)
+  
+
+Rebuttal Figure 3: RT- qPCR of N and E mRNA expression in mice lungs after primary challenge and after rechallenge.  
+
+4. Translation of viral proteins: As a mentioned in the summary, the authors finding that N relocalizes G3BP1 from stress granules to replication transcription complexes to enhance translation is a major strength and a real step forward for the field. However, when digging into the details of their model, the authors propose that the ribopuromycinylation staining colocalizing with NSP3 indicates enhanced translation of viral genes as the emerge from DMVs. While plausible, this is not adequately supported. Looking at Fig. 2C, the protein levels of N itself are not affected much by the I15A, F17A mutation. This should be explained, or the levels of other viral proteins should be examined. Given the normal abundance of N within infected
+
+<--- Page Split --->
+
+cells, viral factors translated from longer subgenomic transcripts may be affected more drastically and validate their model.  
+
+We thank the reviewer for the positive comments on this aspect of our work and the constructive critique. We have strengthened this aspect of the work in the following ways:  
+
+1. We show that total levels of viral proteins at early stages are lower in RATA infected cells (Fig 7d of the revised manuscript and reproduced here as Rebuttal Fig 4). We restricted these analyses to an early time point since differences in RNA replication and transcription confound analyses at later times. At 6 hours post infection, when viral mRNA levels are equivalent (Fig 7c of revised manuscript), we now show that total levels of spike and N protein are lower in RATA infected cells than WT.  
+
+![PLACEHOLDER_10_0]
+  
+
+Rebuttal Figure 4: VeroE6 cells were infected with SARSCoV- 2 WT or RATA mutant at 0.05 MOI for 6h, and cells were lysed for immunoblotting with indicated antibodies. Representative images from three independent experiments are shown. Quantification of western blot was performed using Image J.  
+
+2. We have quantified the number of ribosomes in association with DMVs in WT and RATA infected cells. This is better described in our response to Reviewer 2, point 1e. These new data are presented in Fig 7g of the revised manuscript.  
+
+Minor:  
+
+1. While the microscopy data presented in Fig.5 and Fig 6 (IF and TEM) is convincing, it would be improved by any sort of quantification that the authors would be able to provide. For instance, the number of ribosomes per DMV in Fig. 6C, etc.  
+
+We have quantified co-localisations (Pearson's coefficient) in several places in the manuscript and quantified the number of ribosomes per DMV in a revised figure 7 and supplementary figure 6. This latter point is better described in response to Reviewer 2, point 1e.  
+
+2. The scale bars on all the IF images throughout the manuscript are too small to read.  
+
+We have improved the presentation of microscopy images throughout the manuscript in response to this and similar comments from other reviewers.
+
+<--- Page Split --->
+
+## Reviewer #2 (Remarks to the Author):  
+
+This manuscript addresses the mechanisms by G3BP proteins affect the ability of SARS2 to infect human cells. The work concludes that the G3BP proteins interact with the SARS2 N protein to promote the translation of viral RNAs that are emerging from the membrane bound replication organelles. Although this is potentially an interesting manuscript, as detailed below, significant additional analyses would be required to support the major conclusions.  
+
+This review is from Roy Parker and I would be happy to clarify these comments for the authors if needed.  
+
+## Specific Comments:  
+
+1) The novel conclusion of this work is that a complex of G3BP-N is targeted to the pore complex of double membrane replication organelles to recruit ribosomes to the emerging viral mRNAs. I am not yet convinced of this conclusion given the current data and make the following comments/suggestions.  
+
+a) The first argument for this function is that viruses with an N protein mutation that blocking interact with G3BP proteins showed reduced replication (similar to what was observed in Yang et al., 2024, Cell Reports). However, it remains possible that G3BP inhibits SARS replication and in the absence of N proteins inhibiting G3BP, there is an inhibitory effect on SARS (as suggested at least in part in Burke et al., 2024, Science Advances). I realize the field is split on this issue, but the authors could clarify this issue by examining WT and RATA SARS replication in WT and g3bpΔΔ cells. This would clarify whether G3BP is a host factor, or just limits SARS infection, and how those putative roles are affected by interaction with the N protein. Without clearing demonstrating G3BP is a required host factor it is difficult to argue the N-G3BP protein interaction is required for viral growth.  
+
+We do not believe the field needs to be split on the issue of G3BP's role in SARS- CoV- 2 (or indeed any) virus infection. We believe that G3BP has both antiviral and proviral functions. In the original submission we had not properly cited your Burke et al paper and we have corrected that now.  
+
+It was not our intention to state that G3BP is solely or primarily a pro- viral factor for SARS- CoV- 2 replication. The proviral function of recruiting translational apparatus to the DMVs may not be a 'required' function, but rather an accessory function that promotes efficient viral gene expression and replication without being critical for viral replication. In contrast, our and others' earlier work showed that G3BP really is critical for chikungunya virus replication (Schulte et al., Open Biology 2016, PMID: 27383630; Kim et al., PLoS Pathogens, 2016, PMID: 27509095; Gotte et al., PLoS Pathogens, PMID: 31199850), and viral replication is undetectable in its absence.  
+
+We have now performed SARS- CoV- 2 WT and RATA viral replication analyses in ΔΔGFP and ΔΔ- GFP- G1- WT cells and present the data in Fig 3g of the revised manuscript. These results are better described in response to a major critique of Reviewer 3.  
+
+We have added some text in the manuscript (lines 67, and 294- 297) to clarify that we believe the proviral effect that we describe is important but not critical for replication.  
+
+b) The second argument for this function is that G3BP overlaps with dsRNA and the N protein earlier in infection. This experiment could be improved by i) showing the individual channels (at least in supplement) since it is not possible to assess how significant these overlaps are with
+
+<--- Page Split --->
+
+just one small zoom showing individual channels, and ii) some type of quantification at all the time points. I find it notable that at 12 hours it looks like G3BP is excluded from the area of dsRNA, which is similar to what we observed (Burke et al., 2024, Science Advances). How would exclusion of G3BP from replication areas at this time point fit with the proposed model?  
+
+The G3BP, dsRNA and N protein colocalisation data are now presented in revised Fig S5a, showing individual channels and colocalisation analyses (Pearson's coefficients). The data now more clearly show that in SARS- CoV- 2 WT infection, G3BP1 is recruited to dsRNA foci at early times (3 and 6 hours), but that it is excluded from those sites at later times (12h (and 24h, not shown)).  
+
+We believe that the data fit very well with our proposed model. The very high affinity binding of N protein with G3BP will likely mean that most N protein molecules synthesised in the first hours of infection will bind to G3BP, leading to its stoichiometric neutralisation and consequent block in SG formation. The N- G3BP complexes accumulate around the DMVs where the proviral functions of G3BP are carried out. Later, as more N proteins continue to be synthesised, becoming one of the more abundant viral proteins in the cell, the majority of the N staining is then associated with progeny virus particle assembly.  
+
+c) The third argument for this function of GFP-G3BP is the overlaps of dsRNA, N, and the Nsp3 protein (as a marker of replication organelles). I make two comments on this experiment. i) I am concerned this staining pattern could be affected by the GFP tag on G3BP1. I suggest this possibility because we observed a different distribution of GFP-G3BP and untagged G3BP in SAR infected cells (see Figure below). Because of this difference, we avoided the use of GFP tagged G3BP for SARS experiments. At a minimum, the authors need to show this subcellular distribution of proteins is not affected by the GFP tag. ii) In addition, however this experiment is performed (e.g. GFP or IF), the experiment could be improved by showing the individual channels (at least in supplement) and quantifying the extent of co-localization.  
+
+i) Thank you for sharing your data on (GFP)-G3BP distribution in SARS-CoV-2 infection. We understand the concern and can report that we do not observe any differences in (EGFP)-G3BP1 localisation in parental U2OS or in the U2OS-ΔΔGFP-G1-WT cells after infection with SARS-CoV-2. To illustrate this, in Rebuttal Fig 5 we provide an image of U2OS-ACE2 cells infected with SARS-CoV-2 at MOI 0.5 for 6 hours. Similar to the EGFP-G3BP1 reporter in U2OS-ΔΔGFP-G1-WT cells (see Fig 6c and e of the revised manuscript), endogenous G3BP1 co-localised very strongly with dsRNA and N protein.  
+
+![PLACEHOLDER_12_0]
+  
+
+Rebuttal Figure 5: U2OS-ACE2 cells were infected with SARS-CoV-2 at 0.5 MOI. Cells were fixed at 6 h and stained for endogenous G3BP1 (green), dsRNA (red) and N (grey).  
+
+It is also worth mentioning that we have also used these cells (before ACE2 lentivirus transduction) in our studies with Semliki Forest virus and chikungunya virus subversion of G3BP functions (Panas et al., PLoS Pathogens 2015, PMID: 25658430; Kedersha et al., JCB 2016, PMID: 27022092; Gotte et al., PLoS Pathogens, PMID: 31199850, Gotte et al., J Virol 2020, PMID:
+
+<--- Page Split --->
+
+31941782) and have always found the EGFP- G3BP1 protein to be a faithful reporter for G3BP localisation to viral RNA replication complexes, in comparison with parental U2OS cells and with other cell lines (BHK, Vero, MEF and others).  
+
+To further strengthen this part of the work, we have also now performed the G3BP, N, and the nsp3 co- immunoprecipitation experiments using parental U2OS- ACE2 cells and Vero cells (both with endogenous G3BP expression) to compare with the data from U2OS- ΔΔGFP- G1- WT cells (EGFP- G3BP1 transgene expression), as presented in the original submission. In all cases, the results show that (GFP)G3BP- N- nsp3 complexes are formed in WT SARS- CoV- 2 infection and that the presence of (GFP)G3BP in those complexes is dependent on the RITFG motif in N protein (disrupted in RATA) - see supplementary fig 5e (U2OS- ACE2 cells), Fig 6f (U2OS- ΔΔGFP- G1- WT cells) and Rebuttal Fig 6 showing the data from Vero cells.  
+
+![PLACEHOLDER_13_0]
+  
+
+Rebuttal Figure 6: VeroE6 cells were infected with WT virus or RATA mutant at 0.01 MOI for 24 h. Cells were lysed and immunoprecipitated with GFP or N antibody for immunoblotting as indicated.  
+
+ii) We have improved the presentation of microscopy images throughout the manuscript in response to this and similar comments from other reviewers. Specifically, we now present the dsRNA, N, and the nsp3 colocalisation data with individual channels in Fig 6e of the revised manuscript.  
+
+d) The fourth argument for G3BP promoting local translation is that puromycin labeling overlaps with G3BP IF. It was my understanding that puromycin labeling can no longer be relied upon to identify sites of local translation due to rapid diffusion of released peptides, even when using translation elongation inhibitors in conjunction (Enam et al., 2020, eLife). Given this caveat, additional observations would be needed to make a robust conclusion for G3BP marking the sites of translation. (Can you see ribosome clusters overlapping with G3BP by CLEM or EM with gold labeled antibodies?)  
+
+We are aware of the observations presented in the Enam paper, but we do not believe this to be a confounding issue in our experiments. In our case, we are comparing two similar conditions - localised translation at the DMVs in the presence or absence of G3BP. For the "Enam diffusion" to be a confounding issue, it would have to be occurring in the absence of G3BP (both in the KO cells infected with WT virus and in WT cells infected with RATA), but not in the presence of G3BP (in WT cells infected with WT virus). We consider this improbable.  
+
+Nevertheless, to strengthen this aspect of the work, we have performed the following experiments:  
+
+i) Ribopuromycincylation experiment with only 2 minutes (reduced from 5 minutes in the original submission) of puromycin labelling to restrict potential diffusion of PMY-labelled peptides. These new data confirm that PMY labelling is strong at SARS-CoV-2 dsRNA+ foci, but only when G3BP is recruited there by N-WT. The data are presented in Fig 7a of the revised manuscript. Additionally, we now include data from analyses of correlations of G3BP1-PMY, or G3BP1-N, (Pearson's coefficients) and PMY maximum intensities in Fig 7b.
+
+<--- Page Split --->
+
+ii) To control for mRNA levels at the time of PMY labelling, we have also included qPCR analyses of viral mRNA levels in the same conditions as the PMY experiment. This shows that WT and RATA viral mRNA levels are equivalent at the time of PMY labelling, excluding that the effect could be due to differences in mRNA template availability. These new data are presented in Fig 7c of the revised manuscript.  
+
+iii) We now also show that total levels of viral spike and N proteins are affected by the recruitment of G3BP at early times post infection (6h). The data are presented in Fig 7d of the revised manuscript and are better described in response to Reviewer 1, point 4, above.  
+
+iv) We have quantified the number of ribosomes in association with DMVs in the presence and absence of G3BP, described better in response to the following comment.  
+
+e) The final argument for this function of G3BP is EM imaging showing differences between WT and RATA SARS infection. The key observations are that in RATA mutants and g3bpΔΔ cell lines the ribosomes are more diffuse and less concentrated around the double membrane viral organelles. While a single EM image can be interesting, to make these points robustly, those differences need to be quantified in some manner (beyond just the size of the LVCVs).  
+
+Indeed, our apparent reliance on a single image was a weak point in the original submission, and we are grateful for the opportunity to improve this aspect of the work. The originally presented image had in fact been chosen as representative of over 15 images taken. We have now taken more images of U2OS- ACE2 cells infected with WT or RATA mutant virus and performed blinded quantifications of the numbers of ribosomes associated with DMV membranes in those images. Specifically, images of 66 DMVs in WT and 57 in RATA- infected U2OS- ACE2 cells were shuffled and labelled in alphabetical order (by SL, first author) for quantification of ribosomes per DMV and measurement of DMV circumference (by MG, second author). Final numbers were collated and expressed as the number of ribosomes per \(\mu \mathrm{m}\) of DMV perimeter. The results show that there is a significant decrease in the number of ribosomes in association with DMVs in RATA infection compared to WT. Similar analyses were performed on images from infected U2OS- ΔΔGFP cells and show that, in the absence of G3BP1/2, the difference in number of ribosomes/DMVs is eliminated. These data are now presented in Fig 7g and S6a of the revised manuscript.  
+
+We believe this is strong evidence that the recruitment of G3BP to DMVs by the N protein, leads to more efficient translation of nascent viral mRNAs. Raw data are uploaded as source data file with this submission.  
+
+## Additional Issues:  
+
+2) How do the authors identify the infected cells in Figure 1a/b since the N protein IF is dim at 6 hours. Is the cell making stress granules at 6 hours infected? This should be clarified.  
+
+We believe those cells are infected but are not yet showing strong detectable signal for the N protein. In a new experiment, performed under the same conditions, we could detect dsRNA at earlier stage than N protein. To illustrate this, we include below an image of VeroE6 cells infected with WT SARS- CoV- 2 at 0.5 MOI for 6 hours and stained for G3BP1, dsRNA and N (Rebuttal Fig 7). The field captures cells at different early stages of infection. The indicated cell shows clear SGs, but still very low N protein signal, while the cells above left and above right have strong and intermediate N signals respectively but neither contain any SGs.
+
+<--- Page Split --->
+![PLACEHOLDER_15_0]
+  
+
+Rebuttal Figure 7: Vero E6 cells were infected with WT SARS- CoV- 2 at 0.5 MOl for 6 hours. Cells were fixed and stained for G3BP1 (green), dsRNA (red), N (grey) and Hoechst (blue).  
+
+We have better explained this on line 72 of the revised manuscript.  
+
+3) The specific model the authors put forth predicts that in the RATA virus (or in G3BPΔΔ cell lines), the translation efficiency of the viral RNAs would be low early in infection. This could be directly tested by measuring the rate of viral protein production and the levels of viral RNAs at the same time point and comparing WT to RATA.  
+
+Please see Reviewer 1, point 4, above for a full description of our new data to address this point.
+
+<--- Page Split --->
+
+## Reviewer #3 (Remarks to the Author):  
+
+In this article, Long and colleagues generate a SARS- CoV- 2 mutant (RATA) that lacks the ability to interact with G3BP1/2 proteins, which are RNA- binding proteins that promote stress granule assembly. The RATA mutant virus is attenuated in cell culture models and is less pathogenic in K18- hACE2 transgenic mice. In comparison to WT SARS- CoV- 2, G3BP displays reduced colocalization with dsRNA and N at DMV replication factories in the RATA mutant virus infection. The authors also observe less incorporation of puromycin at DMV in the RATA mutant virus infection, suggesting a reduction in local translation of viral RNA at the DMV. Based on these observations, the authors claim that G3BP1/2 proteins are host factors that are required for SARS- CoV- 2 replication by enhancing localized translation at viral DMV.  
+
+In general, the authors describe interesting cellular biology relating to how SARS- CoV- 2 N modulates G3BP functions during infection, which adds to a growing body of literature. The paper is well- written and the data is of high quality.  
+
+However, their data do not support their primary conclusion that G3BP is a pro- viral protein required for SARS- CoV- 2 replication. Specifically, the observation that SARS- CoV- 2 replicates to similar titers in parental and G3BP- KO cells (data not shown), which is consistent with other studies (Burke et al., RNA 2021 PMID: 34315815), indicates that G3BP is not a pro- viral protein that is required for SARS- CoV- 2 replication. These data strongly argue against their primary model. An alternative explanation for the attenuation of the RATA virus is that the reduction in interaction between G3BP1 and N during RATA mutant virus infection leads to enhanced G3BP1- mediated antiviral activity (i.e., type I IFN response, as suggested in their model) and thus reduce RATA mutant virus replication. However, the authors do not examine if and how the RATA mutant alters antiviral responses, nor do they test if the replication of the RATA mutant would be rescued in G3BP- KO cells as would be expected by this function of N. Because the RATA mutations of N could disrupt a number of putative N functions, it is unclear if altered G3BP interactions contribute to RATA virus attenuation.  
+
+Overall, their data do not support their primary conclusion that G3BP- N interaction is required for optimal viral replication via localization of translation at viral DMV, which weakens the impact of this article.  
+
+We thank the reviewer for the comprehensive critique of our work. Below, we provide point- by- point responses to each of the concerns.  
+
+Importantly, we would like to stress that we believe that G3BP is both an antiviral and a proviral factor. G3BP- dependent SGs are induced very early in infection, before viral protein production is detectable (Fig 1). Later, when N protein sequesters G3BP to RNA replication complexes, the protein then carries out its proviral functions.  
+
+Despite their economical genome coding and preponderance of multifunctional proteins, RNA viruses typically require multiple host factor interactions to carry out functions that cannot be coded for in the relatively small genomes. If one considers the G3BP interaction in this context, it makes sense that the virus sequesters an antiviral protein and uses it for proviral functions.  
+
+We address the major critiques in the reviewer's penultimate paragraph:  
+
+- "the authors do not examine if and how the RATA mutant alters antiviral responses"
+
+<--- Page Split --->
+
+We have now performed qPCR analyses of total mRNA from U2OS- ACE2 cells infected with SARS- CoV- 2 WT or RATA for 12 and 24 hours. The data indicate that IFN transcripts are equal (12h) or lower in number (24h) in RATA compared to WT infected cells. We therefore do not believe that the recruitment of G3BP by N protein in WT SARS- CoV- 2 infection can be implicated in evasion of the IFN response in infected cells, nor that this could be the molecular mechanism of RATA attenuation. These data are now included in Fig S3e of the revised manuscript.  
+
+Further support for this is the observation that the RATA mutant is also attenuated in Vero cells, which lack the ability to produce interferon (Emeny and Morgan, J Gen Virol, 1979, PMID 113494).  
+
+- "nor do they test if the replication of the RATA mutant would be rescued in G3BP-KO cells"  
+
+In response to the suggestion here, and in points 4 and 5 (below), of testing viral replication in G3BP- KO relative to parental cells, we have now performed WT and RATA viral replication analyses in U2OS- ΔΔGFP and U2OS- ΔΔGFP- G1- WT cells and present those data in Fig 3g of the revised manuscript and in Rebuttal Fig 8 below.  
+
+Firstly, to reduce experimental noise caused by unequal expression of EGFP- G3BP and of viral receptor ACE2, we have re- sorted the cell lines based on EGFP expression and on ACE2 expression before performing these analyses. In the interpretation of these data, we would remind the reviewer that we believe that G3BP has both antiviral and proviral effects and, since both are acting on viral replication in such experiment, we would urge caution in drawing conclusions from such a simple readout (infectious viral titre in extracellular medium).  
+
+The data show that in U2OS- ΔΔGFP- G1- WT cells, RATA is greatly attenuated relative to WT virus. This we believe is due to both antiviral effects of G3BP SGs and also the lack of the proviral effect of recruiting translational machinery to DMVs. However, in U2OS- ΔΔGFP (lacking G3BP), RATA replicated to titres much closer to WT, due to the absence of the antiviral effect and proviral effects.  
+
+![PLACEHOLDER_17_0]
+  
+
+Rebuttal Figure 8: U2OS- ΔΔGFP and U2OS- ΔΔGFP- G1- WT cells were sorted for EGFP and ACE2 expression (see methods) and infected with SARS- CoV- 2 WT or RATA at 0.05 MOI. Samples were taken at indicated times and titrated by plaque assay on VeroE6 cells.  
+
+## Specific comments:  
+
+1. Extended Figure 3D. Most infected cells do not contain G3BP1 complexes in cells infected with either RATA or WT virus. Most cells with G3BP1 granules do not appear to be infected. Are these SGs generated through paracrine signaling from infected cells? Also, eIF4G staining is only observed in SARS-CoV-2 infected cells. Does SARS-CoV-2 infection lead to an increase in eIF4G, or is this signal spectral crossover from viral N staining?
+
+<--- Page Split --->
+
+The reviewer is correct that elF4G signal in the relevant image was contaminated with N signal. We had not noticed this and are grateful to the reviewer for pointing it out. We believe the problem arose from our use of an old aliquot of the Santa Cruz anti- elF4G antibody (sc- 133155). We have now repeated this experiment using a new batch of the same antibody and present the data in Fig S3a of the revised manuscript.  
+
+2. Burke et al. 2024 (PMID: 38295168) showed that an N-resistant G3BP1 could cause G3BP1 aggregates to form in SARS-CoV-2 infected cells, and that inhibition of elF4A but not sodium arsenite-induced phosphorylation of elF2-alpha increased G3BP1 interactions with viral RNA in large aggregates containing viral RNA and dsRNA. Because these findings are similar to those made by the authors in this study, the authors should cite accordingly.  
+
+The lack of citation of the Burke et al paper was an embarrassing oversight on our part. We have corrected this now (reference 39). Their work is discussed on lines 198- 200.  
+
+3. Fig. 1A. It is not clear that the cells with SGs are infected since they lack N protein.  
+
+We believe those cells are in a very early stage of infection and are not yet showing detectable signal for the N protein. Please see our response to Reviewer 2, point 2 for a fuller explanation.  
+
+4. Fig. 2G. The authors show that the SARS-CoV-2-RATA mutant is attenuated, as it replicates to lower titers in several cell types. While the authors claim that this is the result of disrupting G3BP1-N interactions required for maximal viral replication capacity, it could also be that this mutation disrupts normal N functions or disrupts interactions with other host proteins. If the attenuation of the RATA mutant is due to disruption of G3BP1, then knockout of G3BP1 would be expected to reduce SARS-CoV-2-WT virus replication capacity similarly. The authors should test if SARS-CoV-2 replicates to lower titers in G3BP-KO cells in comparison to parental cells, and if so, show that rescue of G3BP1 rescues viral replication. Notably, Knockout of G3BP1/2-KO did not reduce SARS-CoV-2 replication in A549 cells (Burke et al., 2024).  
+
+Firstly, we do not believe that the RATA mutation affects interactions with other proteins. This is based on previous work (Kruse et al., Nature Communications 2021; PMID 34799561), where the RATA mutant was originally described (in protein expression constructs, not in replicating  
+
+virus). In that study, quantitative proteomics from GFP-pulldown experiments (reproduced here in Rebuttal Fig 9, left), revealed that the RATA mutation (there named '2A') was highly specific for G3BP1 and 2 binding.  
+
+Rebuttal Figure 9: Quantitative mass spectrometry comparison of YFP tagged SARS-CoV N wt and 2A (RATA) purified from HeLa cells (data reproduced from Kruse et al., 2021, Fig S3c).  
+
+Viral replication assays in the G3BP KO cells are shown in Rebuttal Fig 8, above. Of note, the data presented by Burke et al., and mentioned by the reviewer, show viral titres at only one time point (24h post- infection), at which time, replication should be expected to still be in the logarithmic phase and not yet reached plateau (compare with our data in Vero cells in Fig 3).
+
+<--- Page Split --->
+
+5. Fig. 2G. If the RATA mutant leads to enhanced antiviral activities of G3BP, then it should replicate to equal titers as WT virus in G3BP-KO cells. The authors should consider testing this.  
+
+Viral replication assays presented above, now reveal that indeed the RATA mutant replicates to titres much closer to the WT virus in the U2OS- ΔΔGFP cells.  
+
+6. Lines 182-183. "As expected, N-RATA colocalized neither with G3BP1 nor dsRNA (Fig. 5b), demonstrating that SARS-CoV-2 N recruits G3BP1 to RTC". This statement is misleading based on the data in the figure, as N-RATA does co-localize with dsRNA but not with G3BP1.  
+
+We have corrected this mistake (now on lines 200- 202) and thank the review for pointing it out.  
+
+7. Fig. 6C. It is not obvious that ribosomes are reduced near DMV in the RATA mutant compared to WT virus. The authors should quantify this result.  
+
+We have now performed this quantification and present the results in Fig 7g of the revised manuscript. For a fuller discussion, please see our response to Reviewer 2, comment 1e.
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+## Reviewer #1 (Remarks to the Author):  
+
+The manuscript Long et al. investigates the pro- viral role of G3BP1/2 during SARS- CoV- 2 infection. Previous studies have established that the SARS- CoV- 2 Nucleocapsid (N) protein binds with G3BP1/2, inhibiting stress granule formation (which is believed to be antiviral). However, whether G3BP1/2 is itself pro- or anti- viral is a matter of some debate. To investigate the importance of N:G3BP1/2 binding, the authors use reverse genetics to generate a recombinant virus with a two amino acid substitution in the N protein (N: 115A; F17A). Compared to infection with wild type SARS- CoV- 2, the authors mutant exhibits reduced replication, N:G3BP1/2 binding, and increased stress granule formation.  
+
+Compared to the initial submission, this manuscript has been great improved. All of my major concerns were addressed, specifically 1) my concerns regarding novelty, 2) the specificity of the 115A, F17A mutant compared to F17A alone, 3) the lack of replication data in vivo, and 4) data strengthening their claim regarding an effect on translation of viral proteins. In addition, they have adequately addressed all of my minor critiques.  
+
+As such, I recommend this article is suitable for publication as is.  
+
+Again, we thank the reviewer for the comprehensive review of our work and for this positive recommendation.
+
+<--- Page Split --->
+
+## Reviewer #2 (Remarks to the Author):  
+
+In this revised manuscript, the authors have improved the work. I am supportive of publication but recommend some final alterations be made to the manuscript to strengthen the work and remove ambiguities.  
+
+1) An important experiment is how the WT and RATA virus replicate in WT and G3BPΔΔ cells since this addresses whether G3BP proteins are primarily antiviral or proviral and the nature of the alteration in the RATA mutant. I thank the authors for adding these experiments. The work would be strengthened by clarifying the specific differences observed in g3bpΔΔ cells. As I look at the figure I observe:  
+
+a) The RATA mutant is 10-50 times less replicative than WT virus in WT cell lines, but only \(\sim 3X\) worse that WT in g3bpΔΔ cells, consistent with the authors interpretation that the major effects of this mutation are due to G3BP proteins.  
+
+b) It looks like WT virus is hindered in g3bpΔΔ cells suggesting G3BP proteins can promote replication, but this effect is not rescued by the reintroduction of G3BP1 (assuming I am interpreting the figure correctly).  
+
+I suggest: i) The authors clarify what the differences are for WT and RATA virus in the different cell lines and then ii) describe how they interpret those differences for the functional consequences of the N-G3BP interaction.  
+
+We refer the reviewer to our response to Reviewer 3, below.  
+
+2) The authors interpretation that N-G3BP promotes viral translation could be strengthened by quantifying the puromycin labeling experiments on a single cell level (using data they already have). Since most of the translation at this stage will be viral mRNAs, it would be predicted that either RATA in WT cells, or WT virus in g3bpΔΔ cells, should show overall reduced translation rates.  
+
+In the revised submission, we had included such analyses of the puromycin labelling data in VeroE6 cells infected with WT or RATA (manuscript Fig 7b and in Rebuttal Fig 1A right, below). Since we have observed that the effect on translation is very localised to the sites of viral RNA replication, we had presented maximum PMY intensity/cell since we believe it to be a more appropriate measure than mean intensity/cell, which averages signal over the whole cell. For the reviewer's interest, we here present the mean intensity analyses of the same data (Rebuttal Fig 1A, left). We observe a slightly lower although non-significant mean signal in RATA infected cells compared to WT.  
+
+Furthermore, we have now also similarly analysed the images from WT or RATA infected ΔΔ- GFP- G1- WT cells and included those data in a revised Supplementary Fig 7b and presented in here in Rebuttal Fig 1B. Indeed, the results strengthen our interpretation that viral mRNAs are more efficiently translated in WT than RATA.
+
+<--- Page Split --->
+![PLACEHOLDER_22_0]
+  
+
+Rebuttal Figure 1. Mean and maximum intensity of puromycin staining were calculated in CellProfiler for SARS- CoV- 2 WT or RATA infected VeroE6 (A) or (B) \(\Delta \Delta \mathrm{GFP - G1 - WT}\) cells.  
+
+We also bring to the reviewer's attention, data added in the previous version in response to reviewer 1, point 4. Those data show that whole cell lysates from VeroE6 cells, infected with SARS- CoV- 2 WT (MOI 0.05 for 6h) contain more newly produced Spike and N protein than cells infected with the RATA mutant, despite equivalent viral mRNA levels (Fig 7c,d of the revised manuscript). These data also support that WT viral mRNAs are more efficiently translated than RATA.  
+
+3) It would be appropriate to at least discuss the alternative model wherein G3BP plays a role in promoting virion packaging and release from the cells (Murigneux et al., Nature Communications, 2024). Could both models be true, or might there be a simpler resolution?  
+
+We discussed that on lines 300- 305 of the first revised manuscript and depicted it in Figure 8. Indeed, we believe that both models can be true. Murigneux and colleagues propose (last paragraph of their discussion) "a dual functionality of N/G3BP interactions: on the one hand N sequesters G3BP proteins to prevent antiviral SG formation and to circumvent subsequent antiviral immune responses and on the other hand, the virus hijacks the function of G3BP1/2 to favor production of infectious viral particle". Our work supports that and adds the extra function of facilitating viral mRNA translation at the sites of mRNA production, as depicted in Figure 8.  
+
+4) A terrific experiment would be to show immuno-gold localization of G3BP to the DMV with ribosomes on them. I would not require that for publication, but it would really strengthen the work.  
+
+We agree that this could be very nice to show, but unfortunately, we did not feel that we had the time or resources to do it at this time.
+
+<--- Page Split --->
+
+## Reviewer #3 (Remarks to the Author):  
+
+In this revised manuscript, the authors adequately addressed many reviewer concerns. However, their data in Fig. 3G do not support that G3BP1 is pro- viral during SARS- CoV- 2 infection. Specifically, the authors show that the RATA virus displays a reduced ability to generate plaque forming units in several cell lines, including Vero cells, MA- 104, and U2OS cells. This indicates that the RATA virus is attenuated. However, whether this is due to the inability of N to interact with G3BP1 during RATA infection was unknown.  
+
+To address this, the authors examined growth kinetics of WT and RATA via plaque assays in parental (U2OS- ACE2), G3BP1- KO (AGFP), and rescue (AGFP- G1- WT) U2OS cell lines. Several observations do not support that G3BP1 is a pro- viral host factor required for SARS- CoV- 2 replication based on data in Fig. 3G:  
+
+1. Knockout of G3BP1 in U2OS cells resulted in higher titers of WT virus by 36 hrs. p.i. Thus, G3BP1 is not required for SARS-CoV-2 replication, but in fact could reduce SARS-CoV-2 replication. Moreover, rescue of GFP-G3BP1 in the G3BP-KO cells (AGFP-G1-WT) did not enhance WT SARS-CoV-2 replication kinetics or in increase final PFU titers. This indicates that G3BP1 is not a host factor that enhances SARS-CoV-2 replication.  
+
+2. Knockout of G3BP1 in U2OS cells increased RATA virus replication, suggesting that G3BP1 perturbs RATA virus replication. Notably, RATA virus replication is reduced by GFP-G3BP1 in (AGFP-G1-WT) U2OS cell lines.  
+
+3. The RATA virus is attenuated in G3BP1-KO (AGFP) cells in comparison to WT cells. This indicates that the mutations in N that lead to attenuation of the RATA virus is G3BP1-independent.  
+
+4. Despite the overall higher replication of WT vs. RATA in all the cell lines, the growth kinetics between WT and RATA viruses is similar between parental (U2OS-ACE2) and G3BP1-KO (AGFP) cells.  
+
+In summary, if the interaction between SARS-CoV- 2 N and G3BP1 were required to enhance viral replication, then the authors would have observed a reduction in WT virus replication in G3BP1/2-KO cells to titers equivalent to RATA. Moreover, the reduction in WT virus fitness would be rescued in (AGFP-G1-WT) U2OS cell lines. However, the authors do not observe these effects. Instead, they observe that knockout of G3BP1/2 leads to higher replication of both WT and RATA, and that rescuing G3BP1 expression only reduces RATA. Combined, these data indicate that the RATA mutations attenuate SARS-CoV-2 replication independently of G3BP1/2 interactions, and that the slower replicating RATA virus might be more sensitive to the general antiviral effects of G3BP1 (interferon-independent).  
+
+Indeed, we agree with the reviewer that we cannot claim that G3BP is exclusively a proviral factor. As depicted in Figure 8, we believe that G3BP has both antiviral and proviral functions. The specific function of G3BP is determined by the proteins or RNA with which it interacts. G3BP is antiviral when it is free to induce SG formation, leading to arrest of viral protein translation. However, SARS- CoV- 2 and several other viruses (see citations in the
+
+<--- Page Split --->
+
+manuscript text), sequester the protein to their replication complexes in a way that both inhibits the antiviral functions and drives proviral functions at the sites of mRNA transcription and assembly. This view is shared by Murigneux and colleagues (Nature Communications, 2024 PMID 38245532), quoted above, and our work adds the proviral effect of efficient translation of WT viral mRNAs.  
+
+We have extensively edited the text of the paper, including title and abstract to better express the view that G3BP is both antiviral and proviral.  
+
+In our first revision, we requested caution in the analyses of viral growth replication data in these cell lines and we repeat that here. We believe it is not appropriate to compare viral replication curves across the different cell lines but rather it is better to compare WT and RATA virus replication in each cell line separately. The CRISPR knock out cells and GFP- (G3BP1) reconstitutions were generated 10 years ago, and the cells have been passaged independently since then. In addition, in 2021, all 3 cell lines (U2OS parental, \(\Delta \Delta \mathrm{GFP}\) and \(\Delta \Delta \mathrm{G3BP1 - GFP}\) ) were transduced with ACE2- TMPRSS2- expressing lentiviruses, placed under selection and passaged independently. We have tried to minimise variation caused by unequal expression of these transgenes, but the possibility remains that the viruses might be better able to enter and initiate infection in one cell line than the others.  
+
+Comparing the two viruses, we observe that RATA exhibits 33- fold lower replication compared to WT in U2OS parental cells, 25- fold lower replication compared to WT in \(\Delta \Delta \mathrm{GFP - G1 - WT}\) cells, but only a 4- fold reduction compared to WT in cells lacking G3BP1/2 ( \(\Delta \Delta \mathrm{GFP}\) cells). We believe therefore that the attenuation is largely G3BP1- dependent and is a result of the combination of the antiviral effects of SGs and the loss of G3BP's proviral effects. However, as the reviewer points out (point #3), the 4- fold reduction in \(\Delta \Delta \mathrm{GFP}\) cells suggests some G3BP- independent attenuation of RATA, which we cannot exclude. We discuss this on lines 144- 149 of the revised manuscript.  
+
+We do not believe that RATA is a "slower replicating" virus, since viral RNA levels were equal at 6hpi (Fig 7c), but that the major attenuation effects are seen downstream of the translation of viral mRNAs. Further, we have been careful to define the specificity of the RATA mutation for G3BP (see comment to Reviewer 1, point 2 in first revision) and to show that the mutation does not alter RNA- binding and LLPS properties (figure 5), as well as the binding to other viral structural proteins for assembly (Nature Communications, 2021, PMID: 34799561). Of note, a recent study found that point mutations in the N protein's intrinsically disordered regions (IDRs) can have "nonlocal impact and modulate thermodynamic stability, secondary structure, protein oligomeric state, particle formation, and liquid- liquid phase separation" (Nguyen et al Elife 2024; PMID: 38941236), a possibility we now mention in the text.  
+
+## Additional comments  
+
+Considering the observation that their viral N staining is capable of contaminating other channels, the authors should consider repeating key results with no- primary controls to confirm that any co- localization results are not due to spectral crossover.
+
+<--- Page Split --->
+
+To respond to the reviewer's suggestion, we here present images of U2OS- ACE2 cells infected with SARS- CoV- 2 WT or RATA and stained with a combination of antibodies against G3BP1, dsRNA and N protein (Rebuttal Fig 2, top left), or the same lacking either G3BP1 (bottom left), dsRNA (top right) or N (bottom right). The data show that, using the same antibodies, techniques and hardware used in the paper, no spectral crossover was detected.  
+
+We are now confident that the fluorescence signals in all images are uncontaminated by other channels. For example, to illustrate that N signal is not contaminating other channels, one might examine images from the manuscript of RATA infected cells, where N (Alexa Fluor 647) does not co- stain with G3BP1 (Alexa Fluor 488; Fig 6a), GFP/elF4A (GFP or Alexa Fluor 568; Fig 6c), or GFP (Fig 6e).  
+
+![PLACEHOLDER_25_0]
+
+<center>Rebuttal Figure 2. U2OS-ACE2 cells were infected with SARS-CoV-2 WT MOI 0.5 and cells were fixed and stained with the indicated antibody combinations at 6 hours post infection. </center>
+
+<--- Page Split --->

peer_reviews/supplementary_0_Transparent Peer Review file__ea3baab8c78d31e3169d6916ccba0f903a7ba7d2977a3fd23353f046862522cf/supplementary_0_Transparent Peer Review file__ea3baab8c78d31e3169d6916ccba0f903a7ba7d2977a3fd23353f046862522cf_det.mmd

@@ -0,0 +1,814 @@
+<|ref|>title<|/ref|><|det|>[[72, 53, 296, 80]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[74, 97, 296, 119]]<|/det|>
+Peer Review File  
+
+<|ref|>title<|/ref|><|det|>[[73, 161, 880, 210]]<|/det|>
+# SARS-CoV-2 N protein recruits G3BP to double membrane vesicles to promote translation of viral mRNAs  
+
+<|ref|>text<|/ref|><|det|>[[73, 224, 520, 241]]<|/det|>
+Corresponding Author: Professor Gerald Mclnerney  
+
+<|ref|>text<|/ref|><|det|>[[70, 274, 864, 289]]<|/det|>
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+<|ref|>text<|/ref|><|det|>[[73, 327, 144, 341]]<|/det|>
+Version 0:  
+
+<|ref|>text<|/ref|><|det|>[[73, 354, 219, 368]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[73, 380, 160, 393]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[73, 405, 238, 432]]<|/det|>
+(Remarks to the Author) Summary/Key Results  
+
+<|ref|>text<|/ref|><|det|>[[72, 443, 910, 524]]<|/det|>
+The manuscript Long et al. investigates the pro- viral role of G3BP1/2 during SARS- CoV- 2 infection. Previous studies have established that the SARS- CoV- 2 Nucleocapsid (N) protein binds with G3BP1/2, inhibiting stress granule formation (which is believed to be antiviral). However, whether G3BP1/2 is itself pro- or anti- viral is a matter of some debate. To investigate the importance of N:G3BP1/2 binding, the authors use reverse genetics to generate a recombinant virus with a two amino acid substitution in the N protein (N: 115A; F17A). Compared to infection with wild type SARS- CoV- 2, the authors mutant exhibits reduced replication, N:G3BP1/2 binding, and increased stress granule formation.  
+
+<|ref|>text<|/ref|><|det|>[[72, 535, 907, 589]]<|/det|>
+The most interesting findings of this article are mechanistic. The authors clearly demonstrate that G3BP1 co- localizes with NSP3 on double membrane vesicles and that this interaction requires N:G3BP1 binding. Furthermore, they provide evidence that this results in enhanced translation at RTCs, evidenced by the co- localization of 40S ribosomal subunits, ribopuromycinylation experiments, and the number of ribosomes localized to DMVs by TEM.  
+
+<|ref|>text<|/ref|><|det|>[[72, 600, 897, 641]]<|/det|>
+Overall, this manuscript is both interesting and scientifically sound. In particular, the last 1/3 (Fig. 5 and 6) of the paper linking N:G3BP1 binding with enhanced translation at DMVs is novel. Pro- viral functions of G3BP1 during SARS- CoV- 2 infection have been proposed, but the data presented here are the first to elucidate a concrete mechanism.  
+
+<|ref|>text<|/ref|><|det|>[[72, 653, 912, 732]]<|/det|>
+Despite this article's strengths, there are some major weaknesses. First, the overall novelty of the paper is lessened by the fact that a similar article, Yang et al. 2024, has already been published. In Yang et al., the authors produced a single substitution mutant (F17A) SARS- CoV- 2, and described the effects on in vitro replication, G3BP1/2 binding, phase separation, stress granule formation, and in vivo infection. Given the similarity in the mutants (F17A vs I15A/F17A) and that the results and conclusions are almost identical (they differ only in magnitude), most of the data presented in Figures 2 – 4 are merely a confirmation of prior work. At minimum, the authors need to cite this past work.  
+
+<|ref|>text<|/ref|><|det|>[[72, 744, 914, 810]]<|/det|>
+In conclusion, the mechanistic work present in the figures 5 and 6 of this manuscript are novel, interesting, and important. However, this novelty is harmed by the fact that the data in Figures 2 – 4 is largely a confirmation of a manuscript of another group, which the authors do not cite. The authors can improve their manuscript by a) emphasizing and expanding upon those findings which are novel and b) citing Yang et al. and addressing the minor points of difference between the two studies.  
+
+<|ref|>text<|/ref|><|det|>[[72, 822, 199, 836]]<|/det|>
+Specific Critiques:  
+
+<|ref|>text<|/ref|><|det|>[[72, 848, 118, 861]]<|/det|>
+Major:  
+
+<|ref|>text<|/ref|><|det|>[[72, 874, 890, 914]]<|/det|>
+1. Novelty: Yang et al. have already published on a recombinant SARS-CoV-2 mutant that disrupts N:G3BP1/2 binding. Their mutant contains 1 of the 2 mutations used by the authors (F17A) and show similar results. Thus, Yang et al. has essentially already described the following data included in this manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[72, 926, 556, 940]]<|/det|>
+a. structure of the N:G3BP interface within the NFT2L domain (Fig. 2A)
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 48, 707, 112]]<|/det|>
+b. disruption of N:G3BP1 binding through the F17A mutation (Fig. 2C)  
+c. increased stress granules in cells infected with the F17A mutant compared to WT (Fig. 2D)  
+d. decreased viral replication in vitro in the F17A mutation (Fig. 2G)  
+e. decreased pathogenesis and replication in vivo using a small animal model (Fig. 3B, 3C)  
+f. That G3BP1 facilitates LLPS, which is disrupted by the F17A mutation (Fig. 4)  
+
+<|ref|>text<|/ref|><|det|>[[70, 124, 923, 152]]<|/det|>
+While the mutants are not identical (F17A alone in Yang, vs I15A and F17A in the present manuscript) the same conclusions are reached. At minimum, Yang et. al. should be cited. Overall, this reduces the novelty of the research significantly.  
+
+<|ref|>text<|/ref|><|det|>[[72, 164, 920, 269]]<|/det|>
+2. Specificity of Mutant: While the underlying conclusions of Yang et. al are concordant with the authors work, there is a major difference in the magnitude of the effects seen. For example, Yang et. al. reports \(\sim 1\) -log reduction in virus replication with their mutant (F17A) but the present study reports \(\sim 2\) -log reduction with theirs (I15A, F17A). The Yang et al. manuscript spends a great deal of time demonstrating that the F17A mutation is highly specific, disrupting SARS2 N:G3BP1/2 binding while not affecting SARS2 N's affinity to any other cellular proteins according to mass spectrometry analysis. Thus, one possible explanation for the differences in the magnitude of phenotypes between the two studies is that the addition of I15A results in off target effects. Control experiments addressing the specificity of the double mutant are needed to alleviate these concerns.  
+
+<|ref|>text<|/ref|><|det|>[[72, 281, 924, 333]]<|/det|>
+3. Viral replication in K18-hACE2 mice. Fig. 3 compares the pathogenesis of SARS2 WT vs the I15A, F17A mutant, demonstrating that the I15A, F17A mutation reduces weight loss, H&E staining, and antigen staining relative to WT. The absence of viral lung titer is a striking omission. Viral lung titers between SARS2 WT and the I15A, F17A mutation should be examined to confirm replication differences in vivo.  
+
+<|ref|>text<|/ref|><|det|>[[72, 345, 923, 438]]<|/det|>
+4. Translation of viral proteins: As a mentioned in the summary, the authors finding that N re-localizes G3BP1 from stress granules to replication transcription complexes to enhance translation is a major strength and a real step forward for the field. However, when digging into the details of their model, the authors propose that the ribopuromycinylation staining co-localizing with NSP3 indicates enhanced translation of viral genes as the emerge from DMVs. While plausible, this is not adequately supported. Looking at Fig. 2C, the protein levels of N itself are not affected much by the I15A, F17A mutation. This should be explained, or the levels of other viral proteins should be examined. Given the normal abundance of N within infected cells, viral factors translated from longer subgenomic transcripts may be affected more drastically and validate their model.  
+
+<|ref|>text<|/ref|><|det|>[[72, 450, 119, 462]]<|/det|>
+Minor:  
+
+<|ref|>text<|/ref|><|det|>[[72, 462, 904, 490]]<|/det|>
+1. While the microscopy data presented in Fig.5 and Fig 6 (IF and TEM) is convincing, it would be improved by any sort of quantification that the authors would be able to provide. For instance, the number of ribosomes per DMV in Fig. 6C, etc.  
+
+<|ref|>text<|/ref|><|det|>[[72, 502, 660, 516]]<|/det|>
+2. The scale bars on all the IF images throughout the manuscript are too small to read.  
+
+<|ref|>sub_title<|/ref|><|det|>[[72, 528, 135, 542]]<|/det|>
+## Citations  
+
+<|ref|>text<|/ref|><|det|>[[72, 554, 914, 620]]<|/det|>
+Yang Z, Johnson BA, Meliopoulos VA, Ju X, Zhang P, Hughes MP, Wu J, Koreski KP, Clary JE, Chang TC, Wu G, Hixon J, Duffner J, Wong K, Lemieux R, Lokugamage KG, Alvarado RE, Crocqet-Valdes PA, Walker DH, Plante KS, Plante JA, Weaver SC, Kim HJ, Meyers R, Schultz-Cherry S, Ding Q, Menachery VD, Taylor JP. Interaction between host G3BP and viral nucleocapsid protein regulates SARS-CoV-2 replication and pathogenicity. Cell Rep. 2024 Mar 26;43(3):113965. doi: 10.1016/j.celrep.2024.113965. Epub 2024 Mar 15. PMID: 38492217.  
+
+<|ref|>sub_title<|/ref|><|det|>[[72, 632, 162, 645]]<|/det|>
+## Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[72, 658, 922, 725]]<|/det|>
+(Remarks to the Author) This manuscript addresses the mechanisms by G3BP proteins affect the ability of SARS2 to infect human cells. The work concludes that the G3BP proteins interact with the SARS2 N protein to promote the translation of viral RNAs that are emerging from the membrane bound replication organelles. Although this is potentially an interesting manuscript, as detailed below, significant additional analyses would be required to support the major conclusions.  
+
+<|ref|>text<|/ref|><|det|>[[72, 736, 785, 750]]<|/det|>
+This review is from Roy Parker and I would be happy to clarify these comments for the authors if needed.  
+
+<|ref|>sub_title<|/ref|><|det|>[[72, 762, 211, 776]]<|/det|>
+## Specific Comments:  
+
+<|ref|>text<|/ref|><|det|>[[72, 788, 920, 927]]<|/det|>
+1) The novel conclusion of this work is that a complex of G3BP-N is targeted to the pore complex of double membrane replication organelles to recruit ribosomes to the emerging viral mRNAs. I am not yet convinced of this conclusion given the current data and make the following comments/suggestions. a) The first argument for this function is that viruses with an N protein mutation that blocking interact with G3BP proteins showed reduced replication (similar to what was observed in Yang et al., 2024, Cell Reports). However, it remains possible that G3BP inhibits SARS replication and in the absence of N proteins inhibiting G3BP, there is an inhibitory effect on SARS (as suggested at least in part in Burke et al., 2024, Science Advances). I realize the field is split on this issue, but the authors could clarify this issue by examining WT and RATA SARS replication in WT and g3bpΔ cells. This would clarify whether G3BP is a host factor, or just limits SARS infection, and how those putative roles are affected by interaction with the N protein. Without clearing demonstrating G3BP is a required host factor it is difficult to argue the N-G3BP protein interaction is required for viral growth.  
+
+<|ref|>text<|/ref|><|det|>[[70, 930, 875, 944]]<|/det|>
+b) The second argument for this function is that G3BP overlaps with dsRNA and the N protein earlier in infection. This
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 46, 923, 111]]<|/det|>
+experiment could be improved by i) showing the individual channels (at least in supplement) since it is not possible to assess how significant these overlaps are with just one small zoom showing individual channels, and ii) some type of quantification at all the time points. I find it notable that at 12 hours it looks like G3BP is excluded from the area of dsRNA, which is similar to what we observed (Burke et al., 2024, Science Advances). How would exclusion of G3BP from replication areas at this time point fit with the proposed model?  
+
+<|ref|>text<|/ref|><|det|>[[72, 111, 921, 202]]<|/det|>
+c) The third argument for this function of GFP-G3BP is the overlaps of dsRNA, N, and the Nsp3 protein (as a marker of replication organelles). I make two comments on this experiment. i) I am concerned this staining pattern could be affected by the GFP tag on G3BP1. I suggest this possibility because we observed a different distribution of GFP-G3BP and untagged G3BP in SAR infected cells (see Figure below). Because of this difference, we avoided the use of GFP tagged G3BP for SARS experiments. At a minimum, the authors need to show this subcellular distribution of proteins is not affected by the GFP tag. ii) In addition, however this experiment is performed (e.g. GFP or IF), the experiment could be improved by showing the individual channels (at least in supplement) and quantifying the extent of co-localization.  
+
+<|ref|>text<|/ref|><|det|>[[72, 202, 920, 268]]<|/det|>
+d) The fourth argument for G3BP promoting local translation is that puromycin labeling overlaps with G3BP IF. It was my understanding that puromycin labeling can no longer be relied upon to identify sites of local translation due to rapid diffusion of released peptides, even when using translation elongation inhibitors in conjunction (Enam et al., 2020, eLife). Given this caveat, additional observations would be needed to make a robust conclusion for G3BP marking the sites of translation. (Can you see ribosome clusters overlapping with G3BP by CLEM or EM with gold labeled antibodies?)  
+
+<|ref|>text<|/ref|><|det|>[[72, 268, 921, 321]]<|/det|>
+e) The final argument for this function of G3BP is EM imaging showing differences between WT and RATA SARS infection. The key observations are that in RATA mutants and g3bpΔΔ cell lines the ribosomes are more diffuse and less concentrated around the double membrane viral organelles. While a single EM image can be interesting, to make these points robustly, those differences need to be quantified in some manner (beyond just the size of the LVCVs).  
+
+<|ref|>sub_title<|/ref|><|det|>[[73, 334, 197, 347]]<|/det|>
+## Additional Issues:  
+
+<|ref|>text<|/ref|><|det|>[[70, 359, 891, 387]]<|/det|>
+2) How do the authors identify the infected cells in Figure 1a/b since the N protein IF is dim at 6 hours. Is the cell making stress granules at 6 hours infected? This should be clarified.  
+
+<|ref|>text<|/ref|><|det|>[[72, 397, 920, 438]]<|/det|>
+3) The specific model the authors put forth predicts that in the RATA virus (or in G3BPΔΔ cell lines), the translation efficiency of the viral RNAs would be low early in infection. This could be directly tested by measuring the rate of viral protein production and the levels of viral RNAs at the same time point and comparing WT to RATA.  
+
+<|ref|>text<|/ref|><|det|>[[70, 448, 825, 463]]<|/det|>
+Figure for authors showing differences in endogenous G3BP1 distribution and the distribution of GFP-G3BP1:  
+
+<|ref|>text<|/ref|><|det|>[[72, 475, 833, 503]]<|/det|>
+(Since the website will not let me insert the Figure, I will email it directly to the authors with a copy of my review. Note: I also uploaded to the review site as well.)  
+
+<|ref|>sub_title<|/ref|><|det|>[[73, 528, 162, 541]]<|/det|>
+## Reviewer #3  
+
+<|ref|>text<|/ref|><|det|>[[73, 555, 238, 568]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 567, 923, 660]]<|/det|>
+In this article, Long and colleagues generate a SARS- CoV- 2 mutant (RATA) that lacks the ability to interact with G3BP1/2 proteins, which are RNA- binding proteins that promote stress granule assembly. The RATA mutant virus is attenuated in cell culture models and is less pathogenic in K18- hACE2 transgenic mice. In comparison to WT SARS- CoV- 2, G3BP displays reduced co- localization with dsRNA and N at DMV replication factories in the RATA mutant virus infection. The authors also observe less incorporation of puromycin at DMV in the RATA mutant virus infection, suggesting a reduction in local translation of viral RNA at the DMV. Based on these observations, the authors claim that G3BP1/2 proteins are host factors that are required for SARS- CoV- 2 replication by enhancing localized translation at viral DMV.  
+
+<|ref|>text<|/ref|><|det|>[[72, 672, 880, 700]]<|/det|>
+In general, the authors describe interesting cellular biology relating to how SARS- CoV- 2 N modulates G3BP functions during infection, which adds to a growing body of literature. The paper is well- written and the data is of high quality.  
+
+<|ref|>text<|/ref|><|det|>[[72, 710, 923, 841]]<|/det|>
+However, their data do not support their primary conclusion that G3BP is a pro- viral protein required for SARS- CoV- 2 replication. Specifically, the observation that SARS- CoV- 2 replicates to similar titers in parental and G3BP- KO cells (data not shown), which is consistent with other studies (Burke et al., RNA 2021 PMID: 34315815), indicates that G3BP is not a pro- viral protein that is required for SARS- CoV- 2 replication. These data strongly argue against their primary model. An alternative explanation for the attenuation of the RATA virus is that the reduction in interaction between G3BP1 and N during RATA mutant virus infection leads to enhanced G3BP1- mediated antiviral activity (i.e., type I IFN response, as suggested in their model) and thus reduce RATA mutant virus replication. However, the authors do not examine if and how the RATA mutant alters antiviral responses, nor do they test if the replication of the RATA mutant would be rescued in G3BP- KO cells as would be expected by this function of N. Because the RATA mutations of N could disrupt a number of putative N functions, it is unclear if altered G3BP interactions contribute to RATA virus attenuation.  
+
+<|ref|>text<|/ref|><|det|>[[70, 853, 917, 881]]<|/det|>
+Overall, their data do not support their primary conclusion that G3BP- N interaction is required for optimal viral replication via localization of translation at viral DMV, which weakens the impact of this article.  
+
+<|ref|>text<|/ref|><|det|>[[73, 893, 208, 906]]<|/det|>
+Specific comments:  
+
+<|ref|>text<|/ref|><|det|>[[70, 918, 915, 946]]<|/det|>
+1. Extended Figure 3D. Most infected cells do not contain G3BP1 complexes in cells infected with either RATA or WT virus. Most cells with G3BP1 granules do not appear to be infected. Are these SGs generated through paracrine signaling from
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 46, 917, 75]]<|/det|>
+infected cells? Also, eIF4G staining is only observed in SARS- CoV- 2 infected cells. Does SARS- CoV- 2 infection lead to an increase in eIF4G, or is this signal spectral crossover from viral N staining?  
+
+<|ref|>text<|/ref|><|det|>[[72, 85, 922, 140]]<|/det|>
+2. Burke et al. 2024 (PMID: 38295168) showed that an N-resistant G3BP1 could cause G3BP1 aggregates to form in SARS-CoV-2 infected cells, and that inhibition of eIF4A but not sodium arsenite-induced phosphorylation of eIF2-alpha increased G3BP1 interactions with viral RNA in large aggregates containing viral RNA and dsRNA. Because these findings are similar to those made by the authors in this study, the authors should cite accordingly.  
+
+<|ref|>text<|/ref|><|det|>[[72, 151, 650, 166]]<|/det|>
+3. Fig. 1A. It is not clear that the cells with SGs are infected since they lack N protein.  
+
+<|ref|>text<|/ref|><|det|>[[72, 177, 917, 269]]<|/det|>
+4. Fig. 2G. The authors show that the SARS-CoV-2-RATA mutant is attenuated, as it replicates to lower titers in several cell types. While the authors claim that this is the result of disrupting G3BP1-N interactions required for maximal viral replication capacity, it could also be that this mutation disrupts normal N functions or disrupts interactions with other host proteins. If the attenuation of the RATA mutant is due to disruption of G3BP1, then knockout of G3BP1 would be expected to reduce SARS-CoV-2-WT virus replication capacity similarly. The authors should test if SARS-CoV-2 replicates to lower titers in G3BP-KO cells in comparison to parental cells, and if so, show that rescue of G3BP1 rescues viral replication. Notably, Knockout of G3BP1/2-KO did not reduce SARS-CoV-2 replication in A549 cells (Burke et al., 2024).  
+
+<|ref|>text<|/ref|><|det|>[[70, 281, 904, 310]]<|/det|>
+5. Fig. 2G. If the RATA mutant leads to enhanced antiviral activities of G3BP, then it should replicate to equal titers as WT virus in G3BP-KO cells. The authors should consider testing this.  
+
+<|ref|>text<|/ref|><|det|>[[72, 320, 920, 360]]<|/det|>
+6. Lines 182-183. "As expected, N-RATA colocalized neither with G3BP1 nor dsRNA (Fig. 5b), demonstrating that SARS-CoV-2 N recruits G3BP1 to RTC". This statement is misleading based on the data in the figure, as N-RATA does co-localize with dsRNA but not with G3BP1.  
+
+<|ref|>text<|/ref|><|det|>[[70, 371, 904, 399]]<|/det|>
+7. Fig. 6C. It is not obvious that ribosomes are reduced near DMV in the RATA mutant compared to WT virus. The authors should quantify this result.  
+
+<|ref|>text<|/ref|><|det|>[[72, 450, 144, 463]]<|/det|>
+Version 1:  
+
+<|ref|>text<|/ref|><|det|>[[72, 476, 219, 490]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[72, 503, 160, 516]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[72, 530, 238, 543]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 543, 910, 620]]<|/det|>
+The manuscript Long et al. investigates the pro-viral role of G3BP1/2 during SARS-CoV- 2 infection. Previous studies have established that the SARS-CoV- 2 Nucleocapsid (N) protein binds with G3BP1/2, inhibiting stress granule formation (which is believed to be antiviral). However, whether G3BP1/2 is itself pro- or anti-viral is a matter of some debate. To investigate the importance of N:G3BP1/2 binding, the authors use reverse genetics to generate a recombinant virus with a two amino acid substitution in the N protein (N: 115A; F17A). Compared to infection with wild type SARS-CoV- 2, the authors mutant exhibits reduced replication, N:G3BP1/2 binding, and increased stress granule formation.  
+
+<|ref|>text<|/ref|><|det|>[[72, 631, 897, 685]]<|/det|>
+Compared to the initial submission, this manuscript has been great improved. All of my major concerns were addressed, specifically 1) my concerns regarding novelty, 2) the specificity of the 115A, F17A mutant compared to F17A alone, 3) the lack of replication data in vivo, and 4) data strengthening their claim regarding an effect on translation of viral proteins. In addition, they have adequately addressed all of my minor critiques.  
+
+<|ref|>text<|/ref|><|det|>[[72, 696, 514, 710]]<|/det|>
+As such, I recommend this article is suitable for publication as is.  
+
+<|ref|>text<|/ref|><|det|>[[72, 736, 161, 749]]<|/det|>
+Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[72, 763, 238, 776]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[70, 776, 895, 804]]<|/det|>
+In this revised manuscript, the authors have improved the work. I am supportive of publication but recommend some final alterations be made to the manuscript to strengthen the work and remove ambiguities.  
+
+<|ref|>text<|/ref|><|det|>[[72, 815, 916, 867]]<|/det|>
+1) An important experiment is how the WT and RATA virus replicate in WT and G3BPΔ cells since this addresses whether G3BP proteins are primarily antiviral or proviral and the nature of the alteration in the RATA mutant. I thank the authors for adding these experiments. The work would be strengthened by clarifying the specific differences observed in g3bpΔ cells. As I look at the figure I observe:  
+
+<|ref|>text<|/ref|><|det|>[[70, 867, 920, 944]]<|/det|>
+a) The RATA mutant is 10-50 times less replicative than WT virus in WT cell lines, but only \(\sim 3X\) worse that WT in g3bpΔ cells, consistent with the authors interpretation that the major effects of this mutation are due to G3BP proteins. 
+b) It looks like WT virus is hindered in g3bpΔ cells suggesting G3BP proteins can promote replication, but this effect is not rescued by the reintroduction of G3BP1 (assuming I am interpreting the figure correctly). 
+I suggest: i) The authors clarify what the differences are for WT and RATA virus in the different cell lines and then ii) describe how they interpret those differences for the functional consequences of the N-G3BP interaction.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 59, 915, 112]]<|/det|>
+2) The authors interpretation that N-G3BP promotes viral translation could be strengthened by quantifying the puromycin labeling experiments on a single cell level (using data they already have). Since most of the translation at this stage will be viral mRNAs, it would be predicted that either RATA in WT cells, or WT virus in g3bpΔΔ cells, should show overall reduced translation rates.  
+
+<|ref|>text<|/ref|><|det|>[[72, 124, 918, 166]]<|/det|>
+3) It would be appropriate to at least discuss the alternative model wherein G3BP plays a role in promoting virion packaging and release from the cells (Murigneux et al., Nature Communications, 2024). Could both models be true, or might there be a simpler resolution?  
+
+<|ref|>text<|/ref|><|det|>[[70, 177, 920, 205]]<|/det|>
+4) A terrific experiment would be to show immuno-gold localization of G3BP to the DMV with ribosomes on them. I would not require that for publication, but it would really strengthen the work.  
+
+<|ref|>sub_title<|/ref|><|det|>[[73, 243, 162, 256]]<|/det|>
+## Reviewer #3  
+
+<|ref|>text<|/ref|><|det|>[[73, 269, 238, 282]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 282, 920, 346]]<|/det|>
+In this revised manuscript, the authors adequately addressed many reviewer concerns. However, their data in Fig. 3G do not support that G3BP1 is pro- viral during SARS- CoV- 2 infection. Specifically, the authors show that the RATA virus displays a reduced ability to generate plaque forming units in several cell lines, including Vero cells, MA- 104, and U2OS cells. This indicates that the RATA virus is attenuated. However, whether this is due to the inability of N to interact with G3BP1 during RATA infection was unknown.  
+
+<|ref|>text<|/ref|><|det|>[[72, 358, 904, 399]]<|/det|>
+To address this, the authors examined growth kinetics of WT and RATA via plaque assays in parental (U2OS- ACE2), G3BP1- KO (GFP), and rescue (GFP- G1- WT) U2OS cell lines. Several observations do not support that G3BP1 is a pro- viral host factor required for SARS- CoV- 2 replication based on data in Fig. 3G:  
+
+<|ref|>text<|/ref|><|det|>[[72, 411, 918, 464]]<|/det|>
+1. Knockout of G3BP1 in U2OS cells resulted in higher titers of WT virus by 36 hrs. p.i. Thus, G3BP1 is not required for SARS-CoV-2 replication, but in fact could reduce SARS-CoV-2 replication. Moreover, rescue of GFP-G3BP1 in the G3BP-KO cells (GFP-G1-WT) did not enhance WT SARS-CoV-2 replication kinetics or in increase final PFU titers. This indicates that G3BP1 is not a host factor that enhances SARS-CoV-2 replication.  
+
+<|ref|>text<|/ref|><|det|>[[70, 475, 868, 503]]<|/det|>
+2. Knockout of G3BP1 in U2OS cells increased RATA virus replication, suggesting that G3BP1 perturbs RATA virus replication. Notably, RATA virus replication is reduced by GFP-G3BP1 in (GFP-G1-WT) U2OS cell lines.  
+
+<|ref|>text<|/ref|><|det|>[[70, 515, 920, 543]]<|/det|>
+3. The RATA virus is attenuated in G3BP1-KO (GFP) cells in comparison to WT cells. This indicates that the mutations in N that lead to attenuation of the RATA virus is G3BP1-independent.  
+
+<|ref|>text<|/ref|><|det|>[[70, 555, 888, 582]]<|/det|>
+4. Despite the overall higher replication of WT vs. RATA in all the cell lines, the growth kinetics between WT and RATA viruses is similar between parental (U2OS-ACE2) and G3BP1-KO (GFP) cells.  
+
+<|ref|>text<|/ref|><|det|>[[72, 593, 920, 686]]<|/det|>
+In summary, if the interaction between SARS-CoV- 2 N and G3BP1 were required to enhance viral replication, then the authors would have observed a reduction in WT virus replication in G3BP1/2-KO cells to titers equivalent to RATA. Moreover, the reduction in WT virus fitness would be rescued in (GFP-G1-WT) U2OS cell lines. However, the authors do not observe these effects. Instead, they observe that knockout of G3BP1/2 leads to higher replication of both WT and RATA, and that rescuing G3BP1 expression only reduces RATA. Combined, these data indicate that the RATA mutations attenuate SARS-CoV-2 replication independently of G3BP1/2 interactions, and that the slower replicating RATA virus might be more sensitive to the general antiviral effects of G3BP1 (interferon-independent).  
+
+<|ref|>sub_title<|/ref|><|det|>[[73, 698, 218, 711]]<|/det|>
+## Additional comments  
+
+<|ref|>text<|/ref|><|det|>[[72, 723, 888, 764]]<|/det|>
+Considering the observation that their viral N staining is capable of contaminating other channels, the authors should consider repeating key results with no- primary controls to confirm that any co- localization results are not due to spectral crossover.  
+
+<|ref|>text<|/ref|><|det|>[[73, 802, 144, 815]]<|/det|>
+Version 2:  
+
+<|ref|>text<|/ref|><|det|>[[73, 828, 219, 841]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[73, 854, 161, 867]]<|/det|>
+Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[72, 880, 454, 907]]<|/det|>
+(Remarks to the Author) The authors have adequately addressed my comments.  
+
+<|ref|>text<|/ref|><|det|>[[73, 919, 161, 932]]<|/det|>
+Reviewer #3
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 47, 238, 60]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[70, 60, 911, 88]]<|/det|>
+The manuscript is improved from initial submission, and the authors have addressed many reviewer concerns. Overall, this is a thorough study and thus publication is recommended.  
+
+<|ref|>text<|/ref|><|det|>[[72, 440, 916, 494]]<|/det|>
+Open Access This Peer Review 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.  
+
+<|ref|>text<|/ref|><|det|>[[72, 494, 796, 508]]<|/det|>
+In cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+<|ref|>text<|/ref|><|det|>[[72, 508, 910, 560]]<|/det|>
+The images or other third party material in this Peer Review 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.  
+
+<|ref|>text<|/ref|><|det|>[[72, 559, 618, 572]]<|/det|>
+To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[119, 85, 315, 100]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>sub_title<|/ref|><|det|>[[120, 116, 428, 132]]<|/det|>
+## Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[119, 148, 295, 163]]<|/det|>
+Summary/Key Results  
+
+<|ref|>text<|/ref|><|det|>[[118, 180, 860, 308]]<|/det|>
+The manuscript Long et al. investigates the pro- viral role of G3BP1/2 during SARS- CoV- 2 infection. Previous studies have established that the SARS- CoV- 2 Nucleocapsid (N) protein binds with G3BP1/2, inhibiting stress granule formation (which is believed to be antiviral). However, whether G3BP1/2 is itself pro- or anti- viral is a matter of some debate. To investigate the importance of N:G3BP1/2 binding, the authors use reverse genetics to generate a recombinant virus with a two amino acid substitution in the N protein (N: 115A; F17A). Compared to infection with wild type SARS- CoV- 2, the authors mutant exhibits reduced replication, N:G3BP1/2 binding, and increased stress granule formation.  
+
+<|ref|>text<|/ref|><|det|>[[118, 323, 855, 404]]<|/det|>
+The most interesting findings of this article are mechanistic. The authors clearly demonstrate that G3BP1 co- localizes with NSP3 on double membrane vesicles and that this interaction requires N:G3BP1 binding. Furthermore, they provide evidence that this results in enhanced translation at RTCs, evidenced by the co- localization of 40S ribosomal subunits, ribopuromycincylation experiments, and the number of ribosomes localized to DMVs by TEM.  
+
+<|ref|>text<|/ref|><|det|>[[118, 419, 865, 483]]<|/det|>
+Overall, this manuscript is both interesting and scientifically sound. In particular, the last 1/3 (Fig. 5 and 6) of the paper linking N:G3BP1 binding with enhanced translation at DMVs is novel. Pro- viral functions of G3BP1 during SARS- CoV- 2 infection have been proposed, but the data presented here are the first to elucidate a concrete mechanism.  
+
+<|ref|>text<|/ref|><|det|>[[117, 498, 875, 627]]<|/det|>
+Despite this article's strengths, there are some major weaknesses. First, the overall novelty of the paper is lessened by the fact that a similar article, Yang et al. 2024, has already been published. In Yang et al., the authors produced a single substitution mutant (F17A) SARS- CoV- 2, and described the effects on in vitro replication, G3BP1/2 binding, phase separation, stress granule formation, and in vivo infection. Given the similarity in the mutants (F17A vs I15A/F17A) and that the results and conclusions are almost identical (they differ only in magnitude), most of the data presented in Figures 2 – 4 are merely a confirmation of prior work. At minimum, the authors need to cite this past work.  
+
+<|ref|>text<|/ref|><|det|>[[117, 642, 876, 739]]<|/det|>
+In conclusion, the mechanistic work present in the figures 5 and 6 of this manuscript are novel, interesting, and important. However, this novelty is harmed by the fact that the data in Figures 2 – 4 is largely a confirmation of a manuscript of another group, which the authors do not cite. The authors can improve their manuscript by a) emphasizing and expanding upon those findings which are novel and b) citing Yang et al. and addressing the minor points of difference between the two studies.  
+
+<|ref|>text<|/ref|><|det|>[[118, 754, 715, 770]]<|/det|>
+We thank the reviewer for the positive comments and valuable suggestions.  
+
+<|ref|>text<|/ref|><|det|>[[118, 785, 266, 801]]<|/det|>
+Specific Critiques:  
+
+<|ref|>text<|/ref|><|det|>[[118, 818, 170, 833]]<|/det|>
+Major:  
+
+<|ref|>text<|/ref|><|det|>[[118, 849, 880, 898]]<|/det|>
+1. Novelty: Yang et al. have already published on a recombinant SARS-CoV-2 mutant that disrupts N:G3BP1/2 binding. Their mutant contains 1 of the 2 mutations used by the authors (F17A) and show similar results. Thus, Yang et al. has essentially already described the following
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[119, 85, 381, 100]]<|/det|>
+data included in this manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[118, 115, 850, 213]]<|/det|>
+a. structure of the N:G3BP interface within the NFT2L domain (Fig. 2A)  
+b. disruption of N:G3BP1 binding through the F17A mutation (Fig. 2C)  
+c. increased stress granules in cells infected with the F17A mutant compared to WT (Fig. 2D)  
+d. decreased viral replication in vitro in the F17A mutation (Fig. 2G)  
+e. decreased pathogenesis and replication in vivo using a small animal model (Fig. 3B, 3C)  
+f. That G3BP1 facilitates LLPS, which is disrupted by the F17A mutation (Fig. 4)  
+
+<|ref|>text<|/ref|><|det|>[[119, 227, 829, 276]]<|/det|>
+While the mutants are not identical (F17A alone in Yang, vs I15A and F17A in the present manuscript) the same conclusions are reached. At minimum, Yang et. al. should be cited. Overall, this reduces the novelty of the research significantly.  
+
+<|ref|>text<|/ref|><|det|>[[118, 291, 872, 404]]<|/det|>
+The points about the decreased novelty of parts of our paper after the publication of the Yang paper are well taken. It is an area of great interest and indeed, some of those discoveries listed had been made before Yang, including (a) by Biswal et al., J Mol Biol 2022, PMID: 35240128 and (b) by Kruse et al., Nature Communications 2021; PMID 34799561; Huang et al., 2021, Cell Discovery, PMID: 34400613: Biswal et al., J Mol Biol 2022). The Yang paper was cited in the original manuscript (reference 23) and we have now cited it more prominently and discussed their data on lines 130- 132 in the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[118, 419, 872, 467]]<|/det|>
+A major difference between the studies is our inclusion of the I15A mutation. Our strategy, based on our previous work on viral FG- based motifs binding to G3BP, was to mutate more than one critical residue to confidently predict that no binding would take place in infected cells.  
+
+<|ref|>text<|/ref|><|det|>[[117, 482, 875, 673]]<|/det|>
+The NTF2- like domain of G3BP1 features a long binding groove formed by two \(\alpha\) - helices and two \(\beta\) - sheets, comprising a \(5.6 \AA\) wide groove and a \(3.5 \AA\) narrow groove. In our previous structural work, a dual groove- insertion mode was observed in complexes involving alphavirus nsP3/G3BP1- NTF2L (Schulte et al., Open Biology 2016, PMID: 27383630) and the host protein Caprin1/G3BP1- NTF2L (Schulte et al., Open Biology 2023, PMID: 37161291). Taking N for example in this binding mode, the N- F17 aromatic ring inserts into the aromatic cage at the centre of NTF2L binding groove, stabilised by multiple \(\pi\) - stacking interactions, while the bulky hydrophobic side chain of N- I15 inserts into the small groove, coordinated by G3BP residues L10, V11, and P6. Notably, the Caprin1- Y370A mutation (equivalent to N- I15) significantly reduced Caprin1's binding to G3BP- NTF2L (Schulte et al., Open Biology 2016, PMID: 27383630). Therefore, we postulated that both I15 and F17 in N were crucial for its interaction with G3BP- NTF2L and mutating them to alanine could maximally disrupt this interaction.  
+
+<|ref|>text<|/ref|><|det|>[[118, 688, 860, 785]]<|/det|>
+In the Biswal and Yang papers, the F17A single mutation abolished binding to N in in vitro assays, but a low but detectable level of binding was detected in a smaller study by Huang et al., 2021, Cell Discovery, PMID: 34400613, see their Fig S1C). We were concerned that in infected cells, where there will be other mechanisms for grouping and concentrating viral and cellular components, weak binding affinities might be compensated by high local concentrations of the relevant proteins.  
+
+<|ref|>text<|/ref|><|det|>[[118, 800, 876, 880]]<|/det|>
+Indeed, the data presented by Yang et al, in Figure 6E and F, show that there is weak, but readily detectable colocalisation of G3BP and N- F17A at 24 hours of infection in VeroE6- TMPRSS2 cells, (a detail from their Figure 6F reproduced below in Rebuttal Fig 1). Compare those data with the equivalent experiments from our study with RATA (Figs 6a, c, e, 7a, S3a, S5) where there is negligible colocalisation of N- RATA and G3BP1.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 231, 856, 279]]<|/det|>
+Rebuttal Figure 1: Data reproduced from Yang et al. Cell Reports, figure 6F with red lines and box added to indicate sites of colocalisation of G3BP1 (yellow), viral genomic RNA (magenta) and N- 17A (cyan).  
+
+<|ref|>text<|/ref|><|det|>[[118, 295, 876, 375]]<|/det|>
+This minimal level of recruitment of G3BP by N- F17A is probably insufficient to counteract the antiviral effect, and the conclusions of Yang and colleagues are unaffected. However, it remains possible that a low level of interaction of G3BP and N- F17A, earlier in infection might be able provide some of the proviral effects that might explain the difference in magnitude of the attenuation between F17A (Yang) and I15A+F17A (our work).  
+
+<|ref|>text<|/ref|><|det|>[[118, 392, 870, 455]]<|/det|>
+These data, combined with the specificity of the I15A mutation for G3BP (see our response to comment 2, below), reinforces our decision to mutate both residues to create viral mutant that is truly defective for G3BP interaction. We believe this adds to the novelty of our work and hope the reviewer agrees.  
+
+<|ref|>text<|/ref|><|det|>[[118, 471, 870, 582]]<|/det|>
+In a further change to strengthen the novelty of our work relative to that of Yang and colleagues, we have moved data using virus variants of concern, from supplement to the main figures (new Fig 2). We show, similarly to Yang et al., that the P13L mutation in the N protein of the omicron lineage causes a slight decrease in the binding to G3BP. However, we extend that work with infectious WA- 1 (P13) and XBB.1.5 (L13) viruses, showing that, despite the slightly decreased binding, no difference in SG inhibition is observed, suggesting it unlikely that this mutation might contribute to reduced virulence of omicron variants, as suggested by Yang et al.  
+
+<|ref|>text<|/ref|><|det|>[[118, 598, 875, 678]]<|/det|>
+But, as the reviewer points out, the real novelty in our work lies in the mechanistic work which is "the first to elucidate a concrete mechanism" for G3BP's proviral effects. In response to other comments (below), we have further strengthened those experiments and now believe the work represents an advance to the field. We hope the reviewers and editor agree that the paper is now worthy of publication.  
+
+<|ref|>text<|/ref|><|det|>[[117, 694, 875, 854]]<|/det|>
+2. Specificity of Mutant: While the underlying conclusions of Yang et. al are concordant with the authors work, there is a major difference in the magnitude of the effects seen. For example, Yang et. al. reports \(\sim 1\) -log reduction in virus replication with their mutant (F17A) but the present study reports \(\sim 2\) -log reduction with theirs (I15A, F17A). The Yang et al. manuscript spends a great deal of time demonstrating that the F17A mutation is highly specific, disrupting SARS2 N:G3BP1/2 binding while not affecting SARS2 N's affinity to any other cellular proteins according to mass spectrometry analysis. Thus, one possible explanation for the differences in the magnitude of phenotypes between the two studies is that the addition of I15A results in off target effects. Control experiments addressing the specificity of the double mutant are needed to alleviate these concerns.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 875, 150]]<|/det|>
+We do not believe that the I15A mutation results in off target effects. Our main evidence for this is an experiment performed in our collaborator Jakob Nilsson's lab in Copenhagen, and presented in Kruse et al., Nature Communications 2021; PMID 34799561), figure 2i (reproduced here as Rebuttal Fig 2). The results show that the R14A and I15A double mutant is very specific for G3BP1 and G3BP2. This is in agreement with our observations (mentioned above) that both I15 and F17 are critical for G3BP interaction. Mutation of either residue abolishes G3BP interaction in most experiments, but we chose to mutate both to be sure to exclude any weak interactions.  
+
+<|ref|>text<|/ref|><|det|>[[316, 151, 870, 228]]<|/det|>
+for G3BP1 and G3BP2. This is in agreement with our observations (mentioned above) that both I15 and F17 are critical for G3BP interaction. Mutation of either residue abolishes G3BP interaction in most experiments, but we chose to mutate both to be sure to exclude any weak interactions.  
+
+<|ref|>text<|/ref|><|det|>[[316, 243, 836, 308]]<|/det|>
+Rebuttal Figure 2: Reproduced from Kruse et al., 2021, figure 2i legend - "Quantitative mass spectrometry analysis of YFP- tagged SARS- CoV- 2 N WT or 2A purified from HeLa cells (n = 4 technical replicates)."  
+
+<|ref|>text<|/ref|><|det|>[[118, 339, 877, 419]]<|/det|>
+3. Viral replication in K18-hACE2 mice. Fig. 3 compares the pathogenesis of SARS2 WT vs the I15A, F17A mutant, demonstrating that the I15A, F17A mutation reduces weight loss, H&E staining, and antigen staining relative to WT. The absence of viral lung titer is a striking omission. Viral lung titers between SARS2 WT and the I15A, F17A mutation should be examined to confirm replication differences in vivo.  
+
+<|ref|>text<|/ref|><|det|>[[118, 434, 863, 531]]<|/det|>
+We present RT- qPCR data from fixed lung tissue for viral RNA quantification as a measure of viral load in animals from the primary challenge and from the re- challenge. The results, now included in the revised manuscript as Fig 4c and 4f respectively (and reproduced here in Rebuttal Fig 3), clearly show that the RATA mutant is attenuated for replication in vivo (primary challenge) and that mice previously infected with RATA are better able to control WT virus infection (rechallenge)  
+
+<|ref|>text<|/ref|><|det|>[[118, 546, 861, 594]]<|/det|>
+We did not save fresh- frozen lung tissue from which to quantify infectious viral lung titres, but we believe repeating the experiment to generate live viral titres would be unnecessary and not be in accordance with the "3R" guiding principles for more ethical use of animals (Reduction).  
+
+<|ref|>image<|/ref|><|det|>[[123, 604, 411, 730]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[423, 657, 878, 705]]<|/det|>
+Rebuttal Figure 3: RT- qPCR of N and E mRNA expression in mice lungs after primary challenge and after rechallenge.  
+
+<|ref|>text<|/ref|><|det|>[[118, 770, 866, 897]]<|/det|>
+4. Translation of viral proteins: As a mentioned in the summary, the authors finding that N relocalizes G3BP1 from stress granules to replication transcription complexes to enhance translation is a major strength and a real step forward for the field. However, when digging into the details of their model, the authors propose that the ribopuromycinylation staining colocalizing with NSP3 indicates enhanced translation of viral genes as the emerge from DMVs. While plausible, this is not adequately supported. Looking at Fig. 2C, the protein levels of N itself are not affected much by the I15A, F17A mutation. This should be explained, or the levels of other viral proteins should be examined. Given the normal abundance of N within infected
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 85, 808, 116]]<|/det|>
+cells, viral factors translated from longer subgenomic transcripts may be affected more drastically and validate their model.  
+
+<|ref|>text<|/ref|><|det|>[[118, 132, 827, 164]]<|/det|>
+We thank the reviewer for the positive comments on this aspect of our work and the constructive critique. We have strengthened this aspect of the work in the following ways:  
+
+<|ref|>text<|/ref|><|det|>[[118, 180, 868, 275]]<|/det|>
+1. We show that total levels of viral proteins at early stages are lower in RATA infected cells (Fig 7d of the revised manuscript and reproduced here as Rebuttal Fig 4). We restricted these analyses to an early time point since differences in RNA replication and transcription confound analyses at later times. At 6 hours post infection, when viral mRNA levels are equivalent (Fig 7c of revised manuscript), we now show that total levels of spike and N protein are lower in RATA infected cells than WT.  
+
+<|ref|>image<|/ref|><|det|>[[123, 293, 350, 368]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[371, 291, 866, 371]]<|/det|>
+Rebuttal Figure 4: VeroE6 cells were infected with SARSCoV- 2 WT or RATA mutant at 0.05 MOI for 6h, and cells were lysed for immunoblotting with indicated antibodies. Representative images from three independent experiments are shown. Quantification of western blot was performed using Image J.  
+
+<|ref|>text<|/ref|><|det|>[[118, 402, 864, 450]]<|/det|>
+2. We have quantified the number of ribosomes in association with DMVs in WT and RATA infected cells. This is better described in our response to Reviewer 2, point 1e. These new data are presented in Fig 7g of the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[118, 467, 171, 481]]<|/det|>
+Minor:  
+
+<|ref|>text<|/ref|><|det|>[[118, 483, 874, 530]]<|/det|>
+1. While the microscopy data presented in Fig.5 and Fig 6 (IF and TEM) is convincing, it would be improved by any sort of quantification that the authors would be able to provide. For instance, the number of ribosomes per DMV in Fig. 6C, etc.  
+
+<|ref|>text<|/ref|><|det|>[[118, 546, 874, 593]]<|/det|>
+We have quantified co-localisations (Pearson's coefficient) in several places in the manuscript and quantified the number of ribosomes per DMV in a revised figure 7 and supplementary figure 6. This latter point is better described in response to Reviewer 2, point 1e.  
+
+<|ref|>text<|/ref|><|det|>[[116, 609, 797, 625]]<|/det|>
+2. The scale bars on all the IF images throughout the manuscript are too small to read.  
+
+<|ref|>text<|/ref|><|det|>[[118, 641, 814, 673]]<|/det|>
+We have improved the presentation of microscopy images throughout the manuscript in response to this and similar comments from other reviewers.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[120, 85, 428, 100]]<|/det|>
+## Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[119, 115, 873, 197]]<|/det|>
+This manuscript addresses the mechanisms by G3BP proteins affect the ability of SARS2 to infect human cells. The work concludes that the G3BP proteins interact with the SARS2 N protein to promote the translation of viral RNAs that are emerging from the membrane bound replication organelles. Although this is potentially an interesting manuscript, as detailed below, significant additional analyses would be required to support the major conclusions.  
+
+<|ref|>text<|/ref|><|det|>[[118, 212, 872, 243]]<|/det|>
+This review is from Roy Parker and I would be happy to clarify these comments for the authors if needed.  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 260, 282, 275]]<|/det|>
+## Specific Comments:  
+
+<|ref|>text<|/ref|><|det|>[[118, 291, 877, 355]]<|/det|>
+1) The novel conclusion of this work is that a complex of G3BP-N is targeted to the pore complex of double membrane replication organelles to recruit ribosomes to the emerging viral mRNAs. I am not yet convinced of this conclusion given the current data and make the following comments/suggestions.  
+
+<|ref|>text<|/ref|><|det|>[[117, 355, 877, 515]]<|/det|>
+a) The first argument for this function is that viruses with an N protein mutation that blocking interact with G3BP proteins showed reduced replication (similar to what was observed in Yang et al., 2024, Cell Reports). However, it remains possible that G3BP inhibits SARS replication and in the absence of N proteins inhibiting G3BP, there is an inhibitory effect on SARS (as suggested at least in part in Burke et al., 2024, Science Advances). I realize the field is split on this issue, but the authors could clarify this issue by examining WT and RATA SARS replication in WT and g3bpΔΔ cells. This would clarify whether G3BP is a host factor, or just limits SARS infection, and how those putative roles are affected by interaction with the N protein. Without clearing demonstrating G3BP is a required host factor it is difficult to argue the N-G3BP protein interaction is required for viral growth.  
+
+<|ref|>text<|/ref|><|det|>[[118, 530, 878, 593]]<|/det|>
+We do not believe the field needs to be split on the issue of G3BP's role in SARS- CoV- 2 (or indeed any) virus infection. We believe that G3BP has both antiviral and proviral functions. In the original submission we had not properly cited your Burke et al paper and we have corrected that now.  
+
+<|ref|>text<|/ref|><|det|>[[118, 609, 876, 721]]<|/det|>
+It was not our intention to state that G3BP is solely or primarily a pro- viral factor for SARS- CoV- 2 replication. The proviral function of recruiting translational apparatus to the DMVs may not be a 'required' function, but rather an accessory function that promotes efficient viral gene expression and replication without being critical for viral replication. In contrast, our and others' earlier work showed that G3BP really is critical for chikungunya virus replication (Schulte et al., Open Biology 2016, PMID: 27383630; Kim et al., PLoS Pathogens, 2016, PMID: 27509095; Gotte et al., PLoS Pathogens, PMID: 31199850), and viral replication is undetectable in its absence.  
+
+<|ref|>text<|/ref|><|det|>[[118, 737, 866, 785]]<|/det|>
+We have now performed SARS- CoV- 2 WT and RATA viral replication analyses in ΔΔGFP and ΔΔ- GFP- G1- WT cells and present the data in Fig 3g of the revised manuscript. These results are better described in response to a major critique of Reviewer 3.  
+
+<|ref|>text<|/ref|><|det|>[[118, 802, 875, 833]]<|/det|>
+We have added some text in the manuscript (lines 67, and 294- 297) to clarify that we believe the proviral effect that we describe is important but not critical for replication.  
+
+<|ref|>text<|/ref|><|det|>[[118, 849, 870, 897]]<|/det|>
+b) The second argument for this function is that G3BP overlaps with dsRNA and the N protein earlier in infection. This experiment could be improved by i) showing the individual channels (at least in supplement) since it is not possible to assess how significant these overlaps are with
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 864, 148]]<|/det|>
+just one small zoom showing individual channels, and ii) some type of quantification at all the time points. I find it notable that at 12 hours it looks like G3BP is excluded from the area of dsRNA, which is similar to what we observed (Burke et al., 2024, Science Advances). How would exclusion of G3BP from replication areas at this time point fit with the proposed model?  
+
+<|ref|>text<|/ref|><|det|>[[118, 163, 875, 244]]<|/det|>
+The G3BP, dsRNA and N protein colocalisation data are now presented in revised Fig S5a, showing individual channels and colocalisation analyses (Pearson's coefficients). The data now more clearly show that in SARS- CoV- 2 WT infection, G3BP1 is recruited to dsRNA foci at early times (3 and 6 hours), but that it is excluded from those sites at later times (12h (and 24h, not shown)).  
+
+<|ref|>text<|/ref|><|det|>[[118, 259, 876, 371]]<|/det|>
+We believe that the data fit very well with our proposed model. The very high affinity binding of N protein with G3BP will likely mean that most N protein molecules synthesised in the first hours of infection will bind to G3BP, leading to its stoichiometric neutralisation and consequent block in SG formation. The N- G3BP complexes accumulate around the DMVs where the proviral functions of G3BP are carried out. Later, as more N proteins continue to be synthesised, becoming one of the more abundant viral proteins in the cell, the majority of the N staining is then associated with progeny virus particle assembly.  
+
+<|ref|>text<|/ref|><|det|>[[117, 386, 873, 531]]<|/det|>
+c) The third argument for this function of GFP-G3BP is the overlaps of dsRNA, N, and the Nsp3 protein (as a marker of replication organelles). I make two comments on this experiment. i) I am concerned this staining pattern could be affected by the GFP tag on G3BP1. I suggest this possibility because we observed a different distribution of GFP-G3BP and untagged G3BP in SAR infected cells (see Figure below). Because of this difference, we avoided the use of GFP tagged G3BP for SARS experiments. At a minimum, the authors need to show this subcellular distribution of proteins is not affected by the GFP tag. ii) In addition, however this experiment is performed (e.g. GFP or IF), the experiment could be improved by showing the individual channels (at least in supplement) and quantifying the extent of co-localization.  
+
+<|ref|>text<|/ref|><|det|>[[117, 546, 879, 657]]<|/det|>
+i) Thank you for sharing your data on (GFP)-G3BP distribution in SARS-CoV-2 infection. We understand the concern and can report that we do not observe any differences in (EGFP)-G3BP1 localisation in parental U2OS or in the U2OS-ΔΔGFP-G1-WT cells after infection with SARS-CoV-2. To illustrate this, in Rebuttal Fig 5 we provide an image of U2OS-ACE2 cells infected with SARS-CoV-2 at MOI 0.5 for 6 hours. Similar to the EGFP-G3BP1 reporter in U2OS-ΔΔGFP-G1-WT cells (see Fig 6c and e of the revised manuscript), endogenous G3BP1 co-localised very strongly with dsRNA and N protein.  
+
+<|ref|>image<|/ref|><|det|>[[121, 668, 504, 802]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[526, 707, 872, 787]]<|/det|>
+Rebuttal Figure 5: U2OS-ACE2 cells were infected with SARS-CoV-2 at 0.5 MOI. Cells were fixed at 6 h and stained for endogenous G3BP1 (green), dsRNA (red) and N (grey).  
+
+<|ref|>text<|/ref|><|det|>[[118, 833, 878, 897]]<|/det|>
+It is also worth mentioning that we have also used these cells (before ACE2 lentivirus transduction) in our studies with Semliki Forest virus and chikungunya virus subversion of G3BP functions (Panas et al., PLoS Pathogens 2015, PMID: 25658430; Kedersha et al., JCB 2016, PMID: 27022092; Gotte et al., PLoS Pathogens, PMID: 31199850, Gotte et al., J Virol 2020, PMID:
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 848, 132]]<|/det|>
+31941782) and have always found the EGFP- G3BP1 protein to be a faithful reporter for G3BP localisation to viral RNA replication complexes, in comparison with parental U2OS cells and with other cell lines (BHK, Vero, MEF and others).  
+
+<|ref|>text<|/ref|><|det|>[[118, 148, 872, 276]]<|/det|>
+To further strengthen this part of the work, we have also now performed the G3BP, N, and the nsp3 co- immunoprecipitation experiments using parental U2OS- ACE2 cells and Vero cells (both with endogenous G3BP expression) to compare with the data from U2OS- ΔΔGFP- G1- WT cells (EGFP- G3BP1 transgene expression), as presented in the original submission. In all cases, the results show that (GFP)G3BP- N- nsp3 complexes are formed in WT SARS- CoV- 2 infection and that the presence of (GFP)G3BP in those complexes is dependent on the RITFG motif in N protein (disrupted in RATA) - see supplementary fig 5e (U2OS- ACE2 cells), Fig 6f (U2OS- ΔΔGFP- G1- WT cells) and Rebuttal Fig 6 showing the data from Vero cells.  
+
+<|ref|>image<|/ref|><|det|>[[120, 291, 290, 399]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[306, 323, 866, 371]]<|/det|>
+Rebuttal Figure 6: VeroE6 cells were infected with WT virus or RATA mutant at 0.01 MOI for 24 h. Cells were lysed and immunoprecipitated with GFP or N antibody for immunoblotting as indicated.  
+
+<|ref|>text<|/ref|><|det|>[[118, 435, 863, 499]]<|/det|>
+ii) We have improved the presentation of microscopy images throughout the manuscript in response to this and similar comments from other reviewers. Specifically, we now present the dsRNA, N, and the nsp3 colocalisation data with individual channels in Fig 6e of the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[117, 514, 874, 626]]<|/det|>
+d) The fourth argument for G3BP promoting local translation is that puromycin labeling overlaps with G3BP IF. It was my understanding that puromycin labeling can no longer be relied upon to identify sites of local translation due to rapid diffusion of released peptides, even when using translation elongation inhibitors in conjunction (Enam et al., 2020, eLife). Given this caveat, additional observations would be needed to make a robust conclusion for G3BP marking the sites of translation. (Can you see ribosome clusters overlapping with G3BP by CLEM or EM with gold labeled antibodies?)  
+
+<|ref|>text<|/ref|><|det|>[[118, 642, 875, 737]]<|/det|>
+We are aware of the observations presented in the Enam paper, but we do not believe this to be a confounding issue in our experiments. In our case, we are comparing two similar conditions - localised translation at the DMVs in the presence or absence of G3BP. For the "Enam diffusion" to be a confounding issue, it would have to be occurring in the absence of G3BP (both in the KO cells infected with WT virus and in WT cells infected with RATA), but not in the presence of G3BP (in WT cells infected with WT virus). We consider this improbable.  
+
+<|ref|>text<|/ref|><|det|>[[118, 753, 790, 785]]<|/det|>
+Nevertheless, to strengthen this aspect of the work, we have performed the following experiments:  
+
+<|ref|>text<|/ref|><|det|>[[118, 801, 871, 896]]<|/det|>
+i) Ribopuromycincylation experiment with only 2 minutes (reduced from 5 minutes in the original submission) of puromycin labelling to restrict potential diffusion of PMY-labelled peptides. These new data confirm that PMY labelling is strong at SARS-CoV-2 dsRNA+ foci, but only when G3BP is recruited there by N-WT. The data are presented in Fig 7a of the revised manuscript. Additionally, we now include data from analyses of correlations of G3BP1-PMY, or G3BP1-N, (Pearson's coefficients) and PMY maximum intensities in Fig 7b.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 100, 873, 180]]<|/det|>
+ii) To control for mRNA levels at the time of PMY labelling, we have also included qPCR analyses of viral mRNA levels in the same conditions as the PMY experiment. This shows that WT and RATA viral mRNA levels are equivalent at the time of PMY labelling, excluding that the effect could be due to differences in mRNA template availability. These new data are presented in Fig 7c of the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[118, 196, 853, 243]]<|/det|>
+iii) We now also show that total levels of viral spike and N proteins are affected by the recruitment of G3BP at early times post infection (6h). The data are presented in Fig 7d of the revised manuscript and are better described in response to Reviewer 1, point 4, above.  
+
+<|ref|>text<|/ref|><|det|>[[118, 260, 861, 291]]<|/det|>
+iv) We have quantified the number of ribosomes in association with DMVs in the presence and absence of G3BP, described better in response to the following comment.  
+
+<|ref|>text<|/ref|><|det|>[[118, 306, 864, 386]]<|/det|>
+e) The final argument for this function of G3BP is EM imaging showing differences between WT and RATA SARS infection. The key observations are that in RATA mutants and g3bpΔΔ cell lines the ribosomes are more diffuse and less concentrated around the double membrane viral organelles. While a single EM image can be interesting, to make these points robustly, those differences need to be quantified in some manner (beyond just the size of the LVCVs).  
+
+<|ref|>text<|/ref|><|det|>[[117, 402, 877, 625]]<|/det|>
+Indeed, our apparent reliance on a single image was a weak point in the original submission, and we are grateful for the opportunity to improve this aspect of the work. The originally presented image had in fact been chosen as representative of over 15 images taken. We have now taken more images of U2OS- ACE2 cells infected with WT or RATA mutant virus and performed blinded quantifications of the numbers of ribosomes associated with DMV membranes in those images. Specifically, images of 66 DMVs in WT and 57 in RATA- infected U2OS- ACE2 cells were shuffled and labelled in alphabetical order (by SL, first author) for quantification of ribosomes per DMV and measurement of DMV circumference (by MG, second author). Final numbers were collated and expressed as the number of ribosomes per \(\mu \mathrm{m}\) of DMV perimeter. The results show that there is a significant decrease in the number of ribosomes in association with DMVs in RATA infection compared to WT. Similar analyses were performed on images from infected U2OS- ΔΔGFP cells and show that, in the absence of G3BP1/2, the difference in number of ribosomes/DMVs is eliminated. These data are now presented in Fig 7g and S6a of the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[118, 641, 870, 688]]<|/det|>
+We believe this is strong evidence that the recruitment of G3BP to DMVs by the N protein, leads to more efficient translation of nascent viral mRNAs. Raw data are uploaded as source data file with this submission.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 706, 261, 720]]<|/det|>
+## Additional Issues:  
+
+<|ref|>text<|/ref|><|det|>[[118, 737, 867, 769]]<|/det|>
+2) How do the authors identify the infected cells in Figure 1a/b since the N protein IF is dim at 6 hours. Is the cell making stress granules at 6 hours infected? This should be clarified.  
+
+<|ref|>text<|/ref|><|det|>[[117, 785, 870, 896]]<|/det|>
+We believe those cells are infected but are not yet showing strong detectable signal for the N protein. In a new experiment, performed under the same conditions, we could detect dsRNA at earlier stage than N protein. To illustrate this, we include below an image of VeroE6 cells infected with WT SARS- CoV- 2 at 0.5 MOI for 6 hours and stained for G3BP1, dsRNA and N (Rebuttal Fig 7). The field captures cells at different early stages of infection. The indicated cell shows clear SGs, but still very low N protein signal, while the cells above left and above right have strong and intermediate N signals respectively but neither contain any SGs.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[125, 95, 636, 199]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[655, 100, 874, 213]]<|/det|>
+Rebuttal Figure 7: Vero E6 cells were infected with WT SARS- CoV- 2 at 0.5 MOl for 6 hours. Cells were fixed and stained for G3BP1 (green), dsRNA (red), N (grey) and Hoechst (blue).  
+
+<|ref|>text<|/ref|><|det|>[[118, 228, 644, 244]]<|/det|>
+We have better explained this on line 72 of the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[118, 260, 864, 323]]<|/det|>
+3) The specific model the authors put forth predicts that in the RATA virus (or in G3BPΔΔ cell lines), the translation efficiency of the viral RNAs would be low early in infection. This could be directly tested by measuring the rate of viral protein production and the levels of viral RNAs at the same time point and comparing WT to RATA.  
+
+<|ref|>text<|/ref|><|det|>[[118, 339, 874, 355]]<|/det|>
+Please see Reviewer 1, point 4, above for a full description of our new data to address this point.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[120, 85, 428, 100]]<|/det|>
+## Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 115, 866, 260]]<|/det|>
+In this article, Long and colleagues generate a SARS- CoV- 2 mutant (RATA) that lacks the ability to interact with G3BP1/2 proteins, which are RNA- binding proteins that promote stress granule assembly. The RATA mutant virus is attenuated in cell culture models and is less pathogenic in K18- hACE2 transgenic mice. In comparison to WT SARS- CoV- 2, G3BP displays reduced colocalization with dsRNA and N at DMV replication factories in the RATA mutant virus infection. The authors also observe less incorporation of puromycin at DMV in the RATA mutant virus infection, suggesting a reduction in local translation of viral RNA at the DMV. Based on these observations, the authors claim that G3BP1/2 proteins are host factors that are required for SARS- CoV- 2 replication by enhancing localized translation at viral DMV.  
+
+<|ref|>text<|/ref|><|det|>[[119, 275, 840, 323]]<|/det|>
+In general, the authors describe interesting cellular biology relating to how SARS- CoV- 2 N modulates G3BP functions during infection, which adds to a growing body of literature. The paper is well- written and the data is of high quality.  
+
+<|ref|>text<|/ref|><|det|>[[117, 338, 876, 547]]<|/det|>
+However, their data do not support their primary conclusion that G3BP is a pro- viral protein required for SARS- CoV- 2 replication. Specifically, the observation that SARS- CoV- 2 replicates to similar titers in parental and G3BP- KO cells (data not shown), which is consistent with other studies (Burke et al., RNA 2021 PMID: 34315815), indicates that G3BP is not a pro- viral protein that is required for SARS- CoV- 2 replication. These data strongly argue against their primary model. An alternative explanation for the attenuation of the RATA virus is that the reduction in interaction between G3BP1 and N during RATA mutant virus infection leads to enhanced G3BP1- mediated antiviral activity (i.e., type I IFN response, as suggested in their model) and thus reduce RATA mutant virus replication. However, the authors do not examine if and how the RATA mutant alters antiviral responses, nor do they test if the replication of the RATA mutant would be rescued in G3BP- KO cells as would be expected by this function of N. Because the RATA mutations of N could disrupt a number of putative N functions, it is unclear if altered G3BP interactions contribute to RATA virus attenuation.  
+
+<|ref|>text<|/ref|><|det|>[[118, 562, 862, 609]]<|/det|>
+Overall, their data do not support their primary conclusion that G3BP- N interaction is required for optimal viral replication via localization of translation at viral DMV, which weakens the impact of this article.  
+
+<|ref|>text<|/ref|><|det|>[[118, 626, 861, 657]]<|/det|>
+We thank the reviewer for the comprehensive critique of our work. Below, we provide point- by- point responses to each of the concerns.  
+
+<|ref|>text<|/ref|><|det|>[[118, 673, 867, 737]]<|/det|>
+Importantly, we would like to stress that we believe that G3BP is both an antiviral and a proviral factor. G3BP- dependent SGs are induced very early in infection, before viral protein production is detectable (Fig 1). Later, when N protein sequesters G3BP to RNA replication complexes, the protein then carries out its proviral functions.  
+
+<|ref|>text<|/ref|><|det|>[[118, 753, 867, 816]]<|/det|>
+Despite their economical genome coding and preponderance of multifunctional proteins, RNA viruses typically require multiple host factor interactions to carry out functions that cannot be coded for in the relatively small genomes. If one considers the G3BP interaction in this context, it makes sense that the virus sequesters an antiviral protein and uses it for proviral functions.  
+
+<|ref|>text<|/ref|><|det|>[[118, 833, 689, 848]]<|/det|>
+We address the major critiques in the reviewer's penultimate paragraph:  
+
+<|ref|>text<|/ref|><|det|>[[118, 864, 782, 879]]<|/det|>
+- "the authors do not examine if and how the RATA mutant alters antiviral responses"
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 880, 180]]<|/det|>
+We have now performed qPCR analyses of total mRNA from U2OS- ACE2 cells infected with SARS- CoV- 2 WT or RATA for 12 and 24 hours. The data indicate that IFN transcripts are equal (12h) or lower in number (24h) in RATA compared to WT infected cells. We therefore do not believe that the recruitment of G3BP by N protein in WT SARS- CoV- 2 infection can be implicated in evasion of the IFN response in infected cells, nor that this could be the molecular mechanism of RATA attenuation. These data are now included in Fig S3e of the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[119, 196, 870, 243]]<|/det|>
+Further support for this is the observation that the RATA mutant is also attenuated in Vero cells, which lack the ability to produce interferon (Emeny and Morgan, J Gen Virol, 1979, PMID 113494).  
+
+<|ref|>text<|/ref|><|det|>[[119, 260, 835, 276]]<|/det|>
+- "nor do they test if the replication of the RATA mutant would be rescued in G3BP-KO cells"  
+
+<|ref|>text<|/ref|><|det|>[[118, 291, 874, 355]]<|/det|>
+In response to the suggestion here, and in points 4 and 5 (below), of testing viral replication in G3BP- KO relative to parental cells, we have now performed WT and RATA viral replication analyses in U2OS- ΔΔGFP and U2OS- ΔΔGFP- G1- WT cells and present those data in Fig 3g of the revised manuscript and in Rebuttal Fig 8 below.  
+
+<|ref|>text<|/ref|><|det|>[[118, 370, 870, 466]]<|/det|>
+Firstly, to reduce experimental noise caused by unequal expression of EGFP- G3BP and of viral receptor ACE2, we have re- sorted the cell lines based on EGFP expression and on ACE2 expression before performing these analyses. In the interpretation of these data, we would remind the reviewer that we believe that G3BP has both antiviral and proviral effects and, since both are acting on viral replication in such experiment, we would urge caution in drawing conclusions from such a simple readout (infectious viral titre in extracellular medium).  
+
+<|ref|>text<|/ref|><|det|>[[118, 482, 875, 561]]<|/det|>
+The data show that in U2OS- ΔΔGFP- G1- WT cells, RATA is greatly attenuated relative to WT virus. This we believe is due to both antiviral effects of G3BP SGs and also the lack of the proviral effect of recruiting translational machinery to DMVs. However, in U2OS- ΔΔGFP (lacking G3BP), RATA replicated to titres much closer to WT, due to the absence of the antiviral effect and proviral effects.  
+
+<|ref|>image<|/ref|><|det|>[[118, 585, 393, 715]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[400, 610, 874, 690]]<|/det|>
+Rebuttal Figure 8: U2OS- ΔΔGFP and U2OS- ΔΔGFP- G1- WT cells were sorted for EGFP and ACE2 expression (see methods) and infected with SARS- CoV- 2 WT or RATA at 0.05 MOI. Samples were taken at indicated times and titrated by plaque assay on VeroE6 cells.  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 755, 280, 770]]<|/det|>
+## Specific comments:  
+
+<|ref|>text<|/ref|><|det|>[[118, 785, 864, 864]]<|/det|>
+1. Extended Figure 3D. Most infected cells do not contain G3BP1 complexes in cells infected with either RATA or WT virus. Most cells with G3BP1 granules do not appear to be infected. Are these SGs generated through paracrine signaling from infected cells? Also, eIF4G staining is only observed in SARS-CoV-2 infected cells. Does SARS-CoV-2 infection lead to an increase in eIF4G, or is this signal spectral crossover from viral N staining?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 875, 164]]<|/det|>
+The reviewer is correct that elF4G signal in the relevant image was contaminated with N signal. We had not noticed this and are grateful to the reviewer for pointing it out. We believe the problem arose from our use of an old aliquot of the Santa Cruz anti- elF4G antibody (sc- 133155). We have now repeated this experiment using a new batch of the same antibody and present the data in Fig S3a of the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[118, 180, 873, 260]]<|/det|>
+2. Burke et al. 2024 (PMID: 38295168) showed that an N-resistant G3BP1 could cause G3BP1 aggregates to form in SARS-CoV-2 infected cells, and that inhibition of elF4A but not sodium arsenite-induced phosphorylation of elF2-alpha increased G3BP1 interactions with viral RNA in large aggregates containing viral RNA and dsRNA. Because these findings are similar to those made by the authors in this study, the authors should cite accordingly.  
+
+<|ref|>text<|/ref|><|det|>[[118, 275, 872, 307]]<|/det|>
+The lack of citation of the Burke et al paper was an embarrassing oversight on our part. We have corrected this now (reference 39). Their work is discussed on lines 198- 200.  
+
+<|ref|>text<|/ref|><|det|>[[118, 322, 787, 338]]<|/det|>
+3. Fig. 1A. It is not clear that the cells with SGs are infected since they lack N protein.  
+
+<|ref|>text<|/ref|><|det|>[[118, 354, 864, 386]]<|/det|>
+We believe those cells are in a very early stage of infection and are not yet showing detectable signal for the N protein. Please see our response to Reviewer 2, point 2 for a fuller explanation.  
+
+<|ref|>text<|/ref|><|det|>[[117, 402, 879, 547]]<|/det|>
+4. Fig. 2G. The authors show that the SARS-CoV-2-RATA mutant is attenuated, as it replicates to lower titers in several cell types. While the authors claim that this is the result of disrupting G3BP1-N interactions required for maximal viral replication capacity, it could also be that this mutation disrupts normal N functions or disrupts interactions with other host proteins. If the attenuation of the RATA mutant is due to disruption of G3BP1, then knockout of G3BP1 would be expected to reduce SARS-CoV-2-WT virus replication capacity similarly. The authors should test if SARS-CoV-2 replicates to lower titers in G3BP-KO cells in comparison to parental cells, and if so, show that rescue of G3BP1 rescues viral replication. Notably, Knockout of G3BP1/2-KO did not reduce SARS-CoV-2 replication in A549 cells (Burke et al., 2024).  
+
+<|ref|>text<|/ref|><|det|>[[118, 563, 870, 603]]<|/det|>
+Firstly, we do not believe that the RATA mutation affects interactions with other proteins. This is based on previous work (Kruse et al., Nature Communications 2021; PMID 34799561), where the RATA mutant was originally described (in protein expression constructs, not in replicating  
+
+<|ref|>text<|/ref|><|det|>[[298, 605, 857, 675]]<|/det|>
+virus). In that study, quantitative proteomics from GFP-pulldown experiments (reproduced here in Rebuttal Fig 9, left), revealed that the RATA mutation (there named '2A') was highly specific for G3BP1 and 2 binding.  
+
+<|ref|>text<|/ref|><|det|>[[299, 706, 864, 753]]<|/det|>
+Rebuttal Figure 9: Quantitative mass spectrometry comparison of YFP tagged SARS-CoV N wt and 2A (RATA) purified from HeLa cells (data reproduced from Kruse et al., 2021, Fig S3c).  
+
+<|ref|>text<|/ref|><|det|>[[118, 801, 872, 865]]<|/det|>
+Viral replication assays in the G3BP KO cells are shown in Rebuttal Fig 8, above. Of note, the data presented by Burke et al., and mentioned by the reviewer, show viral titres at only one time point (24h post- infection), at which time, replication should be expected to still be in the logarithmic phase and not yet reached plateau (compare with our data in Vero cells in Fig 3).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 84, 866, 116]]<|/det|>
+5. Fig. 2G. If the RATA mutant leads to enhanced antiviral activities of G3BP, then it should replicate to equal titers as WT virus in G3BP-KO cells. The authors should consider testing this.  
+
+<|ref|>text<|/ref|><|det|>[[117, 132, 866, 163]]<|/det|>
+Viral replication assays presented above, now reveal that indeed the RATA mutant replicates to titres much closer to the WT virus in the U2OS- ΔΔGFP cells.  
+
+<|ref|>text<|/ref|><|det|>[[118, 180, 866, 228]]<|/det|>
+6. Lines 182-183. "As expected, N-RATA colocalized neither with G3BP1 nor dsRNA (Fig. 5b), demonstrating that SARS-CoV-2 N recruits G3BP1 to RTC". This statement is misleading based on the data in the figure, as N-RATA does co-localize with dsRNA but not with G3BP1.  
+
+<|ref|>text<|/ref|><|det|>[[118, 243, 865, 259]]<|/det|>
+We have corrected this mistake (now on lines 200- 202) and thank the review for pointing it out.  
+
+<|ref|>text<|/ref|><|det|>[[117, 275, 870, 307]]<|/det|>
+7. Fig. 6C. It is not obvious that ribosomes are reduced near DMV in the RATA mutant compared to WT virus. The authors should quantify this result.  
+
+<|ref|>text<|/ref|><|det|>[[118, 323, 828, 355]]<|/det|>
+We have now performed this quantification and present the results in Fig 7g of the revised manuscript. For a fuller discussion, please see our response to Reviewer 2, comment 1e.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[119, 85, 315, 101]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 120, 435, 137]]<|/det|>
+## Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 154, 858, 294]]<|/det|>
+The manuscript Long et al. investigates the pro- viral role of G3BP1/2 during SARS- CoV- 2 infection. Previous studies have established that the SARS- CoV- 2 Nucleocapsid (N) protein binds with G3BP1/2, inhibiting stress granule formation (which is believed to be antiviral). However, whether G3BP1/2 is itself pro- or anti- viral is a matter of some debate. To investigate the importance of N:G3BP1/2 binding, the authors use reverse genetics to generate a recombinant virus with a two amino acid substitution in the N protein (N: 115A; F17A). Compared to infection with wild type SARS- CoV- 2, the authors mutant exhibits reduced replication, N:G3BP1/2 binding, and increased stress granule formation.  
+
+<|ref|>text<|/ref|><|det|>[[118, 311, 868, 398]]<|/det|>
+Compared to the initial submission, this manuscript has been great improved. All of my major concerns were addressed, specifically 1) my concerns regarding novelty, 2) the specificity of the 115A, F17A mutant compared to F17A alone, 3) the lack of replication data in vivo, and 4) data strengthening their claim regarding an effect on translation of viral proteins. In addition, they have adequately addressed all of my minor critiques.  
+
+<|ref|>text<|/ref|><|det|>[[118, 415, 644, 432]]<|/det|>
+As such, I recommend this article is suitable for publication as is.  
+
+<|ref|>text<|/ref|><|det|>[[118, 450, 875, 484]]<|/det|>
+Again, we thank the reviewer for the comprehensive review of our work and for this positive recommendation.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[119, 85, 435, 101]]<|/det|>
+## Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[119, 120, 872, 172]]<|/det|>
+In this revised manuscript, the authors have improved the work. I am supportive of publication but recommend some final alterations be made to the manuscript to strengthen the work and remove ambiguities.  
+
+<|ref|>text<|/ref|><|det|>[[118, 189, 872, 276]]<|/det|>
+1) An important experiment is how the WT and RATA virus replicate in WT and G3BPΔΔ cells since this addresses whether G3BP proteins are primarily antiviral or proviral and the nature of the alteration in the RATA mutant. I thank the authors for adding these experiments. The work would be strengthened by clarifying the specific differences observed in g3bpΔΔ cells. As I look at the figure I observe:  
+
+<|ref|>text<|/ref|><|det|>[[118, 293, 848, 345]]<|/det|>
+a) The RATA mutant is 10-50 times less replicative than WT virus in WT cell lines, but only \(\sim 3X\) worse that WT in g3bpΔΔ cells, consistent with the authors interpretation that the major effects of this mutation are due to G3BP proteins.  
+
+<|ref|>text<|/ref|><|det|>[[118, 363, 856, 415]]<|/det|>
+b) It looks like WT virus is hindered in g3bpΔΔ cells suggesting G3BP proteins can promote replication, but this effect is not rescued by the reintroduction of G3BP1 (assuming I am interpreting the figure correctly).  
+
+<|ref|>text<|/ref|><|det|>[[118, 432, 810, 484]]<|/det|>
+I suggest: i) The authors clarify what the differences are for WT and RATA virus in the different cell lines and then ii) describe how they interpret those differences for the functional consequences of the N-G3BP interaction.  
+
+<|ref|>text<|/ref|><|det|>[[118, 502, 612, 518]]<|/det|>
+We refer the reviewer to our response to Reviewer 3, below.  
+
+<|ref|>text<|/ref|><|det|>[[118, 536, 870, 623]]<|/det|>
+2) The authors interpretation that N-G3BP promotes viral translation could be strengthened by quantifying the puromycin labeling experiments on a single cell level (using data they already have). Since most of the translation at this stage will be viral mRNAs, it would be predicted that either RATA in WT cells, or WT virus in g3bpΔΔ cells, should show overall reduced translation rates.  
+
+<|ref|>text<|/ref|><|det|>[[117, 641, 872, 780]]<|/det|>
+In the revised submission, we had included such analyses of the puromycin labelling data in VeroE6 cells infected with WT or RATA (manuscript Fig 7b and in Rebuttal Fig 1A right, below). Since we have observed that the effect on translation is very localised to the sites of viral RNA replication, we had presented maximum PMY intensity/cell since we believe it to be a more appropriate measure than mean intensity/cell, which averages signal over the whole cell. For the reviewer's interest, we here present the mean intensity analyses of the same data (Rebuttal Fig 1A, left). We observe a slightly lower although non-significant mean signal in RATA infected cells compared to WT.  
+
+<|ref|>text<|/ref|><|det|>[[118, 798, 874, 867]]<|/det|>
+Furthermore, we have now also similarly analysed the images from WT or RATA infected ΔΔ- GFP- G1- WT cells and included those data in a revised Supplementary Fig 7b and presented in here in Rebuttal Fig 1B. Indeed, the results strengthen our interpretation that viral mRNAs are more efficiently translated in WT than RATA.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[125, 81, 875, 275]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[118, 295, 853, 329]]<|/det|>
+Rebuttal Figure 1. Mean and maximum intensity of puromycin staining were calculated in CellProfiler for SARS- CoV- 2 WT or RATA infected VeroE6 (A) or (B) \(\Delta \Delta \mathrm{GFP - G1 - WT}\) cells.  
+
+<|ref|>text<|/ref|><|det|>[[118, 348, 873, 451]]<|/det|>
+We also bring to the reviewer's attention, data added in the previous version in response to reviewer 1, point 4. Those data show that whole cell lysates from VeroE6 cells, infected with SARS- CoV- 2 WT (MOI 0.05 for 6h) contain more newly produced Spike and N protein than cells infected with the RATA mutant, despite equivalent viral mRNA levels (Fig 7c,d of the revised manuscript). These data also support that WT viral mRNAs are more efficiently translated than RATA.  
+
+<|ref|>text<|/ref|><|det|>[[118, 470, 873, 522]]<|/det|>
+3) It would be appropriate to at least discuss the alternative model wherein G3BP plays a role in promoting virion packaging and release from the cells (Murigneux et al., Nature Communications, 2024). Could both models be true, or might there be a simpler resolution?  
+
+<|ref|>text<|/ref|><|det|>[[117, 540, 879, 679]]<|/det|>
+We discussed that on lines 300- 305 of the first revised manuscript and depicted it in Figure 8. Indeed, we believe that both models can be true. Murigneux and colleagues propose (last paragraph of their discussion) "a dual functionality of N/G3BP interactions: on the one hand N sequesters G3BP proteins to prevent antiviral SG formation and to circumvent subsequent antiviral immune responses and on the other hand, the virus hijacks the function of G3BP1/2 to favor production of infectious viral particle". Our work supports that and adds the extra function of facilitating viral mRNA translation at the sites of mRNA production, as depicted in Figure 8.  
+
+<|ref|>text<|/ref|><|det|>[[118, 696, 844, 748]]<|/det|>
+4) A terrific experiment would be to show immuno-gold localization of G3BP to the DMV with ribosomes on them. I would not require that for publication, but it would really strengthen the work.  
+
+<|ref|>text<|/ref|><|det|>[[118, 766, 878, 800]]<|/det|>
+We agree that this could be very nice to show, but unfortunately, we did not feel that we had the time or resources to do it at this time.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[119, 85, 435, 101]]<|/det|>
+## Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 120, 860, 223]]<|/det|>
+In this revised manuscript, the authors adequately addressed many reviewer concerns. However, their data in Fig. 3G do not support that G3BP1 is pro- viral during SARS- CoV- 2 infection. Specifically, the authors show that the RATA virus displays a reduced ability to generate plaque forming units in several cell lines, including Vero cells, MA- 104, and U2OS cells. This indicates that the RATA virus is attenuated. However, whether this is due to the inability of N to interact with G3BP1 during RATA infection was unknown.  
+
+<|ref|>text<|/ref|><|det|>[[118, 241, 864, 311]]<|/det|>
+To address this, the authors examined growth kinetics of WT and RATA via plaque assays in parental (U2OS- ACE2), G3BP1- KO (AGFP), and rescue (AGFP- G1- WT) U2OS cell lines. Several observations do not support that G3BP1 is a pro- viral host factor required for SARS- CoV- 2 replication based on data in Fig. 3G:  
+
+<|ref|>text<|/ref|><|det|>[[118, 328, 876, 416]]<|/det|>
+1. Knockout of G3BP1 in U2OS cells resulted in higher titers of WT virus by 36 hrs. p.i. Thus, G3BP1 is not required for SARS-CoV-2 replication, but in fact could reduce SARS-CoV-2 replication. Moreover, rescue of GFP-G3BP1 in the G3BP-KO cells (AGFP-G1-WT) did not enhance WT SARS-CoV-2 replication kinetics or in increase final PFU titers. This indicates that G3BP1 is not a host factor that enhances SARS-CoV-2 replication.  
+
+<|ref|>text<|/ref|><|det|>[[118, 433, 866, 486]]<|/det|>
+2. Knockout of G3BP1 in U2OS cells increased RATA virus replication, suggesting that G3BP1 perturbs RATA virus replication. Notably, RATA virus replication is reduced by GFP-G3BP1 in (AGFP-G1-WT) U2OS cell lines.  
+
+<|ref|>text<|/ref|><|det|>[[118, 504, 856, 556]]<|/det|>
+3. The RATA virus is attenuated in G3BP1-KO (AGFP) cells in comparison to WT cells. This indicates that the mutations in N that lead to attenuation of the RATA virus is G3BP1-independent.  
+
+<|ref|>text<|/ref|><|det|>[[118, 575, 872, 627]]<|/det|>
+4. Despite the overall higher replication of WT vs. RATA in all the cell lines, the growth kinetics between WT and RATA viruses is similar between parental (U2OS-ACE2) and G3BP1-KO (AGFP) cells.  
+
+<|ref|>text<|/ref|><|det|>[[118, 645, 879, 802]]<|/det|>
+In summary, if the interaction between SARS-CoV- 2 N and G3BP1 were required to enhance viral replication, then the authors would have observed a reduction in WT virus replication in G3BP1/2-KO cells to titers equivalent to RATA. Moreover, the reduction in WT virus fitness would be rescued in (AGFP-G1-WT) U2OS cell lines. However, the authors do not observe these effects. Instead, they observe that knockout of G3BP1/2 leads to higher replication of both WT and RATA, and that rescuing G3BP1 expression only reduces RATA. Combined, these data indicate that the RATA mutations attenuate SARS-CoV-2 replication independently of G3BP1/2 interactions, and that the slower replicating RATA virus might be more sensitive to the general antiviral effects of G3BP1 (interferon-independent).  
+
+<|ref|>text<|/ref|><|det|>[[118, 819, 866, 907]]<|/det|>
+Indeed, we agree with the reviewer that we cannot claim that G3BP is exclusively a proviral factor. As depicted in Figure 8, we believe that G3BP has both antiviral and proviral functions. The specific function of G3BP is determined by the proteins or RNA with which it interacts. G3BP is antiviral when it is free to induce SG formation, leading to arrest of viral protein translation. However, SARS- CoV- 2 and several other viruses (see citations in the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 85, 842, 170]]<|/det|>
+manuscript text), sequester the protein to their replication complexes in a way that both inhibits the antiviral functions and drives proviral functions at the sites of mRNA transcription and assembly. This view is shared by Murigneux and colleagues (Nature Communications, 2024 PMID 38245532), quoted above, and our work adds the proviral effect of efficient translation of WT viral mRNAs.  
+
+<|ref|>text<|/ref|><|det|>[[118, 189, 820, 222]]<|/det|>
+We have extensively edited the text of the paper, including title and abstract to better express the view that G3BP is both antiviral and proviral.  
+
+<|ref|>text<|/ref|><|det|>[[118, 241, 866, 415]]<|/det|>
+In our first revision, we requested caution in the analyses of viral growth replication data in these cell lines and we repeat that here. We believe it is not appropriate to compare viral replication curves across the different cell lines but rather it is better to compare WT and RATA virus replication in each cell line separately. The CRISPR knock out cells and GFP- (G3BP1) reconstitutions were generated 10 years ago, and the cells have been passaged independently since then. In addition, in 2021, all 3 cell lines (U2OS parental, \(\Delta \Delta \mathrm{GFP}\) and \(\Delta \Delta \mathrm{G3BP1 - GFP}\) ) were transduced with ACE2- TMPRSS2- expressing lentiviruses, placed under selection and passaged independently. We have tried to minimise variation caused by unequal expression of these transgenes, but the possibility remains that the viruses might be better able to enter and initiate infection in one cell line than the others.  
+
+<|ref|>text<|/ref|><|det|>[[118, 433, 867, 573]]<|/det|>
+Comparing the two viruses, we observe that RATA exhibits 33- fold lower replication compared to WT in U2OS parental cells, 25- fold lower replication compared to WT in \(\Delta \Delta \mathrm{GFP - G1 - WT}\) cells, but only a 4- fold reduction compared to WT in cells lacking G3BP1/2 ( \(\Delta \Delta \mathrm{GFP}\) cells). We believe therefore that the attenuation is largely G3BP1- dependent and is a result of the combination of the antiviral effects of SGs and the loss of G3BP's proviral effects. However, as the reviewer points out (point #3), the 4- fold reduction in \(\Delta \Delta \mathrm{GFP}\) cells suggests some G3BP- independent attenuation of RATA, which we cannot exclude. We discuss this on lines 144- 149 of the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[118, 591, 870, 781]]<|/det|>
+We do not believe that RATA is a "slower replicating" virus, since viral RNA levels were equal at 6hpi (Fig 7c), but that the major attenuation effects are seen downstream of the translation of viral mRNAs. Further, we have been careful to define the specificity of the RATA mutation for G3BP (see comment to Reviewer 1, point 2 in first revision) and to show that the mutation does not alter RNA- binding and LLPS properties (figure 5), as well as the binding to other viral structural proteins for assembly (Nature Communications, 2021, PMID: 34799561). Of note, a recent study found that point mutations in the N protein's intrinsically disordered regions (IDRs) can have "nonlocal impact and modulate thermodynamic stability, secondary structure, protein oligomeric state, particle formation, and liquid- liquid phase separation" (Nguyen et al Elife 2024; PMID: 38941236), a possibility we now mention in the text.  
+
+<|ref|>sub_title<|/ref|><|det|>[[120, 800, 295, 816]]<|/det|>
+## Additional comments  
+
+<|ref|>text<|/ref|><|det|>[[119, 835, 838, 886]]<|/det|>
+Considering the observation that their viral N staining is capable of contaminating other channels, the authors should consider repeating key results with no- primary controls to confirm that any co- localization results are not due to spectral crossover.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 85, 870, 170]]<|/det|>
+To respond to the reviewer's suggestion, we here present images of U2OS- ACE2 cells infected with SARS- CoV- 2 WT or RATA and stained with a combination of antibodies against G3BP1, dsRNA and N protein (Rebuttal Fig 2, top left), or the same lacking either G3BP1 (bottom left), dsRNA (top right) or N (bottom right). The data show that, using the same antibodies, techniques and hardware used in the paper, no spectral crossover was detected.  
+
+<|ref|>text<|/ref|><|det|>[[117, 189, 872, 277]]<|/det|>
+We are now confident that the fluorescence signals in all images are uncontaminated by other channels. For example, to illustrate that N signal is not contaminating other channels, one might examine images from the manuscript of RATA infected cells, where N (Alexa Fluor 647) does not co- stain with G3BP1 (Alexa Fluor 488; Fig 6a), GFP/elF4A (GFP or Alexa Fluor 568; Fig 6c), or GFP (Fig 6e).  
+
+<|ref|>image<|/ref|><|det|>[[117, 312, 868, 797]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[117, 808, 860, 843]]<|/det|>
+<center>Rebuttal Figure 2. U2OS-ACE2 cells were infected with SARS-CoV-2 WT MOI 0.5 and cells were fixed and stained with the indicated antibody combinations at 6 hours post infection. </center>
+
+<--- Page Split --->

peer_reviews/supplementary_0_Transparent Peer Review file__ea70799eb1d530d0b6bfa2b1a6bd3034e6bc63286fb046e26e168ba379e2933c/images_list.json

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

peer_reviews/supplementary_0_Transparent Peer Review file__ea70799eb1d530d0b6bfa2b1a6bd3034e6bc63286fb046e26e168ba379e2933c/supplementary_0_Transparent Peer Review file__ea70799eb1d530d0b6bfa2b1a6bd3034e6bc63286fb046e26e168ba379e2933c.mmd

@@ -0,0 +1,664 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+# Isotope Encoded Spatial Biology Identifies Plaque-Age-Dependent Maturation and Synaptic Loss in an Alzheimer's Disease Mouse Model  
+
+Corresponding Author: Dr Jörg Hanrieder  
+
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+Version 0:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+Understanding the evolution of Aβ in Alzheimer's etiology is crucial for elucidating the mechanisms underlying disease progression, and this study partially addresses this aspect. The authors applied the iSILK technique to track amyloid plaque formation, maturation, neurotoxicity, and synaptic loss in a spatiotemporal way. The authors employ the AppNL- F/NL- F aged mouse model to examine how plaque formation progresses over time. The study presents an innovative approach by integrating mass spectrometry imaging (MALDI MSI), spatial transcriptomics, and hyperspectral microscopy to assess plaque heterogeneity and its associated effects.  
+
+However, critical technical and conceptual issues need to be addressed before this manuscript can be considered for publication.  
+
+1. One major limitation of this study is the lack of consideration for known post-translational modifications (PTMs) of Aβ, which are known to influence its aggregation and pathological progression. The centroid mass values of MALDI MSI (Table S1) do not reflect known Aβ PTMs, which should result in characteristic mass shifts. The authors should respond the following questions regarding the peak assignment of their MALDI MS spectra.  
+
+1) Why were these known PTMs not detected in the dataset? 
+2) Does the dataset contain any mass peaks that could suggest age-dependent PTM changes in Aβ?  
+
+2. The authors should explain why no background peptide signals are observed in the spectrum. Additionally, they should present the MS/MS spectra to confirm the presence of amyloid beta and assess whether other peptides were detected.  
+
+3. The authors claim that older amyloid plaques exhibit increased neurotoxicity and synaptic loss, as demonstrated through iSILK labeling, transcriptomics, and fluorescence imaging. While the data suggest a correlation between plaque age and synaptic dysfunction, it is not yet clear how this study advances our understanding beyond prior work that has already reported synapse loss in the vicinity of amyloid plaques.  
+
+To fully support their conclusions, the authors should clarify what new biological insights this study provides compared to previous findings on plaque-induced toxicity.  
+
+## Reviewer #2  
+
+(Remarks to the Author)  
+
+Dear Editor and authors,  
+
+I have reviewed the paper "Isotope Encoded Spatial Biology Identifies Amyloid Plaque- Age- Dependent Structural Maturation, Synaptic Loss, and Increased Toxicity" by Hanrieder and co- workers. The paper aims to understand the initiation and progression of Aβ aggregate accumulation and correlations with CNS cell responses by spatial transcriptomics in the proximity of early and late Aβ aggregates. Chemical time stamps were induced using pulse chase of food supplemented with 15N- isotope that label newly produced protein facilitating using a method called Imaging of stable isotope labelling kinetics (iSILK). As a complementary time stamp, the maturity of the Aβ- plaque structures was monitored by LCO fluorescence technology developed by researchers at Linkoping University, Sweden.  
+
+Spatial transcriptomics together with iSILK and LCO/immunofluorescence structure correlations allowed an unprecedented
+
+<--- Page Split --->
+
+correlation of proximal cell response to the plaque maturity stage. The study concluded that mature Aβ plaques, in comparison with early plaques, regardless of chronological mouse age, were associated with changes suggesting synaptic toxicity responses. Early plaque formations showed increased immune response gene expression. This methodological approach puts within reach more information on differentiated cellular responses to plaque fibril structure at different plaque development stages and cell types (microglia, astrocytes, and diverse neuronal populations).  
+
+Overall the impression of the innovative application of different integrated techniques and the presentation of the paper is very positive.  
+
+There are some of points that should be modified and clarified in the revised version to improve the paper:  
+
+1. The choice of the APPNL-F knock-in mouse in this study was clever, because this mouse makes almost exclusively Aβ1-42, allowing the authors to specifically iSILK-monitor Aβ1-42 species using MALDI-ToF imaging. This feature should be more clearly stated in the description of the selection of the mouse model. Suggestively at Results under the header "iSILK delineates spatial and structural patterns of plaque formation and maturation" ... after the sentence in the first paragraph "This gradual increase in plaque pathology with age more closely resembles the human disease". Add something like this: "The choice of the APPNL-F knock-in mouse in this study provided another biochemical advantage. Because this mouse makes almost exclusively deposited Aβ1-42, it allows us to specifically iSILK-monitor Aβ1-42 species using MALDI-ToF imaging".  
+
+2. Under the header "Amyloid plaque maturation is characterized by continuous fibrilization with age" There is a description of the LCO and especially qFTAA and hFTAA. The reference here is good but insufficient. LCO reference 21 is ok for general purposes but then the following (see below) should be added in the description. While the references to the original publications on this method are present in the paper (later in the discussion), from reading this result presentation it appears that the description comes at face-value from the present study.  
+
+So suggested placement of references: "This is enabled by the difference in affinity of the two LCO probes, q-FTAA and h- FTAA, towards amyloid aggregates. Specifically, q-FTAA preferentially binds to mature and compact beta-pleated aggregates, while h-FTAA binds to less compact, yet still beta- pleated aggregates (reference: Now ref 34 in the list of references Nystrom, S. et al. Evidence for age-dependent in vivo conformational rearrangement within Abeta amyloid deposits). Due to their different emission profiles, the LCO probes can be spatially delineated using hyperspectral fluorescent microscopy. Here, the ratio of the LCO maxima (500 nm for q- FTAA / 580 nm for h-FTAA) is used to express preferential binding of either of the two LCO probes used, whereby an increase in 500 nm intensity is indicative of increased q-FTAA binding and therefore, increased structural maturity of the amyloid fibrils (Fig. 2B). (reference: Now ref 35 Rasmussen, J. et al. Amyloid polymorphisms constitute distinct clouds of conformational variants in different etiological subtypes of Alzheimer's disease. + Now ref 45 Parvin, F. et al. Divergent Age-Dependent Conformational Rearrangement within Abeta Amyloid Deposits in APP23, APPPS1, and App(NL-F) Mice.).  
+
+3. The spacial transcriptomics data should be clarified. It is not clear how the reference data (baseline normalized expression) were obtained for making the volcano-plots of overexpression and decreased expression in Fig. 3C and 3F? In the same context: How to interpret the GeoMx CSV files when there is no reference data set included. Was this from already published databases? If so, the used numbers from this reference data set should be included. They can be very useful for others looking for other genes and to derive the data associated with the volcano plot (discussed above). The contrast (samples vs reference) is not obvious from the CSV-files. It appears to be only numbers listed for 10 and 18 month APPNL-F mice.  
+
+Furthermore, the link given:  
+
+https://hanriederlab.shinyapps.io/PlaqueAgeTranscriptomics/  
+
+can be very useful if it more clearly described how it was obtained and what cutoff values are used for significance. Is the x- axis plaque age (in months?) based on iSILK or q-FTAA/h-FTAA fluorescence and what is Log Cnorm? A clear description of how to use the link should be included in the paper with more descriptions in the methods section.  
+
+4. The list of acknowledgements for funding is very long. But there is no acknowledgement or information for where the authors obtained the LCOs q-FTAA and h-FTAA which plays an important role in the paper. Since these molecules appear not to be commercially available it needs to be specified. If they were synthesized in their own lab this should be stated in the materials and methods.  
+
+## Reviewer #3  
+
+(Remarks to the Author)  
+
+In this manuscript, Wood & Dulewicz et al. describe the combination of mass spectrometry imaging (MSI) and hyperspectral imaging to measure Ab42 plaque age and morphology in a mouse model of Alzheimer's disease. Using a pulse- chase strategy where mice are fed with a 15N- enriched diet, they were able to distinguish nascent and aged plaques. Hyperspectral imaging with oligothiophene chemistry further demonstrates changes in plaque morphology correlated to age. They further combine this approach with single- plaque GeoMx spatial transcriptomics to identify correlates with age inferred from their MSI approach.
+
+<--- Page Split --->
+
+Overall, the experiments are executed quite well, and the methods and data will be useful for the spatial and Alzheimer's biology community. In the future, I also see great value in combining these techniques with higher- resolution spatial transcriptomics or multiomics to understand the molecular underpinning of these processes. While I find their iSILK MSI technology to be technically impressive, especially when combined with these orthogonal measurements, I believe the richness of their data could have been used to better extract biologically novel conclusions. The manuscript would also benefit from more extended analyses, embedding with more specific knowledge in literature, and better bookkeeping of the collected data. If the authors could address or comment on these points, I believe the manuscript will be significantly improved and I would be supportive of publication.  
+
+## Major comments  
+
+1. The use of MSI pulse chase is quite inspired. I agree that 15/14N ratio is a good proxy for plaque age. Here are some points the authors may want to consider:  
+
+a. It seems like a missed opportunity for developing a numerical index or 'clock' here for plaque age. This could be used to assign ages to plaques without MSI, for instance. If hyperspectral imaging or morphology could be used to do a regression analysis, I think that could be immensely useful. I understand that this may be challenging to establish rigorously, but could be a point of discussion for future work.  
+
+b. It would be interesting for the authors to look at how the MSI and hyperspectral imaging replicates within and across animals. Are these correlations sufficiently strong and hence, biologically robust across animals? A breakdown of Fig. 2e by animal for example could be helpful.  
+
+c. Could the authors briefly comment on the feasibility and cost of deploying this technology? What is the feasibility as well of MSI and LCO imaging on the same section?  
+
+2. I have some general reservations about the GeoMx analysis. Given the richness of the dataset, the analysis should give correspondingly rich and novel biological insights. Broadly, are there any unexpected findings here?  
+
+a. Based on the sequencing, the authors suggest dysregulation of synaptic, immune and metabolic genes proximal to plaques, in a manner correlating with age on adjacent sections. Rather than a change in regulatory patterns, is it possible that these changes are explained instead by differences in cell population composition? Much of this is also driven by FOV (or rather, AOI) size and how many cells are expected to be captured.  
+
+i. Neuronal mRNA is typically enriched in the soma, with a number of exceptions, and often, in an activity-dependent manner. Thus, depletion of neurons might be an equally fair explanation of apparent decrease in synaptic gene expression observed in the GO analysis. This is likewise for immune/metabolic genes – as recently observed with high-resolution spatial data in 5XFAD mice. The authors could consider using morphological (Nissl, silver, DAPI etc.) staining or further immunofluorescence to quantify the cell populations here. Further markers for DAA/DAMs might also be helpful.  
+
+ii. In a similar vein, are there any changes in mRNAs known to be translocated to synapses, versus those known to be somalocalized?  
+
+iii. Do the authors observe upregulation of cryptic genes associated with nonnative cell types in these plaques? For instance, cell type markers of unexpected immune cells.  
+
+iv. Can the authors further embed some of these findings with more recent literature on functional changes in astrocytes and microglia?  
+
+b. The statistics on correlations between the GeoMx and plaque age should consider multiple testing correction if possible (q-val or FDR perhaps). My understanding is that raw p-values are shown. Are these conclusions still robust?  
+
+c. While the viewer is helpful, for some example genes, could the authors directly show the scatterplots demonstrating the correlations between gene expression and plaque age in the main text? Regressions with the corresponding coefficients might also be helpful.  
+
+d. Could the authors also provide some commentary on the challenges of hyperspectral imaging prior to GeoMx on the same tissue slice? As they mention, serial sections preclude the analysis of nascent small plaques, which would certainly be biologically fascinating. Do the authors see this as a major technical limitation? To my knowledge, there are several groups combining multimodal imaging and GeoMx imaging and sequencing on the same sections as well, so I am interested in just hearing some perspective on this – though a experimental demonstration would be most impressive.  
+
+3. There are some ways the data could also be better organized and bookkept, for the purpose of transparency:  
+
+a. Can the authors enumerate the number of plaques and their age distribution captured by MSI within and across animals?
+
+<--- Page Split --->
+
+b. Figure S1 is insufficiently described to let me understand the structure of the data. A more thorough caption and explanation would be helpful, including a breakdown by source animal rather than just age.  
+
+c. Could the authors clarify what is shown in Fig. S2b? I am not sure how to interpret these values. It might be more useful to show PCAs instead.  
+
+a. Could we have a complete description of all GeoMx FOVs imaged, and their corresponding 15/14N ratios, morphologies and hyperspectral ratios, anatomical localization, number of unique genes and transcripts sequenced, etc? It would also be important to know the source animals for each FOV. This is an essential quality control metric for readers to make critical judgments about the data. This could for example be in the form of a table (for small numbers of FOVs), or a heatmap. It will give me a better idea of how powered the conclusions are.  
+
+## Minor comments  
+
+1. It might be helpful for the authors to further comment on the applicability of their method in conjunction with higher-resolution spatial transcriptomics methods (e.g. Visium HD, MERFISH, Xenium etc.) or proteomics (CODEX, DVP etc.) in future work. Integrating iSILK with the growing landscape of these tools is certainly a point of interest for many spatial biologists.  
+
+2. Some general commentary of the limitations of their approach might be helpful.  
+
+3. Fig. 1a scale bar value is not provided. The imaging modality should also be stated.  
+
+4. As a matter of preference, it might be better for authors to use perceptually uniform colormaps instead of spectral (e.g. in Fig. 1f). It may be hard for some readers to see.  
+
+5. Legends in some panels (e.g. Fig. 1e, 2c) also are rather small and hard to see.  
+
+6. Gene and GO annotations for Fig. 3c-h are also hard to read.  
+
+7. Could the MSI for Fig. 3b be decomposed into 15/14N for clarity?  
+
+8. In "spanning across two consecutive 12mm sections" did the authors means 12 micron?  
+
+Version 1:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+I have now reviewed the revised manuscript, "Isotope Encoded Spatial Biology Identifies Amyloid Plaque-Age-Dependent Structural Maturation, Synaptic Loss, and Increased Toxicity," along with the authors' point-by-point rebuttal to the concerns I raised in my initial review.  
+
+The authors have addressed all of my previous concerns by conducting significant new experiments to validate their findings and strengthen their claims. The new high-sensitivity LC-MS/MS analysis to investigate post-translational modifications and the inclusion of MS/MS spectra for sequence confirmation have fully resolved my initial questions regarding the data.  
+
+As a final point of scientific interest, I would like to offer one minor point for the author's consideration. In the MS/MS spectrum for Aβ 1- 42 shown in Supplementary Figure 1F, it is interesting to note the dominance of the b- ion series, while the corresponding y- ion series is almost absent. A brief comment on this observation in the Results section or the caption of the figure could enhance the spectral interpretation of the fragmentation behavior.  
+
+Reviewer #2  
+
+(Remarks to the Author)  
+
+The authors have made appropriate clarifications of all my previously raised points in the first round of detailed review and the paper has been improved. Congratulations to very interesting and important work. I support publication as soon as possible.  
+
+Reviewer #3  
+
+(Remarks to the Author)  
+
+The presentation of the data in the revised manuscript makes it much more straightforward to read and interpret the data in
+
+<--- Page Split --->
+
+my opinion. Here are some remaining minor comments for the authors, though they are mostly either presentation issues or nominal analyses that might provide some further insights if the authors choose to perform them. Nevertheless, I believe they have satisfied my major concerns, and I am supportive of publication.  
+
+## Minor Comments  
+
+1. In their discussion of limitations of other high-res ST approaches, they might note that the CosMx 18k now offers near-transcriptome coverage – though at markedly lower gene-wise sensitivity  
+2. I would appreciate if the authors could do one additional analysis – to highlight novel gene-gene correlation modules that arise across from analysis across plaques (e.g. with WGCNA or similar).  
+3. The authors might comment on whether their observed gene expression changes signify some T cell infiltration (or any other evidence for this in their animals) versus microglial activation due to overlap in GO annotations. This was what I meant by cryptic expression in my previous review and I apologize for the misunderstanding.  
+4. The statement that APP NL-F is strictly better than NL-G-F as it reflects human pathology (Line 387-389) is too strongly made.  
+
+5. Scale bars missing in Fig. 3A, 4A.  
+
+6. Labels for GFAP, Ab, and SYTO colors are missing in Fig. 3B, Supp Fig. 5, 6.  
+
+7. I am not sure why the data must be duplicated as Supp Fig. 3D – would be preferable to instead include these panels in Fig. 3B?  
+
+8. The bottom axis labels for Fig. 3C, D are clipped.  
+
+Typographical  
+
+1. Line 190 "oligothiophene (LCO)" > "oligothiophenes (LCOs)  
+
+2. Line 429 "information" > "information"  
+
+Open Access This Peer Review 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 cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+The images or other third party material in this Peer Review 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 https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+
+## Dear Reviewers  
+
+Please find attached with this resubmission our revised manuscript "Isotope Encoded Spatial Biology Identifies Amyloid Plaque- Age- Dependent Structural Maturation, Synaptic Loss, and Increased Toxicity" by et al. We are grateful for the overall positive, thorough and constructive feedback provided, which helped significantly to both clarify and improve the manuscript substantially.  
+
+Please find below a detailed point- by- point response to the comments raised by the reviewers. The corresponding changes have been highlighted in an additional manuscript file (article_tracked_changes).  
+
+With best regards Jörg Hanrieder  
+
+## Reviewer #1:  
+
+Comment 1: "One major limitation of this study is the lack of consideration for known post- translational modifications (PTMs) of \(\mathsf{A}\beta\) , which are known to influence its aggregation and pathological progression. The centroid mass values of MALDI MSI (Table S1) do not reflect known \(\mathsf{A}\beta\) PTMs, which should result in characteristic mass shifts. The authors should respond to the following questions regarding the peak assignment of their MALDI MS spectra."  
+
+(1) Why were these known PTMs not detected in the dataset?  
+
+We concur that this is an important point. Unfortunately, it is a limitation of this mouse model that no prominent \(\mathsf{A}\beta\) truncations or PTMs are observed, specifically \(pE - x\) and PSer8- x (Saito et al. 2014). To underpin this observation, we performed additional, in- depth ex situ mass spectrometric experiments to investigate potential sequence truncations/modifications that might be suppressed or below the detection limit in the MALDI MSI analyses.  
+
+Specifically, we performed two elaborate ex situ analyses of brain tissue extracts, whole proteomic profiling of purified fibrils and immune- precipitation LC- MS/MS of amyloid peptides.  
+
+To this end, we first immune- precipitated \(\mathsf{A}\beta\) peptides from \(\mathsf{A}\mathsf{p}\mathsf{p}^{\mathsf{N L - F}}\) brain extracts and analyzed the samples by LC- MS/MS on a high- resolution Orbitrap mass spectrometer (Supplementary Fig. 1). Resulting data was searched in PEAKS (de novo peptide sequencing) for characterizing potential posttranslational modifications, including phosphorylation, deamidation, oxidation and pyroglutamation. The analysis confirmed the presence of \(\mathsf{A}\beta\) 1- 38, 1- 39, and 1- 42, with 1- 42 being the most prominent species. Following sequence analyses, no PTMs were observed for either species. Please find an excerpt from the obtained PEAKS results, now detailed in Supplementary Fig. 1B, in the table below:
+
+<--- Page Split --->
+
+
+Table 1: Excerpt from PEAKS results of the anti-Aβ IP-LC-MS/MS experiment.   
+
+<table><tr><td>Gene name</td><td>Peptide</td><td>Mass</td><td>Length</td><td>ppm</td><td>m/z</td><td>z</td><td>RT</td><td>PTM</td></tr><tr><td rowspan="3">APP</td><td>DAEFRHDSGYEVHHQKLVFFAEDVGS</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>NKGAIIGLMVGG</td><td>4129,012</td><td>1-38</td><td>3,1</td><td>826,81213</td><td>5</td><td>43,179</td><td>/</td></tr><tr><td>DAEFRHDSGYEVHHQKLVFFAEDVGS</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td rowspan="2">APP</td><td>NKGAIIGLMVGGV</td><td>4228,08</td><td>1-39</td><td>2,5</td><td>846,62537</td><td>5</td><td>44,834</td><td>/</td></tr><tr><td>DAEFRHDSGYEVHHQKLVFFAEDVGS</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>APP</td><td>NKGAIIGLMVGGVVIA</td><td>4511,27</td><td>1-42</td><td>1,1</td><td>903,26215</td><td>5</td><td>68,445</td><td>/</td></tr></table>  
+
+The obtained RAW data and results have been uploaded to the PRIDE repository under the accession number PXD063759.  
+
+Second, we analyzed fibrils purified from NLF brain. For this we used proteomics as the enzymatic digestion step (trypsin) provides increased sensitivity towards potentially truncated species. In contrast, this approach does not necessarily provide exact annotation of N- terminal modification for different C- terminal species (i.e x- 40, x- 42). Similarly to the IP- MS analyses, no PTMs were observed for the Aβ fragments. The corresponding raw data is accessible on the PRIDE repository under the accession number MSV000092311.  
+
+All experiments described above are now detailed in methods section "Ex situ Aβ Analysis" and have been incorporated into the main text on page 25 line 24.  
+
+(2) Does the dataset contain any mass peaks that could suggest age-dependent PTM changes in Aβ?  
+
+We thank the reviewer for this relevant question. Neither our MALDI MSI nor LC- MS/MS experiments (see Table above and Supplementary Fig. 1) suggest any presence of PTM for Aβ.  
+
+In our MALDI MS data, the phosphorylation would be detectable as the mass shift is +98 (H2PO2) or +80 (PO3), while a shift in 15N can be maximal +55 (as there are only 55 nitrogen in the Aβ1- 42 sequence). Similarly, the 3pE-x or 11pE-x modification would require any peak close to the unmodified 3-x and 11-x, neither of which is observed.  
+
+For the raw LC- MS/MS data, please refer to the PRIDE repository (accession numbers PXD063759 and MSV000092311), and for the raw MALDI MSI data, see the Shiny app: (https://maciejdulewiczgu.shinyapps.io/MALDI_GEOMX_VOLCANO/), where we present the data in an easily accessible format for the public.  
+
+Please see Supplementary Fig.1.  
+
+Comment 2: "The authors should explain why no background peptide signals are observed in the spectrum.  
+
+The focus of the current work was on imaging amyloid peptides that are inherently difficult to detect in situ due to their hydrophobic properties and highly aggregated state. Consequently, we have been tuning our protocols to enhance the amyloid signal, which requires extensive washes and formic acid hydrolysis. This extensive sample
+
+<--- Page Split --->
+
+preparation leads, in turn, to a reduced signal of other less aggregated, endogenous peptides and small proteins. We performed control experiments with and without formic acid to demonstrate the effects of FA retrieval on amyloid signal, as shown below in Figure 1.  
+
+This is now clarified in the methods on page 22 line 20 and in a newly created Supplementary Fig. 2.  
+
+"Additionally, they should present the MS/MS spectra to confirm the presence of amyloid beta and assess whether other peptides were detected."  
+
+We thank the reviewer for this comment and agree that MS/MS confirmation of \(\mathsf{A}\beta\) presence is important. Due to the limited sensitivity of direct in situ MALDI MS/MS, we employed extensive ex situ analyses for \(\mathsf{A}\beta\) sequence validation as described above under Comment 1. Briefly, we performed anti- \(\mathsf{A}\beta\) immunoprecipitation followed by LC- MS/MS as well as LC- MS/MS- based proteomics of purified amyloid fibrils, all on \(\mathsf{A}\beta^{\mathsf{N L - }}\) \(F\) mouse brain tissue extracts. Here, LC- MS/MS confirmed the presence of \(\mathsf{A}\beta 1 - 42\) along with \(\mathsf{A}\beta 1 - 38\) and \(\mathsf{A}\beta 1 - 39\) . A representative MS/MS spectrum of \(\mathsf{A}\beta 1 - 42\) is shown in Supplementary Figure 1F.  
+
+Our MALDI MSI data predominantly show \(\mathsf{A}\beta 1 - 42\) within plaques, with \(\mathsf{A}\beta 1 - 38\) detected in only two ROIs/plaques (low S/N ratio), while 1- 39 was not detected in any MALDI MSI ROI. This further highlights that \(\mathsf{A}\beta 1 - 42\) is indeed the predominant \(\mathsf{A}\beta\) species. The identification of \(\mathsf{A}\beta 1 - 38\) and \(\mathsf{A}\beta 1 - 39\) by LC- MS/MS is not unexpected and rather highlights the high sensitivity of this method (as compared to direct in situ detection via MALDI MSI).  
+
+Please see Supplementary Figure 1.  
+
+Comment 3: "The authors claim that older amyloid plaques exhibit increased neurotoxicity and synaptic loss, as demonstrated through iSILK labeling, transcriptomics, and fluorescence imaging. While the data suggest a correlation between plaque age and synaptic dysfunction, it is not yet clear how this study advances our understanding beyond prior work that has already reported synapse loss in the vicinity of amyloid plaques. To fully support their conclusions, the authors should clarify what new biological insights this study provides compared to previous findings on plaque- induced toxicity."  
+
+We appreciate this comment and would like to offer an extended explanation and discussion. Most importantly, the methods described allow us to decouple plaque age from chronological age of the mice. This brings several advantages that would not be discernible with static methods.  
+
+1) Most importantly, this reveals and validates that plaques indeed affect synapses and that persisting plaque ageing/maturation is associated with proximal synaptic dysfunction and loss, rather than plaques exerting immediate, acute toxicity. In static experiments one could not discern the age of individual plaques within one mouse/age group. Hence it would not be possible to assess whether the differences in plaque phenotypes and their effects (for example on synapses) are a result of plaque maturation or whether they represent different subgroups of plaques within a single tissue. Our methods clearly allow for this distinction.
+
+<--- Page Split --->
+
+2) Even more important, on the transcriptomic scale, the effect of plaque maturation is easily masked when comparing plaques with each other across the same mouse ages. For this, we further demonstrate the advantage of correlation of similar, spatially refined entities as compared to both bulk RNAseq and static spatial transcriptomics (ST) experiments. This is demonstrated in the more sensitive detection of synaptic genes that vary across plaques within the same mouse. To highlight this, we provide additional data demonstrating how both spatial transcriptomics (as compared to bulk brain extract transcriptomics), as well as dynamic spatial transcriptomics, improve sensitivity and specificity to plaque and plaque age molecular processes (please see Supplementary Fig. 10).  
+
+3) We further demonstrate conceptual novelty in integrating chemical imaging with spatial transcriptomics. Of note, this is demonstrated by the combination of functional LCO microscopy with iSILK and ST, showing that plaque maturation involves the continuous fibrillization at the plaque.  
+
+4) An additional advance is related to the technical innovation, implementing the iSILK method in old mice. While challenging, we succeeded in following plaque formation from before plaques form (6-10mo) during plaque growth at old ages (18mo). This showed that for this knock in model, plaques form as small cores in the cortex consisting of 1-42.  
+
+To clarify the novelty of our biological insights, we have added the following sentences to the discussion and embedded references to the most recent and impactful spatial transcriptomics studies in the field (page 15 line 18):  
+
+"Whereas findings from other impactful spatial transcriptomics studies on AD brain tissue (Chen et al. 2020, Mallach et al. 2024) have also supported accumulated rather than acute plaque toxicity, they were unable to directly link plaque age with increasing synaptic dysfunction, entirely decoupled from chronological mouse age. In fact, many studies have focused on the spatial rather than temporal aspect of plaque toxicity (Wood et al. 2022, Johnston et al. 2025) as well as focused on subsets of genes associated with e.g. glial activation in response to plaque pathology. Our study, in contrast, allowed us to evaluate the effects of maturing plaques independent of mouse age on a whole transcriptomic scale, with key findings being validated with direct immunological stainings. Additionally, the mouse model used in this study more accurately reflects human AD as this combines WT Abeta pathology together with ageing, which contrasts with the NLGF mouse model, on which most prior conclusions have been based."
+
+<--- Page Split --->
+
+## Reviewer #2:  
+
+Dear Editor and authors, I have reviewed the paper "Isotope Encoded Spatial Biology Identifies Amyloid Plaque- Age- Dependent Structural Maturation, Synaptic Loss, and Increased Toxicity" by Hanrieder and co- workers. The paper aims to understand the initiation and progression of Aβ aggregate accumulation and correlations with CNS cell responses by spatial transcriptomics in the proximity of early and late Aβ aggregates. Chemical time stamps were induced using pulse chase of food supplemented with 15N- isotope that label newly produced protein facilitating using a method called Imaging of stable isotope labelling kinetics (iSILK). As a complementary time stamp, the maturity of the Aβ- plaque structures was monitored by LCO fluorescence technology developed by researchers at Linkoping University, Sweden. Spatial transcriptomics together with iSILK and LCO/immunofluorescence structure correlations allowed an unprecedented correlation of proximal cell response to the plaque maturity stage. The study concluded that mature Aβ plaques, in comparison with early plaques, regardless of chronological mouse age, were associated with changes suggesting synaptic toxicity responses. Early plaque formations showed increased immune response gene expression. This methodological approach puts within reach more information on differentiated cellular responses to plaque fibril structure at different plaque development stages and cell types (microglia, astrocytes, and diverse neuronal populations). Overall the impression of the innovative application of different integrated techniques and the presentation of the paper is very positive. There are some of points that should be modified and clarified in the revised version to improve the paper:  
+
+Comment 1: "The choice of the APP NL- F knock- in mouse in this study was clever, because this mouse makes almost exclusively Aβ1- 42, allowing the authors to specifically iSILK- monitor Aβ1- 42 species using MALDI- ToF imaging. This feature should be more clearly stated in the description of the selection of the mouse model. Suggestively, at Results under the header 'iSILK delineates spatial and structural patterns of plaque formation and maturation' - after the sentence in the first paragraph 'This gradual increase in plaque pathology with age more closely resembles the human disease' - add something like: "The choice of the APP NL- F knock- in mouse in this study provided another biochemical advantage. Because this mouse makes almost exclusively deposited Aβ1- 42, it allows us to specifically iSILK- monitor Aβ1- 42 species using MALDI- ToF imaging."  
+
+We are grateful for these positive comments and the constructive feedback. We concur with this comment regarding the choice of the mouse model. As suggested, a corresponding statement, commenting on the biochemical advantage of the mouse model, has been incorporated in the results. In addition, we have performed additional ex situ LC- MS/MS experiments confirming Aβ1- 42 to be the predominantly produced and deposited Aβ species. Please see Supplementary Fig. 1, Methods section "Ex situ Aβ Analysis" and Results page 25 line 24.  
+
+Comment 2: \\*"Under the header 'Amyloid plaque maturation is characterized by continuous fibrillization with age' there is a description of the LCO probes q- FTAA and h- FTAA. The reference here is good but insufficient. LCO reference 21 is ok for general purposes but the following should be added in the description (with appropriate references):
+
+<--- Page Split --->
+
+'This is enabled by the difference in affinity of the two LCO probes, \(q\) - FTAA and h- FTAA, towards amyloid aggregates. Specifically, \(q\) - FTAA preferentially binds to mature and compact \(\beta\) - pleated aggregates, while h- FTAA binds to less compact, yet still \(\beta\) - pleated aggregates (reference: now ref 34, Nyström et al. Evidence for age- dependent in vivo conformational rearrangement within \(\mathrm{A}\beta\) amyloid deposits). Due to their different emission profiles, the LCO probes can be spatially delineated using hyperspectral fluorescent microscopy. Here, the ratio of the LCO maxima (500 nm for \(q\) - FTAA / 580 nm for h- FTAA) is used to express preferential binding of either of the two LCO probes, whereby an increase in 500 nm intensity is indicative of increased \(q\) - FTAA binding and therefore increased structural maturity of the amyloid fibrils (Fig. 2B).'  
+
+And then add references for these statements, e.g., ref 35 Rasmussen et al. (Amyloid polymorphisms...), and ref 45 Parvin et al. (Divergent Age- Dependent Conformational Rearrangement...).\*\*  
+
+We appreciate these suggestions and revised this part accordingly (page 8, lines 18).  
+
+Comment 3: \*\*The spatial transcriptomics data should be clarified. It is not clear how the reference data (baseline normalized expression) were obtained for making the volcano plots of overexpression and decreased expression in Fig. 3C and 3F. In the same context: How does one interpret the GeoMx CSV files when there is no reference data set included? Was this from an already published database? If so, the values from this reference data set should be included, as they can be very useful for others looking at other genes and to derive the data underlying the volcano plot. The contrast (samples vs reference) is not obvious from the CSV files – it appears to be only numbers listed for 10- month and 18- month \(App^{NL - F}\) mice.  
+
+We agree with this important comment and would like to further explain what is shown in the data. The volcano plots 3C and 3F do not represent differential expression analysis. They show genes that were significantly correlating with plaque age, both positively and negatively for the respective mouse age group (10mo: 3C; 18mo: 3D). Consequently, there is no reference data set to refer to.  
+
+We acknowledge that this is not clearly presented in the manuscript. We therefore extended the description of the data in Figure legend 3. Additionally, we adjusted the wording in the main text on page 10 line 14 for clarification: "Volcano plots for both 10- month (Fig. 3C) and 18- month- old (Fig. 3D) mice demonstrated that gene expression levels had significant positive and negative correlations with increasing plaque age."  
+
+Furthermore, the link given (the Shiny app: https://hanriederlab.shinyapps.io/PlaqueAgeTranscriptomics/) can be very useful if it is more clearly described how it was obtained and what cutoff values are used for significance.  
+
+We agree and provide a detailed description of the data accessible in the Shiny App both in the methods and additionally detailed in a Readme file to highlight the details of the data and the analysis they originate from. Additionally, we updated the methods section "MALDI MSI - GeoMx Data Analysis" for increased clarity, including cut- off values. The data accessible via the ShinyApp is briefly explained under the Data
+
+<--- Page Split --->
+
+Availability section in the manuscript. Also, we curated all GeoMx and MALDI data under a new ShinyApp address. Please see: https://maciejulewiczgu.shinyapps.io/MALDI_GEOMX_VOLCANO/  
+
+Is the x- axis "plaque age" (in months?) based on iSILK or q- FTAA/h- FTAA fluorescence, and what is "Log Cnorm"? A clear description of how to use the link should be included in the paper, with more details in the methods section."\*  
+
+We concur that this is critical and update the description of these annotations throughout all presented Figures. In brief, the x- axis values are representing measures of plaque age and are obtained from calculating the width of the MALDI mass peak for Ab 1- 42. Here the width (calculated as full width at half maximum, FWHM) is a measure of peak broadening and consequently an indication of 15N incorporation and plaque age, respectively.  
+
+In Figures 1- 4 as well as in the Shiny app, the x- axis is now referred to as "Nitrogen index LP", which is explained in the main text on page 10 lines 9 and page 30 line2 and visualized in the updated Fig. 1 E.  
+
+We also included a more detailed explanation of the nitrogen index calculation under the header "Calculation of nitrogen index in linear positive mode" in the methods section.  
+
+Comment 4: "The list of acknowledgements for funding is very long. But there is no acknowledgement or information for where the authors obtained the LCOs q- FTAA and h- FTAA, which play an important role in the paper. Since these molecules do not appear to be commercially available, it needs to be specified. If they were synthesized in their own lab this should be stated in the Materials and Methods."  
+
+We received the probes as a kind gift. This information was updated in the methods and acknowledgements.
+
+<--- Page Split --->
+
+## Reviewer #3:  
+
+In this manuscript, Wood & Dulewicz et al. describe the combination of mass spectrometry imaging (MSI) and hyperspectral imaging to measure Ab42 plaque age and morphology in a mouse model of Alzheimer's disease. Using a pulse- chase strategy where mice are fed with a 15N- enriched diet, they were able to distinguish nascent and aged plaques. Hyperspectral imaging with oligothiophene chemistry further demonstrates changes in plaque morphology correlated to age. They further combine this approach with single- plaque GeoMx spatial transcriptomics to identify correlates with age inferred from their MSI approach.  
+
+Overall, the experiments are executed quite well, and the methods and data will be useful for the spatial and Alzheimer's biology community. In the future, I also see great value in combining these techniques with higher- resolution spatial transcriptomics or multicomics to understand the molecular underpinning of these processes. While I find their iSILK MSI technology to be technically impressive, especially when combined with these orthogonal measurements, I believe the richness of their data could have been used to better extract biologically novel conclusions. The manuscript would also benefit from more extended analyses, embedding with more specific knowledge in literature, and better bookkeeping of the collected data.  
+
+If the authors could address or comment on these points, I believe the manuscript will be significantly improved and I would be supportive of publication.  
+
+We are grateful for the positive feedback and appreciate the constructive comments that helped to significantly improve the manuscript. Please see our detailed responses below.  
+
+## Major Comments:  
+
+Comment 1: "The use of MSI pulse- chase is quite inspired. I agree that the 15N/14N ratio is a good proxy for plaque age. Here are some points the authors may want to consider:  
+
+a. It seems like a missed opportunity for developing a numerical index or 'clock' for plaque age. This could be used to assign ages to plaques without MSI, for instance. If hyperspectral imaging or morphology could be used to do a regression analysis, I think that could be immensely useful. I understand that this may be challenging to establish rigorously, but it could be a point of discussion for future work."  
+
+This is a very important point, and we would like to clarify some methodological aspects as we concur and were in fact aiming to use the LCO as surrogate markers of plaque age. In detail, we performed two correlative spatial experiments (Fig. 1D- E):  
+
+1) iSILK (MALDI MSI) in conjunction with spatial transcriptomics (Fig. 1E) to investigate plaque age-specific changes in gene expression.  
+
+2) iSILK (MALDI MSI) in combination with LCO hyperspectral microscopy (Fig. 1D; though the LCO microscopy was not performed together with GeoMx for limited feasibility reasons, we further detail below under Comment 2d.)  
+
+The rationale for combining iSILK (MALDI MSI) with LCO microscopy was to both identify whether aged plaques are characterized with changes in fibrillation and,
+
+<--- Page Split --->
+
+exactly as suggested by your comment, to have a surrogate marker of plaque age that is easier to implement with follow- up experiments and does not require costly and time- consuming SILK labelling experiments. (Indeed, the LCO approach has been used for the IHC experiments described in Fig. 4, experimental overview in Fig. 1F)  
+
+Please refer to an updated Figure 1 for a detailed methodological overview. As suggested by the reviewer, we incorporated this in the discussion on page 19 line 4, highlighting the future potential of the LCO q/h emission ratio.  
+
+b. "It would be interesting for the authors to look at how the MSI and hyperspectral imaging replicate within and across animals. Are these correlations sufficiently strong and hence biologically robust across animals? A breakdown of Fig. 2E by animal, for example, could be helpful."  
+
+We thank the reviewer for raising this important question. In Fig. 2E, the correlation between MSI and LCO hyperspectral imaging was originally presented for each individual mouse \((n = 3\) , with each plot corresponding to one mouse). This showed that the correlation is biologically robust across animals included in the study \((R = 0.64 - 0.89, p< 0.05)\)  
+
+We acknowledge that this was not clearly described and improperly visualized in the original submission.  
+
+To improve clarity, we have now combined all plaques from the three mice into a single plot (Fig. 2E) and revised the figure caption accordingly. Individual mice are distinguished by color, and plaque location (hippocampal vs. cortical) as indicated by different shapes. Both the combined trend (Fig. 2F) and the individual mouse correlations (now reported in the main text on page 9, line 10, Supplementary Fig. 4A- C) are consistent and support the same conclusions:  
+
+a) the hyperspectral ratio, reflects plaque age, b) the plaque core is older than the plaque periphery and c) cortical plaques are, on average, older than hippocampal plaques.  
+
+c. "Could the authors briefly comment on the feasibility and cost of deploying this technology? What is the feasibility as well of MSI and LCO imaging on the same section?"  
+
+We appreciate this comment as there are certain challenges and limitations that come with the various spatial techniques that need to be considered when planning the experiments.  
+
+A significant cost factor for iSILK is the stable isotope diet (ca \(\) 12000/kg)\$. This is particular true for the long labelling experiments needed in the present study as those mice develop amyloid pathology gradually with age.  
+
+Each mouse requires 1g labelled diet per day. We labelled each mouse for 4months (ca 122days ie 122g/mice, \(n = 7\) ) requiring 1kg label for all 7 mice. In addition, there are of course breeding costs and most importantly costs for personell as these experiments pan over 1- 2years.  
+
+In addition, buffers and probe reagents constitute a major cost for the associated GeoMx experiments. GeoMx kits and buffers for two glasses amount to ca. \(\) 3000\(/glass slide (3 - 4 sections/glass). Finally, sequencing costs to quantify the released probes amount to ca.\) \ \(3000\) /index plate. (96 AOI/plate). As those latter costs are highly dependable on inhouse availability and deals, we merely mentioned the
+
+<--- Page Split --->
+
+consumables (diet and glasses/GeoMx run in the Methods section/Experimental Design and Spatial Transcriptomics).  
+
+A further challenge includes further the compatibility of the different techniques for correlative spatial biology/multimics in a single section. The LCO microscopy requires mounting with cover slips which prevents to do LCO imaging prior to MALDI (or GeoMx). In turn, the MALDI sample preparation and acquisition settings for peptide imaging lead to tissue distortions that prevent subsequent fluorescent microscopy. (please see Kaya et al 2017 PMID: 28318232). Consequently, the MALDI/LCO as well as the MALDI/GeoMx analyses had to be carried out on sequential sections.  
+
+We further discuss compatibility challenges in the limitations (page 19 line 8) and provide more detail on the design of our correlative experiments in a revised Fig 1.  
+
+Comment 2: "I have some general reservations about the GeoMx analysis. Given the richness of the dataset, the analysis should give correspondingly rich and novel biological insights. Broadly, are there any unexpected findings here?  
+
+We thank the reviewer for raising this concern regarding the richness of the results.  
+
+While the dataset is indeed rich, capturing a total of 19962 RNA transcripts, the lack of even more distinct and rich changes can be related to the design of the study and the research question we wanted to address i.e. Is the toxicity of plaques an immediate acute phenomenon in the area that they occupy or do plaques continue to cause ongoing increasing toxicity as they mature over time.  
+
+Previous studies on plaque associated changes in gene expression did not consider the time component on a single plaque age level and in addition rather focus on differential changes between plaques and non- plaque regions. Indeed, as Supplementary Figure 10 demonstrates, a direct spatial transcriptomics comparison between plaque- associated and non- plaque- associated areas revealed distinct up- or downregulation of disease- associated astrocytic and synaptic genes. These marked changes likely reflect the stark biological contrast between two fundamentally different environments: one area containing neurotoxic plaques and one area without plaque pathology.  
+
+Our study, in contrast, aimed to explore how gene expression correlates with increasing plaque age, independently of the animals' chronological age (something warranted by our iSILK paradigm). Given the relative similarity between differentially aged though biochemically similar cored plaques our plaque centric GeoMx analyses was naturally expected to reveal rather subtle biological differences as compared to analyses where plaque changes are compared to non- plaque regions.  
+
+It was very interesting and surprising that the major groupings affected in this incremental way, correlating with plaque age, were the synaptic genes.  
+
+To our knowledge, we are the first to demonstrate this continuous decline in synaptic gene expression with increasing plaque age on a (whole) transcriptomic scale. Thus, we can show that the well- known synaptic loss at plaques is not an immediate acute
+
+<--- Page Split --->
+
+effect but rather a slow ongoing infliction of stress and damage as the plaque matures over time. This is now more clearly presented and compared to relevant literature in the discussion on page15 line 18.  
+
+a. Based on the sequencing, the authors suggest dysregulation of synaptic, immune, and metabolic genes proximal to plaques, correlating with plaque age on adjacent sections. Rather than a change in regulatory patterns, is it possible that these changes are explained instead by differences in cell population composition? Much of this is also driven by FOV (AOI) size and how many cells are captured.  
+
+i. Neuronal mRNA is typically enriched in the soma (with some exceptions, often in an activity-dependent manner). Thus, depletion of neurons might be an equally fair explanation of the apparent decrease in synaptic gene expression observed in the GO analysis. This likewise applies for immune/metabolic genes – as recently observed with high-resolution spatial data in 5xFAD mice. The authors could consider using morphological (Nissl, silver, DAPI, etc.) staining or further immunofluorescence to quantify the cell populations here. Further markers for DAA/DAMs might also be helpful.  
+
+We concur and would like to offer our perspective along with additional data analyses addressing this concern, which is that differential cellular composition of the AOI can yield artefacts rather than true biological changes.  
+
+Delineating cell population heterogeneity is in fact a limitation of GeoMx as cell- type- specific analysis at the single- plaque level is currently not practically feasible using this method. The number of transcripts from individual cell types surrounding a single plaque within a finite AOI is insufficient to support high- quality transcriptomic profiling. Alternatively, pooling cell types from multiple plaques would result in a loss of spatiotemporal resolution, including the ability to associate transcriptomic data with individual plaque age. Additionally, GeoMx experiments are limited by the number of morphological markers that can be used simultaneously (typically 3 to 4), which restricts cell- type identification, particularly when pathological structures like Aβ plaques, as in our study, are also being imaged. Consequently, we were limited by the morphology markers we managed to implement (Aβ, cell count (SYTO, similar to DAPI) and GFAP).  
+
+With this we were however able to evaluate any plaque age associated changes in AOI cell count (SYTO), AOI size and GFAP immunoreactivity.  
+
+First, there was no significant correlation (P>0.05) between AOI size or SYTO cell count with plaque age (Supplementary Fig. 8), suggesting that transcriptional differences are unlikely to be a result of systematic differences in the number of cells surrounding aging plaques.  
+
+Second, to estimate the number of astrocytes near plaques within the AOIs, we quantified the GFAP signal area for each AOI. Notably, correlation analysis between
+
+<--- Page Split --->
+
+Gfap mRNA expression and GFAP IHC signal showed strong concordance (R=0.95- 0.97, P<0.0001; Supplementary Fig. 9 A-B), indicating that Gfap transcript levels indeed reflect corresponding protein abundance. 
+
+To investigate whether astrocyte numbers change with increasing plaque age, we correlated the GFAP IHC signal with plaque age. However, no significant correlation was observed (P>0.05, Supplementary Fig. 9 C-D), suggesting that astrocyte numbers do not systematically vary with plaque age. However, since GFAP is generally regarded as a marker of proliferation rather than activation, this lack of correlation does not necessarily imply an absence of astrocytic reactivity to plaques. It is possible that astrogliosis (i.e., astroglial proliferation) is more prominent during initial Aβ plaque formation and may diminish as plaques age. It is also important to stress that we might be underpowered to detect meaningful differences in astrogliosis with increasing plaque age. 
+
+A discussion of the this limitation to not capture potential cell-type heterogeneity in the proximity of aging plaques as well as the Gfap IHC has been added to the discussion (page 17 line 7) and in the limitation (page 19 line 8). 
+
+## ii. In a similar vein, are there any changes in mRNAs known to be translocated to synapses, versus those known to be soma-localized? 
+
+This is a very interesting point as it would be interesting whether the lower number of synaptic genes with plaque age is a consequence of synapse specific impairment or global neuronal loss around plaques. 
+
+We therefore analyzed mRNAs that were significantly positively or negatively correlated with plaque age to assess their subcellular localization (soma vs. neuropil, i.e., synaptic regions), For this we used data from Glock et al. (2021) (PMID: 34670838) that identified 807 neuropil-translocated (28%) and 2945 (72%) soma-located RNAs (3.5-times more soma- than neuropil-localized transcripts). 
+
+Our data show a similar pattern: across both positively and negatively correlated gene sets, we found 4–5 times more soma- located than neuropil-translocated RNAs (please see Table below). 
+
+<table><tr><td>mouse age (directionality of correlation)</td><td>Localization</td><td>Number of Genes (n)</td><td>Percentage of all significantly correlated RNAs in respective group (%)</td></tr><tr><td rowspan="2">10 m (positive)</td><td>neuropil</td><td>7</td><td>20</td></tr><tr><td>soma</td><td>28</td><td>80</td></tr><tr><td rowspan="2">10m (negative)</td><td>neuropil</td><td>35</td><td>16,6</td></tr><tr><td>soma</td><td>176</td><td>83,4</td></tr><tr><td rowspan="2">18 m (positive)</td><td>neuropil</td><td>24</td><td>17,8</td></tr><tr><td>soma</td><td>111</td><td>82,2</td></tr></table>
+
+<--- Page Split --->
+
+
+<table><tr><td rowspan="2">18 m (negative)</td><td>neuropil</td><td>55</td><td>25,3</td></tr><tr><td>soma</td><td>162</td><td>74,7</td></tr></table>
+
+An exception to this trend was observed in genes negatively correlated with plaque age in 18-month-old mice, where 25% were neuropil-translocated (2.9-fold more soma-than neuropil-localized RNAs).
+
+In comparison, only 16% of negatively correlated genes were neuropil-translocated in 10-month-old mice (a 5-fold difference).
+
+This could suggest that with increasing plaque age, synaptic loss becomes more pronounced relative to general neuronal loss.
+
+However, we emphasize that this interpretation is speculative due to several limitations: (1) the classification of RNA localization is based on a single study, as transcriptome-wide RNA translocation is not well-characterized; (2) we cannot rule out (post-mortem) RNA diffusion or other artifacts affecting localization; and (3) the data do not allow for robust statistical comparison.
+
+We included a comment on this in the discussion. (page 16 line 12)
+
+**iii.** Do the authors observe upregulation of **cryptic** **genes** associated with non-native cell types in these plaques? For instance, cell type markers of unexpected immune cells.
+
+We thank the reviewer for this interesting question. According to Bruker Spatial Biology, GeoMx probes are designed not to target cryptic genes, as they are based on the RefSeq transcriptome, which includes only well-characterized transcripts.
+
+**iv.** Can the authors further embed some of these findings with more recent literature on **functional** **changes** **in** **astrocytes** **and** **microglia?"**
+
+We agree and in order to address this, present correlation data for disease associated microglial and astroglial genes in both Fig 3 and Supplementary Fig. 7 in the Results and discuss those data with respect to the current literature (page 17 line 7).
+
+**Comment 2b:** "The statistics on correlations between the GeoMx data and plaque age should consider multiple testing correction if possible (q-value or FDR perhaps). My understanding is that raw p-values are shown. Are these conclusions still robust?"
+
+The reviewer is correct: the volcano plots in Figures 3C and 3F display unadjusted p-values. Correlations were performed across a total of 34 plaques, which limits statistical power. Consequently, the resulting p-values do not survive transcriptome-wide FDR correction. A key limitation of applying correction, such as the commonly used Benjamini-Hochberg method, for multiple comparisons in genome-wide **correlation** **analyses** rather than the common simple pairwise comparisons (across 19,963 genes) is that the threshold for statistical significance becomes extremely stringent. Specifically, only near-perfect correlations (r≈0.9) will survive correction.
+
+<--- Page Split --->
+
+We fully acknowledge the reviewer's concern regarding robustness. However, we believe our broader findings (namely, the decrease in synaptic gene expression and the increase in inflammatory and metabolic gene expression with increasing plaque age) remain robust.  
+
+This conclusion is supported by the consistent correlation patterns observed not in isolated genes, but across entire groups of synaptic, inflammatory, and metabolism- related genes, suggesting biologically meaningful and coherent changes.  
+
+Importantly, all GO analyses presented in the manuscript were performed using FDR- corrected p- values.  
+
+We have now explicitly noted this limitation (limited statistical power) in the discussion section (page 19, line 8) and specified in the methods section (page 33 line 6) and Figure 3 caption that uncorrected p- values are displayed for the correlation analyses.  
+
+Comment 2c: "While the viewer is helpful, for some example genes, could the authors directly show the scatterplots demonstrating the correlations between gene expression and plaque age in the main text? Regressions with the corresponding coefficients might also be helpful."  
+
+We concur and include representative scatterplots for different synaptic (Nptx2, Dlg4) and DAA/DAM genes (Anxa1, Axl, Csf1, Ctsd) in Figure 3 F,G,I,J,L,M,O,P and Supplementary Fig. 7.  
+
+Comment 2d: "Could the authors provide some commentary on the challenges of hyperspectral imaging prior to GeoMx on the same tissue slice? As they mention, serial sections preclude the analysis of nascent small plaques, which would certainly be biologically fascinating. Do the authors see this as a major technical limitation? There are groups combining multimodal imaging and GeoMx on the same sections as well, so I am interested in hearing some perspective on this – though an experimental demonstration would be most impressive."  
+
+As rightfully acknowledged by the reviewer, the LCO microscopy provides the possibility to serve as surrogate marker for plaque maturation/age and provides additional biophysical insight on plaque constitution. Consequently, it would be desirable to perform LCO and GeoMx within the same tissue rather than or in addition to, using a correlative approach using sequential sections. The same applies for combining MALDI and GeoMx.  
+
+However, LCO- based hyperspectral microscopy requires the use of mounting media and cover slips to afford an accurate readout of the spectral data and linear unmixing, respectively. Therefore, these experiments cannot be performed prior to the GeoMx (or MALDI) experiments on the same tissue section. (please see our response to your comment 1c)  
+
+We naturally investigated whether LCO staining could be performed after MALDI or GeoMx as well as whether MALDI and GeoMx could be performed on the same tissue. For MALDI, the combination with LCO microscopy and/or GeoMx is not possible as the sample prep and MSI peptide experiments leads to significant tissue distortion. (Of note, this is specific for amyloid peptide MSI. MALDI or DESI MSI of lipids could theoretically be interfaced with direct LCO imaging or GeoMx, as those MSI methods show no tissue distortion, please see Kaya et al 2017 doi:10.1021/acs.analchem.7b00313/PMID: 28318232)
+
+<--- Page Split --->
+
+Regarding LCO analysis after GeoMx: this approach while feasible has as obvious limitations with respect to AOI selection as the LCO data should preferably guide AOI selection prior to the GeoMx experiment and not retrospectively. Further, as mentioned under Comment 1a, the GeoMx approach is limited with respect to emission wavelengths used for fluorescent imaging as the UV range (<420nm) cannot be used due to photocleavable tags used that absorb in that range. In addition, the LCO dyes emit rather broadly between 450- 620nm, which leads to interference with morphology markers used in the GeoMx experiment that emit in this range. These challenges are now discussed in the limitation section at the end of the discussion. (page 19 line 8)  
+
+Comment 3: "There are some ways the data could be better organized and bookkept, for transparency:  
+
+a. Can the authors enumerate the number of plaques and their age distribution captured by MSI within and across animals?  
+
+The number of animals was \(n = 3\) (10mo) and \(n = 4\) (18mo). The number of plaques per animal was \(N = 5 - 6\) . We provide this information in the Methods (Animal experimental design) as well as summarize all plaque ROI (MALDI/LCO) and AOI (MALDI/GeoMx) data in a new Supplementary Table 1-5.  
+
+b. Figure S1 is insufficiently described to let me understand the structure of the data. A more thorough caption and explanation would be helpful, including a breakdown by source animal rather than just age.  
+
+We concur and updated the Supplementary Tables (now Suppl. Tab 1- 5) accordingly as detailed in Comment 3a.  
+
+c. Could the authors clarify what is shown in Fig. S2b? I am not sure how to interpret these values. It might be more useful to show PCAs instead.  
+
+We appreciate this comment. The Figure aims to illustrate the advantage of spatial transcriptomics on the single plaque level vs bulk tissue RNA seq.  
+
+Here, the plots aim to illustrate a measure of quantification for synaptic and DAA genes that are significantly changed in response to plaque pathology. The y- axis thereby represents the first principal component of the synaptic and DAA genes, respectively, and serves as a representative abundance value. The results show that spatial analysis allows to delineate plaque specific decrease in synaptic genes and increase in DAA genes, something which is otherwise convoluted in bulk tissue analysis.  
+
+We updated the legend (now Supplementary Figure 10) accordingly for improved clarity.  
+
+d. Could we have a complete description of all GeoMx FOVs imaged, and their corresponding 15N/14N ratios, morphologies and hyperspectral ratios, anatomical localization, number of unique genes and transcripts sequenced, etc? It would also be important to know the source animals for each FOV.  
+
+This could be in the form of a table or heatmap. It will give a better idea of how powered the conclusions are."
+
+<--- Page Split --->
+
+We agree with this suggestion and curate these data in comprehensive Supplementary Fig. 5- 6 and Supplementary Tables 3- 5.  
+
+## Minor Comments:  
+
+Minor Comments:Minor 1: "It might be helpful for the authors to comment further on the applicability of their method in conjunction with higher- resolution spatial transcriptomics methods (e.g., Visium HD, MERFISH, Xenium, etc.) or proteomics (CODEX, IMC, etc.) in future work. Integrating iSILK with the growing landscape of these tools is certainly of interest to many spatial biologists."  
+
+We agree this important point. Unfortunately, the MALDI sample preparation and acquisition impacts tissue morphology and prevents a direct interfacing with other spatial techniques on the same tissue. Similar to the approach used here iSILK can be used on sequential tissues in concert with those techniques.  
+
+this is certainly of interest as those other spatial transcriptomics/proteomics methods would give a better resolution of transcripts and proteins towards plaques. As discussed we were faced with the tradeoff between sensitivity and spatial resolution/single cell specificity but agree that follow up studies expanding towards these techniques would be very valuable.  
+
+We discuss the compatibility of iSILK with spatial biology techniques in the limitations section.  
+
+Minor 2: "Some general commentary on the limitations of their approach might be helpful."  
+
+We included Limitations comment at the end of the Discussion (page 19).  
+
+Minor 3: "Fig. 1a scale bar value is not provided. The imaging modality should also be stated."  
+
+We provide this value in the legend.  
+
+Minor 4: "As a matter of preference, it might be better for authors to use perceptually uniform colormaps instead of 'spectral' (e.g., in Fig. 1f). It may be hard for some readers to see."  
+
+We attempted to use a uniform color map instead of spectral coloring, as suggested by the reviewer.  
+
+A side- by- side comparison of both color maps (see below) demonstrates that subtle changes in \(^{14}\mathrm{N} / ^{15}\mathrm{N}\) enrichment are more readily discernible to the human eye employing a spectral color map.
+
+<--- Page Split --->
+![PLACEHOLDER_21_0]
+  
+
+For this reason, we have opted to retain our original color scheme, as it better highlights the features of interest in the data.  
+
+Minor 5: "Legends in some panels (e.g., Fig. 1e, 2c) are rather small and hard to see."  
+
+We increased the font size for the respective panels.  
+
+Minor 6: "Gene and GO annotations for Fig. 3c- h are also hard to read."  
+
+We increased the font size for the respective panels and rearranged the text to better accommodate the figure dimensions.  
+
+Minor 7: "Could the MSI for Fig. 3b be decomposed into 15N and 14N for clarity?"  
+
+We provide decomposed images in Fig 3b (as overlay) and in the SI (Supplementary Fig. 3D)  
+
+Minor 8: "In 'spanning across two consecutive 12mm sections' did the authors mean 12 micron?"  
+
+The sections are 12 μm in thickness as rightfully pointed out by the reviewer. This has now been amended accordingly.
+
+<--- Page Split --->
+
+Please find attached with this final resubmission our revised manuscript "Isotope Encoded Spatial Biology Identifies Amyloid Plaque-Age-Dependent Structural Maturation and Synaptic Loss" by Woods, Dulewicz et al.  
+
+In this revision, we have addressed the remaining comments, including performing minor additional analyses (WCGNA) and making the requested textual and figure modifications. The corresponding changes are detailed in the point- by- point responses below.  
+
+With best regards, Jörg Hanrieder  
+
+## Reviewer #1:  
+
+I have now reviewed the revised manuscript, "Isotope Encoded Spatial Biology Identifies Amyloid Plaque- Age- Dependent Structural Maturation, Synaptic Loss, and Increased Toxicity," along with the authors' point- by- point rebuttal to the concerns I raised in my initial review.  
+
+The authors have addressed all of my previous concerns by conducting significant new experiments to validate their findings and strengthen their claims. The new high- sensitivity LC- MS/MS analysis to investigate post- translational modifications and the inclusion of MS/MS spectra for sequence confirmation have fully resolved my initial questions regarding the data.  
+
+As a final point of scientific interest, I would like to offer one minor point for the author's consideration. In the MS/MS spectrum for Aβ 1- 42 shown in Supplementary Figure 1F, it is interesting to note the dominance of the b- ion series, while the corresponding y- ion series is almost absent. A brief comment on this observation in the Results section or the caption of the figure could enhance the spectral interpretation of the fragmentation behavior.  
+
+## Response:  
+
+We thank the reviewer for raising this point and added the following statement to the figure description of Supplementary Figure 1: "The b- ion series (charge retained on N- terminal fragment upon peptide fragmentation) dominates in several Aβ isoforms because the N- terminal region contains multiple charge carriers (e.g., Arg5, Lys16). In contrast, the C- terminal region of Aβ peptides largely lacks such charge carriers, hence, y- ions are less abundant, and fragment spectra are dominated by the b- ions."  
+
+## Reviewer #2:  
+
+The authors have made appropriate clarifications of all my previously raised points in
+
+<--- Page Split --->
+
+the first round of detailed review and the paper has been improved. Congratulations to very interesting and important work. I support publication as soon as possible.  
+
+## Response:  
+
+We sincerely thank the reviewer for their thorough evaluation of our manuscript and their positive feedback. We are pleased that the revisions addressed the previously raised points.
+
+<--- Page Split --->
+
+## Reviewer #3:  
+
+The presentation of the data in the revised manuscript makes it much more straightforward to read and interpret the data in my opinion. Here are some remaining minor comments for the authors, though they are mostly either presentation issues or nominal analyses that might provide some further insights if the authors choose to perform them. Nevertheless, I believe they have satisfied my major concerns, and I am supportive of publication.  
+
+## Minor Comments  
+
+1. In their discussion of limitations of other high-res ST approaches, they might note that the CosMx 18k now offers near-transcriptome coverage – though at markedly lower gene-wise sensitivity  
+
+## Response:  
+
+We thank the reviewer for this helpful suggestion. We have revised the manuscript to acknowledge that CosMx now provides near whole-transcriptome coverage and have added a note on the associated limitation of reduced gene-wise sensitivity. Please see p20line10- 14  
+
+2. I would appreciate if the authors could do one additional analysis – to highlight novel gene-gene correlation modules that arise across from analysis across plaques (e.g. with WGCNA or similar).  
+
+## Response:  
+
+We thank the reviewer for this suggestion and performed WGCNA to explore gene- gene correlation networks separately in 10- and 18- month- old mice (see methods section 'Weighted Gene Co- expression Network Analysis (WGCNA)'). For ease of computation and interpretation, we limited the genes included in network construction to those that significantly correlated with our trait of interest, i.e. plaque age. While the specific filtering criterion was tailored to our study, the use of pre- filtering approaches prior to WGCNA is widely used and described in the literature (Deng et al. 2015, Lin et al. 2018, Zuo et al. 2018). We have deposited our results (gene cluster membership and GO analysis of each cluster) to the Shiny app, where users can explore genes of interest in a detailed manner. Please see: https://maciejdulewiczgu.shinyapps.io/MALDI_GEOMX_VOLCANO/  
+
+We also included the WGCNA results in a new Supplementary Figure (SI Fig 8), which illustrates the correlation between the identified clusters and plaque age, highlighting selected clusters of interest along with their associated GO terms. The corresponding results are described in the manuscript. In brief, we observed that
+
+<--- Page Split --->
+
+gene- gene correlation clusters relating to (glutamatergic) synaptic processes correlated negatively with plaque age, in line with our previously reported results. Additionally, we found clusters enriched in glial proteins (e.g. H2- k1) to be positively correlated with plaque age.  
+
+Please see Results: p11line18 and Methods p35  
+
+3. The authors might comment on whether their observed gene expression changes signify some T cell infiltration (or any other evidence for this in their animals) versus microglial activation due to overlap in GO annotations. This was what I meant by cryptic expression in my previous review and I apologize for the misunderstanding.  
+
+## Response:  
+
+We thank the reviewer for this clarification and have expanded the Discussion to address the potential contribution of T cell infiltration. We now note that although T cell- associated ontologies were significantly enriched, the driving genes are also broadly expressed in microglia, and canonical T cell markers (CD3, CD4, CD8) were not significantly changed. Therefore, in this dataset, potential T cell infiltration cannot be reliably distinguished from microglial activation. We added a comment in the discussion. p18line11  
+
+4. The statement that APP NL-F is strictly better than NL-G-F as it reflects human pathology (Line 387-389) is too strongly made.  
+
+## Response:  
+
+We agree with the reviewer that the original statement was too strong and have revised the text to avoid implying that the APP NL-F model is strictly better or overstating its relevance to human pathology over the NL-G-F model.  
+
+5. Scale bars missing in Fig. 3A, 4A.  
+
+## Response:  
+
+We thank the reviewer for noting this omission. Scale bars have now been added to both Fig. 3A and Fig. 4A.  
+
+6. Labels for GFAP, Ab, and SYTO colors are missing in Fig. 3B, Supp Fig. 5, 6.  
+
+## Response:  
+
+We thank the reviewer for pointing this out. Labels for GFAP, Ab, and SYTO channel colors have now been added to Fig. 3B, Supplementary Fig. 5, and Supplementary Fig. 6.  
+
+7. I am not sure why the data must be duplicated as Supp Fig. 3D – would be preferable to instead include these panels in Fig. 3B?
+
+<--- Page Split --->
+
+## Response:  
+
+Response:We agree with the reviewer that duplication was unnecessary. We have re- arranged Fig. 3B to incorporate the relevant panels and have removed Supplementary Fig. 3D to avoid redundancy.  
+
+8. The bottom axis labels for Fig. 3C, D are clipped.  
+
+## Response:  
+
+Response:We thank the reviewer for noting this. The bottom axis labels for Fig. 3C and 3D have been corrected to ensure they are fully visible in the revised version of the manuscript.  
+
+Typographical  
+
+Typographical1. Line 190 "oligothiophene (LCO)" > "oligothiophenes (LCOs)2. Line 429 "information" > "information"  
+
+## Response:  
+
+Response:We thank the reviewer for spotting these errors. Both typographical issues have been corrected in the revised manuscript.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Transparent Peer Review file__ea70799eb1d530d0b6bfa2b1a6bd3034e6bc63286fb046e26e168ba379e2933c/supplementary_0_Transparent Peer Review file__ea70799eb1d530d0b6bfa2b1a6bd3034e6bc63286fb046e26e168ba379e2933c_det.mmd

@@ -0,0 +1,937 @@
+<|ref|>title<|/ref|><|det|>[[72, 53, 295, 80]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[74, 96, 297, 119]]<|/det|>
+Peer Review File  
+
+<|ref|>title<|/ref|><|det|>[[73, 153, 888, 222]]<|/det|>
+# Isotope Encoded Spatial Biology Identifies Plaque-Age-Dependent Maturation and Synaptic Loss in an Alzheimer's Disease Mouse Model  
+
+<|ref|>text<|/ref|><|det|>[[73, 235, 430, 252]]<|/det|>
+Corresponding Author: Dr Jörg Hanrieder  
+
+<|ref|>text<|/ref|><|det|>[[70, 274, 864, 289]]<|/det|>
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+<|ref|>text<|/ref|><|det|>[[73, 326, 144, 340]]<|/det|>
+Version 0:  
+
+<|ref|>text<|/ref|><|det|>[[73, 353, 219, 367]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[73, 379, 160, 393]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[73, 405, 238, 418]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 418, 922, 499]]<|/det|>
+Understanding the evolution of Aβ in Alzheimer's etiology is crucial for elucidating the mechanisms underlying disease progression, and this study partially addresses this aspect. The authors applied the iSILK technique to track amyloid plaque formation, maturation, neurotoxicity, and synaptic loss in a spatiotemporal way. The authors employ the AppNL- F/NL- F aged mouse model to examine how plaque formation progresses over time. The study presents an innovative approach by integrating mass spectrometry imaging (MALDI MSI), spatial transcriptomics, and hyperspectral microscopy to assess plaque heterogeneity and its associated effects.  
+
+<|ref|>text<|/ref|><|det|>[[70, 497, 874, 524]]<|/det|>
+However, critical technical and conceptual issues need to be addressed before this manuscript can be considered for publication.  
+
+<|ref|>text<|/ref|><|det|>[[72, 535, 916, 590]]<|/det|>
+1. One major limitation of this study is the lack of consideration for known post-translational modifications (PTMs) of Aβ, which are known to influence its aggregation and pathological progression. The centroid mass values of MALDI MSI (Table S1) do not reflect known Aβ PTMs, which should result in characteristic mass shifts. The authors should respond the following questions regarding the peak assignment of their MALDI MS spectra.  
+
+<|ref|>text<|/ref|><|det|>[[73, 589, 760, 616]]<|/det|>
+1) Why were these known PTMs not detected in the dataset? 
+2) Does the dataset contain any mass peaks that could suggest age-dependent PTM changes in Aβ?  
+
+<|ref|>text<|/ref|><|det|>[[70, 627, 904, 655]]<|/det|>
+2. The authors should explain why no background peptide signals are observed in the spectrum. Additionally, they should present the MS/MS spectra to confirm the presence of amyloid beta and assess whether other peptides were detected.  
+
+<|ref|>text<|/ref|><|det|>[[72, 665, 914, 720]]<|/det|>
+3. The authors claim that older amyloid plaques exhibit increased neurotoxicity and synaptic loss, as demonstrated through iSILK labeling, transcriptomics, and fluorescence imaging. While the data suggest a correlation between plaque age and synaptic dysfunction, it is not yet clear how this study advances our understanding beyond prior work that has already reported synapse loss in the vicinity of amyloid plaques.  
+
+<|ref|>text<|/ref|><|det|>[[70, 719, 900, 745]]<|/det|>
+To fully support their conclusions, the authors should clarify what new biological insights this study provides compared to previous findings on plaque-induced toxicity.  
+
+<|ref|>sub_title<|/ref|><|det|>[[73, 769, 161, 782]]<|/det|>
+## Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[73, 795, 240, 808]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[73, 809, 240, 821]]<|/det|>
+Dear Editor and authors,  
+
+<|ref|>text<|/ref|><|det|>[[72, 821, 923, 914]]<|/det|>
+I have reviewed the paper "Isotope Encoded Spatial Biology Identifies Amyloid Plaque- Age- Dependent Structural Maturation, Synaptic Loss, and Increased Toxicity" by Hanrieder and co- workers. The paper aims to understand the initiation and progression of Aβ aggregate accumulation and correlations with CNS cell responses by spatial transcriptomics in the proximity of early and late Aβ aggregates. Chemical time stamps were induced using pulse chase of food supplemented with 15N- isotope that label newly produced protein facilitating using a method called Imaging of stable isotope labelling kinetics (iSILK). As a complementary time stamp, the maturity of the Aβ- plaque structures was monitored by LCO fluorescence technology developed by researchers at Linkoping University, Sweden.  
+
+<|ref|>text<|/ref|><|det|>[[70, 925, 911, 940]]<|/det|>
+Spatial transcriptomics together with iSILK and LCO/immunofluorescence structure correlations allowed an unprecedented
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[73, 46, 915, 113]]<|/det|>
+correlation of proximal cell response to the plaque maturity stage. The study concluded that mature Aβ plaques, in comparison with early plaques, regardless of chronological mouse age, were associated with changes suggesting synaptic toxicity responses. Early plaque formations showed increased immune response gene expression. This methodological approach puts within reach more information on differentiated cellular responses to plaque fibril structure at different plaque development stages and cell types (microglia, astrocytes, and diverse neuronal populations).  
+
+<|ref|>text<|/ref|><|det|>[[72, 124, 900, 152]]<|/det|>
+Overall the impression of the innovative application of different integrated techniques and the presentation of the paper is very positive.  
+
+<|ref|>text<|/ref|><|det|>[[73, 163, 796, 179]]<|/det|>
+There are some of points that should be modified and clarified in the revised version to improve the paper:  
+
+<|ref|>text<|/ref|><|det|>[[72, 190, 915, 295]]<|/det|>
+1. The choice of the APPNL-F knock-in mouse in this study was clever, because this mouse makes almost exclusively Aβ1-42, allowing the authors to specifically iSILK-monitor Aβ1-42 species using MALDI-ToF imaging. This feature should be more clearly stated in the description of the selection of the mouse model. Suggestively at Results under the header "iSILK delineates spatial and structural patterns of plaque formation and maturation" ... after the sentence in the first paragraph "This gradual increase in plaque pathology with age more closely resembles the human disease". Add something like this: "The choice of the APPNL-F knock-in mouse in this study provided another biochemical advantage. Because this mouse makes almost exclusively deposited Aβ1-42, it allows us to specifically iSILK-monitor Aβ1-42 species using MALDI-ToF imaging".  
+
+<|ref|>text<|/ref|><|det|>[[72, 306, 916, 372]]<|/det|>
+2. Under the header "Amyloid plaque maturation is characterized by continuous fibrilization with age" There is a description of the LCO and especially qFTAA and hFTAA. The reference here is good but insufficient. LCO reference 21 is ok for general purposes but then the following (see below) should be added in the description. While the references to the original publications on this method are present in the paper (later in the discussion), from reading this result presentation it appears that the description comes at face-value from the present study.  
+
+<|ref|>text<|/ref|><|det|>[[72, 371, 918, 516]]<|/det|>
+So suggested placement of references: "This is enabled by the difference in affinity of the two LCO probes, q-FTAA and h- FTAA, towards amyloid aggregates. Specifically, q-FTAA preferentially binds to mature and compact beta-pleated aggregates, while h-FTAA binds to less compact, yet still beta- pleated aggregates (reference: Now ref 34 in the list of references Nystrom, S. et al. Evidence for age-dependent in vivo conformational rearrangement within Abeta amyloid deposits). Due to their different emission profiles, the LCO probes can be spatially delineated using hyperspectral fluorescent microscopy. Here, the ratio of the LCO maxima (500 nm for q- FTAA / 580 nm for h-FTAA) is used to express preferential binding of either of the two LCO probes used, whereby an increase in 500 nm intensity is indicative of increased q-FTAA binding and therefore, increased structural maturity of the amyloid fibrils (Fig. 2B). (reference: Now ref 35 Rasmussen, J. et al. Amyloid polymorphisms constitute distinct clouds of conformational variants in different etiological subtypes of Alzheimer's disease. + Now ref 45 Parvin, F. et al. Divergent Age-Dependent Conformational Rearrangement within Abeta Amyloid Deposits in APP23, APPPS1, and App(NL-F) Mice.).  
+
+<|ref|>text<|/ref|><|det|>[[72, 527, 919, 620]]<|/det|>
+3. The spacial transcriptomics data should be clarified. It is not clear how the reference data (baseline normalized expression) were obtained for making the volcano-plots of overexpression and decreased expression in Fig. 3C and 3F? In the same context: How to interpret the GeoMx CSV files when there is no reference data set included. Was this from already published databases? If so, the used numbers from this reference data set should be included. They can be very useful for others looking for other genes and to derive the data associated with the volcano plot (discussed above). The contrast (samples vs reference) is not obvious from the CSV-files. It appears to be only numbers listed for 10 and 18 month APPNL-F mice.  
+
+<|ref|>text<|/ref|><|det|>[[73, 620, 260, 633]]<|/det|>
+Furthermore, the link given:  
+
+<|ref|>text<|/ref|><|det|>[[73, 633, 491, 647]]<|/det|>
+https://hanriederlab.shinyapps.io/PlaqueAgeTranscriptomics/  
+
+<|ref|>text<|/ref|><|det|>[[73, 647, 915, 687]]<|/det|>
+can be very useful if it more clearly described how it was obtained and what cutoff values are used for significance. Is the x- axis plaque age (in months?) based on iSILK or q-FTAA/h-FTAA fluorescence and what is Log Cnorm? A clear description of how to use the link should be included in the paper with more descriptions in the methods section.  
+
+<|ref|>text<|/ref|><|det|>[[72, 698, 911, 751]]<|/det|>
+4. The list of acknowledgements for funding is very long. But there is no acknowledgement or information for where the authors obtained the LCOs q-FTAA and h-FTAA which plays an important role in the paper. Since these molecules appear not to be commercially available it needs to be specified. If they were synthesized in their own lab this should be stated in the materials and methods.  
+
+<|ref|>sub_title<|/ref|><|det|>[[73, 815, 161, 828]]<|/det|>
+## Reviewer #3  
+
+<|ref|>text<|/ref|><|det|>[[73, 841, 238, 853]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 853, 920, 932]]<|/det|>
+In this manuscript, Wood & Dulewicz et al. describe the combination of mass spectrometry imaging (MSI) and hyperspectral imaging to measure Ab42 plaque age and morphology in a mouse model of Alzheimer's disease. Using a pulse- chase strategy where mice are fed with a 15N- enriched diet, they were able to distinguish nascent and aged plaques. Hyperspectral imaging with oligothiophene chemistry further demonstrates changes in plaque morphology correlated to age. They further combine this approach with single- plaque GeoMx spatial transcriptomics to identify correlates with age inferred from their MSI approach.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 46, 912, 153]]<|/det|>
+Overall, the experiments are executed quite well, and the methods and data will be useful for the spatial and Alzheimer's biology community. In the future, I also see great value in combining these techniques with higher- resolution spatial transcriptomics or multiomics to understand the molecular underpinning of these processes. While I find their iSILK MSI technology to be technically impressive, especially when combined with these orthogonal measurements, I believe the richness of their data could have been used to better extract biologically novel conclusions. The manuscript would also benefit from more extended analyses, embedding with more specific knowledge in literature, and better bookkeeping of the collected data. If the authors could address or comment on these points, I believe the manuscript will be significantly improved and I would be supportive of publication.  
+
+<|ref|>sub_title<|/ref|><|det|>[[73, 177, 188, 191]]<|/det|>
+## Major comments  
+
+<|ref|>text<|/ref|><|det|>[[72, 202, 890, 230]]<|/det|>
+1. The use of MSI pulse chase is quite inspired. I agree that 15/14N ratio is a good proxy for plaque age. Here are some points the authors may want to consider:  
+
+<|ref|>text<|/ref|><|det|>[[72, 241, 911, 295]]<|/det|>
+a. It seems like a missed opportunity for developing a numerical index or 'clock' here for plaque age. This could be used to assign ages to plaques without MSI, for instance. If hyperspectral imaging or morphology could be used to do a regression analysis, I think that could be immensely useful. I understand that this may be challenging to establish rigorously, but could be a point of discussion for future work.  
+
+<|ref|>text<|/ref|><|det|>[[72, 306, 918, 347]]<|/det|>
+b. It would be interesting for the authors to look at how the MSI and hyperspectral imaging replicates within and across animals. Are these correlations sufficiently strong and hence, biologically robust across animals? A breakdown of Fig. 2e by animal for example could be helpful.  
+
+<|ref|>text<|/ref|><|det|>[[70, 358, 905, 386]]<|/det|>
+c. Could the authors briefly comment on the feasibility and cost of deploying this technology? What is the feasibility as well of MSI and LCO imaging on the same section?  
+
+<|ref|>text<|/ref|><|det|>[[70, 410, 905, 438]]<|/det|>
+2. I have some general reservations about the GeoMx analysis. Given the richness of the dataset, the analysis should give correspondingly rich and novel biological insights. Broadly, are there any unexpected findings here?  
+
+<|ref|>text<|/ref|><|det|>[[72, 449, 911, 502]]<|/det|>
+a. Based on the sequencing, the authors suggest dysregulation of synaptic, immune and metabolic genes proximal to plaques, in a manner correlating with age on adjacent sections. Rather than a change in regulatory patterns, is it possible that these changes are explained instead by differences in cell population composition? Much of this is also driven by FOV (or rather, AOI) size and how many cells are expected to be captured.  
+
+<|ref|>text<|/ref|><|det|>[[72, 513, 911, 580]]<|/det|>
+i. Neuronal mRNA is typically enriched in the soma, with a number of exceptions, and often, in an activity-dependent manner. Thus, depletion of neurons might be an equally fair explanation of apparent decrease in synaptic gene expression observed in the GO analysis. This is likewise for immune/metabolic genes – as recently observed with high-resolution spatial data in 5XFAD mice. The authors could consider using morphological (Nissl, silver, DAPI etc.) staining or further immunofluorescence to quantify the cell populations here. Further markers for DAA/DAMs might also be helpful.  
+
+<|ref|>text<|/ref|><|det|>[[70, 592, 920, 619]]<|/det|>
+ii. In a similar vein, are there any changes in mRNAs known to be translocated to synapses, versus those known to be somalocalized?  
+
+<|ref|>text<|/ref|><|det|>[[70, 631, 861, 659]]<|/det|>
+iii. Do the authors observe upregulation of cryptic genes associated with nonnative cell types in these plaques? For instance, cell type markers of unexpected immune cells.  
+
+<|ref|>text<|/ref|><|det|>[[70, 670, 916, 699]]<|/det|>
+iv. Can the authors further embed some of these findings with more recent literature on functional changes in astrocytes and microglia?  
+
+<|ref|>text<|/ref|><|det|>[[70, 710, 905, 737]]<|/det|>
+b. The statistics on correlations between the GeoMx and plaque age should consider multiple testing correction if possible (q-val or FDR perhaps). My understanding is that raw p-values are shown. Are these conclusions still robust?  
+
+<|ref|>text<|/ref|><|det|>[[72, 749, 904, 789]]<|/det|>
+c. While the viewer is helpful, for some example genes, could the authors directly show the scatterplots demonstrating the correlations between gene expression and plaque age in the main text? Regressions with the corresponding coefficients might also be helpful.  
+
+<|ref|>text<|/ref|><|det|>[[72, 801, 920, 867]]<|/det|>
+d. Could the authors also provide some commentary on the challenges of hyperspectral imaging prior to GeoMx on the same tissue slice? As they mention, serial sections preclude the analysis of nascent small plaques, which would certainly be biologically fascinating. Do the authors see this as a major technical limitation? To my knowledge, there are several groups combining multimodal imaging and GeoMx imaging and sequencing on the same sections as well, so I am interested in just hearing some perspective on this – though a experimental demonstration would be most impressive.  
+
+<|ref|>text<|/ref|><|det|>[[70, 892, 830, 906]]<|/det|>
+3. There are some ways the data could also be better organized and bookkept, for the purpose of transparency:  
+
+<|ref|>text<|/ref|><|det|>[[70, 917, 914, 932]]<|/det|>
+a. Can the authors enumerate the number of plaques and their age distribution captured by MSI within and across animals?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[70, 46, 852, 75]]<|/det|>
+b. Figure S1 is insufficiently described to let me understand the structure of the data. A more thorough caption and explanation would be helpful, including a breakdown by source animal rather than just age.  
+
+<|ref|>text<|/ref|><|det|>[[70, 86, 916, 114]]<|/det|>
+c. Could the authors clarify what is shown in Fig. S2b? I am not sure how to interpret these values. It might be more useful to show PCAs instead.  
+
+<|ref|>text<|/ref|><|det|>[[72, 126, 914, 193]]<|/det|>
+a. Could we have a complete description of all GeoMx FOVs imaged, and their corresponding 15/14N ratios, morphologies and hyperspectral ratios, anatomical localization, number of unique genes and transcripts sequenced, etc? It would also be important to know the source animals for each FOV. This is an essential quality control metric for readers to make critical judgments about the data. This could for example be in the form of a table (for small numbers of FOVs), or a heatmap. It will give me a better idea of how powered the conclusions are.  
+
+<|ref|>sub_title<|/ref|><|det|>[[72, 217, 188, 230]]<|/det|>
+## Minor comments  
+
+<|ref|>text<|/ref|><|det|>[[72, 243, 901, 296]]<|/det|>
+1. It might be helpful for the authors to further comment on the applicability of their method in conjunction with higher-resolution spatial transcriptomics methods (e.g. Visium HD, MERFISH, Xenium etc.) or proteomics (CODEX, DVP etc.) in future work. Integrating iSILK with the growing landscape of these tools is certainly a point of interest for many spatial biologists.  
+
+<|ref|>text<|/ref|><|det|>[[72, 307, 627, 321]]<|/det|>
+2. Some general commentary of the limitations of their approach might be helpful.  
+
+<|ref|>text<|/ref|><|det|>[[72, 333, 661, 348]]<|/det|>
+3. Fig. 1a scale bar value is not provided. The imaging modality should also be stated.  
+
+<|ref|>text<|/ref|><|det|>[[70, 359, 904, 387]]<|/det|>
+4. As a matter of preference, it might be better for authors to use perceptually uniform colormaps instead of spectral (e.g. in Fig. 1f). It may be hard for some readers to see.  
+
+<|ref|>text<|/ref|><|det|>[[72, 398, 631, 413]]<|/det|>
+5. Legends in some panels (e.g. Fig. 1e, 2c) also are rather small and hard to see.  
+
+<|ref|>text<|/ref|><|det|>[[72, 424, 504, 439]]<|/det|>
+6. Gene and GO annotations for Fig. 3c-h are also hard to read.  
+
+<|ref|>text<|/ref|><|det|>[[72, 450, 534, 465]]<|/det|>
+7. Could the MSI for Fig. 3b be decomposed into 15/14N for clarity?  
+
+<|ref|>text<|/ref|><|det|>[[72, 476, 689, 491]]<|/det|>
+8. In "spanning across two consecutive 12mm sections" did the authors means 12 micron?  
+
+<|ref|>text<|/ref|><|det|>[[72, 528, 145, 542]]<|/det|>
+Version 1:  
+
+<|ref|>text<|/ref|><|det|>[[72, 555, 219, 568]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[72, 580, 160, 594]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[72, 606, 238, 619]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 619, 923, 660]]<|/det|>
+I have now reviewed the revised manuscript, "Isotope Encoded Spatial Biology Identifies Amyloid Plaque-Age-Dependent Structural Maturation, Synaptic Loss, and Increased Toxicity," along with the authors' point-by-point rebuttal to the concerns I raised in my initial review.  
+
+<|ref|>text<|/ref|><|det|>[[72, 671, 923, 712]]<|/det|>
+The authors have addressed all of my previous concerns by conducting significant new experiments to validate their findings and strengthen their claims. The new high-sensitivity LC-MS/MS analysis to investigate post-translational modifications and the inclusion of MS/MS spectra for sequence confirmation have fully resolved my initial questions regarding the data.  
+
+<|ref|>text<|/ref|><|det|>[[72, 723, 920, 777]]<|/det|>
+As a final point of scientific interest, I would like to offer one minor point for the author's consideration. In the MS/MS spectrum for Aβ 1- 42 shown in Supplementary Figure 1F, it is interesting to note the dominance of the b- ion series, while the corresponding y- ion series is almost absent. A brief comment on this observation in the Results section or the caption of the figure could enhance the spectral interpretation of the fragmentation behavior.  
+
+<|ref|>text<|/ref|><|det|>[[72, 802, 160, 815]]<|/det|>
+Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[72, 828, 238, 841]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 841, 909, 881]]<|/det|>
+The authors have made appropriate clarifications of all my previously raised points in the first round of detailed review and the paper has been improved. Congratulations to very interesting and important work. I support publication as soon as possible.  
+
+<|ref|>text<|/ref|><|det|>[[72, 893, 160, 906]]<|/det|>
+Reviewer #3  
+
+<|ref|>text<|/ref|><|det|>[[72, 920, 238, 932]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[70, 932, 905, 947]]<|/det|>
+The presentation of the data in the revised manuscript makes it much more straightforward to read and interpret the data in
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 47, 923, 88]]<|/det|>
+my opinion. Here are some remaining minor comments for the authors, though they are mostly either presentation issues or nominal analyses that might provide some further insights if the authors choose to perform them. Nevertheless, I believe they have satisfied my major concerns, and I am supportive of publication.  
+
+<|ref|>sub_title<|/ref|><|det|>[[72, 101, 193, 113]]<|/det|>
+## Minor Comments  
+
+<|ref|>text<|/ref|><|det|>[[70, 113, 920, 230]]<|/det|>
+1. In their discussion of limitations of other high-res ST approaches, they might note that the CosMx 18k now offers near-transcriptome coverage – though at markedly lower gene-wise sensitivity  
+2. I would appreciate if the authors could do one additional analysis – to highlight novel gene-gene correlation modules that arise across from analysis across plaques (e.g. with WGCNA or similar).  
+3. The authors might comment on whether their observed gene expression changes signify some T cell infiltration (or any other evidence for this in their animals) versus microglial activation due to overlap in GO annotations. This was what I meant by cryptic expression in my previous review and I apologize for the misunderstanding.  
+4. The statement that APP NL-F is strictly better than NL-G-F as it reflects human pathology (Line 387-389) is too strongly made.  
+
+<|ref|>text<|/ref|><|det|>[[72, 243, 322, 256]]<|/det|>
+5. Scale bars missing in Fig. 3A, 4A.  
+
+<|ref|>text<|/ref|><|det|>[[72, 256, 617, 270]]<|/det|>
+6. Labels for GFAP, Ab, and SYTO colors are missing in Fig. 3B, Supp Fig. 5, 6.  
+
+<|ref|>text<|/ref|><|det|>[[70, 270, 910, 297]]<|/det|>
+7. I am not sure why the data must be duplicated as Supp Fig. 3D – would be preferable to instead include these panels in Fig. 3B?  
+
+<|ref|>text<|/ref|><|det|>[[72, 297, 425, 310]]<|/det|>
+8. The bottom axis labels for Fig. 3C, D are clipped.  
+
+<|ref|>text<|/ref|><|det|>[[72, 323, 174, 336]]<|/det|>
+Typographical  
+
+<|ref|>text<|/ref|><|det|>[[72, 336, 498, 350]]<|/det|>
+1. Line 190 "oligothiophene (LCO)" > "oligothiophenes (LCOs)  
+
+<|ref|>text<|/ref|><|det|>[[72, 350, 340, 363]]<|/det|>
+2. Line 429 "information" > "information"  
+
+<|ref|>text<|/ref|><|det|>[[72, 715, 915, 768]]<|/det|>
+Open Access This Peer Review 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.  
+
+<|ref|>text<|/ref|><|det|>[[72, 768, 796, 781]]<|/det|>
+In cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+<|ref|>text<|/ref|><|det|>[[72, 781, 910, 833]]<|/det|>
+The images or other third party material in this Peer Review 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.  
+
+<|ref|>text<|/ref|><|det|>[[72, 833, 618, 846]]<|/det|>
+To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 83, 263, 99]]<|/det|>
+## Dear Reviewers  
+
+<|ref|>text<|/ref|><|det|>[[117, 100, 883, 183]]<|/det|>
+Please find attached with this resubmission our revised manuscript "Isotope Encoded Spatial Biology Identifies Amyloid Plaque- Age- Dependent Structural Maturation, Synaptic Loss, and Increased Toxicity" by et al. We are grateful for the overall positive, thorough and constructive feedback provided, which helped significantly to both clarify and improve the manuscript substantially.  
+
+<|ref|>text<|/ref|><|det|>[[118, 182, 882, 232]]<|/det|>
+Please find below a detailed point- by- point response to the comments raised by the reviewers. The corresponding changes have been highlighted in an additional manuscript file (article_tracked_changes).  
+
+<|ref|>text<|/ref|><|det|>[[118, 247, 278, 280]]<|/det|>
+With best regards Jörg Hanrieder  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 305, 303, 328]]<|/det|>
+## Reviewer #1:  
+
+<|ref|>text<|/ref|><|det|>[[117, 329, 882, 428]]<|/det|>
+Comment 1: "One major limitation of this study is the lack of consideration for known post- translational modifications (PTMs) of \(\mathsf{A}\beta\) , which are known to influence its aggregation and pathological progression. The centroid mass values of MALDI MSI (Table S1) do not reflect known \(\mathsf{A}\beta\) PTMs, which should result in characteristic mass shifts. The authors should respond to the following questions regarding the peak assignment of their MALDI MS spectra."  
+
+<|ref|>text<|/ref|><|det|>[[118, 427, 670, 444]]<|/det|>
+(1) Why were these known PTMs not detected in the dataset?  
+
+<|ref|>text<|/ref|><|det|>[[117, 459, 882, 558]]<|/det|>
+We concur that this is an important point. Unfortunately, it is a limitation of this mouse model that no prominent \(\mathsf{A}\beta\) truncations or PTMs are observed, specifically \(pE - x\) and PSer8- x (Saito et al. 2014). To underpin this observation, we performed additional, in- depth ex situ mass spectrometric experiments to investigate potential sequence truncations/modifications that might be suppressed or below the detection limit in the MALDI MSI analyses.  
+
+<|ref|>text<|/ref|><|det|>[[118, 558, 882, 607]]<|/det|>
+Specifically, we performed two elaborate ex situ analyses of brain tissue extracts, whole proteomic profiling of purified fibrils and immune- precipitation LC- MS/MS of amyloid peptides.  
+
+<|ref|>text<|/ref|><|det|>[[117, 607, 883, 755]]<|/det|>
+To this end, we first immune- precipitated \(\mathsf{A}\beta\) peptides from \(\mathsf{A}\mathsf{p}\mathsf{p}^{\mathsf{N L - F}}\) brain extracts and analyzed the samples by LC- MS/MS on a high- resolution Orbitrap mass spectrometer (Supplementary Fig. 1). Resulting data was searched in PEAKS (de novo peptide sequencing) for characterizing potential posttranslational modifications, including phosphorylation, deamidation, oxidation and pyroglutamation. The analysis confirmed the presence of \(\mathsf{A}\beta\) 1- 38, 1- 39, and 1- 42, with 1- 42 being the most prominent species. Following sequence analyses, no PTMs were observed for either species. Please find an excerpt from the obtained PEAKS results, now detailed in Supplementary Fig. 1B, in the table below:
+
+<--- Page Split --->
+<|ref|>table<|/ref|><|det|>[[115, 99, 911, 236]]<|/det|>
+<|ref|>table_caption<|/ref|><|det|>[[115, 82, 815, 99]]<|/det|>
+Table 1: Excerpt from PEAKS results of the anti-Aβ IP-LC-MS/MS experiment.   
+
+<table><tr><td>Gene name</td><td>Peptide</td><td>Mass</td><td>Length</td><td>ppm</td><td>m/z</td><td>z</td><td>RT</td><td>PTM</td></tr><tr><td rowspan="3">APP</td><td>DAEFRHDSGYEVHHQKLVFFAEDVGS</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>NKGAIIGLMVGG</td><td>4129,012</td><td>1-38</td><td>3,1</td><td>826,81213</td><td>5</td><td>43,179</td><td>/</td></tr><tr><td>DAEFRHDSGYEVHHQKLVFFAEDVGS</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td rowspan="2">APP</td><td>NKGAIIGLMVGGV</td><td>4228,08</td><td>1-39</td><td>2,5</td><td>846,62537</td><td>5</td><td>44,834</td><td>/</td></tr><tr><td>DAEFRHDSGYEVHHQKLVFFAEDVGS</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>APP</td><td>NKGAIIGLMVGGVVIA</td><td>4511,27</td><td>1-42</td><td>1,1</td><td>903,26215</td><td>5</td><td>68,445</td><td>/</td></tr></table>  
+
+<|ref|>text<|/ref|><|det|>[[118, 249, 883, 284]]<|/det|>
+The obtained RAW data and results have been uploaded to the PRIDE repository under the accession number PXD063759.  
+
+<|ref|>text<|/ref|><|det|>[[115, 298, 883, 414]]<|/det|>
+Second, we analyzed fibrils purified from NLF brain. For this we used proteomics as the enzymatic digestion step (trypsin) provides increased sensitivity towards potentially truncated species. In contrast, this approach does not necessarily provide exact annotation of N- terminal modification for different C- terminal species (i.e x- 40, x- 42). Similarly to the IP- MS analyses, no PTMs were observed for the Aβ fragments. The corresponding raw data is accessible on the PRIDE repository under the accession number MSV000092311.  
+
+<|ref|>text<|/ref|><|det|>[[115, 429, 881, 463]]<|/det|>
+All experiments described above are now detailed in methods section "Ex situ Aβ Analysis" and have been incorporated into the main text on page 25 line 24.  
+
+<|ref|>text<|/ref|><|det|>[[115, 492, 881, 527]]<|/det|>
+(2) Does the dataset contain any mass peaks that could suggest age-dependent PTM changes in Aβ?  
+
+<|ref|>text<|/ref|><|det|>[[115, 541, 881, 591]]<|/det|>
+We thank the reviewer for this relevant question. Neither our MALDI MSI nor LC- MS/MS experiments (see Table above and Supplementary Fig. 1) suggest any presence of PTM for Aβ.  
+
+<|ref|>text<|/ref|><|det|>[[115, 592, 881, 658]]<|/det|>
+In our MALDI MS data, the phosphorylation would be detectable as the mass shift is +98 (H2PO2) or +80 (PO3), while a shift in 15N can be maximal +55 (as there are only 55 nitrogen in the Aβ1- 42 sequence). Similarly, the 3pE-x or 11pE-x modification would require any peak close to the unmodified 3-x and 11-x, neither of which is observed.  
+
+<|ref|>text<|/ref|><|det|>[[115, 672, 881, 740]]<|/det|>
+For the raw LC- MS/MS data, please refer to the PRIDE repository (accession numbers PXD063759 and MSV000092311), and for the raw MALDI MSI data, see the Shiny app: (https://maciejdulewiczgu.shinyapps.io/MALDI_GEOMX_VOLCANO/), where we present the data in an easily accessible format for the public.  
+
+<|ref|>text<|/ref|><|det|>[[118, 754, 415, 772]]<|/det|>
+Please see Supplementary Fig.1.  
+
+<|ref|>text<|/ref|><|det|>[[115, 786, 881, 821]]<|/det|>
+Comment 2: "The authors should explain why no background peptide signals are observed in the spectrum.  
+
+<|ref|>text<|/ref|><|det|>[[117, 836, 881, 903]]<|/det|>
+The focus of the current work was on imaging amyloid peptides that are inherently difficult to detect in situ due to their hydrophobic properties and highly aggregated state. Consequently, we have been tuning our protocols to enhance the amyloid signal, which requires extensive washes and formic acid hydrolysis. This extensive sample
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 82, 883, 149]]<|/det|>
+preparation leads, in turn, to a reduced signal of other less aggregated, endogenous peptides and small proteins. We performed control experiments with and without formic acid to demonstrate the effects of FA retrieval on amyloid signal, as shown below in Figure 1.  
+
+<|ref|>text<|/ref|><|det|>[[118, 164, 883, 199]]<|/det|>
+This is now clarified in the methods on page 22 line 20 and in a newly created Supplementary Fig. 2.  
+
+<|ref|>text<|/ref|><|det|>[[118, 214, 880, 248]]<|/det|>
+"Additionally, they should present the MS/MS spectra to confirm the presence of amyloid beta and assess whether other peptides were detected."  
+
+<|ref|>text<|/ref|><|det|>[[117, 263, 883, 395]]<|/det|>
+We thank the reviewer for this comment and agree that MS/MS confirmation of \(\mathsf{A}\beta\) presence is important. Due to the limited sensitivity of direct in situ MALDI MS/MS, we employed extensive ex situ analyses for \(\mathsf{A}\beta\) sequence validation as described above under Comment 1. Briefly, we performed anti- \(\mathsf{A}\beta\) immunoprecipitation followed by LC- MS/MS as well as LC- MS/MS- based proteomics of purified amyloid fibrils, all on \(\mathsf{A}\beta^{\mathsf{N L - }}\) \(F\) mouse brain tissue extracts. Here, LC- MS/MS confirmed the presence of \(\mathsf{A}\beta 1 - 42\) along with \(\mathsf{A}\beta 1 - 38\) and \(\mathsf{A}\beta 1 - 39\) . A representative MS/MS spectrum of \(\mathsf{A}\beta 1 - 42\) is shown in Supplementary Figure 1F.  
+
+<|ref|>text<|/ref|><|det|>[[117, 395, 883, 494]]<|/det|>
+Our MALDI MSI data predominantly show \(\mathsf{A}\beta 1 - 42\) within plaques, with \(\mathsf{A}\beta 1 - 38\) detected in only two ROIs/plaques (low S/N ratio), while 1- 39 was not detected in any MALDI MSI ROI. This further highlights that \(\mathsf{A}\beta 1 - 42\) is indeed the predominant \(\mathsf{A}\beta\) species. The identification of \(\mathsf{A}\beta 1 - 38\) and \(\mathsf{A}\beta 1 - 39\) by LC- MS/MS is not unexpected and rather highlights the high sensitivity of this method (as compared to direct in situ detection via MALDI MSI).  
+
+<|ref|>text<|/ref|><|det|>[[119, 509, 444, 526]]<|/det|>
+Please see Supplementary Figure 1.  
+
+<|ref|>text<|/ref|><|det|>[[117, 541, 881, 675]]<|/det|>
+Comment 3: "The authors claim that older amyloid plaques exhibit increased neurotoxicity and synaptic loss, as demonstrated through iSILK labeling, transcriptomics, and fluorescence imaging. While the data suggest a correlation between plaque age and synaptic dysfunction, it is not yet clear how this study advances our understanding beyond prior work that has already reported synapse loss in the vicinity of amyloid plaques. To fully support their conclusions, the authors should clarify what new biological insights this study provides compared to previous findings on plaque- induced toxicity."  
+
+<|ref|>text<|/ref|><|det|>[[118, 689, 881, 755]]<|/det|>
+We appreciate this comment and would like to offer an extended explanation and discussion. Most importantly, the methods described allow us to decouple plaque age from chronological age of the mice. This brings several advantages that would not be discernible with static methods.  
+
+<|ref|>text<|/ref|><|det|>[[117, 771, 881, 903]]<|/det|>
+1) Most importantly, this reveals and validates that plaques indeed affect synapses and that persisting plaque ageing/maturation is associated with proximal synaptic dysfunction and loss, rather than plaques exerting immediate, acute toxicity. In static experiments one could not discern the age of individual plaques within one mouse/age group. Hence it would not be possible to assess whether the differences in plaque phenotypes and their effects (for example on synapses) are a result of plaque maturation or whether they represent different subgroups of plaques within a single tissue. Our methods clearly allow for this distinction.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 98, 883, 265]]<|/det|>
+2) Even more important, on the transcriptomic scale, the effect of plaque maturation is easily masked when comparing plaques with each other across the same mouse ages. For this, we further demonstrate the advantage of correlation of similar, spatially refined entities as compared to both bulk RNAseq and static spatial transcriptomics (ST) experiments. This is demonstrated in the more sensitive detection of synaptic genes that vary across plaques within the same mouse. To highlight this, we provide additional data demonstrating how both spatial transcriptomics (as compared to bulk brain extract transcriptomics), as well as dynamic spatial transcriptomics, improve sensitivity and specificity to plaque and plaque age molecular processes (please see Supplementary Fig. 10).  
+
+<|ref|>text<|/ref|><|det|>[[117, 278, 882, 346]]<|/det|>
+3) We further demonstrate conceptual novelty in integrating chemical imaging with spatial transcriptomics. Of note, this is demonstrated by the combination of functional LCO microscopy with iSILK and ST, showing that plaque maturation involves the continuous fibrillization at the plaque.  
+
+<|ref|>text<|/ref|><|det|>[[117, 360, 882, 444]]<|/det|>
+4) An additional advance is related to the technical innovation, implementing the iSILK method in old mice. While challenging, we succeeded in following plaque formation from before plaques form (6-10mo) during plaque growth at old ages (18mo). This showed that for this knock in model, plaques form as small cores in the cortex consisting of 1-42.  
+
+<|ref|>text<|/ref|><|det|>[[117, 459, 882, 510]]<|/det|>
+To clarify the novelty of our biological insights, we have added the following sentences to the discussion and embedded references to the most recent and impactful spatial transcriptomics studies in the field (page 15 line 18):  
+
+<|ref|>text<|/ref|><|det|>[[116, 541, 882, 755]]<|/det|>
+"Whereas findings from other impactful spatial transcriptomics studies on AD brain tissue (Chen et al. 2020, Mallach et al. 2024) have also supported accumulated rather than acute plaque toxicity, they were unable to directly link plaque age with increasing synaptic dysfunction, entirely decoupled from chronological mouse age. In fact, many studies have focused on the spatial rather than temporal aspect of plaque toxicity (Wood et al. 2022, Johnston et al. 2025) as well as focused on subsets of genes associated with e.g. glial activation in response to plaque pathology. Our study, in contrast, allowed us to evaluate the effects of maturing plaques independent of mouse age on a whole transcriptomic scale, with key findings being validated with direct immunological stainings. Additionally, the mouse model used in this study more accurately reflects human AD as this combines WT Abeta pathology together with ageing, which contrasts with the NLGF mouse model, on which most prior conclusions have been based."
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[117, 101, 303, 124]]<|/det|>
+## Reviewer #2:  
+
+<|ref|>text<|/ref|><|det|>[[115, 124, 881, 486]]<|/det|>
+Dear Editor and authors, I have reviewed the paper "Isotope Encoded Spatial Biology Identifies Amyloid Plaque- Age- Dependent Structural Maturation, Synaptic Loss, and Increased Toxicity" by Hanrieder and co- workers. The paper aims to understand the initiation and progression of Aβ aggregate accumulation and correlations with CNS cell responses by spatial transcriptomics in the proximity of early and late Aβ aggregates. Chemical time stamps were induced using pulse chase of food supplemented with 15N- isotope that label newly produced protein facilitating using a method called Imaging of stable isotope labelling kinetics (iSILK). As a complementary time stamp, the maturity of the Aβ- plaque structures was monitored by LCO fluorescence technology developed by researchers at Linkoping University, Sweden. Spatial transcriptomics together with iSILK and LCO/immunofluorescence structure correlations allowed an unprecedented correlation of proximal cell response to the plaque maturity stage. The study concluded that mature Aβ plaques, in comparison with early plaques, regardless of chronological mouse age, were associated with changes suggesting synaptic toxicity responses. Early plaque formations showed increased immune response gene expression. This methodological approach puts within reach more information on differentiated cellular responses to plaque fibril structure at different plaque development stages and cell types (microglia, astrocytes, and diverse neuronal populations). Overall the impression of the innovative application of different integrated techniques and the presentation of the paper is very positive. There are some of points that should be modified and clarified in the revised version to improve the paper:  
+
+<|ref|>text<|/ref|><|det|>[[115, 500, 881, 682]]<|/det|>
+Comment 1: "The choice of the APP NL- F knock- in mouse in this study was clever, because this mouse makes almost exclusively Aβ1- 42, allowing the authors to specifically iSILK- monitor Aβ1- 42 species using MALDI- ToF imaging. This feature should be more clearly stated in the description of the selection of the mouse model. Suggestively, at Results under the header 'iSILK delineates spatial and structural patterns of plaque formation and maturation' - after the sentence in the first paragraph 'This gradual increase in plaque pathology with age more closely resembles the human disease' - add something like: "The choice of the APP NL- F knock- in mouse in this study provided another biochemical advantage. Because this mouse makes almost exclusively deposited Aβ1- 42, it allows us to specifically iSILK- monitor Aβ1- 42 species using MALDI- ToF imaging."  
+
+<|ref|>text<|/ref|><|det|>[[117, 697, 883, 814]]<|/det|>
+We are grateful for these positive comments and the constructive feedback. We concur with this comment regarding the choice of the mouse model. As suggested, a corresponding statement, commenting on the biochemical advantage of the mouse model, has been incorporated in the results. In addition, we have performed additional ex situ LC- MS/MS experiments confirming Aβ1- 42 to be the predominantly produced and deposited Aβ species. Please see Supplementary Fig. 1, Methods section "Ex situ Aβ Analysis" and Results page 25 line 24.  
+
+<|ref|>text<|/ref|><|det|>[[117, 828, 881, 911]]<|/det|>
+Comment 2: \\*"Under the header 'Amyloid plaque maturation is characterized by continuous fibrillization with age' there is a description of the LCO probes q- FTAA and h- FTAA. The reference here is good but insufficient. LCO reference 21 is ok for general purposes but the following should be added in the description (with appropriate references):
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 81, 884, 245]]<|/det|>
+'This is enabled by the difference in affinity of the two LCO probes, \(q\) - FTAA and h- FTAA, towards amyloid aggregates. Specifically, \(q\) - FTAA preferentially binds to mature and compact \(\beta\) - pleated aggregates, while h- FTAA binds to less compact, yet still \(\beta\) - pleated aggregates (reference: now ref 34, Nyström et al. Evidence for age- dependent in vivo conformational rearrangement within \(\mathrm{A}\beta\) amyloid deposits). Due to their different emission profiles, the LCO probes can be spatially delineated using hyperspectral fluorescent microscopy. Here, the ratio of the LCO maxima (500 nm for \(q\) - FTAA / 580 nm for h- FTAA) is used to express preferential binding of either of the two LCO probes, whereby an increase in 500 nm intensity is indicative of increased \(q\) - FTAA binding and therefore increased structural maturity of the amyloid fibrils (Fig. 2B).'  
+
+<|ref|>text<|/ref|><|det|>[[117, 245, 883, 295]]<|/det|>
+And then add references for these statements, e.g., ref 35 Rasmussen et al. (Amyloid polymorphisms...), and ref 45 Parvin et al. (Divergent Age- Dependent Conformational Rearrangement...).\*\*  
+
+<|ref|>text<|/ref|><|det|>[[116, 311, 875, 329]]<|/det|>
+We appreciate these suggestions and revised this part accordingly (page 8, lines 18).  
+
+<|ref|>text<|/ref|><|det|>[[115, 344, 881, 491]]<|/det|>
+Comment 3: \*\*The spatial transcriptomics data should be clarified. It is not clear how the reference data (baseline normalized expression) were obtained for making the volcano plots of overexpression and decreased expression in Fig. 3C and 3F. In the same context: How does one interpret the GeoMx CSV files when there is no reference data set included? Was this from an already published database? If so, the values from this reference data set should be included, as they can be very useful for others looking at other genes and to derive the data underlying the volcano plot. The contrast (samples vs reference) is not obvious from the CSV files – it appears to be only numbers listed for 10- month and 18- month \(App^{NL - F}\) mice.  
+
+<|ref|>text<|/ref|><|det|>[[116, 507, 881, 591]]<|/det|>
+We agree with this important comment and would like to further explain what is shown in the data. The volcano plots 3C and 3F do not represent differential expression analysis. They show genes that were significantly correlating with plaque age, both positively and negatively for the respective mouse age group (10mo: 3C; 18mo: 3D). Consequently, there is no reference data set to refer to.  
+
+<|ref|>text<|/ref|><|det|>[[115, 606, 881, 705]]<|/det|>
+We acknowledge that this is not clearly presented in the manuscript. We therefore extended the description of the data in Figure legend 3. Additionally, we adjusted the wording in the main text on page 10 line 14 for clarification: "Volcano plots for both 10- month (Fig. 3C) and 18- month- old (Fig. 3D) mice demonstrated that gene expression levels had significant positive and negative correlations with increasing plaque age."  
+
+<|ref|>text<|/ref|><|det|>[[115, 720, 881, 787]]<|/det|>
+Furthermore, the link given (the Shiny app: https://hanriederlab.shinyapps.io/PlaqueAgeTranscriptomics/) can be very useful if it is more clearly described how it was obtained and what cutoff values are used for significance.  
+
+<|ref|>text<|/ref|><|det|>[[116, 802, 881, 886]]<|/det|>
+We agree and provide a detailed description of the data accessible in the Shiny App both in the methods and additionally detailed in a Readme file to highlight the details of the data and the analysis they originate from. Additionally, we updated the methods section "MALDI MSI - GeoMx Data Analysis" for increased clarity, including cut- off values. The data accessible via the ShinyApp is briefly explained under the Data
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 81, 881, 133]]<|/det|>
+Availability section in the manuscript. Also, we curated all GeoMx and MALDI data under a new ShinyApp address. Please see: https://maciejulewiczgu.shinyapps.io/MALDI_GEOMX_VOLCANO/  
+
+<|ref|>text<|/ref|><|det|>[[117, 164, 881, 214]]<|/det|>
+Is the x- axis "plaque age" (in months?) based on iSILK or q- FTAA/h- FTAA fluorescence, and what is "Log Cnorm"? A clear description of how to use the link should be included in the paper, with more details in the methods section."\*  
+
+<|ref|>text<|/ref|><|det|>[[117, 230, 881, 328]]<|/det|>
+We concur that this is critical and update the description of these annotations throughout all presented Figures. In brief, the x- axis values are representing measures of plaque age and are obtained from calculating the width of the MALDI mass peak for Ab 1- 42. Here the width (calculated as full width at half maximum, FWHM) is a measure of peak broadening and consequently an indication of 15N incorporation and plaque age, respectively.  
+
+<|ref|>text<|/ref|><|det|>[[117, 329, 881, 378]]<|/det|>
+In Figures 1- 4 as well as in the Shiny app, the x- axis is now referred to as "Nitrogen index LP", which is explained in the main text on page 10 lines 9 and page 30 line2 and visualized in the updated Fig. 1 E.  
+
+<|ref|>text<|/ref|><|det|>[[117, 379, 881, 426]]<|/det|>
+We also included a more detailed explanation of the nitrogen index calculation under the header "Calculation of nitrogen index in linear positive mode" in the methods section.  
+
+<|ref|>text<|/ref|><|det|>[[117, 442, 881, 523]]<|/det|>
+Comment 4: "The list of acknowledgements for funding is very long. But there is no acknowledgement or information for where the authors obtained the LCOs q- FTAA and h- FTAA, which play an important role in the paper. Since these molecules do not appear to be commercially available, it needs to be specified. If they were synthesized in their own lab this should be stated in the Materials and Methods."  
+
+<|ref|>text<|/ref|><|det|>[[117, 540, 880, 572]]<|/det|>
+We received the probes as a kind gift. This information was updated in the methods and acknowledgements.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[117, 84, 302, 106]]<|/det|>
+## Reviewer #3:  
+
+<|ref|>text<|/ref|><|det|>[[115, 107, 881, 240]]<|/det|>
+In this manuscript, Wood & Dulewicz et al. describe the combination of mass spectrometry imaging (MSI) and hyperspectral imaging to measure Ab42 plaque age and morphology in a mouse model of Alzheimer's disease. Using a pulse- chase strategy where mice are fed with a 15N- enriched diet, they were able to distinguish nascent and aged plaques. Hyperspectral imaging with oligothiophene chemistry further demonstrates changes in plaque morphology correlated to age. They further combine this approach with single- plaque GeoMx spatial transcriptomics to identify correlates with age inferred from their MSI approach.  
+
+<|ref|>text<|/ref|><|det|>[[115, 239, 881, 386]]<|/det|>
+Overall, the experiments are executed quite well, and the methods and data will be useful for the spatial and Alzheimer's biology community. In the future, I also see great value in combining these techniques with higher- resolution spatial transcriptomics or multicomics to understand the molecular underpinning of these processes. While I find their iSILK MSI technology to be technically impressive, especially when combined with these orthogonal measurements, I believe the richness of their data could have been used to better extract biologically novel conclusions. The manuscript would also benefit from more extended analyses, embedding with more specific knowledge in literature, and better bookkeeping of the collected data.  
+
+<|ref|>text<|/ref|><|det|>[[117, 386, 880, 420]]<|/det|>
+If the authors could address or comment on these points, I believe the manuscript will be significantly improved and I would be supportive of publication.  
+
+<|ref|>text<|/ref|><|det|>[[118, 434, 881, 485]]<|/det|>
+We are grateful for the positive feedback and appreciate the constructive comments that helped to significantly improve the manuscript. Please see our detailed responses below.  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 499, 309, 518]]<|/det|>
+## Major Comments:  
+
+<|ref|>text<|/ref|><|det|>[[118, 533, 881, 583]]<|/det|>
+Comment 1: "The use of MSI pulse- chase is quite inspired. I agree that the 15N/14N ratio is a good proxy for plaque age. Here are some points the authors may want to consider:  
+
+<|ref|>text<|/ref|><|det|>[[145, 584, 882, 682]]<|/det|>
+a. It seems like a missed opportunity for developing a numerical index or 'clock' for plaque age. This could be used to assign ages to plaques without MSI, for instance. If hyperspectral imaging or morphology could be used to do a regression analysis, I think that could be immensely useful. I understand that this may be challenging to establish rigorously, but it could be a point of discussion for future work."  
+
+<|ref|>text<|/ref|><|det|>[[118, 697, 881, 748]]<|/det|>
+This is a very important point, and we would like to clarify some methodological aspects as we concur and were in fact aiming to use the LCO as surrogate markers of plaque age. In detail, we performed two correlative spatial experiments (Fig. 1D- E):  
+
+<|ref|>text<|/ref|><|det|>[[117, 763, 880, 797]]<|/det|>
+1) iSILK (MALDI MSI) in conjunction with spatial transcriptomics (Fig. 1E) to investigate plaque age-specific changes in gene expression.  
+
+<|ref|>text<|/ref|><|det|>[[118, 811, 881, 861]]<|/det|>
+2) iSILK (MALDI MSI) in combination with LCO hyperspectral microscopy (Fig. 1D; though the LCO microscopy was not performed together with GeoMx for limited feasibility reasons, we further detail below under Comment 2d.)  
+
+<|ref|>text<|/ref|><|det|>[[118, 877, 881, 910]]<|/det|>
+The rationale for combining iSILK (MALDI MSI) with LCO microscopy was to both identify whether aged plaques are characterized with changes in fibrillation and,
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 82, 883, 150]]<|/det|>
+exactly as suggested by your comment, to have a surrogate marker of plaque age that is easier to implement with follow- up experiments and does not require costly and time- consuming SILK labelling experiments. (Indeed, the LCO approach has been used for the IHC experiments described in Fig. 4, experimental overview in Fig. 1F)  
+
+<|ref|>text<|/ref|><|det|>[[117, 164, 883, 216]]<|/det|>
+Please refer to an updated Figure 1 for a detailed methodological overview. As suggested by the reviewer, we incorporated this in the discussion on page 19 line 4, highlighting the future potential of the LCO q/h emission ratio.  
+
+<|ref|>text<|/ref|><|det|>[[117, 230, 881, 297]]<|/det|>
+b. "It would be interesting for the authors to look at how the MSI and hyperspectral imaging replicate within and across animals. Are these correlations sufficiently strong and hence biologically robust across animals? A breakdown of Fig. 2E by animal, for example, could be helpful."  
+
+<|ref|>text<|/ref|><|det|>[[117, 312, 883, 395]]<|/det|>
+We thank the reviewer for raising this important question. In Fig. 2E, the correlation between MSI and LCO hyperspectral imaging was originally presented for each individual mouse \((n = 3\) , with each plot corresponding to one mouse). This showed that the correlation is biologically robust across animals included in the study \((R = 0.64 - 0.89, p< 0.05)\)  
+
+<|ref|>text<|/ref|><|det|>[[117, 395, 881, 428]]<|/det|>
+We acknowledge that this was not clearly described and improperly visualized in the original submission.  
+
+<|ref|>text<|/ref|><|det|>[[117, 428, 883, 526]]<|/det|>
+To improve clarity, we have now combined all plaques from the three mice into a single plot (Fig. 2E) and revised the figure caption accordingly. Individual mice are distinguished by color, and plaque location (hippocampal vs. cortical) as indicated by different shapes. Both the combined trend (Fig. 2F) and the individual mouse correlations (now reported in the main text on page 9, line 10, Supplementary Fig. 4A- C) are consistent and support the same conclusions:  
+
+<|ref|>text<|/ref|><|det|>[[117, 526, 881, 576]]<|/det|>
+a) the hyperspectral ratio, reflects plaque age, b) the plaque core is older than the plaque periphery and c) cortical plaques are, on average, older than hippocampal plaques.  
+
+<|ref|>text<|/ref|><|det|>[[117, 590, 881, 641]]<|/det|>
+c. "Could the authors briefly comment on the feasibility and cost of deploying this technology? What is the feasibility as well of MSI and LCO imaging on the same section?"  
+
+<|ref|>text<|/ref|><|det|>[[117, 656, 881, 706]]<|/det|>
+We appreciate this comment as there are certain challenges and limitations that come with the various spatial techniques that need to be considered when planning the experiments.  
+
+<|ref|>text<|/ref|><|det|>[[117, 707, 881, 756]]<|/det|>
+A significant cost factor for iSILK is the stable isotope diet (ca \(\) 12000/kg)\$. This is particular true for the long labelling experiments needed in the present study as those mice develop amyloid pathology gradually with age.  
+
+<|ref|>text<|/ref|><|det|>[[117, 756, 881, 821]]<|/det|>
+Each mouse requires 1g labelled diet per day. We labelled each mouse for 4months (ca 122days ie 122g/mice, \(n = 7\) ) requiring 1kg label for all 7 mice. In addition, there are of course breeding costs and most importantly costs for personell as these experiments pan over 1- 2years.  
+
+<|ref|>text<|/ref|><|det|>[[117, 821, 881, 903]]<|/det|>
+In addition, buffers and probe reagents constitute a major cost for the associated GeoMx experiments. GeoMx kits and buffers for two glasses amount to ca. \(\) 3000\(/glass slide (3 - 4 sections/glass). Finally, sequencing costs to quantify the released probes amount to ca.\) \ \(3000\) /index plate. (96 AOI/plate). As those latter costs are highly dependable on inhouse availability and deals, we merely mentioned the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 82, 883, 116]]<|/det|>
+consumables (diet and glasses/GeoMx run in the Methods section/Experimental Design and Spatial Transcriptomics).  
+
+<|ref|>text<|/ref|><|det|>[[116, 131, 883, 264]]<|/det|>
+A further challenge includes further the compatibility of the different techniques for correlative spatial biology/multimics in a single section. The LCO microscopy requires mounting with cover slips which prevents to do LCO imaging prior to MALDI (or GeoMx). In turn, the MALDI sample preparation and acquisition settings for peptide imaging lead to tissue distortions that prevent subsequent fluorescent microscopy. (please see Kaya et al 2017 PMID: 28318232). Consequently, the MALDI/LCO as well as the MALDI/GeoMx analyses had to be carried out on sequential sections.  
+
+<|ref|>text<|/ref|><|det|>[[117, 279, 883, 313]]<|/det|>
+We further discuss compatibility challenges in the limitations (page 19 line 8) and provide more detail on the design of our correlative experiments in a revised Fig 1.  
+
+<|ref|>text<|/ref|><|det|>[[117, 344, 881, 395]]<|/det|>
+Comment 2: "I have some general reservations about the GeoMx analysis. Given the richness of the dataset, the analysis should give correspondingly rich and novel biological insights. Broadly, are there any unexpected findings here?  
+
+<|ref|>text<|/ref|><|det|>[[118, 410, 857, 427]]<|/det|>
+We thank the reviewer for raising this concern regarding the richness of the results.  
+
+<|ref|>text<|/ref|><|det|>[[117, 442, 883, 525]]<|/det|>
+While the dataset is indeed rich, capturing a total of 19962 RNA transcripts, the lack of even more distinct and rich changes can be related to the design of the study and the research question we wanted to address i.e. Is the toxicity of plaques an immediate acute phenomenon in the area that they occupy or do plaques continue to cause ongoing increasing toxicity as they mature over time.  
+
+<|ref|>text<|/ref|><|det|>[[116, 541, 883, 690]]<|/det|>
+Previous studies on plaque associated changes in gene expression did not consider the time component on a single plaque age level and in addition rather focus on differential changes between plaques and non- plaque regions. Indeed, as Supplementary Figure 10 demonstrates, a direct spatial transcriptomics comparison between plaque- associated and non- plaque- associated areas revealed distinct up- or downregulation of disease- associated astrocytic and synaptic genes. These marked changes likely reflect the stark biological contrast between two fundamentally different environments: one area containing neurotoxic plaques and one area without plaque pathology.  
+
+<|ref|>text<|/ref|><|det|>[[116, 705, 883, 803]]<|/det|>
+Our study, in contrast, aimed to explore how gene expression correlates with increasing plaque age, independently of the animals' chronological age (something warranted by our iSILK paradigm). Given the relative similarity between differentially aged though biochemically similar cored plaques our plaque centric GeoMx analyses was naturally expected to reveal rather subtle biological differences as compared to analyses where plaque changes are compared to non- plaque regions.  
+
+<|ref|>text<|/ref|><|det|>[[116, 803, 883, 835]]<|/det|>
+It was very interesting and surprising that the major groupings affected in this incremental way, correlating with plaque age, were the synaptic genes.  
+
+<|ref|>text<|/ref|><|det|>[[116, 836, 883, 885]]<|/det|>
+To our knowledge, we are the first to demonstrate this continuous decline in synaptic gene expression with increasing plaque age on a (whole) transcriptomic scale. Thus, we can show that the well- known synaptic loss at plaques is not an immediate acute
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 81, 883, 150]]<|/det|>
+effect but rather a slow ongoing infliction of stress and damage as the plaque matures over time. This is now more clearly presented and compared to relevant literature in the discussion on page15 line 18.  
+
+<|ref|>text<|/ref|><|det|>[[115, 180, 881, 264]]<|/det|>
+a. Based on the sequencing, the authors suggest dysregulation of synaptic, immune, and metabolic genes proximal to plaques, correlating with plaque age on adjacent sections. Rather than a change in regulatory patterns, is it possible that these changes are explained instead by differences in cell population composition? Much of this is also driven by FOV (AOI) size and how many cells are captured.  
+
+<|ref|>text<|/ref|><|det|>[[147, 279, 883, 410]]<|/det|>
+i. Neuronal mRNA is typically enriched in the soma (with some exceptions, often in an activity-dependent manner). Thus, depletion of neurons might be an equally fair explanation of the apparent decrease in synaptic gene expression observed in the GO analysis. This likewise applies for immune/metabolic genes – as recently observed with high-resolution spatial data in 5xFAD mice. The authors could consider using morphological (Nissl, silver, DAPI, etc.) staining or further immunofluorescence to quantify the cell populations here. Further markers for DAA/DAMs might also be helpful.  
+
+<|ref|>text<|/ref|><|det|>[[118, 458, 881, 510]]<|/det|>
+We concur and would like to offer our perspective along with additional data analyses addressing this concern, which is that differential cellular composition of the AOI can yield artefacts rather than true biological changes.  
+
+<|ref|>text<|/ref|><|det|>[[116, 523, 883, 723]]<|/det|>
+Delineating cell population heterogeneity is in fact a limitation of GeoMx as cell- type- specific analysis at the single- plaque level is currently not practically feasible using this method. The number of transcripts from individual cell types surrounding a single plaque within a finite AOI is insufficient to support high- quality transcriptomic profiling. Alternatively, pooling cell types from multiple plaques would result in a loss of spatiotemporal resolution, including the ability to associate transcriptomic data with individual plaque age. Additionally, GeoMx experiments are limited by the number of morphological markers that can be used simultaneously (typically 3 to 4), which restricts cell- type identification, particularly when pathological structures like Aβ plaques, as in our study, are also being imaged. Consequently, we were limited by the morphology markers we managed to implement (Aβ, cell count (SYTO, similar to DAPI) and GFAP).  
+
+<|ref|>text<|/ref|><|det|>[[118, 737, 881, 772]]<|/det|>
+With this we were however able to evaluate any plaque age associated changes in AOI cell count (SYTO), AOI size and GFAP immunoreactivity.  
+
+<|ref|>text<|/ref|><|det|>[[118, 786, 881, 854]]<|/det|>
+First, there was no significant correlation (P>0.05) between AOI size or SYTO cell count with plaque age (Supplementary Fig. 8), suggesting that transcriptional differences are unlikely to be a result of systematic differences in the number of cells surrounding aging plaques.  
+
+<|ref|>text<|/ref|><|det|>[[115, 869, 881, 902]]<|/det|>
+Second, to estimate the number of astrocytes near plaques within the AOIs, we quantified the GFAP signal area for each AOI. Notably, correlation analysis between
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 881, 132]]<|/det|>
+Gfap mRNA expression and GFAP IHC signal showed strong concordance (R=0.95- 0.97, P<0.0001; Supplementary Fig. 9 A-B), indicating that Gfap transcript levels indeed reflect corresponding protein abundance. 
+
+<|ref|>text<|/ref|><|det|>[[115, 134, 883, 297]]<|/det|>
+To investigate whether astrocyte numbers change with increasing plaque age, we correlated the GFAP IHC signal with plaque age. However, no significant correlation was observed (P>0.05, Supplementary Fig. 9 C-D), suggesting that astrocyte numbers do not systematically vary with plaque age. However, since GFAP is generally regarded as a marker of proliferation rather than activation, this lack of correlation does not necessarily imply an absence of astrocytic reactivity to plaques. It is possible that astrogliosis (i.e., astroglial proliferation) is more prominent during initial Aβ plaque formation and may diminish as plaques age. It is also important to stress that we might be underpowered to detect meaningful differences in astrogliosis with increasing plaque age. 
+
+<|ref|>text<|/ref|><|det|>[[115, 313, 883, 362]]<|/det|>
+A discussion of the this limitation to not capture potential cell-type heterogeneity in the proximity of aging plaques as well as the Gfap IHC has been added to the discussion (page 17 line 7) and in the limitation (page 19 line 8). 
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 395, 881, 428]]<|/det|>
+## ii. In a similar vein, are there any changes in mRNAs known to be translocated to synapses, versus those known to be soma-localized? 
+
+<|ref|>text<|/ref|><|det|>[[115, 445, 883, 493]]<|/det|>
+This is a very interesting point as it would be interesting whether the lower number of synaptic genes with plaque age is a consequence of synapse specific impairment or global neuronal loss around plaques. 
+
+<|ref|>text<|/ref|><|det|>[[115, 511, 883, 592]]<|/det|>
+We therefore analyzed mRNAs that were significantly positively or negatively correlated with plaque age to assess their subcellular localization (soma vs. neuropil, i.e., synaptic regions), For this we used data from Glock et al. (2021) (PMID: 34670838) that identified 807 neuropil-translocated (28%) and 2945 (72%) soma-located RNAs (3.5-times more soma- than neuropil-localized transcripts). 
+
+<|ref|>text<|/ref|><|det|>[[115, 609, 883, 657]]<|/det|>
+Our data show a similar pattern: across both positively and negatively correlated gene sets, we found 4–5 times more soma- located than neuropil-translocated RNAs (please see Table below). 
+
+<|ref|>table<|/ref|><|det|>[[115, 690, 874, 906]]<|/det|>
+<table><tr><td>mouse age (directionality of correlation)</td><td>Localization</td><td>Number of Genes (n)</td><td>Percentage of all significantly correlated RNAs in respective group (%)</td></tr><tr><td rowspan="2">10 m (positive)</td><td>neuropil</td><td>7</td><td>20</td></tr><tr><td>soma</td><td>28</td><td>80</td></tr><tr><td rowspan="2">10m (negative)</td><td>neuropil</td><td>35</td><td>16,6</td></tr><tr><td>soma</td><td>176</td><td>83,4</td></tr><tr><td rowspan="2">18 m (positive)</td><td>neuropil</td><td>24</td><td>17,8</td></tr><tr><td>soma</td><td>111</td><td>82,2</td></tr></table>
+
+<--- Page Split --->
+<|ref|>table<|/ref|><|det|>[[116, 81, 875, 128]]<|/det|>
+
+<table><tr><td rowspan="2">18 m (negative)</td><td>neuropil</td><td>55</td><td>25,3</td></tr><tr><td>soma</td><td>162</td><td>74,7</td></tr></table>
+
+<|ref|>text<|/ref|><|det|>[[115, 160, 883, 206]]<|/det|>
+An exception to this trend was observed in genes negatively correlated with plaque age in 18-month-old mice, where 25% were neuropil-translocated (2.9-fold more soma-than neuropil-localized RNAs).
+
+<|ref|>text<|/ref|><|det|>[[117, 209, 883, 238]]<|/det|>
+In comparison, only 16% of negatively correlated genes were neuropil-translocated in 10-month-old mice (a 5-fold difference).
+
+<|ref|>text<|/ref|><|det|>[[115, 242, 883, 271]]<|/det|>
+This could suggest that with increasing plaque age, synaptic loss becomes more pronounced relative to general neuronal loss.
+
+<|ref|>text<|/ref|><|det|>[[115, 290, 883, 370]]<|/det|>
+However, we emphasize that this interpretation is speculative due to several limitations: (1) the classification of RNA localization is based on a single study, as transcriptome-wide RNA translocation is not well-characterized; (2) we cannot rule out (post-mortem) RNA diffusion or other artifacts affecting localization; and (3) the data do not allow for robust statistical comparison.
+
+<|ref|>text<|/ref|><|det|>[[117, 389, 719, 402]]<|/det|>
+We included a comment on this in the discussion. (page 16 line 12)
+
+<|ref|>text<|/ref|><|det|>[[115, 421, 881, 468]]<|/det|>
+**iii.** Do the authors observe upregulation of **cryptic** **genes** associated with non-native cell types in these plaques? For instance, cell type markers of unexpected immune cells.
+
+<|ref|>text<|/ref|><|det|>[[117, 487, 883, 534]]<|/det|>
+We thank the reviewer for this interesting question. According to Bruker Spatial Biology, GeoMx probes are designed not to target cryptic genes, as they are based on the RefSeq transcriptome, which includes only well-characterized transcripts.
+
+<|ref|>text<|/ref|><|det|>[[115, 568, 883, 599]]<|/det|>
+**iv.** Can the authors further embed some of these findings with more recent literature on **functional** **changes** **in** **astrocytes** **and** **microglia?"**
+
+<|ref|>text<|/ref|><|det|>[[117, 618, 883, 664]]<|/det|>
+We agree and in order to address this, present correlation data for disease associated microglial and astroglial genes in both Fig 3 and Supplementary Fig. 7 in the Results and discuss those data with respect to the current literature (page 17 line 7).
+
+<|ref|>text<|/ref|><|det|>[[115, 684, 883, 730]]<|/det|>
+**Comment 2b:** "The statistics on correlations between the GeoMx data and plaque age should consider multiple testing correction if possible (q-value or FDR perhaps). My understanding is that raw p-values are shown. Are these conclusions still robust?"
+
+<|ref|>text<|/ref|><|det|>[[117, 750, 883, 878]]<|/det|>
+The reviewer is correct: the volcano plots in Figures 3C and 3F display unadjusted p-values. Correlations were performed across a total of 34 plaques, which limits statistical power. Consequently, the resulting p-values do not survive transcriptome-wide FDR correction. A key limitation of applying correction, such as the commonly used Benjamini-Hochberg method, for multiple comparisons in genome-wide **correlation** **analyses** rather than the common simple pairwise comparisons (across 19,963 genes) is that the threshold for statistical significance becomes extremely stringent. Specifically, only near-perfect correlations (r≈0.9) will survive correction.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 82, 883, 149]]<|/det|>
+We fully acknowledge the reviewer's concern regarding robustness. However, we believe our broader findings (namely, the decrease in synaptic gene expression and the increase in inflammatory and metabolic gene expression with increasing plaque age) remain robust.  
+
+<|ref|>text<|/ref|><|det|>[[115, 149, 883, 198]]<|/det|>
+This conclusion is supported by the consistent correlation patterns observed not in isolated genes, but across entire groups of synaptic, inflammatory, and metabolism- related genes, suggesting biologically meaningful and coherent changes.  
+
+<|ref|>text<|/ref|><|det|>[[115, 198, 880, 230]]<|/det|>
+Importantly, all GO analyses presented in the manuscript were performed using FDR- corrected p- values.  
+
+<|ref|>text<|/ref|><|det|>[[117, 230, 881, 280]]<|/det|>
+We have now explicitly noted this limitation (limited statistical power) in the discussion section (page 19, line 8) and specified in the methods section (page 33 line 6) and Figure 3 caption that uncorrected p- values are displayed for the correlation analyses.  
+
+<|ref|>text<|/ref|><|det|>[[117, 295, 881, 362]]<|/det|>
+Comment 2c: "While the viewer is helpful, for some example genes, could the authors directly show the scatterplots demonstrating the correlations between gene expression and plaque age in the main text? Regressions with the corresponding coefficients might also be helpful."  
+
+<|ref|>text<|/ref|><|det|>[[118, 377, 881, 428]]<|/det|>
+We concur and include representative scatterplots for different synaptic (Nptx2, Dlg4) and DAA/DAM genes (Anxa1, Axl, Csf1, Ctsd) in Figure 3 F,G,I,J,L,M,O,P and Supplementary Fig. 7.  
+
+<|ref|>text<|/ref|><|det|>[[115, 459, 881, 576]]<|/det|>
+Comment 2d: "Could the authors provide some commentary on the challenges of hyperspectral imaging prior to GeoMx on the same tissue slice? As they mention, serial sections preclude the analysis of nascent small plaques, which would certainly be biologically fascinating. Do the authors see this as a major technical limitation? There are groups combining multimodal imaging and GeoMx on the same sections as well, so I am interested in hearing some perspective on this – though an experimental demonstration would be most impressive."  
+
+<|ref|>text<|/ref|><|det|>[[115, 591, 881, 689]]<|/det|>
+As rightfully acknowledged by the reviewer, the LCO microscopy provides the possibility to serve as surrogate marker for plaque maturation/age and provides additional biophysical insight on plaque constitution. Consequently, it would be desirable to perform LCO and GeoMx within the same tissue rather than or in addition to, using a correlative approach using sequential sections. The same applies for combining MALDI and GeoMx.  
+
+<|ref|>text<|/ref|><|det|>[[115, 689, 881, 771]]<|/det|>
+However, LCO- based hyperspectral microscopy requires the use of mounting media and cover slips to afford an accurate readout of the spectral data and linear unmixing, respectively. Therefore, these experiments cannot be performed prior to the GeoMx (or MALDI) experiments on the same tissue section. (please see our response to your comment 1c)  
+
+<|ref|>text<|/ref|><|det|>[[115, 772, 883, 903]]<|/det|>
+We naturally investigated whether LCO staining could be performed after MALDI or GeoMx as well as whether MALDI and GeoMx could be performed on the same tissue. For MALDI, the combination with LCO microscopy and/or GeoMx is not possible as the sample prep and MSI peptide experiments leads to significant tissue distortion. (Of note, this is specific for amyloid peptide MSI. MALDI or DESI MSI of lipids could theoretically be interfaced with direct LCO imaging or GeoMx, as those MSI methods show no tissue distortion, please see Kaya et al 2017 doi:10.1021/acs.analchem.7b00313/PMID: 28318232)
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 81, 884, 265]]<|/det|>
+Regarding LCO analysis after GeoMx: this approach while feasible has as obvious limitations with respect to AOI selection as the LCO data should preferably guide AOI selection prior to the GeoMx experiment and not retrospectively. Further, as mentioned under Comment 1a, the GeoMx approach is limited with respect to emission wavelengths used for fluorescent imaging as the UV range (<420nm) cannot be used due to photocleavable tags used that absorb in that range. In addition, the LCO dyes emit rather broadly between 450- 620nm, which leads to interference with morphology markers used in the GeoMx experiment that emit in this range. These challenges are now discussed in the limitation section at the end of the discussion. (page 19 line 8)  
+
+<|ref|>text<|/ref|><|det|>[[115, 277, 880, 311]]<|/det|>
+Comment 3: "There are some ways the data could be better organized and bookkept, for transparency:  
+
+<|ref|>text<|/ref|><|det|>[[115, 327, 880, 361]]<|/det|>
+a. Can the authors enumerate the number of plaques and their age distribution captured by MSI within and across animals?  
+
+<|ref|>text<|/ref|><|det|>[[117, 375, 881, 443]]<|/det|>
+The number of animals was \(n = 3\) (10mo) and \(n = 4\) (18mo). The number of plaques per animal was \(N = 5 - 6\) . We provide this information in the Methods (Animal experimental design) as well as summarize all plaque ROI (MALDI/LCO) and AOI (MALDI/GeoMx) data in a new Supplementary Table 1-5.  
+
+<|ref|>text<|/ref|><|det|>[[115, 457, 880, 508]]<|/det|>
+b. Figure S1 is insufficiently described to let me understand the structure of the data. A more thorough caption and explanation would be helpful, including a breakdown by source animal rather than just age.  
+
+<|ref|>text<|/ref|><|det|>[[115, 522, 881, 557]]<|/det|>
+We concur and updated the Supplementary Tables (now Suppl. Tab 1- 5) accordingly as detailed in Comment 3a.  
+
+<|ref|>text<|/ref|><|det|>[[115, 572, 881, 606]]<|/det|>
+c. Could the authors clarify what is shown in Fig. S2b? I am not sure how to interpret these values. It might be more useful to show PCAs instead.  
+
+<|ref|>text<|/ref|><|det|>[[115, 621, 881, 655]]<|/det|>
+We appreciate this comment. The Figure aims to illustrate the advantage of spatial transcriptomics on the single plaque level vs bulk tissue RNA seq.  
+
+<|ref|>text<|/ref|><|det|>[[115, 655, 881, 755]]<|/det|>
+Here, the plots aim to illustrate a measure of quantification for synaptic and DAA genes that are significantly changed in response to plaque pathology. The y- axis thereby represents the first principal component of the synaptic and DAA genes, respectively, and serves as a representative abundance value. The results show that spatial analysis allows to delineate plaque specific decrease in synaptic genes and increase in DAA genes, something which is otherwise convoluted in bulk tissue analysis.  
+
+<|ref|>text<|/ref|><|det|>[[115, 754, 881, 788]]<|/det|>
+We updated the legend (now Supplementary Figure 10) accordingly for improved clarity.  
+
+<|ref|>text<|/ref|><|det|>[[115, 802, 881, 869]]<|/det|>
+d. Could we have a complete description of all GeoMx FOVs imaged, and their corresponding 15N/14N ratios, morphologies and hyperspectral ratios, anatomical localization, number of unique genes and transcripts sequenced, etc? It would also be important to know the source animals for each FOV.  
+
+<|ref|>text<|/ref|><|det|>[[115, 869, 881, 901]]<|/det|>
+This could be in the form of a table or heatmap. It will give a better idea of how powered the conclusions are."
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 99, 881, 133]]<|/det|>
+We agree with this suggestion and curate these data in comprehensive Supplementary Fig. 5- 6 and Supplementary Tables 3- 5.  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 148, 310, 166]]<|/det|>
+## Minor Comments:  
+
+<|ref|>text<|/ref|><|det|>[[116, 167, 881, 250]]<|/det|>
+Minor Comments:Minor 1: "It might be helpful for the authors to comment further on the applicability of their method in conjunction with higher- resolution spatial transcriptomics methods (e.g., Visium HD, MERFISH, Xenium, etc.) or proteomics (CODEX, IMC, etc.) in future work. Integrating iSILK with the growing landscape of these tools is certainly of interest to many spatial biologists."  
+
+<|ref|>text<|/ref|><|det|>[[116, 265, 881, 333]]<|/det|>
+We agree this important point. Unfortunately, the MALDI sample preparation and acquisition impacts tissue morphology and prevents a direct interfacing with other spatial techniques on the same tissue. Similar to the approach used here iSILK can be used on sequential tissues in concert with those techniques.  
+
+<|ref|>text<|/ref|><|det|>[[116, 333, 881, 413]]<|/det|>
+this is certainly of interest as those other spatial transcriptomics/proteomics methods would give a better resolution of transcripts and proteins towards plaques. As discussed we were faced with the tradeoff between sensitivity and spatial resolution/single cell specificity but agree that follow up studies expanding towards these techniques would be very valuable.  
+
+<|ref|>text<|/ref|><|det|>[[116, 412, 881, 445]]<|/det|>
+We discuss the compatibility of iSILK with spatial biology techniques in the limitations section.  
+
+<|ref|>text<|/ref|><|det|>[[116, 464, 881, 498]]<|/det|>
+Minor 2: "Some general commentary on the limitations of their approach might be helpful."  
+
+<|ref|>text<|/ref|><|det|>[[118, 513, 771, 531]]<|/det|>
+We included Limitations comment at the end of the Discussion (page 19).  
+
+<|ref|>text<|/ref|><|det|>[[116, 546, 881, 579]]<|/det|>
+Minor 3: "Fig. 1a scale bar value is not provided. The imaging modality should also be stated."  
+
+<|ref|>text<|/ref|><|det|>[[118, 596, 437, 612]]<|/det|>
+We provide this value in the legend.  
+
+<|ref|>text<|/ref|><|det|>[[116, 628, 881, 678]]<|/det|>
+Minor 4: "As a matter of preference, it might be better for authors to use perceptually uniform colormaps instead of 'spectral' (e.g., in Fig. 1f). It may be hard for some readers to see."  
+
+<|ref|>text<|/ref|><|det|>[[116, 694, 881, 727]]<|/det|>
+We attempted to use a uniform color map instead of spectral coloring, as suggested by the reviewer.  
+
+<|ref|>text<|/ref|><|det|>[[116, 727, 881, 777]]<|/det|>
+A side- by- side comparison of both color maps (see below) demonstrates that subtle changes in \(^{14}\mathrm{N} / ^{15}\mathrm{N}\) enrichment are more readily discernible to the human eye employing a spectral color map.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[115, 80, 570, 360]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[115, 378, 883, 413]]<|/det|>
+For this reason, we have opted to retain our original color scheme, as it better highlights the features of interest in the data.  
+
+<|ref|>text<|/ref|><|det|>[[115, 427, 884, 446]]<|/det|>
+Minor 5: "Legends in some panels (e.g., Fig. 1e, 2c) are rather small and hard to see."  
+
+<|ref|>text<|/ref|><|det|>[[118, 460, 587, 479]]<|/det|>
+We increased the font size for the respective panels.  
+
+<|ref|>text<|/ref|><|det|>[[115, 493, 770, 512]]<|/det|>
+Minor 6: "Gene and GO annotations for Fig. 3c- h are also hard to read."  
+
+<|ref|>text<|/ref|><|det|>[[115, 526, 884, 562]]<|/det|>
+We increased the font size for the respective panels and rearranged the text to better accommodate the figure dimensions.  
+
+<|ref|>text<|/ref|><|det|>[[115, 575, 883, 595]]<|/det|>
+Minor 7: "Could the MSI for Fig. 3b be decomposed into 15N and 14N for clarity?"  
+
+<|ref|>text<|/ref|><|det|>[[115, 608, 884, 644]]<|/det|>
+We provide decomposed images in Fig 3b (as overlay) and in the SI (Supplementary Fig. 3D)  
+
+<|ref|>text<|/ref|><|det|>[[115, 673, 883, 708]]<|/det|>
+Minor 8: "In 'spanning across two consecutive 12mm sections' did the authors mean 12 micron?"  
+
+<|ref|>text<|/ref|><|det|>[[115, 722, 883, 758]]<|/det|>
+The sections are 12 μm in thickness as rightfully pointed out by the reviewer. This has now been amended accordingly.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 83, 839, 139]]<|/det|>
+Please find attached with this final resubmission our revised manuscript "Isotope Encoded Spatial Biology Identifies Amyloid Plaque-Age-Dependent Structural Maturation and Synaptic Loss" by Woods, Dulewicz et al.  
+
+<|ref|>text<|/ref|><|det|>[[118, 159, 855, 234]]<|/det|>
+In this revision, we have addressed the remaining comments, including performing minor additional analyses (WCGNA) and making the requested textual and figure modifications. The corresponding changes are detailed in the point- by- point responses below.  
+
+<|ref|>text<|/ref|><|det|>[[118, 255, 283, 290]]<|/det|>
+With best regards, Jörg Hanrieder  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 312, 244, 329]]<|/det|>
+## Reviewer #1:  
+
+<|ref|>text<|/ref|><|det|>[[118, 349, 870, 423]]<|/det|>
+I have now reviewed the revised manuscript, "Isotope Encoded Spatial Biology Identifies Amyloid Plaque- Age- Dependent Structural Maturation, Synaptic Loss, and Increased Toxicity," along with the authors' point- by- point rebuttal to the concerns I raised in my initial review.  
+
+<|ref|>text<|/ref|><|det|>[[118, 444, 876, 538]]<|/det|>
+The authors have addressed all of my previous concerns by conducting significant new experiments to validate their findings and strengthen their claims. The new high- sensitivity LC- MS/MS analysis to investigate post- translational modifications and the inclusion of MS/MS spectra for sequence confirmation have fully resolved my initial questions regarding the data.  
+
+<|ref|>text<|/ref|><|det|>[[118, 558, 872, 671]]<|/det|>
+As a final point of scientific interest, I would like to offer one minor point for the author's consideration. In the MS/MS spectrum for Aβ 1- 42 shown in Supplementary Figure 1F, it is interesting to note the dominance of the b- ion series, while the corresponding y- ion series is almost absent. A brief comment on this observation in the Results section or the caption of the figure could enhance the spectral interpretation of the fragmentation behavior.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 693, 222, 708]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[118, 710, 875, 822]]<|/det|>
+We thank the reviewer for raising this point and added the following statement to the figure description of Supplementary Figure 1: "The b- ion series (charge retained on N- terminal fragment upon peptide fragmentation) dominates in several Aβ isoforms because the N- terminal region contains multiple charge carriers (e.g., Arg5, Lys16). In contrast, the C- terminal region of Aβ peptides largely lacks such charge carriers, hence, y- ions are less abundant, and fragment spectra are dominated by the b- ions."  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 843, 243, 860]]<|/det|>
+## Reviewer #2:  
+
+<|ref|>text<|/ref|><|det|>[[118, 880, 875, 898]]<|/det|>
+The authors have made appropriate clarifications of all my previously raised points in
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 83, 866, 120]]<|/det|>
+the first round of detailed review and the paper has been improved. Congratulations to very interesting and important work. I support publication as soon as possible.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 141, 222, 157]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[118, 160, 864, 214]]<|/det|>
+We sincerely thank the reviewer for their thorough evaluation of our manuscript and their positive feedback. We are pleased that the revisions addressed the previously raised points.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 84, 243, 101]]<|/det|>
+## Reviewer #3:  
+
+<|ref|>text<|/ref|><|det|>[[118, 120, 863, 234]]<|/det|>
+The presentation of the data in the revised manuscript makes it much more straightforward to read and interpret the data in my opinion. Here are some remaining minor comments for the authors, though they are mostly either presentation issues or nominal analyses that might provide some further insights if the authors choose to perform them. Nevertheless, I believe they have satisfied my major concerns, and I am supportive of publication.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 255, 274, 271]]<|/det|>
+## Minor Comments  
+
+<|ref|>text<|/ref|><|det|>[[118, 273, 867, 329]]<|/det|>
+1. In their discussion of limitations of other high-res ST approaches, they might note that the CosMx 18k now offers near-transcriptome coverage – though at markedly lower gene-wise sensitivity  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 350, 222, 366]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[118, 368, 876, 443]]<|/det|>
+We thank the reviewer for this helpful suggestion. We have revised the manuscript to acknowledge that CosMx now provides near whole-transcriptome coverage and have added a note on the associated limitation of reduced gene-wise sensitivity. Please see p20line10- 14  
+
+<|ref|>text<|/ref|><|det|>[[118, 463, 872, 520]]<|/det|>
+2. I would appreciate if the authors could do one additional analysis – to highlight novel gene-gene correlation modules that arise across from analysis across plaques (e.g. with WGCNA or similar).  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 541, 222, 556]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[115, 558, 870, 800]]<|/det|>
+We thank the reviewer for this suggestion and performed WGCNA to explore gene- gene correlation networks separately in 10- and 18- month- old mice (see methods section 'Weighted Gene Co- expression Network Analysis (WGCNA)'). For ease of computation and interpretation, we limited the genes included in network construction to those that significantly correlated with our trait of interest, i.e. plaque age. While the specific filtering criterion was tailored to our study, the use of pre- filtering approaches prior to WGCNA is widely used and described in the literature (Deng et al. 2015, Lin et al. 2018, Zuo et al. 2018). We have deposited our results (gene cluster membership and GO analysis of each cluster) to the Shiny app, where users can explore genes of interest in a detailed manner. Please see: https://maciejdulewiczgu.shinyapps.io/MALDI_GEOMX_VOLCANO/  
+
+<|ref|>text<|/ref|><|det|>[[118, 822, 857, 898]]<|/det|>
+We also included the WGCNA results in a new Supplementary Figure (SI Fig 8), which illustrates the correlation between the identified clusters and plaque age, highlighting selected clusters of interest along with their associated GO terms. The corresponding results are described in the manuscript. In brief, we observed that
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 84, 866, 159]]<|/det|>
+gene- gene correlation clusters relating to (glutamatergic) synaptic processes correlated negatively with plaque age, in line with our previously reported results. Additionally, we found clusters enriched in glial proteins (e.g. H2- k1) to be positively correlated with plaque age.  
+
+<|ref|>text<|/ref|><|det|>[[118, 160, 553, 177]]<|/det|>
+Please see Results: p11line18 and Methods p35  
+
+<|ref|>text<|/ref|><|det|>[[118, 198, 868, 272]]<|/det|>
+3. The authors might comment on whether their observed gene expression changes signify some T cell infiltration (or any other evidence for this in their animals) versus microglial activation due to overlap in GO annotations. This was what I meant by cryptic expression in my previous review and I apologize for the misunderstanding.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 294, 222, 309]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[117, 312, 875, 443]]<|/det|>
+We thank the reviewer for this clarification and have expanded the Discussion to address the potential contribution of T cell infiltration. We now note that although T cell- associated ontologies were significantly enriched, the driving genes are also broadly expressed in microglia, and canonical T cell markers (CD3, CD4, CD8) were not significantly changed. Therefore, in this dataset, potential T cell infiltration cannot be reliably distinguished from microglial activation. We added a comment in the discussion. p18line11  
+
+<|ref|>text<|/ref|><|det|>[[118, 464, 844, 500]]<|/det|>
+4. The statement that APP NL-F is strictly better than NL-G-F as it reflects human pathology (Line 387-389) is too strongly made.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 522, 222, 537]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[118, 540, 830, 594]]<|/det|>
+We agree with the reviewer that the original statement was too strong and have revised the text to avoid implying that the APP NL-F model is strictly better or overstating its relevance to human pathology over the NL-G-F model.  
+
+<|ref|>text<|/ref|><|det|>[[118, 615, 444, 632]]<|/det|>
+5. Scale bars missing in Fig. 3A, 4A.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 655, 222, 670]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[118, 673, 866, 708]]<|/det|>
+We thank the reviewer for noting this omission. Scale bars have now been added to both Fig. 3A and Fig. 4A.  
+
+<|ref|>text<|/ref|><|det|>[[115, 729, 828, 747]]<|/det|>
+6. Labels for GFAP, Ab, and SYTO colors are missing in Fig. 3B, Supp Fig. 5, 6.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 769, 222, 784]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[118, 787, 869, 840]]<|/det|>
+We thank the reviewer for pointing this out. Labels for GFAP, Ab, and SYTO channel colors have now been added to Fig. 3B, Supplementary Fig. 5, and Supplementary Fig. 6.  
+
+<|ref|>text<|/ref|><|det|>[[118, 862, 810, 898]]<|/det|>
+7. I am not sure why the data must be duplicated as Supp Fig. 3D – would be preferable to instead include these panels in Fig. 3B?
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 104, 223, 120]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[118, 122, 872, 178]]<|/det|>
+Response:We agree with the reviewer that duplication was unnecessary. We have re- arranged Fig. 3B to incorporate the relevant panels and have removed Supplementary Fig. 3D to avoid redundancy.  
+
+<|ref|>text<|/ref|><|det|>[[118, 197, 579, 216]]<|/det|>
+8. The bottom axis labels for Fig. 3C, D are clipped.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 237, 223, 253]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[118, 255, 879, 311]]<|/det|>
+Response:We thank the reviewer for noting this. The bottom axis labels for Fig. 3C and 3D have been corrected to ensure they are fully visible in the revised version of the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[118, 331, 247, 349]]<|/det|>
+Typographical  
+
+<|ref|>text<|/ref|><|det|>[[118, 350, 674, 387]]<|/det|>
+Typographical1. Line 190 "oligothiophene (LCO)" > "oligothiophenes (LCOs)2. Line 429 "information" > "information"  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 406, 223, 422]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[118, 424, 831, 461]]<|/det|>
+Response:We thank the reviewer for spotting these errors. Both typographical issues have been corrected in the revised manuscript.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Transparent Peer Review file__ea7b89490d5070a7f709cdfa17261c6cb6597b8fd9fd5e9522fa48388126481f/images_list.json

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

peer_reviews/supplementary_0_Transparent Peer Review file__ea7b89490d5070a7f709cdfa17261c6cb6597b8fd9fd5e9522fa48388126481f/supplementary_0_Transparent Peer Review file__ea7b89490d5070a7f709cdfa17261c6cb6597b8fd9fd5e9522fa48388126481f.mmd

@@ -0,0 +1,834 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+# Emergence and Global Spread of a Dominant Multidrug-Resistant Clade in Acinetobacter baumannii  
+
+Corresponding Author: Professor Zhemin Zhou  
+
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+Version 0:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+Summary  
+
+The authors sequence 100 genomes from Suzhou and Wenzhou and add 15,643 publicly available genomes. They create one of the most extensive data sets for A. baumannii to study transmission of the International Clones. Although I see the manuscript has some interesting analyses, the results are not properly framed in terms of much of the knowledge about the genomic epidemiology of A. baumannii generated in the last few years (see major comment 1). More importantly, some results are not really novel (see major comment 2). Finally, there are some methodological issues (see major comment 3) that need to be sorted out. The authors want to work on a revised version taking into account the minor but especially the major comments.  
+
+## Major comments  
+
+1. Some parts of the manuscript are out of proper context, in terms of the molecular epidemiology of A. baumannii. For instance, what is stated about the poor characterization of the population structure and temporal dynamic of the ICs is not accurate. Many studies have characterized the genomic epidemiology of ICs; see this recent comment about it (https://journals.asm.org/doi/10.1128/mbio.02520-23). Of note, currently, the two most "relevant" ICs are IC2 and IC5; not IC2 and IC1. Also, see the last point in minor comments. Another example is the resistance analysis, the authors conducted an in silico analysis of the resistance but they failed to put their findings in the context of previous studies (most of them coming out just 2 or 3 years ago). The authors can use the search query "resistome[TI] AND baumannii[TI]" on PubMed to see some of them.  
+
+2. Unfortunately, some of the findings are not strictly novel. One of the main findings of the manuscript, namely that ESL (IC2 + IC1) represents \(70\%\) of the genomes, is almost identical to the finding of the 2019 Micro Gen paper by Nigro and Hamidian (Ref 19 in the authors' manuscript). Another important finding (as stated by the authors), the relevance of recombination as a major driver in introducing genetic diversity, was made before for IC1 in a 2016 Micro Gen paper (Ref 7 in the authors' manuscript) and a PNAS paper from 2011 (ref below).  
+
+https://pubmed.ncbi.nlm.nih.gov/21825119/  
+
+3. There are important methodological issues that need to be addressed. First, are you sure there are 2,266 core genes? This number of core genes seems very high to me. Previous studies using much fewer genomes (just hundreds) and less diverse in terms of ICs have found a core genome much smaller (less than 1000 core genes). Are these strict core genes or softcore genes? Secondly, the temporal signal is really, really weak \(r2 < 0.1\) ! — given this, I don't think the tMRCA estimate is reliable at all. This can also be appreciated in the estimated substitution rate, which is very low. Previous estimates for this species have found estimates in the order of NxE-6.  
+
+4. Although I agree that A. baumannii should be considered under the One Health approach (see refs below about it), the authors did not present any data or analysis to this end.  
+
+https://pubmed.ncbi.nlm.nih.gov/35544213/
+
+<--- Page Split --->
+
+https://pubmed.ncbi.nlm.nih.gov/36150399/  
+
+5. I do not see a natural connection between the global database and the genomes sequenced from East China. The authors want to put more effort into connecting those two sections or maybe the authors want to write another paper with the genomes from East China.  
+
+6. Regarding the newly sequenced genomes from East China, the following terms were not found in BioProject: PRJCA016507, CNP0005359.  
+
+Minor comments  
+
+Please mention the coverage and identity per cent in predicting the antibiotic and virulence genes.  
+
+Supplementary Table 1, please add the name of the strains next to the RefSeq assembly identifier  
+
+Line 52: "multiple drugs" reads better than "multi- drugs"  
+
+Line 100: should be "its single- locus variants"  
+
+Lines 110- 119: how does this compare to previous studies on the resistome of A. baumannii  
+
+Line 286: These days there are at least 11 ICs. IC10 and IC11 have been recently described, below are the studies describing them.  
+
+IC10: https://www.biorxiv.org/content/10.1101/2023.10.09.561570v1  
+
+IC11: https://pubmed.ncbi.nlm.nih.gov/37244424/  
+
+(Remarks on code availability)  
+
+## Reviewer #2  
+
+(Remarks to the Author)  
+
+This paper performs a large genomics analysis to evaluate the emergence of pandemic lineages of A. baumannii. Overall, the paper describes the acquisition of ARGs and virulence genes that could help explain the dominance of the 2.5.6 "variant". The recombination analysis is interesting and the location of recombination events could reveal information on the continuing evolution of this pathogen.  
+
+There were a number of inconsistencies in the manuscript that I try to point out below. The authors talk about limiting analyses to only portions of the genome or by excluding recombination, without a thorough justification. Information on how timing analyses can be performed on a highly recombinant organism are not clear and need to be justified.  
+
+It was also not clear to me that the data from the 100 genomes have been deposited in sequence repositories. Another broad comment is that software versions need to be added throughout and the Capy code should be improved to make it useful for researchers; no readme is available and the code is not annotated in a way that makes it easy to follow. There were a number of typos and incorrect links to files and tables that should have been corrected in previous drafts. A list of questions and comments that will improve the manuscript include:  
+
+L1: Variant is an odd language choice. Why no lineage or sub- lineage? L84: How was the SDI calculated? I don't see those methods in the manuscript L90: Could this not also be due to sampling bias? L93: Where is the accession information for the 100 sequenced genomes? L98: Not sure what "proxies to natural populations" means. Please clarify. L99: CCs "each" corresponded L100: "and its single- locus"? Do you mean and its single- locus variants? L102: How do you "build" a subset. I would rewrite for clarity L109: Is "super- lineage" a term that is commonly used? I would just say that it constitutes a monophyletic lineage L111: I would reword. You didn't detect "resistance", you detected "predicted resistance" L114: When I sort the column for colistin in SD1, I see five entries with a "1". I see no "1"s in the column for tigecycline. L117: antimicrobial "usage" L137: high "levels" L138: "the" recombination rate L147: Seems like cherry picking to avoid portions of the genome that show homoplasy. The phylogenetic history of A. baumannii is also shaped by recombination, correct? How would the results change by including the entire core genome? L149: Why are only 21 shown? Why would you use "variant" and not "lineage" or another phylogenetic term? L151: Clade 2.2 is polyphyletic. Why would this not be 2 different clades? L164: My impression is that the r2 value makes this analysis inappropriate
+
+<--- Page Split --->
+
+L171: resulting "in" the formation L192: "gyrA" L194: macrolide"s" L200: B-lactam"s" L232: do you mean enhanced global dissemination "of" highly recombining IC/CCs? L244: How do we identify the ESL in Figure 4a? L253: Fig. 4c? L268: Supplementary Table 3 has information for blaADC alleles and not metagenomes L269: Where are the methods and citation for Metaplan4? L321: What does "SNP- based barcoding" mean? I've never heard that term and unclear of how it differs from other SNP- based approaches L393: All genomes were "annotated" L395: How did you handle mismatches to established alleles? L411: Where are these genomes listed? L413: Where is the accession information for the 100 genomes sequenced in this study? L423: All "sequence" data. I don't see sequence data in Table S2. Do you mean Table S3? L432: Table with the identities of the 2,266 core genes? L436: How was missing data handled? If \(>95\%\) presence was used, there must be missing genes in some genomes L446: What was the size of the effective core genome? That seems like a small number of SNPs considering the diversity of A. baumannii L454: What's the justification for removing SNPs "imported by recombination" L460: Citation and how you tried to use hierBAPS needed L512: Usage information needs to be added to the Github repository L515: minimap2 does not call SNPs, correct?  
+
+Figure 1: axis labels should be added to panels c and d Figure 1b: these pie charts are too small to see, even when zoomed in Table S4 claims to have used "abracadabra", but this method is not referenced in the text  
+
+(Remarks on code availability) The code is not very useful in its current state. Improvements could include a README, a walk through, and annotated code. A field in the parser reads: "usage="CHIPICHIPI CHAPACHAPA DUBIDUBI DABADABA"). Not sure what to make of that.  
+
+Version 1:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+First, I want to congratulate the authors for all the effort to answer my comments. They have done a good job. However, there are still a couple of things they can do to answer my major comments 3 (the mol dating section) and 4 slightly better.  
+
+Considering the mol dating issue, the date- randomization test was a good step forward but I guess to fully get around this, and given that the dating of IC2 is an important part of the manuscript, I would suggest the authors employ alternative strategies to TempEst and BEAST2 to conduct the molecular dating analysis. So, that the authors can see if using these alternative approaches the results of the dating of IC2 are robust. There are optimization methods based on maximum likelihood — for instance, node dating, treader and TreeTime — or Bayesian inference (BactDating) that can be used to get independent estimates of the tMRCA for IC2. These methods can deal with larger datasets because they assume that the tree has previously been produced.  
+
+As for the One Health issue (major comment 4), the authors have conducted new analyses to address this issue but I did not find any panel "d" in Supplementary Figure 5 — a plot that, given the rebuttal letter, should show ARG carriages in A. baumannii from non- clinical environments. Maybe you forgot to upload the most recent version of Supplementary Figure 5.  
+
+Finally, I concur with the authors that the definition of ICs in A. baumannii is rather arbitrary and ambiguous. They are right in that the use of clones in A. baumannii is very peculiar — and does not abide by the common definition of clone — yet would suggest they still tie up their epidemic lineages to the ICs (or rather to the Pasteur MLST scheme) just because potential legacy issues.  
+
+(Remarks on code availability)  
+
+Reviewer #2  
+
+(Remarks to the Author)
+
+<--- Page Split --->
+
+The authors have made significant improvements to this manuscript compared to the original version. There are still 2 issues that remain that I think need further attention. One is the filtering of recombination in the tree. In their rebuttal, the authors argue that this method has been used in "almost all" genomic analyses in the last 2 decades. This is false and studies that perform these analyses are typically not justified in doing so. The authors themselves describe how recombination shapes emerging lineages of A. baumannii. A phylogeny is a model, typically based on statistics (maximum likelihood) that fits the data to a simple framework. Based on my reading, the authors remove \(87\%\) of the data to fit their data to this model, which will obviously change the results.  
+
+The second major issue is the use of the timed tree. The authors argue that a temporal signal is justified based on the datarandomisation test. However, the BEAST2 documentation also states that: "One limitation of the date- randomisation test is that in some circumstances the test can fail to reject data sets with no temporal structure". Although it is difficult to determine if your data represents such a circumstance, the uncertainty of the test, based on the very low linear regression correlation, suggests that the timed tree analysis is speculative at best and should be framed in that context.  
+
+Some other comments, including more detailed comments that address these two concerns include:  
+
+L1: clade "of"? clade "within"? L76: I would indicate early that this is a term that you have created and what it means L118: This supertree will likely contain recombination. How does this agree then with your SNP based trees that remove recombination? L188: Your justification for using this test was on the data- randomisation test, correct? Why isn't this listed here? L226: If recombination is the driving force for evolution, why would you filter it out of your SNP- based phylogenies? L461: MLST "profiles were" L470: "USEARCH" is typically capitalized L491: How is the reader supposed to identify your data in Table S1? Maybe list "this study" somewhere to guide the reader to new data generated in this study L499: How is the reader supposed to know what DTy is? the authors should describe what this does. L501: What is the justification for using \(>94\%\) here and \(>95\) of genomes? L510: "de- recombination", as far as I know, is not a phrase that is used. L522: Did the authors verify that all recombinant SNPs had been removed? L525: So the authors removed 158,208 recombinant SNPs? By doing this, the authors have removed \(87\%\) of the diversity in the dataset. This seems neither justified nor warranted L540: Yet the BEAST documentation mentions that this test can fail to reject datasets with no temporal structure. This seems to be a major flaw in reporting this, especially based on the poor R2 correlation L586: based on "a" SNP  
+
+(Remarks on code availability)  
+
+Version 2:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author) The authors have addressed my comments. Congratulations to them on all their hard work.  
+
+(Remarks on code availability)  
+
+Reviewer #3  
+
+(Remarks to the Author)  
+
+The manuscript entitled "Emergence and Global Spread of a Dominant Multidrug- Resistant Clade in Acinetobacter baumannii" reports the analysis of a Clade of A. baumannii.  
+
+Authors have addressed all comments raised by reviewers however I think are still a few issues to be clarified.  
+
+My general and specific comments include:  
+
+- Apart from IC1 and IC2 (GC1 or GC2, or WW1 and WW2), other ICs/GCs/WWs are less common and less known to most readers, please use the sequence type (ST) equivalent so readers can follow the text easier.  
+
+- Line 251-269 under the Stepwise acquisition of ARGs in the ESL section: could you please comment on the fact that many known ARGs, including oxa23, in Ab are either in chromosomal genomic islands (acquired by an ancestral strain with evidence published on different STs elsewhere) and acquired via plasmids? Tn2006, AbaR4 and variants of Tn6167 in ST2 are the classic examples.
+
+<--- Page Split --->
+
+- Line 263: First reference 12 is an incorrect reference and not suitable for this statement. In fact, this information is also incorrect and if it has been mentioned in ref 12, this reference needs an Erratum. blaADC (ampC) is an intrinsic gene present in all A. baumannii strains and can barely cause any detectable resistance phenotype unless there is an ISAba1 copy present upstream of this gene (which gives it a strong promote that enhances the expression) making it resistant to 3rd-generation cephalosporins. Please fix this, where mentioned in the text and include an appropriate reference.  
+
+- Big parts of this study examines the evolution of resistance while it is now known that plasmids play a crucial role in AMR trafficking in A. baumannii. To me it's a bit strange that there is no mention of one of the most important elements that drives resistance in this paper.  
+
+(Remarks on code availability) N/A  
+
+Open Access This Peer Review 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 cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source. The images or other third party material in this Peer Review 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 https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+
+## Responses to Reviewers  
+
+Reviewer #1 (Remarks to the Author):  
+
+## Summary  
+
+The authors sequence 100 genomes from Suzhou and Wenzhou and add 15,643 publicly available genomes. They create one of the most extensive data sets for A. baumannii to study transmission of the International Clones. Although I see the manuscript has some interesting analyses, the results are not properly framed in terms of much of the knowledge about the genomic epidemiology of A. baumannii generated in the last few years (see major comment 1). More importantly, some results are not really novel (see major comment 2). Finally, there are some methodological issues (see major comment 3) that need to be sorted out. The authors want to work on a revised version taking into account the minor but especially the major comments.  
+
+Many thanks for your helpful comments. We have made modifications accordingly and substantially rewritten the manuscript to highlight the novelty of our findings. We also added discussions about the methods that were used in the revised manuscript, and hope that will solve the problems. Please find the point- by- point responses below.  
+
+## Major comments  
+
+1. Some parts of the manuscript are out of proper context, in terms of the molecular epidemiology of A. baumannii. For instance, what is stated about the poor characterization of the population structure and temporal dynamic of the ICs is not accurate. Many studies have characterized the genomic epidemiology of ICs; see this recent comment about it (https://journals.asm.org/doi/10.1128/mbio.02520-23). Of note, currently, the two most "relevant" ICs are IC2 and IC5; not IC2 and IC1. Also, see the last point in minor comments. Another example is the resistome analysis, the authors conducted an in silico analysis of the resistome but they failed to put their findings in the context of previous studies (most of them coming out just 2 or 3 years ago). The authors can use the search query "resistome[TI] AND baumannii[TI]" on PubMed to see some of them.  
+
+A: Many thanks for pointing out many relevant studies. We acknowledged that the detection of ICs is important and now included the citations in the manuscript. Particularly, we modified the introduction to include these important studies (lines 59- 65), which reads:  
+
+'Several international clones (ICs) of A. baumannii have been identified through restriction
+
+<--- Page Split --->
+
+fragment length polymorphism (RFLP) \(^4\) and multi- locus sequence typing (MLST) \(^{5 - 7}\) . The most prominent ICs are IC2, associated with the majority of nosocomial infections internationally \(^8\) , and IC1, responsible for infection outbreaks in casualties returned from the Middle East conflicts \(^9\) . Furthermore, IC5 accounts for \(>50\%\) of infections in Latin America \(^{10,11}\) . The resistance, virulome, and epidemiology of many ICs have also been investigated \(^{12,13}\) .  
+
+However, as proposed in the discussion, we are not fully convinced by the definitions of many ICs. For example, the mBio paper invented IC9 because it was detected in four different countries. We identified at least 61 ST complexes (CCs) that have been found in four or more countries based on the global genomic dataset. These could have all been designated as new ICs. Therefore, we are not sure whether the international detection of these ICs is of clinical importance. Furthermore, they are not real "clones", because some ICs can have high levels of genetic diversities. Therefore, we proposed "epidemic lineage" in the manuscript as a replacement for "international clone". Based on the genomic data, we designed only epidemic lineages 1 (IC1) and 2 (IC2), which both exhibited extensive geographic distributions and high levels of AMRs.  
+
+Particularly, thanks for pointing out IC5 as one of the clinically relevant lineages recently. Based on our data, this IC is primarily isolated from American countries, exhibiting some sort of geographic restrictions (Supplementary Fig. 1d). Additionally, IC5 strains encoded \(\sim 13\) ARGs, substantially lower than the amount in IC1 (17 ARGs) and IC2 (19 ARGs). We have added sentences in lines 315- 317 to discuss IC5:  
+
+'For example, IC5 has been proposed as a major clone in some studies \(^{10}\) . Our genomic data showed that it was rarely isolated outside of South America and encoded relatively lower levels of ARGs.'  
+
+We also thank you for pointing out many papers on the resistance of A. baumannii. While the majority of these studies are based on fewer amount of strains (100- 1500 strains), they offered invaluable resources for both clinical and microbiological studies. As shown above, many of these invaluable citations have been added to the introduction section. Meanwhile, A. baumannii is a divergent species with many distinct lineages. We believe that it is worthwhile to give a comprehensive overview of its genetic landscape based on ten- fold more genomes. Furthermore, we focused on the prevalent epidemic super- lineage, which allowed a detailed investigation of its evolution, transmission, and ARG accumulation.  
+
+2. Unfortunately, some of the findings are not strictly novel. One of the main findings of the manuscript, namely that ESL (IC2 + IC1) represents 70% of the genomes, is almost identical to the
+
+<--- Page Split --->
+
+finding of the 2019 Micro Gen paper by Nigro and Hamidian (Ref 19 in the authors' manuscript). Another important finding (as stated by the authors), the relevance of recombination as a major driver in introducing genetic diversity, was made before for IC1 in a 2016 Micro Gen paper (Ref 7 in the authors' manuscript) and a PNAS paper from 2011 (ref below).  
+
+A: Many thanks again for pointing out many citations. As we described in the results, it is not too surprising that the predominance of IC1 and IC2 has been known, as The MLST data hosted in the pubmlst.org database showed the same phenomenon. What is new is the close genetic relationship between these two predominant ICs. According to the phylogeny, these two ICs formed one superlineage, which separated from other lineages from almost the root of the tree. This is quite important because it indicates that IC1 and IC2 might share some properties (maybe the recombination hotspots that were described) that made them successful in clinical settings. Furthermore, we described their detailed evolutionary history and global transmission, prompting an association between their emergence and the post- war use of antimicrobials.  
+
+We cited many previous studies that described the predominance of IC2 and IC1 in the discussion section (lines 321- 324), which reads:  
+
+'Given their clinical importance and global prevalence, as evidenced by various studies'7,19,29, we suggest designating the clade comprising IC1 and IC2 as the epidemic super- lineage, setting it apart from populations with more localized distributions and fewer ARGs.'  
+
+7. Holt, K. et al. Five decades of genome evolution in the globally distributed, extensively antibiotic-resistant Acinetobacter baumannii global clone 1. Microbial Genomics 2, (2016).  
+
+19. Zarrilli, R., Pournaras, S., Giannoulì, M. & Tsakris, A. Global evolution of multidrug-resistant Acinetobacter baumannii clonal lineages. Int J Antimicrob Agents 41, 11–19 (2013).  
+
+29. Khuntayaporn, P. et al. Predominance of international clone 2 multidrug-resistant Acinetobacter baumannii clinical isolates in Thailand: a nationwide study. Ann Clin Microbiol Antimicrob 20, 19 (2021).  
+
+Secondly, we also appreciated that many papers, including the two mentioned by the reviewer, have described the high level of recombination in A. baumannii. We focused more on the selective stress that resulted in such high- level recombination and discussed the genetic repository of the imported genes. For example, we reported three recombination hotspots in the ESL lineage, which were all associated with virulence and host adaptations. Such a phenomenon is few reported in bacterial pathogens. Recombination is often described as neutral genetic events, e.g., in Helicobacter pylori [https://www.pnas.org/doi/10.1073/pnas.251396098], Vibrio parahaemolyticus [https://pubmed.ncbi.nlm.nih.gov/31235840/], and Salmonella enterica [https://pubmed.ncbi.nlm.nih.gov/23637636/]. The exceptions are the extracellular polysaccharides
+
+<--- Page Split --->
+
+synthesis genes, which were reported under selection in Streptococcus pneumoniae [https://pubmed.ncbi.nlm.nih.gov/28595308/]. Some publications identified more recombination hotspots and attributed them as outcomes of flanking mobile genetic elements rather than selective events (https://www.nature.com/articles/ncomms4956). The only analysis we found that is similar is an analysis in Campylobacter jejuni, which identified a zinc uptake gene cluster as recombination hotspots. To highlight the importance of this, we have now included a new paragraph in the discussion lines 354- 365, which reads:  
+
+'Recombination has been primarily considered neutral in bacterial evolution<sup>36</sup>. Reported recombination "hotspots" in bacterial genomes were primarily driven by flanking mobile genetic elements<sup>37</sup> or genes associated with extracellular polysaccharides<sup>38</sup>. Our findings highlighted the role of selection in shaping recombination hotspots in A. baumannii, which include not only capsular gene clusters but also many other genes responsible for host/environmental adaptations (Fig. 2). Notably, HRB2 and HRB3 were both involved in genes responsible for iron uptake, which is essential for the in vivo survival of not only A. baumannii<sup>39</sup> but also many other pathogens<sup>40</sup>. Similar recombination hotspots have also been reported in Campylobacter jejuni<sup>41</sup>, which experienced frequent recombination in its zinc uptake system genes (znuABC), likely associated with adaptation to chicken caeca and other low- zinc environments.'  
+
+3. There are important methodological issues that need to be addressed. First, are you sure there are 2,266 core genes? This number of core genes seems very high to me. Previous studies using much fewer genomes (just hundreds) and less diverse in terms of ICs have found a core genome much smaller (less than 1000 core genes). Are these strict core genes or softcore genes?  
+
+A: Sorry for the insufficient description in the manuscript. First, you are right. We only identified 307 strict core genes (100% presence) in the 5,824 representative genomes. We decided to use "softcore" genes with \(\geq 95\%\) presence in the representatives because I previously found that the strict core genes kept decreasing with the number of genomes (https://genome.cshlp.org/content/30/11/1667). We described this in the method section and have now decided also modify the result section (lines 113- 115), which reads:  
+
+'We selected 5,824 genomes with pairwise average nucleotide identities (ANIs) of \(< 99.5\%\) to establish a representative set of A. baumannii and identified 2,266 soft core genes shared by at least 95% (5,533/5,824) of the representatives (Supplementary Data 2).'  
+
+Secondly, the temporal signal is really, really weak \(\mathrm{r}2< 0.1\) ! — given this, I don't think the tMRCA estimate is reliable at all. This can also be appreciated in the estimated substitution rate, which is very low. Previous estimates for this species have found estimates in the order of NxE- 6.
+
+<--- Page Split --->
+
+A: We agree that the \(r^2\) value of the root-to-tip regression is very low. We were puzzled by this in many previous analyses and consulted Simon Ho, one of the developers of the BEAST software. He pointed us to the date randomization test, which was described in [https://doi.org/10.1093/molbev/msv056] by Ho and [https://doi.org/10.1093/nar/gky783] by Xavier Didelot. It captures very low levels of temporal signal by comparing the real data to randomly permutated data. We used the date randomization test in this project and summarized its results in Supplementary Fig. 3c. For clarity, we added one sentence in the discussion which reads (lines 366- 367):  
+
+'We detected significant temporal signals in IC2 but not IC1 using date randomization tests, a sensitive approach that allows for high variations of substitution rates<sup>42</sup>.  
+
+We also modified the method part with additional descriptions in lines 536- 540, which reads:  
+
+'We evaluated the accumulation of variations in both the ESL and the two of its lineages (IC1 and IC2) using both the TempEst v1.5.3<sup>79</sup> and tested the reliability of the temporal signal by date randomisation test<sup>42</sup> which analyzed multiple date- randomised replicate datasets after randomly reassigning the isolation dates of the genomes. IC2 is the only group that had significant signals (Supplementary Fig. 4b and c).'  
+
+Additionally, thank you for pointing out previous studies on the mutation rates in A. baumannii. We acknowledged that the "Genome- scale rates of evolutionary change in bacteria" publication (https://doi.org/10.1099/mgen.0.000094) gave an estimate of \(1 - 3 \times 10^{- 6}\) substitutions per site year<sup>- 1</sup> for A. baumannii based on a few isolates collected over a short period. Also in the manuscript, they identified a strong time- dependent decrease in substitution rate in almost all bacterial species. Furthermore, the early estimates of substitution rates were based on much fewer strains with limited genetic diversities.  
+
+We successfully reproduced the substitution rates from two previous research efforts (IC1: https://doi.org/10.1099/mgen.0.000052; IC2: https://doi.org/10.1099/mgen.0.000050) using their original sets of genomes. Notably, the IC1 study included only genomes from Clade 1.4, and the IC2 study included only genomes from Clade 2.4. We found that, when all the genomes from these two clades were used, the linear associations diminished, with \(r^2\) values of nearly zero (Supplementary Fig. 8d). We believe that, with at least 100- fold more international strains included, our dating analysis is a better estimate of the evolutionary dynamics of the ESL. We have now added a new paragraph in discussion (lines 367- 376), which reads:  
+
+'Our results estimated an overall substitution rate of IC2 to \(8.16 \pm 2.19 \times 10^{- 8}\) , substantially lower than previous estimates of \(1 - 3 \times 10^{- 6}\) substitutions per site year<sup>- 1</sup> based on 24- 75 IC1 or IC2
+
+<--- Page Split --->
+
+isolates \(^{7,43}\) . We managed to reproduce the same clock rates using their original sets of strains (Supplementary Fig. 8a). However, the estimates of clock rates became lower, with greater uncertainty, when additional isolates from the same lineages were included (Supplementary Fig. 8d). Furthermore, it has been well- acknowledged that the substitution rates reduced when genomes were more divergent \(^{44}\) . Our incorporation of all known genetic diversity in the IC2 lineage likely has enabled a more accurate estimate of its emergence.  
+
+4. Although I agree that \(A\) . baumannii should be considered under the One Health approach (see refs below about it), the authors did not present any data or analysis to this end.  
+
+Thank you for the invaluable comments. We have now substantially revised the manuscript to look into the One Health concept from three aspects. First, based on public genomes, we found lower ARG carriages in \(A\) . baumannii from non- clinical environments (new Supplementary Fig. 5d). Second, based on metagenomes sampled from inpatients and those from the community in China, we found that the ESL accounted for \(>90\%\) of the isolates in the clinical setting but only rarely found in the non- clinical environments. Finally, we looked into the gene flows between the ESL and the non- epidemic lineages, and found that the latter behaves as a genetic repository, facilitating the evolution of the epidemic populations via frequent homologous recombination.  
+
+The discussion about the "One Health" concept was separated into three sections in the discussion, which read:  
+
+(lines 332- 335): 'Additionally, based on both genomic and metagenomic data, we showed that the expansion and predominance of the ESL was restricted to clinical environment, and the samples from non- clinical sources exhibited greater genetic diversities (Supplementary Fig. 1a).  
+
+\\*\\*\\*\\*  
+
+(lines 392- 400): We developed an integrated bioinformatics pipeline, Capybara, which could be applied to both genomic and metagenomic reads. Our application of this tool to both clinical and community metagenomes revealed a predominance of ESL, particularly Clade 2.5.6, in clinical settings other than communities. Substantially lower frequencies of ICs in non- clinical samples have also been found in genomic data and many other studies \(^{46,47}\) . Similar findings have also been reported in \(K\) . pneumoniae \(^{30}\) and \(S\) . aureus \(^{48}\) , questioning the efficiency of the "One Health" strategy due to the genetic separation of epidemic lineages causing most infectious diseases and the remaining environmental/animal populations.  
+
+\\*\\*\\*\\*  
+
+(lines 427- 435): Altogether, we propose a gene- centred view of the 'One Health' paradigm, where genes, rather than the bacteria themselves, are transferred between different
+
+<--- Page Split --->
+
+environments through homologous recombination or horizontal genetic transfer. This model might apply to the evolution of other MDR pathogens, emphasizing the critical role of environmental populations in preserving genetic diversity for both ARGs and virulence genes. Such genetic diversity is integral to the selective process that drives the evolution of the ESL and the emergence of successful epidemic clades. Furthermore, the ARGs and virulence genes might also spread into the non- clinical environments through a similar pathway, potentially threatening public health. '  
+
+5. I do not see a natural connection between the global database and the genomes sequenced from East China. The authors want to put more effort into connecting those two sections or maybe the authors want to write another paper with the genomes from East China.  
+
+Thank you for the suggestion. We decided to keep the Chinese data and use them as evidence for the "One Health" concept (see above). Furthermore, our Chinese genomes proved the prevalence of Clade 2.5.6 in China and were used to evaluate the performance of Capybara, the genotyping tool. We added two sentences in the result, one in lines 202- 205 that reads:  
+
+'The East China strains sequenced in this study revealed the co- circulation of three primary clades of 2.5.6 (57 strains), 2.4.13 (21), and 2.4.6 (10) in China in 2023, highlighting the sustained prevalence of 2.5.6 since its emergence (Supplementary Figs. 5a and c).'  
+
+And in lines 213- 216 which reads:  
+
+'To ease the use of the genotyping scheme, we developed a tool named Capybara (See Methods) to automatically classify strains into lineages, clusters, and clades based on short sequencing reads, and assessed its efficacy using the 100 strains sequenced in this study (Fig. 4a), confirming consistent clade assignments.'  
+
+## 6. Regarding the newly sequenced genomes from East China, the following terms were not found in BioProject: PRJCA016507, CNP0005359.  
+
+A: Apologizes for the confusion. The accession numbers PRJCA016507 and CNP0005359 corresponded to two genomic databases in China, respectively. PRJCA016507 was stored in CNCB (https://ngdc.cncb.ac.cn/bioproject/browse/PRJCA016507), and CNP0005359 was stored in CNGB (https://db.cngb.org/search/project/CNP0005359/). We have modified our data availability to match the correct source.  
+
+Minor comments
+
+<--- Page Split --->
+
+Please mention the coverage and identity percent in predicting the antibiotic and virulence genes.  
+
+A: Agreed. We revised the method part to demonstrate coverage and identity for the identification of genes, which reads (lines 462- 465):  
+
+'Antibiotic resistance genes in each genome were predicted using AMRFinderPlus v3.11.26 \(^{59}\) , and the virulence genes were predicted based on BLASTp searches against the VFDB 2022 release \(^{60}\) , with \(\geq 90\%\) identity and \(\geq 60\%\) coverage, respectively.'  
+
+Supplementary Table 1, please add the name of the strains next to the RefSeq assembly identifier  
+
+A: Modified.  
+
+Line 52: "multiple drugs" reads better than "multi- drugs"  
+
+A: Modified.  
+
+Line 100: should be "its single- locus variants"  
+
+A: Modified.  
+
+Lines 110- 119: how does this compare to previous studies on the resistome of A. baumannii  
+
+A: Our estimates of the distribution of resistance genes are largely consistent with current studies for the ICs. Moreover, we identified additional, highly resistant CCs, many of which are worth further investigation. We have added sentences in the results to reflect these new findings, which read:  
+
+'In addition to the ICs, we predicted three major CCs ( \(\geq 20\) genomes) with averagely \(\geq 15\) ARGs and five major CCs with \(\geq 80\%\) carbapenem resistance (Supplementary Data 2).'  
+
+And we discussed their importance and potential sampling bias in the discussion (lines 317- 319), which reads:  
+
+'In contrast, our analysis revealed seven non- IC populations that exhibited high levels of predicted ARGs ( \(\geq 15\) ) or carbapenem genes ( \(\geq 80\%\) ), or both, which could be due to sampling bias but worth further investigations.'
+
+<--- Page Split --->
+
+## Line 286: These days there are at least 11 ICs. IC10 and IC11 have been recently described, below are the studies describing them.  
+
+A: Thanks for reminding us of these two new ICs. We have added them to the manuscript, with modifications in Figs. 1 and 4 and Supplementary Fig. 1.  
+
+Reviewer #2 (Remarks to the Author):  
+
+This paper performs a large genomics analysis to evaluate the emergence of pandemic lineages of A. baumannii. Overall, the paper describes the acquisition of ARGs and virulence genes that could help explain the dominance of the \(2.5.6\) "variant". The recombination analysis is interesting, and the location of recombination events could reveal information on the continuing evolution of this pathogen.  
+
+There were a number of inconsistencies in the manuscript that I try to point out below. The authors talk about limiting analyses to only portions of the genome or by excluding recombination, without a thorough justification. Information on how timing analyses can be performed on a highly recombinant organism are not clear and need to be justified.  
+
+It was also not clear to me that the data from the 100 genomes have been deposited in sequence repositories. Another broad comment is that software versions need to be added throughout and the Capy code should be improved to make it useful for researchers; no readme is available and the code is not annotated in a way that makes it easy to follow. There were a number of typos and incorrect links to files and tables that should have been corrected in previous drafts. A list of questions and comments that will improve the manuscript include:  
+
+## L1: Variant is an odd language choice. Why no lineage or sub-lineage?  
+
+We used a three- level clustering scheme, similar to S. Paratyphi and Typhi SNP barcode described previously (https://www.nature.com/articles/ncomms12827). We reserved lineage for the \(1^{\text{st}}\) level because it is approximately equal to ST complexes. The \(2^{\text{nd}}\) level has now been renamed as "cluster" because many of them were polyphyletic. This leaves the \(3^{\text{rd}}\) level as "clade".
+
+<--- Page Split --->
+
+## L84: How was the SDI calculated? I don't see those methods in the manuscript.  
+
+A: We have now revised the method section (lines 600- 604) to include the following details:  
+
+## "Calculation of Simpson Diversity Index of STs  
+
+Simpson Diversity Index (SDI) of sequence types (STs) for each bacteria, was calculated as \(\mathrm{SDI} = 1 - \sum_{i = 1}^{n}p_{i}^{2}\) , where \(n\) was the total number of STs and \(p_{i}\) was the total number of detection rate of the \(i\) - th ST. The closer the SDI value is to 1, the more evenly distributed the species is in different STs, with no obvious preference."  
+
+## L90: Could this not also be due to sampling bias?  
+
+A: Thanks for the comment. This could well be a sampling bias, and that is why we redo the analysis based on genomic data, which also assigned \(\sim 70\%\) of the records to only two predominant lineages. Furthermore, IC2 strains accounted for 99 of the 100 Chinese strains that were genomic sequenced, and \(91\%\) (64/70) of the metagenomes from inpatients. While we have not done the same analyses for other ESKAPE bacteria here, we calculated the population structure of \(K\) . pneumoniae at the genomic level in a separate project (https://doi.org/10.1101/2024.04.16.24305880), which exhibited a much higher level of genetic diversity.  
+
+Even so, I would like to state that the seemly prevalence of IC2 and IC1 could be a sampling bias towards clinical samples. Only one of the 26 community samples was associated with IC2 in China. More analysis is certainly needed to investigate the natural genetic diversity of \(A\) . baumannii, especially those outside of the hospitals.  
+
+## L93: Where is the accession information for the 100 sequenced genomes?  
+
+A: Sorry for that, we now added the accession number "PRJNA1112767" in the data availability part and modified the method part describing them. All the sequenced genomes are under that BioProject.  
+
+## L98: Not sure what "proxies to natural populations" means. Please clarify.  
+
+A: Sorry for the inaccurate expression. We modified the sentence to read:  
+
+"We utilized eBURST to cluster the STs into 371 clonal complexes (CCs), including eleven that had been previously designated as international clones (ICs) \(^{7,18 - 23}\) ."
+
+<--- Page Split --->
+
+## L99: CCs "each" corresponded.  
+
+A: The sentence has been completely rewritten. See above.  
+
+L100: "and its single- locus"? Do you mean and its single- locus variants?  
+
+A: Modified.  
+
+## L102: How do you "build" a subset. I would rewrite for clarity  
+
+A: The sentence has been revised to read:  
+
+'We selected 5,824 genomes with pairwise average nucleotide identities (ANIs) of \(< 99.5\%\) to establish a representative set of \(A\) . baumannii and identified 2,266 soft core genes shared by at least \(95\%\) (5,533/5,824) of the representatives (Supplementary Data 2).'  
+
+## L109: Is "super-lineage" a term that is commonly used? I would just say that it constitutes a monophyletic lineage  
+
+A: We previously used the term "super- lineage" in Salmonella enterica (https://pubmed.ncbi.nlm.nih.gov/33614977) to describe monophyletic lineage consisting of multiple ST complexes, because the ST complexes are often described as lineages in publications. Super- lineages are super- sets of the lineages (ST complexes). We re- use the super- lineage in \(A\) . baumannii as well, to highlight the fact that this branch consists of two ICs, and leave the "lineage" name for IC1 (Lineage 1) and IC2 (lineage 2). Additional sentences were added to explain the word use (lines 120- 123):  
+
+'Remarkably, the top two ICs (IC2 and IC1) were genetically closely related, forming a monophyletic clade in the supertree. This finding reveals a larger- scale population structure above the CCs in \(A\) . baumannii, which we designated as a 'super- lineage' to differentiate it from the lineages formed by CCs.'  
+
+L111: I would reword. You didn't detect "resistance", you detected "predicted resistance"  
+
+A: Revised.  
+
+L114: When I sort the column for colistin in SD1, I see five entries with a "1". I see no "1"s in
+
+<--- Page Split --->
+
+## the column for tigecycline.  
+
+A: Sorry for the mistake. Supplementary Data 1 has been revised.  
+
+## L117: antimicrobial "usage"  
+
+A: Revised.  
+
+## L137: high "levels"  
+
+A: Revised.  
+
+## L138: "the" recombination rate  
+
+A: Revised.  
+
+L147: Seems like cherry picking to avoid portions of the genome that show homoplasy. The phylogenetic history of A. baumannii is also shaped by recombination, correct? How would the results change by including the entire core genome?  
+
+A: Agreed. We updated Supplementary Fig. 3 with a tree based on the entire core genome and those based on non- recombinant SNPs in each region. We found that The LRB retained \(66.1\%\) (23,524/35,609) of the non- recombinant SNPs in the core genome. Furthermore, the whole core genome tree and the LRB tree are very similar, with a normalized Robinson- Foulds (NRF) distance of only 0.1, compared to the NRF distance of \(>0.8\) for other regions. In practice, we found that the greater homoplastic SNPs in other regions reduced the confidence of resulting phylogeny, and decided to use only the LRB tree.  
+
+Ideally, both recombination and mutations can be incorporated in the estimation of the phylogeny. However, multiple tries found that this will lead to an extremely complicated model (https://pubmed.ncbi.nlm.nih.gov/20923983/) which is untractable in the calculation. Therefore, it has been a general practice in bacterial genomics that the SNPs imported by recombination need to be removed before phylogenetic analysis (see below for details).  
+
+We revised the manuscript in lines 164- 170, which reads:  
+
+'We found that the mutational phylogeny from the LRB was broadly consistent with the tree reconstructed from the whole core genome (Supplementary Fig. 2d), with a normalized Robinson- Foulds (NRF) distance of 0.1. Whereas trees from other regions were all different (NRF distance \(>0.8\) ). Given these findings, we utilized the mutational phylogeny from the LRB,
+
+<--- Page Split --->
+
+which encompassed \(66.1\%\) (23,524/35,609) of the mutational SNPs and exhibited the lowest level of homoplasy, as a representative of the ESL's phylogenetic history.  
+
+## L149: Why are only 21 shown? Why would you use "variant" and not "lineage" or another phylogenetic term?  
+
+A: Modified and all clades are shown in the figure now.  
+
+## L151: Clade 2.2 is polyphyletic. Why would this not be 2 different clades?  
+
+A: You are right. Actually, many of the "Clades" were not monophyletic. These "Clades" were artifacts designed to summarize the mid- term evolutionary dynamics of the bacteria. However, the word "Clade" was misleading. We checked the GenoTyphi, a similar SNP barcode scheme designed for Salmonella Typhi, and found that they named these polyphyletic levels as "Clusters", which is simply a group of similar items. We designed to use the same designation here and renamed the three levels to be "lineages" (equal to ST complexes), "clusters" (polyphyletic), and "clades" (monophyletic).  
+
+## L164: My impression is that the r2 value makes this analysis inappropriate  
+
+A: We agree that the \(\mathrm{r}^2\) value is a bit low. And that is why we used date randomization for further confirmation of the temporal signal. See response for major comments 3 from reviewer 1 for details.  
+
+## L171: resulting "in" the formation  
+
+A: Revised.  
+
+## L192: "gyrA"  
+
+A: Revised.  
+
+## L194: macrolide"s"  
+
+A: Revised.
+
+<--- Page Split --->
+
+## L200: B-lactam"s"  
+
+A: Revised.  
+
+L232: do you mean enhanced global dissemination "of" highly recombining IC/CCs?  
+
+A: Revised.  
+
+L244: How do we identify the ESL in Figure 4a?  
+
+A: Agreed. New Fig. 4a with figure legends updated.  
+
+L253: Fig. 4c?  
+
+A: Revised.  
+
+L268: Supplementary Table 3 has information for blaADC alleles and not metagenomes  
+
+A: Revised.  
+
+L269: Where are the methods and citation for Metaphlan4?  
+
+A: Revised.  
+
+L321: What does "SNP- based barcoding" mean? I've never heard that term and unclear of how it differs from other SNP- based approaches  
+
+A: SNP- based barcoding, or, SNP barcoding, refers to a typing scheme that assigns strains into lineages/clusters/clades in the phylogeny based on their shared SNPs. The first SNP barcode scheme seems to be established in Mycobacterium tuberculosis (https://www.nature.com/articles/ncomms5812), using 62 SNPs to separate the species into seven (or eight including M. bovis) lineages and 55 sublineages. The scheme was later extended to include more SNPs and sublineages (https://pubmed.ncbi.nlm.nih.gov/33317631). Similar schemes have also been established for Salmonella Typhi (https://www.nature.com/articles/ncomms12827), S.
+
+<--- Page Split --->
+
+Paratyphi A (https://www.nature.com/articles/s41467- 022- 35587- 6), and Shigella sonnei (https://www.nature.com/articles/s41467- 021- 22700- 4). These schemes have been adopted globally for genome- based surveillance of these pathogens. The SNP barcode scheme differs from standard SNP- based phylogeny by its use of only a subset of SNPs that were consistent with phylogeny, thus reducing calculation requirements and minimizing errors due to inaccurate choices of bioinformatic pipelines. The SNP barcode scheme also gives hierarchical, digital designations of the lineages and clades, facilitating population genetic analysis of the pathogens (https://elifesciences.org/articles/85867). Notably, a similar clade designation system seems to have been widely used for the influenza H5N1 virus since 2007 (https://pubmed.ncbi.nlm.nih.gov/18598616/).  
+
+## L393: All genomes were "annotated"  
+
+A: Revised.  
+
+## L395: How did you handle mismatches to established alleles?  
+
+A: We only accept sequences that were \(100\%\) identical to an existing allele. It will otherwise be considered as new alleles, and the corresponding ST will be designed as "new". We identified 386 genomes with new Pasteur STs and 2928 strains with new Oxford STs, as deposited in Supplementary Data 1.  
+
+## L411: Where are these genomes listed?  
+
+A: We have modified Supplementary Data 1 and added a column named "Representative genome" there.  
+
+## L413: Where is the accession information for the 100 genomes sequenced in this study?  
+
+A: Sorry, we now added the accession number "PRJNA1112767" in the data availability part and modified the method part describing them.  
+
+L423: All "sequence" data. I don't see sequence data in Table S2. Do you mean Table S3?
+
+<--- Page Split --->
+
+A: Revised.  
+
+## L432: Table with the identities of the 2,266 core genes?  
+
+A: We have added a new Supplementary Data with all 2,266 core genes.  
+
+## L436: How was missing data handled? If \(>95\%\) presence was used, there must be missing genes in some genomes  
+
+A: Yes, the majority of the soft- core genes were not \(100\%\) present. cgMLSA uses ASTRID and ASTRAL, two highly- cited algorithms for merging gene trees into a super- tree in the presence of missing data. First, ASTRID converted each gene tree into a distance matrix and averaged the matrices into one. Certain levels of missing data will be filled based on matrices from other genes. Then, ASTRAL was used to evaluate and update the super- tree. Internally, ASTRAL uses quartets, namely subtrees of four tips to compare different trees, which also allows a certain level of missing data.  
+
+## L446: What was the size of the effective core genome? That seems like a small number of SNPs considering the diversity of A. baumannii  
+
+A: Sorry, you are right. We mistakenly wrote the number of SNPs in the LRB regions. There were a total of 181,732 SNPs in the non- repetitive core genome of 2.35 MB (60.2% of the reference genome). We have revised the sentence in lines 511- 513 to read: 'We aligned all genomes in the ESL onto a reference genome [MDR- TJ: GCF_000187205.2; from the hospital, 2013, China72] using EToKi align, which identified a total of 181,732 SNPs in the 2.35 MB non- repetitive core genome.'  
+
+## L454: What's the justification for removing SNPs "imported by recombination"  
+
+A: Phylogeny is a diagram that depicts the evolutionary history of an organism. All phylogenetic reconstruction algorithms, including ML, MP, NJ, and Bayesian approaches, estimate evolutionary history based on the assumption that mutations occur randomly and are derived vertically to the descendants. Recombination, however, imported distantly related DNA fragments into the
+
+<--- Page Split --->
+
+chromosome, resulting in numerous sequence variations in small regions. Including these SNPs in phylogenetic analysis resulted in two problems: (1) extra- long branch lengths, and (2) inaccurate topology when the same DNA fragments were imported into unrelated strains. Thus, removing recombinational SNPs before phylogenetic analysis has been widely used in almost all genomic analyses in the least two decades (https://pubmed.ncbi.nlm.nih.gov/17151252/, https://pubmed.ncbi.nlm.nih.gov/19115014/). Many bioinformatic pipelines have also been developed and widely used (https://pubmed.ncbi.nlm.nih.gov/25414349/, https://pubmed.ncbi.nlm.nih.gov/25675341/). We have added additional sentences in the discussion to cover this topic, which reads (lines 339- 349):  
+
+'Convergent recombination between two lineages in the dataset will produce spurious similarities between them, thereby scrambling phylogenetic signal<sup>34</sup>. Meanwhile, recombination events with external, genetically less related organisms could introduce multiple polymorphisms into the population, confounding accurate measures of branch lengths<sup>35</sup>. We estimated the frequencies of externally imported regions in the core genome of the ESL using three distinct approaches of RecHMM, Gubbins, and ClonalFrameML and the frequencies of convergent recombination in each region by the frequencies of incongruent phylogenetic partitions. The analysis revealed the presence of an LRB region with the least frequencies of both external and convergent recombination, suitable for further phylogenetic analysis.'  
+
+## L460: Citation and how you tried to use hierBAPS needed  
+
+A: Thanks for the comment. We added a detailed description of how hierBAPS was used, as well as its primary clusters. The paragraph now reads:  
+
+'Furthermore, hierBAPS<sup>83</sup> was employed for population assignments of ESL with parameters "max.depth=2, n.pops=50", which failed to generate meaningful clusters, possibly due to the high recombination level of the ESL (Supplementary Fig. 2e).'  
+
+## L512: Usage information needs to be added to the Github repository  
+
+A: We have substantially improved the document in GitHub.
+
+<--- Page Split --->
+
+## L515: minimap2 does not call SNPs, correct?  
+
+A: Thanks. We have revised the whole paragraph that reads:  
+
+'Capybara is a generalized genotyping pipeline based on an SNP barcoding scheme, namely a set of 80 pre- calculated SNPs (Supplementary Table 4) that separate the ESL into clusters and clades. Capybara accepts the files containing sequencing reads as parameters, and (1) Align all reads onto a reference genome [MDR- TJ: GCF_000187205.2] by minimap2 \(^{84}\) and call base variants using SAMtools \(^{85}\) and bcftools \(^{86}\) . (2) It then compares the called nucleotides with the set of pre- calculated SNPs and assigns the read sets to the corresponding clusters and clades based on the matching SNPs. '  
+
+Figure 1: axis labels should be added to panels c and d  
+
+A: Revised.  
+
+Figure 1b: these pie charts are too small to see, even when zoomed in  
+
+A: Revised.  
+
+Table S4 claims to have used "abracadabra", but this method is not referenced in the text.  
+
+A: Revised. "abracadabra" was the old name of "Capybara".  
+
+Reviewer #2 (Remarks on code availability):  
+
+The code is not very useful in its current state. Improvements could include a README, a walk through, and annotated code. A field in the parser reads: "usage='CHIPICHIPI CHAPACHAPA DUBIDUBI DABADABA')". Not sure what to make of that.  
+
+A: Sorry, this was a meme. We have rewritten the codes with proper annotations and a clear README document.
+
+<--- Page Split --->
+
+\*\* See Nature Portfolio's author and referees' website at www.nature.com/authors for information about policies, services and author benefits.  
+
+This email has been sent through the Springer Nature Tracking System NY- 610A- NPG&MTS  
+
+Confidentiality Statement:  
+
+This e- mail is confidential and subject to copyright. Any unauthorised use or disclosure of its contents is prohibited. If you have received this email in error please notify our Manuscript Tracking System Helpdesk team at http://platformsupport.nature.com.  
+
+Details of the confidentiality and pre- publicity policy may be found here http://www.nature.com/authors/policies/confidentiality.html  
+
+Privacy Policy | Update Profile  
+
+DISCLAIMER: This e- mail is confidential and should not be used by anyone who is not the original intended recipient. If you have received this e- mail in error please inform the sender and delete it from your mailbox or any other storage mechanism. Springer Nature Limited does not accept liability for any statements made which are clearly the sender's own and not expressly made on behalf of Springer Nature Ltd or one of their agents.  
+
+Please note that Springer Nature Limited and their agents and affiliates do not accept any responsibility for viruses or malware that may be contained in this e- mail or its attachments and it is your responsibility to scan the e- mail and attachments (if any).  
+
+Springer Nature Ltd. Registered office: The Campus, 4 Crinan Street, London, N1 9XW. Registered Number: 00785998 England.
+
+<--- Page Split --->
+
+## RESPONSES TO REVIEWERS  
+
+## Reviewer #1  
+
+Q1. First, I want to congratulate the authors for all the effort to answer my comments. They have done a good job. However, there are still a couple of things they can do to answer my major comments 3 (the mol dating section) and 4 slightly better.  
+
+- Thank you. We now included additional BactDating and BEAST2 analyses (Supplementary Figure 5) to cover comment 3, and added Supplementary Figure 6d to cover comment 4. Please find the details below.  
+
+Q2. Considering the mol dating issue, the date-randomization test was a good step forward but I guess to fully get around this, and given that the dating of IC2 is an important part of the manuscript, I would suggest the authors employ alternative strategies to TempEst and BEAST2 to conduct the molecular dating analysis. So, that the authors can see if using these alternative approaches the results of the dating of IC2 are robust. There are optimization methods based on maximum likelihood — for instance, node.dating, treedater and TreeTime— or Bayesian inference (BactDating) that can be used to get independent estimates of the tMRCA for IC2. These methods can deal with larger datasets because they assume that the tree has previously been produced.  
+
+- Thanks very much for the comments. We used TreeTime as the primary dating approach because it is the only approach that we found to work efficiently on \(>8000\) genomes included in the IC2 lineage. Furthermore, we started to run BEAST2 on the same dataset since the first revision but did not report those results because the analyses failed to compete in 3 months. Following your suggestion, we attempted to use node.dating, treedater, and BactDating. Both node.dating and treedater failed to run, likely due to difficulties in handling such a large dataset, or possibly due to our limited expertise in using these tools. Fortunately, BactDating completed its run successfully, though it took 12 weeks to finish. Additionally, the BEAST2 runs we initiated earlier have now converged, after a total runtime of 6-10 months. We summarized all results in Supplementary Figure 6. We used five distinct models for each of BactDating and BEAST2, and all ten models generated consistent results than those from TreeTime. Specifically,
+
+<--- Page Split --->
+
+the tMRCAs for both the entire IC2 lineage and Clade 2.5.6 show remarkable consistency across the different methods, underscoring the robustness and approach- independence of our date estimates.  
+
+We now revised the manuscripts to reflect the use of different approaches in Lines 229- 235: "To further validate our analysis, we conducted additional analyses using BactDating and BEAST2, each with five distinct models, on the IC2 lineage (Supplementary Figs. 5a and b). All results exhibited high consistency with those from TreeTime. Specifically, the tMRCAs averaged 1935 (1908- 1953) for BactDating and 1944 (1925- 1961) for BEAST2 (Supplementary Fig. 5c). Similarly, all runs consistently placed the origin of Clade 2.5.6 around 2006 (CI 2002- 2008) (Supplementary Fig. 5d), corroborating the recent emergence of this clade."  
+
+And discussed the efficiency differences among the three approaches in discussion (Lines 401- 409):  
+
+"We also compared three TreeTime results with those from BactDating and BEAST2 runs. All eleven runs generated comparable estimates of the tMRCA for both IC2 and the clade 2.5.6, demonstrating the robustness of the analysis. Notably, we incorporated recombinant events in BactDating analysis, which did not significantly change the date estimates. However, the three pipelines differed in their efficiencies. The TreeTime handles \(\sim 8,000\) IC2 genomes in 30 minutes. In contrast, it took one to three months and six to ten months for BactDating and BEAST2 on the same dataset, respectively. Thus, we decided to use the TreeTime result, which offers a promising solution for other, large- scale population genetic analyses."  
+
+## Also Supplementary Figure 5:
+
+<--- Page Split --->
+![PLACEHOLDER_26_0]
+
+
+<--- Page Split --->
+
+Q3. As for the One Health issue (major comment 4), the authors have conducted new analyses to address this issue but I did not find any panel “d” in Supplementary Figure 5 — a plot that, given the rebuttal letter, should show ARG carriages in A. baumannii from non- clinical environments. Maybe you forgot to upload the most recent version of Supplementary Figure 5.  
+
+- We apologize for this missing figure. We now revised Supplementary Figure 5 (now Supplementary Figure 6) as below:  
+
+![PLACEHOLDER_27_0]
+
+
+<--- Page Split --->
+
+Q4. Finally, I concur with the authors that the definition of ICs in A. baumannii is rather arbitrary and ambiguous. They are right in that the use of clones in A. baumannii is very peculiar — and does not abide by the common definition of clone— yet would suggest they still tie up their epidemic lineages to the ICs (or rather to the Pasteur MLST scheme) just because potential legacy issues.  
+
+- Thanks for the comments.  
+
+## Reviewer #2  
+
+Q1. The authors have made significant improvements to this manuscript compared to the original version. There are still 2 issues that remain that I think need further attention. One is the filtering of recombination in the tree. In their rebuttal, the authors argue that this method has been used in "almost all" genomic analyses in the last 2 decades. This is false and studies that perform these analyses are typically not justified in doing so. The authors themselves describe how recombination shapes emerging lineages of A. baumannii. A phylogeny is a model, typically based on statistics (maximum likelihood) that fits the data to a simple framework. Based on my reading, the authors remove \(87\%\) of the data to fit their data to this model, which will obviously change the results.  
+
+- Thank you for your feedback. We understand that the reviewer has raised concerns about the selection of recombination and SNPs. We apologize for any ambiguity in our initial submission. We are committed to revising the sections pertaining to recombination and SNP selection to ensure clarity and rigor. We also seek your valuable insights to help us refine these aspects of our work. Your guidance is greatly appreciated as we aim to address these concerns effectively.  
+
+On the matter of whether recombination removal should be incorporated into the temporal phylogenetic analysis, it's important to note that the approach of "recombination filtering within the tree" is a common practice in studies on A. baumannii and other bacterial genomes. As an illustration, a paper by Kathryn Holt et al. in Microbial Genomics (see Box1) employs Gubbins for recombination filtering before constructing a phylogenetic tree for dating analysis. They removed the majority of variations and kept only 2000 non-recombinant SNPs for the phylogenetic tree.
+
+<--- Page Split --->
+
+Box1  
+
+(Kathryn Holt et al., 2016)  
+
+![PLACEHOLDER_29_0]
+  
+
+Consider another instance: the research by Liu et al., featured in Nature Communications in November 2024 (see Box2), where they analyzed a complete genome sequence of 3,709,099 base pairs, subsequently eliminating recombinant SNPs to retain a mere 120 variable sites. In contrast, our study has rigorously filtered the data, resulting in 181,732 SNPs for temporal analysis. We are confident that our approach yields a more precise and compelling temporal resolution compared to the study by Liu et al., which relied on only 120 SNPs.  
+
+Box2  
+
+(Liu et al., 2024)  
+
+Bioinformatic analyses  
+
+Sequence reads were trimmed, assembled using Unicycler v0.4.8 and assessed for quality. MLST was used to determine multi- locus sequence types with the Pasteur and Oxford typing schemes19,20. Typing of capsular polysaccharide (KL) was conducted with Kaptive 2.0 and Bautype 1.021. AMRFinder v1 was used to identify antimicrobial resistance genes22. Plasmid replicons were typed using the Hamidian lab typing scheme23.  
+
+Snippy v4.4.5 (https://github.com/tseemann/snippy) was used to align Illumina reads against a corresponding reference hybrid assembly (A. baumannii DETAB- R21, GenBank accession GCA 024205285.1) and generate a core genome alignment (3,709,099 bp). Polymorphic sites (120 polymorphic sites) were extracted with Gubbins v2.4.0, excluding those that were predicted to occur via recombination24. Phylogenies were constructed from these polymorphic sites using IQtree v2.0.3. Divergence dating was undertaken with the least- squares method implemented by IQTree v2.0.3, using the previously generated tree, Gubbins fasta file, and a GTR + G model25,26. SNP- distances were calculated from the Gubbins- filtered polymorphic sites file using SNP- dists 0.6.3 (https://github.com/tseemann/snp- dists).  
+
+In the case of other bacterial species like Klebsiella pneumoniae and Salmonella enterica, researchers typically initiate their analysis by the detection and removal of recombinant SNPs. The remaining SNPs are then utilized to construct the phylogenetic tree that traces the temporal
+
+<--- Page Split --->
+
+evolution of the bacteria, as demonstrated in the following table by Lucy van Dorp et al. (2019) and Caisey V. Pulford et al. (2020) (see Table 1 of responses).  
+
+Xavier Didelot introduced a manuscript named "A Scalable Analytical Approach from Bacterial Genomes to Epidemiology" in 2022, where he summarized the workflow as below:  
+
+![PLACEHOLDER_30_0]
+  
+
+Highlighting the importance of a "recombination corrected phylogeny".  
+
+Meanwhile, we agree with the reviewer that it is worth trying to incorporate recombination events in the second step, "dated phylogeny". Xavier offered a function in his BactDating package, which loads in ClonalFrameML results to help the date estimates. Technically, this allowed all variations to be used. Therefore, we incorporated recombination events in the BactDating analyses here, which exhibited minimum impacts on the results (Supplementary Figure 5). Therefore, we did not change the main results, but described the incorporation of recombination in the discussion (lines 403- 405):  
+
+"Notably, we incorporated recombinant events in BactDating analysis, which did not significantly change the date estimates."  
+
+And described the way to do so in the methods (Lines 586- 595):  
+
+"Additionally, we run BactDating and BEAST2 on the IC2 lineage. Specifically, BactDating incorporated recombinant events in date estimates by loading clonalframeML results using the loadCFML() function. We ran BactDating with alternative models of "arc", "carc", "negbin", "mixedgamma", and "relaxedgamma", and ran BEAST2 with a "GTR" substitution model and alternative clock models of "Relaxed LogNormal", "Relaxed Exponential", "Relax Uniform", "Strict Clock", and "Random Local Clock". For each run, the chain length was set to 1e12 or
+
+<--- Page Split --->
+
+when the ESS exceeded 200 (Supplementary Fig. 5). The BactDating runs finished in 1- 3 months, whereas the BEAST2 runs lasted for 6- 10 months. "
+
+Table 1 of responses. Cases of recombinant removal and SNPs for time-tree building.
+
+<table><tr><td>Article</td><td>Title</td><td>Publisher</td><td>Bacteria</td><td>Recombination<br>removal?</td><td>SNPs for time-tree building</td></tr><tr><td>1 (Kathryn Holt<br>et al.)</td><td>Five decades of genome<br>evolution in the globally<br>distributed, extensively<br>antibiotic-resistant<br>Acinetobacter baumannii<br>global clone 1</td><td>Microbial<br>Genomics</td><td>Acinetobacter<br>baumannii</td><td>Yes, use Gubbins</td><td>1800 SNPs</td></tr><tr><td>2 (Haiyang Liu et<br>al.)</td><td>Longitudinal genomics<br>reveals carbapenem-<br>resistant Acinetobacter<br>baumannii population<br>changes with emergence of highly resistant ST164 clone</td><td>Nature<br>Communications</td><td>Acinetobacter<br>baumannii</td><td>Yes, use Gubbins</td><td>120 SNPs</td></tr><tr><td>3 (Lucia Graña-<br>Miraglia et al.)</td><td>Origin of OXA-23 Variant<br>OXA-239 from a Recently<br>Emerged Lineage of<br>Acinetobacter baumannii<br>International Clone V</td><td>mSphere</td><td>Acinetobacter<br>baumannii</td><td>Yes, 383 SGFs<br>had evidence of<br>recombination<br>and were<br>discarded</td><td>388<br>nonrecombinant single-gene<br>families (SGFs)</td></tr><tr><td>4 (Lucy van Dorp<br>et al.)</td><td>Rapid phenotypic evolution<br>in multidrug-resistant<br>Klebsiella pneumoniae<br>hospital outbreak strains</td><td>Microbial<br>Genomics</td><td>Klebsiella<br>pneumoniae</td><td>Yes, use bcftools</td><td>~15,000 SNPs</td></tr><tr><td>5 (Caisey V<br>Pulford et al.)</td><td>Stepwise evolution of<br>Salmonella Typhimurium<br>ST313 causing bloodstream<br>infection in Africa</td><td>Nature<br>Microbiology</td><td>Salmonella<br>Typhimurium</td><td>Yes, use Gubbins</td><td>12,013 SNPs</td></tr><tr><td>This paper</td><td>Emergence and Global<br>Spread of a Dominant<br>Multidrug-Resistant Clade<br>in Acinetobacter baumannii</td><td>/</td><td>Acinetobacter<br>baumannii</td><td>Yes, both<br>RecHMM,<br>ClonalFrameML<br>and Gubbins</td><td>23,524 SNPs</td></tr></table>
+
+<--- Page Split --->
+
+Q2. The second major issue is the use of the timed tree. The authors argue that a temporal signal is justified based on the data-randomisation test. However, the BEAST2 documentation also states that: "One limitation of the date-randomisation test is that in some circumstances the test can fail to reject data sets with no temporal structure". Although it is difficult to determine if your data represents such a circumstance, the uncertainty of the test, based on the very low linear regression correlation, suggests that the timed tree analysis is speculative at best and should be framed in that context.  
+
+- Thanks very much for the comments. We now added ten additional temporal analyses using BEAST2 or BactDating. All analyses generated highly consistent date estimates. Please find details in our answers to Reviewer 1.  
+
+Additionally, our estimated R of 0.28 is not specifically low. Rossi obtained an R- value of 0.33 for \*Mycobacterium bovis\* (https://doi.org/10.3389/fvets.2023.1086001); Dupas had an R- value of 0.21 for \*Xylella fastidiosa\* (https://doi.org/10.1038/s42003-023-04499-6). The temporal signal can change with the size of the population, especially when there is a significant difference in the numbers.  
+
+In my practice, it is quite normal that the R- value will decrease in very large- scale datasets, which is likely due to the dense sampling of recent isolates. To demonstrate that, we subdivided the IC2 isolates into bins of 10 years, and randomly selected up to 10 isolates from each bin. The temporal signals of the resulting subtrees substantially improved, with the R- value averaging 0.56. This simulation likely explained the better R- values obtained from earlier, smaller- scaled studies. We did not incorporate this result in the manuscript, because I think additional simulations and calculations are needed to explain the phenomenon in greater detail. Instead, I added sentences in the discussion (lines 469- 475) that read:  
+
+"A limitation of this work is the absence of phenotypic tests for AMR, which limited our ability to detect novel, undocumented drug resistance mechanisms. Furthermore, the involved strains were predominantly collected in clinical environments over the past 20 years and may not accurately reflect earlier epidemiological patterns. Particularly, the oversampling of recent
+
+<--- Page Split --->
+
+isolates also reduced the significance of the temporal signal. An equal sampling of the strains along with time would substantially improve the R- value in the root- to- tip regression. "  
+
+Some other comments, including more detailed comments that address these two concerns include:  
+
+L1: clade "of"? clade "within"?  
+
+- Corrected. We used "within" instead. Thanks.  
+
+L76: I would indicate early that this is a term that you have created and what it means - Thanks for the comments. The concept of the IC (International Clone), also known as the GC (Global Clone), was originally introduced as the "European Clone" by Dijkshoorn et al. in 1996 and by Nemec et al. in 2004 to describe two pan- European clones (IC1 and IC2, see Box4). As more geographically widespread clones were identified, the concept broadened, and it eventually became known simply as IC. We have inserted these two papers into references.  
+
+(Dijkshoorn et al., 1996)  
+
+The most striking finding of our study was the distinction of four groups of strains that were highly similar by at least one genomic and one other typing method (Table 2), which could be explained by a common clonal origin (28). Two groups (groups I and II) mainly comprised strains from outbreaks at different locations in northwestern Europe. Whether these  
+
+(Nemec et al., 2004)  
+
+![PLACEHOLDER_33_0]
+  
+
+L118: This supertree will likely contain recombination. How does this agree then with your SNP based trees that remove recombination?
+
+<--- Page Split --->
+
+- You are correct that the supertree does include recombination. To minimize their impact, we chose a supertree approach, which involved estimating 2,266 individual gene trees—one for each core gene—and then summarizing these into a single supertree.  
+
+As described earlier, recombination typically introduces many mutations in short DNA fragments. Consequently, while some genes may be heavily influenced by recombination, the majority remain unaffected and retain their original phylogenetic topology. The supertree algorithm we used operates by taking majority votes on the topologies of the individual gene trees, a method known to be robust against the effects of recombination. Tandy Warnow summarized this in https://doi.org/10.1098/rstb.2021.0244, although in different technical terms because of her eukaryotic background.  
+
+Using this approach, we successfully recovered most clades in the supertree. However, it is important to note that the core genes used in supertree represent only \(50\%\) of the chromosomal regions, limiting the resolution for detailed phylogenetic inference. Additionally, the majority vote approach may fail in branches where recombination is exceptionally prevalent. For these reasons, a standard core genome alignment and SNP- based phylogeny remain more suitable for analyzing genetically closely related genomes, such as the ESL.  
+
+## L188: Your justification for using this test was on the data-randomisation test, correct? Why isn't this listed here?  
+
+- We apologize for this unclear description. The root-to-tip regression method, shown in Supplementary Figure 4b, is one of the earliest and most classical approaches for evaluating temporal signals in phylogenetic trees. It is widely used due to its straightforward assumption: if substitutions accumulate at a constant rate over time, strains sampled later are expected to exhibit more genetic variation than those sampled earlier. However, this method is less effective when substitution rates vary over time, a scenario known as a relaxed molecular clock.  
+
+To address this limitation, alternative models, such as the relaxed lognormal and random local clock, were developed to accommodate variable substitution rates. While these models are more flexible, they have their own drawbacks: they can produce results even in the absence of a true temporal signal, leading to potential false positives.
+
+<--- Page Split --->
+
+Therefore, the state- of- the- art, date- randomisation test was invented. This method provides a non- parametric framework to compare the temporal signal in the real dataset to those in randomly permuted datasets. While, as the reviewers proposed, the date- randomization test is not without limitations, it remains the most sensitive and statistically robust approach currently available for detecting temporal signals. We showed the date- randomisation results in Supplementary Figure 4c.  
+
+## L226: If recombination is the driving force for evolution, why would you filter it out of your SNP-based phylogenies?  
+
+- Thanks. Please find our responses above.  
+
+## L461: MLST "profiles were"  
+
+- corrected. New line 497.  
+
+## L470: "USEARCH" is typically capitalized  
+
+- corrected. New line 506.  
+
+L491: How is the reader supposed to identify your data in Table S1? Maybe list "this study" somewhere to guide the reader to new data generated in this study  
+
+- Thanks for the comment. We have added a list of "this study" to show the strains from this study annotated by STAR \(\star\) . Please see the new Table S1.  
+
+L499: How is the reader supposed to know what DTY is? the authors should describe what this does.  
+
+- Thanks. Please see the new line 534 to 536 which read:  
+
+"We calculated the presence of pan genes in the representative genomes using the DTY \(^{66}\) , which identifies and extracts homologous genes in genomes based on the reference sequences."
+
+<--- Page Split --->
+
+## L501: What is the justification for using \(>94\%\) here and \(>95\) of genomes?  
+
+- We apologize for this unclear description. It seems to be a common practice to use \(95\%\) presence as a threshold of the core genome. Specifically, we chose genes that were shared by 5,533 of the 5,824 representative genomes. While it is always worth discussing whether more stringent or relaxed criteria should be used, we found that the resulting 2,266 core genes gave enough genetic signals for the separation of major populations in A. baumannii.  
+
+The \(>94\%\) intact CDS threshold was more of a common practice for core genome multi- locus sequence typing (cgMLST) schemes. We evaluated the impact of different values and found that \(94\%\) served a balance between the genetic stability of the genes and the number of final core genes, and found that, compared to the \(95\%\) cutoff, \(94\%\) allowed the inclusion of 260 more genes. We initially wanted to include the building of a cgMLST scheme for A. baumannii in the manuscript, but later took it out because this would dilute our main findings, namely the stepwise evolution of IC2.  
+
+We have now added a description in the methods (lines 537- 540) that reads:  
+
+"we extracted a subset of 2,266 soft core genes by selecting each gene that was (1) present in \(\geq 95\%\) of genomes and (2) maintained intact open reading frames in \(\geq 94\%\) of its alleles, thresholds that were derived from our early practice for core genome multi- locus sequence typing (cgMLST) schemes \(^{53}\) ."  
+
+L510: "de- recombination", as far as I know, is not a phrase that is used.  
+
+- Thanks for the point. It was changed to "Removal of Recombination"  
+
+## L522: Did the authors verify that all recombinant SNPs had been removed?  
+
+- It is almost impossible to remove all recombinant SNPs, particularly for those occurring between genetically extremely closely related strains. Luckily, those closely related recombination will not significantly influence the phylogenetic estimates because they will not result in many SNPs. Current edge-cutting methods all focus on recombination from external sources, which will bring in lots of mutations in very short segments. Here we compared results
+
+<--- Page Split --->
+
+from three distinct approaches, RecHMM, Gubbins, and ClonalFrameML, and found that they all gave similar results.  
+
+L525: So the authors removed 158,208 recombinant SNPs? By doing this, the authors have removed \(87\%\) of the diversity in the dataset. This seems nether justified nor warranted  
+
+- Thanks for the comment. Please see the answers above.  
+
+L540: Yet the BEAST documentation mentions that this test can fail to reject datasets with no temporal structure. This seems to be a major flaw in reporting this, especially based on the poor R2 correlation  
+
+- Thanks for the comment. Please see the answers to reviewer 1.  
+
+## L586: based on "a" SNP  
+
+- corrected.  
+
+## Additional grammar corrections.  
+
+- As both reviewers are concerned, we did a thorough proofreading by an English speaker and made substantial revisions to the manuscript.
+
+<--- Page Split --->
+
+## RESPONSES TO REVIEWERS  
+
+## Reviewer #1  
+
+The authors have addressed my comments. Congratulations to them on all their hard work.  
+
+- We sincerely thank you for your feedback enhancing the quality of our work.  
+
+## Reviewer #3  
+
+1. Apart from IC1 and IC2 (GC1 or GC2, or WW1 and WW2), other ICs/GCs/WWs are less common and less known to most readers, please use the sequence type (ST) equivalent so readers can follow the text easier.  
+
+-Corrected, we have added the clonal complex (CC) named by ST for each IC in the first time of mention. We also included detailed information, including STs of all strains, in Supplementary Data 1. As an example, please see the new line 169-171 as follows.  
+
+"The ESL accounted for \(65\%\) (Africa & North America) to \(90\%\) (East Asia & Oceania) of the isolates from every geographic region except South America, where IC5 (CC79, namely ST79 clonal complex) and IC7 (CC25) were predominant (Fig. 1d)."  
+
+2. Line 251-269 under the Stepwise acquisition of ARGs in the ESL section: could you please comment on the fact that many known ARGs, including oxa23, in Ab are either in chromosomal genomic islands (acquired by an ancestral strain with evidence published on different STs elsewhere) and acquired via plasmids? Tn2006, AbaR4 and variants of Tn6167 in ST2 are the classic examples.  
+
+- Thank you for your advice. We evaluated 168 complete genomes of the ESL lineage and found blaOXA-23 in 111 of them. Notably, we found that the majority of blaOXA-23 genes were associated with Tn/IS elements, which could result in breaks in draft assemblies. As a result, it is impossible to evaluate the replicon association of the blaOXA-23 genes in most of the draft genomes. So we decided to limit our analyses to only the complete ones. Please see the new line 256-260, as well as the new supplementary figure 8.  
+
+"To assess the role of mobile elements in this incremental acquisition of resistance, we evaluated 168 complete ESL genomes, finding blaOXA-23 in 111 of them (Supplementary Fig. 8). Notably, only \(23\%\) (26/111) blaOXA-23 was transferred by plasmids. In contrast, Tn/IS elements account for \(93\%\) (103/111) of blaOXA-23 carriage, with the majority being Tn2007 and Tn6166."
+
+<--- Page Split --->
+![PLACEHOLDER_39_0]
+  
+
+Supplementary figure 8. The locations of \(bla_{\mathrm{OXA - 23}}\) within ESL.  
+
+3. Line 263: First reference 12 is an incorrect reference and not suitable for this statement. In fact, this information is also incorrect and if it has been mentioned in ref 12, this reference needs an Erratum. blaADC (ampC) is an intrinsic gene present in all A. baumannii strains and can barely cause any detectable resistance phenotype unless there is an ISAbal copy present upstream of this gene (which gives it a strong promote that enhances the expression) making it resistant to 3rd-generation cephalosporins. Please fix this, where mentioned in the text and include an appropriate reference.  
+
+- Thanks for the comment. We have rewritten the sentences and updated the new references. Please see the new lines 267-270. We also removed the old Supplementary Figure 8 and Supplementary Data 5, because they are irrelevant now.  
+
+"Many IC2 strains have developed resistance to \(\beta\) - lactamase inhibitors like sulbactam due to the presence of \(bla_{\mathrm{TEM - 1}}\) gene<sup>24</sup>, and the MRCA of Cluster 2.5 acquired the ftsI-A515V mutation that has been associated with resistance to sulbactam-durlobactam<sup>25</sup>."
+
+<--- Page Split --->
+
+4. Big parts of this study examines the evolution of resistance while it is now known that plasmids play a crucial role in AMR trafficking in A. baumannii. To me it's a bit strange that there is no mention of one of the most important elements that drives resistance in this paper.  
+
+-We have indeed found some interesting facts about the transposon/plasmids of A. baumannii. Instead of carrying by plasmids (such as in Klebsiella pneumoniae), the blaOXA-23 in A. baumannii was primarily carried by transposons. We added the following paragraph in the discussion to discuss the role of plasmid/transposons in AMR transmission. Please see the new line 465-472  
+
+"In this study, we also highlighted the significant role of plasmids and transposons in the dissemination of antimicrobial resistance. Notably, by analyzing 162 complete A. baumannii genomes, we discovered that the spread of blaOXA-23 genes in IC2 and IC1 was primarily mediated by transposons. In our dataset, 92% (103/111) of the blaOXA-23 genes were associated with transposon (Tn/IS) elements, particularly Tn6166 and Tn2007, whereas only 23% (26/111) were located in plasmids. Furthermore, 69% (18/26) of the plasmid-carrying blaOXA-23 genes were also associated with transposons, underscoring the important role of transposons in the dissemination of blaOXA-23 genes."
+
+<--- Page Split --->

peer_reviews/supplementary_0_Transparent Peer Review file__eadaf414198d42d185872205bbb0c7524665a53d6db7bd0ab2e4abf679bfc7b7/images_list.json

@@ -0,0 +1,175 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Figure R1: Overall accuracy comparison of the effects of Leiden and Louvain clustering to TACIT performance in PCF-CRC, PCF-HI, and MERFISH.",
+    "footnote": [],
+    "bbox": [
+      [
+        258,
+        131,
+        728,
+        336
+      ]
+    ],
+    "page_idx": 9
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_1.jpg",
+    "caption": "Figure R3: (a) Proportion of cell type predictions between TACIT and SCINA (do not allow unknown cells). (b) Overall accuracy comparison between SCINA (do not allow unknown cells) and TACIT in the PCF-HI datasets.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 9
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_2.jpg",
+    "caption": "Figure R4: Recall, Precision, and F1 scores by sample, comparing the PCF-CRC (140 samples) and PCF-HI (64 samples) datasets across four methods: TACIT, CELESTA, Louvain, and SCINA.",
+    "footnote": [],
+    "bbox": [
+      [
+        115,
+        126,
+        884,
+        305
+      ]
+    ],
+    "page_idx": 10
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_3.jpg",
+    "caption": "Figure R5: (a) Boxplot showing the number of genes per cell type for each simulation. (b) Boxplot of accuracy scores (5 bootstraps) for each simulation.",
+    "footnote": [],
+    "bbox": [
+      [
+        130,
+        241,
+        860,
+        451
+      ]
+    ],
+    "page_idx": 11
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_4.jpg",
+    "caption": "Figure R2: (a) Number of clusters across different resolutions (0.1-1.9) for the Leiden and Louvain methods. The red dotted line indicates the expected number of cell types. (b) Comparison of average entropy score at varying resolutions using Leiden and Louvain. (c) Comparison of purity score at different resolutions using Leiden and Louvain.",
+    "footnote": [],
+    "bbox": [
+      [
+        115,
+        123,
+        880,
+        283
+      ]
+    ],
+    "page_idx": 12
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_5.jpg",
+    "caption": "Figure R6: (a) Bar plot comparing F1 scores of recent cell type annotation methods for spatial proteomics, including TACIT, Astrir, CELESTA (from the Spatial-ID paper), STELLAR, and TYPEx, across the PCF-HI and PCF-CRC datasets. (b) Bar plot comparing F1 scores of TACIT with recently published cell type annotation methods for spatial transcriptomics data using MERFISH dataset.",
+    "footnote": [],
+    "bbox": [
+      [
+        120,
+        272,
+        872,
+        450
+      ]
+    ],
+    "page_idx": 13
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_6.jpg",
+    "caption": "Figure R7: (a) Heatmap of the mean expression of protein markers in TACIT predictions for the PCF-CRC dataset, displaying cells matched with the reference (left) and cells not matched with the reference (right). (b) Heatmap of the mean expression of protein markers in TACIT predictions for the PCF-HI dataset, showing cells matched with the reference (left) and cells not matched with the reference (right).",
+    "footnote": [],
+    "bbox": [
+      [
+        123,
+        180,
+        880,
+        504
+      ]
+    ],
+    "page_idx": 17
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_7.jpg",
+    "caption": "Figure R8: Average proportion of cells per cluster at each resolution.",
+    "footnote": [],
+    "bbox": [
+      [
+        208,
+        145,
+        787,
+        392
+      ]
+    ],
+    "page_idx": 18
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_8.jpg",
+    "caption": "Figure R9: UMAP of scRNA-seq with cell type expert manual annotations (left) and Xenium (spatial transcriptomics) with Seurat transfer annotations (right).",
+    "footnote": [],
+    "bbox": [
+      [
+        155,
+        272,
+        840,
+        544
+      ]
+    ],
+    "page_idx": 19
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_9.jpg",
+    "caption": "Figure R10: Run time comparison. All analyses were performed on a system with an Intel(R) Core(TM) Ultra 9 185H 2.30 GHz processor and 64 GB of RAM. Note unsupervised clustering methods such as Leiden and Louvain clustering require additional manual annotation steps to convert clusters into cell types, which can be time-consuming. Recent reports from Brbic, M. et al. 2022, Nature Methods reported that processing four tissues ( \\(\\sim 160,000\\) cells) in a typical PCF dataset with a 48-marker panel can take over 25 hours for clustering, merging, re-clustering, sub-clustering, and cell type assignment based on marker expressions and spatial locations within PCF images.",
+    "footnote": [],
+    "bbox": [
+      [
+        120,
+        404,
+        868,
+        631
+      ]
+    ],
+    "page_idx": 20
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_10.jpg",
+    "caption": "Figure R1: Comparison of the number of clusters, entropy score, and purity score, among Louvain, Leiden, and TACIT methods across resolutions ranging from 0.005 to 50 using PCFCRC datasets (R1a-c) and MERFISH datasets (R1d-f).",
+    "footnote": [],
+    "bbox": [
+      [
+        135,
+        133,
+        844,
+        510
+      ]
+    ],
+    "page_idx": 21
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_11.jpg",
+    "caption": "Figure R2: Evaluating Segmentation Impact in TACIT Annotations on PCF Datasets. (a) Following TACIT annotations using three different methods—Watershed, stardist, and Cellpose version 3—we used Voronoi plots to reconstruct the slides and analyze the variations in reconstructions attributable to different segmentation methods. The Watershed model resulted in a reconstruction with a dense population of ductal cells, while the stardist and Cellpose version 3 models produced reconstructions that more closely represented the architecture of acinar/myoepithelial cells and ducts within the salivary lobules. (b) Each model identified varying",
+    "footnote": [],
+    "bbox": [
+      [
+        128,
+        461,
+        863,
+        745
+      ]
+    ],
+    "page_idx": 22
+  }
+]

peer_reviews/supplementary_0_Transparent Peer Review file__eadaf414198d42d185872205bbb0c7524665a53d6db7bd0ab2e4abf679bfc7b7/supplementary_0_Transparent Peer Review file__eadaf414198d42d185872205bbb0c7524665a53d6db7bd0ab2e4abf679bfc7b7.mmd

@@ -0,0 +1,756 @@
+
+# Deconvolution of Cell Types and States in Spatial Multimics Utilizing TACIT  
+
+Corresponding Author: Professor Jinze Liu  
+
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+Version 0:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+Huyhn et al. present Threshold- based Assignment of Cell Types from Multiplexed Imaging Data (TACIT), a method for annotating cell types based on known cell marker signatures and expression profiles. TACIT is an unsupervised algorithm built on Louvain clustering followed by fitting a "Cell Type Relevance score" regression model to best separate cells into discrete cell types. The authors compare TACIT to select similar algorithms and compare across technologies including proteomic PhenoCycler and transcriptomic MERFISH and Xenium and tissue types including colorectal cancer, healthy intestine, and salivary glands with graft- versus- host disease. Overall, the manuscript was adequately written and clearly described the method and offered a favorable view of the algorithm compared to others. However there are some major points which hamper enthusiasm for the report. One of the major issues is the benchmarking, which appears to compare algorithms in an unfair manner, especially the Louvain algorithm. Additionally, the quantitative results may need to be refined based on the lack of a ground truth, perhaps using, for example, alternative Silhouette- based measures.  
+
+Detailed issues are as follows:  
+
+How are values chosen for relevance of markers defining cell types? Is this arbitrary?  
+
+TACIT attempts to cluster highly homogenous cell communities by choosing a Louvain resolution to contain small numbers of cells. There are two issues with this assumption. First, Louvain clusters are highly connected which is not necessarily true which led to the development of the Leiden method. Second, if the assumption is that the cells are homogenous then the cutoff should not be cluster size but rather cell similarity within clusters by using something akin to Silhouette.  
+
+TACIT learns a threshold to separate positive signals and background noise - - does this hold true for highly homogeneous data? What about heterogeneous data of a same cell type such as a tumor?  
+
+For the benchmarking, as there is no real ground truth, the authors should also include an analysis with a measure quantifying feature similarity such as entropy, purity, or Silhouette.  
+
+How sensitive is TACIT to shifting parameters? For example, how much does varying the resolution of the initial Louvain clustering change the results? Any
+
+<--- Page Split --->
+
+other parameters should be checked as well.  
+
+CELESTA and SCINA were not included in the PCF- HI dataset comparison due to too many "Other" annotations. However, algorithms have customizable parameters to account for this - for example, SCINA has an "allow_unknown" parameter than can be set to 0 if this is too much of an issue to limit comparison.  
+
+The authors mention significantly higher recall, precision, and F1 scores, but there is no listed p- value at that point in the text. A solution to this would be to split the data by sample and have a score for each sample to compare.  
+
+Why was TACIT always compared with Louvain following the benchmark? Especially when it seems like CELESTA was better performing than their method for Louvain.  
+
+Minor: Figure 1e "i" layout is confusing and should be clearer in the figure or have a better notation.  
+
+There are typos in the manuscript (e.g. pg. 8: "transcriptomics., providing")  
+
+Page 9: "release, highlighting" -> "release, highlighting"  
+
+The authors compared Louvain with their algorithm based on the UMAP embedding, which is highly controversial and not recommended as UMAP is known to destroy distances between observations. Furthermore, they most likely did not use a resolution on par with their own algorithm so the comparison may not be fair. Lastly, they only looked at the top 5 markers instead of all of the a priori known markers to define each cluster, unlike their own algorithm. In all, I am not entirely convinced that the Louvain method they compared to was fair, and if corrected, how much TACIT would add to this existing pipeline.  
+
+(Remarks on code availability)  
+
+The code is deposited on Code Ocean, although I could not review the Github repository as it was private. The authors present the code on Code Ocean as a large monolithic file and not all functions are documented, this could be improved. Furthermore, there appears to be no README, tutorial, or vignette, hampering transparency, reproducibility, and ease- of- use. These would absolutely need to be added before moving forward with any publication. Running the code with a reproducible run did not appear to successfully output anything, perhaps better documentation for this run would help as well.  
+
+## Reviewer #2  
+
+(Remarks to the Author) Comments to NCOMMS- 24- 38058- T.  
+
+The author developed TACIT, an unsupervised algorithm for cell annotation in spatial transcriptomics and proteomics. Benchmark analysis showed that TACIT outperformed existing methods in accuracy and scalability when tested on five datasets. The authors further revealed new phenotypes in inflammatory gland diseases and identified under- and overrepresented immune cell types through the integration of spatial transcriptomics and proteomics. However, substantial revisions are still required to enhance the clarity, robustness, and impact of your algorithm and findings.  
+
+Major concerns:  
+
+1. The TACIT method appears to be heavily constrained by the signature. Is there a metric to evaluate the efficiency of markers in distinguishing cell types, thereby assisting in the design of a marker panel? What if the markers are insufficient for differentiation?  
+
+2. In scenarios where signatures are subject to noise, an investigation into the robustness of the TACIT method is warranted, particularly considering the addition or subtraction of markers. Moreover, determining the minimum number of markers required for accurate cell type recognition would be beneficial.  
+
+3. The performance of the Louvain method critically depends on clustering resolution, and misclassifications could potentially stem from clustering inaccuracies. When evaluating, would changes in the resolution of the Louvain method help improve performance? Also, it would be insightful to evaluate the outcomes when utilizing pre-existing cluster assignments (if available) directly.  
+
+4. It is imperative to conduct a comparative analysis of TACIT against contemporary methodologies, notably those employing deep learning frameworks such as SPACEL, and emergent single-cell RNA sequencing (scRNA-seq) techniques exemplified by ACT.
+
+<--- Page Split --->
+
+5. Certain cell types had very low F1 scores. A detailed inquiry is warranted to elucidate which cell type exhibits the peak predictive accuracy and conversely, which demonstrates the nadir, along with an explanation grounded in the underlying characteristics or limitations of the dataset or methodology.  
+
+6. When performing MC clustering, how does one determine the relationship between resolution and the average number of cells per cluster (0.1% to 0.5%)?  
+
+7. Results > Section 3: An investigation into the disparity of outcomes when leveraging the top 5 markers as signatures versus established markers is essential. This comparison will shed light on the potential advantages or shortcomings of each strategy.  
+
+8. Results > Section 3: While scRNA-seq has successfully identified cell subtypes, subsequent application of Seurat's transfer method fails to recapitulate these distinctions. An analysis is required to uncover the cause of this discrepancy.  
+
+9. The analytical focus on cell states appears cursory and lacks specificity. Moreover, the identification of cell states is not inherently a feature of the TACIT methodology. In light of these considerations, it may be prudent to reassess the inclusion of 'cell state' in the title to accurately reflect the scope and findings of the study.  
+
+10. Results > Section 6: The consistency between the results of spatial proteomics and transcriptomics data is not high. What are the reasons for poor consistency? Which one should be trusted more, or is it necessary to use matched markers and integrate multi-omics data? Due to poor consistency, it is recommended to visualize markers and directly compare the visualization results of markers.  
+
+Minor concerns:  
+
+1. There are minor errors present throughout the text. Please thoroughly review the entire document to correct these and similar mistakes.  
+
+Methods, Segmented Regression Model: "This was determined by the lowest Akaike Information Criterion (AIC) score achieved among the three models the three models"  
+
+Figure 6i: "B Cells: Cell State (+)"  
+
+Figure 5: The text does not match the figure, the caption does not correspond to the figure, and the captions are incomplete.  
+
+Figure 1f: Note the parentheses that distinguish 'Clean cells' from 'Mixed cells', and there is an issue with the representation of 'Mixed cells' (One "Clean cell' among them).  
+
+Section 4, Figs. 4c-e only represent the results of one dataset, not corresponding to the two datasets mentioned in the text.  
+
+2. How is the Bootstrap performed?  
+
+3. The main text sections do not mention Astrograph, yet it is involved in the abstract. Please revise the relevant content accordingly.  
+
+4. It is suggested to revise the section titles, as several sections lack informative headings, making it unclear what each section entails based on the section title alone.  
+
+(Remarks on code availability) Invalid link.  
+
+Reviewer #3  
+
+(Remarks to the Author)  
+
+Despite the advances in spatial multiomics technologies, identifying cell types and states remains challenging due to issues like segmentation noise, signal bleed- through, and limited molecular markers. Traditional unsupervised clustering methods struggle with sparse marker set, while deep learning algorithms require extensive and diverse training data to be effective. To address these challenges, the authors developed TACIT, an unsupervised algorithm that used cell- marker expression profiles to accurately assign cell identities. TACIT took as input a CELLxFEATURE matrix, and a CELLTYPExMARKER matrix, and performed cell type annotation in two rounds: first, it clustered cells into microclusters and calculated Cell Type Relevance (CTR) scores for each cell to label it with high or low relevance to specific cell types. Subsequently, a deconvolution step using KNN was applied to annotate the cell types of individual cells.  
+
+Extensive experiments showed TACIT's effectiveness in cell type annotation, outperforming CELESTA, SCINA, and Louvain across multiple spatial proteomics and transcriptomics datasets. Additionally, TACIT identified T cell subtypes in an unpublished Xenium dataset, which were crucial for understanding Sjogren's disease. In Graft- versus- Host Disease samples, TACIT identified more cell types with higher accuracy, including key ones missed by Louvain. It effectively handled segmentation errors in nascent tertiary lymphoid structures (TLS), accurately identifying adaptive and innate immune cells. Furthermore, TACIT could integrate spatial transcriptomics and proteomics data, achieving higher agreement in cell type identification compared to using only proteomics markers versus only transcriptomic markers, revealing crucial differences in cell states for optimizing immunotherapy treatments.
+
+<--- Page Split --->
+
+Strengths:  
+
+1. The paper presented a novel, unsupervised method "TACIT" for cell type identification, addressing limitations in existing methods. Uniquely, it could integrate data from multiple modalities (spatial transcriptomics and proteomics).  
+
+2. The authors conducted thorough experiments on a variety of datasets, including various spatial proteomics and transcriptomic dataset across different tissue types (brain, intestine, gland) and species (human, mouse). Extensive benchmarking with existing methods (CELESTA, SCINA, Louvain, Seurat transfer) was performed to demonstrate its superiority.  
+
+3. TACIT illustrated its clinical value by identifying significant immune cell markers relevant for diseases like Sjögren's disease and Graft-versus-Host Disease (GvHD).  
+
+Weaknesses:  
+
+1. The TACIT's performance heavily depended on the quality and comprehensiveness of the marker panels used for different cell types. However, the author did not seem to explain the detailed process of selecting markers. In practice, if there is no expert knowledge especially for some rare cell types, the author should explain how to ensure markers are accurate. Alternatively, if some low-quality markers were included in TACIT's signature matrix, did TACIT's performance remain robust?  
+
+2. In TACIT's algorithm, the effectiveness of using KNN for the deconvolution should depend on the number of "clean cells" from previous "cell type categorization" step, which were used as anchors. How can we ensure there are enough "clean cells" when TACIT is applied to other organs and species? If there are too few "clean cells", the cell type annotation will definitely perform poorly.  
+
+3. Although the authors claimed TACIT had a promising result in integrating multimodal data, its performance was only demonstrated using transcriptomics and proteomics. Newly emerging modalities like spatial epigenomics and metabolomics were not thoroughly tested in the paper. Further validation on a broader range of spatial omics technologies would strengthen its applicability and increase the impact of TACIT.  
+
+4. Although the author discussed TACIT's memory usage, they should also mention TACIT's computation time and compare it with other benchmarking methods on real or simulated data to demonstrate TACIT's practicality.  
+
+5. There are some writing and formatting issues.  
+
+a) In Figure 2 legend, this description should be changed to (b, h) "(e,k) UMAP representations with cell type delineations, showing the clustering of cells in a two-dimensional space ... Identifying cell types " b) In Figure 2 legend, (f,i) should be (c,i) "(f,i) Heatmaps comparing the mean marker values for each cell type identified by TACIT and other existing methods. TACIT's heatmaps exhibit distinct and clear unique marker expressions for each cell type, with a diagonal pattern that highlights its precise cell type identification capabilities. " c) The resolution of Figure 3c is very low, making it difficult to read the names of the markers.  
+
+(Remarks on code availability)  
+
+Version 1:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author) I thank the authors for their effort and thorough response to my concerns.  
+
+The main concern I have is still the benchmarking, as while the authors did implement other metrics such as purity and entropy, they did so in a way that may be misleading. For example, in Figure R2a increasing resolution resulted in more clusters which is contrary to the idea of resolution where an increase should have less clusters. In addition, in Figure R2a,b they show a range of resolution values for Leiden and Louvain but not for TACIT. This is odd, especially when in Extended Data 2 they have a range for TACIT. In addition, that range was from 0.005 to 150, but we do not know what TACIT used in this benchmark, it should be the same as Louvain at least as it uses Louvain as part of the algorithm. I understand the authors' point that it would be unfeasible to annotate manually this many clusters, but entropy purity do not require a cluster annotation so that is not the intended purpose of this benchmark.  
+
+Lastly, I noticed some typos in the newly included text, some proofreading may be required before publication.
+
+<--- Page Split --->
+
+(Remarks on code availability) As the GitHub repository is not public, I am unable to review that repository's code quality, vignettes, README, etc.  
+
+I still did not see any of that documentation in the Code Ocean repository as well.  
+
+Reviewer #2  
+
+(Remarks to the Author) Most of my concerns have been resolved. However, three minor issues still need to be addressed.  
+
+Minor concerns:  
+
+Minor concerns:1. Results, section: Multimodal Cell Identification with TACIT: "... capturing both antibody intensities from PCF and count values from Xenium (Figs. 6b, c)." Should it be "Figs. 6b, 6c" for consistency?2. Figure 6i, Legend: "B Cells::Cell State (+)". There seems to be a duplicated colon ("."). This appears to be the same issue as in the legend of Extended Data 9b.3. Methods, Segmented Regression Model: "This was determined by the lowest Akaike Information Criterion (AIC) score achieved among the three models the three models...". The phrase "the three models" is repeated. Please revise for clarity.  
+
+(Remarks on code availability)  
+
+The code appears to be well- written and meets the standards for reproducibility and usability. Below are the detailed points of evaluation:  
+
+Reproducibility:  
+
+The results of the paper are largely reproducible using the provided code. The implementation is consistent with the methodology described in the paper, and the code includes all necessary scripts to reproduce the key findings.  
+
+Usability:  
+
+The repository includes a comprehensive README file that provides clear instructions for installation and running the application. It outlines the required dependencies, setup steps, and example commands for running experiments, which is helpful.  
+
+The repository structure is logical and user- friendly, allowing users to locate files and resources without confusion.  
+
+Installation and Execution:  
+
+The installation process is straightforward, and all dependencies are either listed in a requirements file or documented explicitly. I was able to install the necessary packages and execute the code without encountering significant issues. Sample datasets and configurations are provided, enabling users to quickly test the implementation without needing to prepare custom inputs initially.  
+
+Community Usability:  
+
+The code serves as a valuable resource for the community, particularly due to its clear structure and adherence to reproducibility practices.  
+
+Overall, the code is functional, well- documented, and reproducible, making it a reliable and usable resource for researchers and practitioners in the field.  
+
+Reviewer #3  
+
+(Remarks to the Author)  
+
+We appreciate the authors' detailed and thoughtful responses, which have addressed most of the primary concerns raised in our initial review. The comprehensive explanation of TACIT's marker panel selection pipeline, the adaptive approach to clean cell identification effectively clarify its strengths and address key questions about its applicability and reliability. Additionally, the clarification of TACIT's performance across diverse datasets and its computational scalability underscore its promise as a versatile tool for spatial omics analyses.  
+
+While the authors' responses have resolved many issues, there remain two areas where further clarification or exploration would enhance the robustness and generalizability of TACIT:  
+
+Benchmarking Against Additional Methods: While TACIT has been compared to Louvain clustering, additional benchmarking against other widely used or emerging tools in spatial omics, such as Spatial- ID or Cell2location, would provide a broader context for its strengths and limitations. We noticed that the authors have provided additional comparative analyses in their replies to Reviewer #1. However, the results of these comparisons are crucial to fully appreciating TACIT's
+
+<--- Page Split --->
+
+performance relative to existing tools. As such, the authors should consider providing these benchmarking results directly in the main text of the manuscript rather than relegating them to supplementary figures.  
+
+Segmentation Dependencies: As the authors mentioned in the replies, the reliance on accurate segmentation masks remains a critical dependency for TACIT's performance. Although the authors propose strategies to mitigate errors, such as using a human- in- the- loop Cellpose3 model, further validation is needed to evaluate how segmentation inaccuracies affect cell type annotation, for example in highly complex or crowded regions like tertiary lymphoid structures (TLS). This dependency may limit TACIT's applicability in cases where segmentation tools underperform or lack sufficient training for the tissue type in question.  
+
+Addressing these points would not only strengthen the manuscript but also provide the community with greater confidence in TACIT's utility across a wide range of spatial omics applications. Thank you for your continued efforts in refining and improving this promising tool.  
+
+(Remarks on code availability) Codes are not fully released.  
+
+Version 2:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author) I thank the authors for their additional comparisons and believe they have fully addressed my concerns.  
+
+(Remarks on code availability) The code is monolithic and filled with global variables which may cause issues with future maintenance, but does install and run.  
+
+Reviewer #3  
+
+(Remarks to the Author) The authors answered all my concerns and I have no more questions.  
+
+(Remarks on code availability)  
+
+Open Access This Peer Review 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
+
+<--- Page Split --->
+
+credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.  
+
+In cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source. The images or other third party material in this Peer Review 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 https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+
+## Contents of point-by-point responses to review comments for NCOMMS-24-38058-T  
+
+Reviewer #1: Page 2Reviewer #2: Page 10Reviewer #3: Page 19  
+
+We sincerely thank the reviewers for their insightful and constructive feedback on our manuscript. Their detailed comments have provided us with valuable guidance to improve both the clarity and scientific rigor of our work. In response, we have carefully considered each point and made significant revisions to address the concerns raised. These changes, along with additional analyses, have strengthened the manuscript and better highlight the contributions of TACIT in the context of cell type annotation. Below, we outline our responses to the reviewers' comments, describing the revisions made and the new insights gained because of their suggestions. We hope these revisions meet the reviewers' expectations and enhance the overall quality of our work.  
+
+Below is a summary of major newly added results in the manuscript:  
+
+Figure 1: Moved the threshold derivation for cell type relevance section right after the Figure 1e. Re- numbered all the subfigures to enhance clarity. Figures 2d and 2j: Replaced the original average Recall, precision, and F1 scores with the distribution of scores of individual samples. Figure 4a: changed PECAM to PECAM1 Extended Data 4: Relocated the original Extended Data 4 to Extended Data 6. Added benchmarking studies comparing TACIT with a range of existing algorithms for cell type annotation using both spatial proteomics datasets (PCF- CRC, PCF- HI) and spatial transcriptomics datasets (MERFISH). Additionally, included the performance of Louvain and Leiden clustering in terms of entropy and purity scores across a range of resolution levels (from 0.1 to 1.9). Time comparisons are also included in these panels. Extended Data 5: Moved the original Extended Data 5 to Extended Data 7. Added UMAP visualization of scRNA- seq and Xenium Annotations to explain the challenges associated with Xenium data in comparison to scRNA- seq data. RMarkdown Guide: Added a step- by- step tutorial on how to run TACIT using the PCF- CRC dataset.
+
+<--- Page Split --->
+
+## Reviewer #1 (Remarks to the Author):  
+
+Huynh et al. present Threshold- based Assignment of Cell Types from Multiplexed Imaging Data (TACIT), a method for annotating cell types based on known cell marker signatures and expression profiles. TACIT is an unsupervised algorithm built on Louvain clustering followed by fitting a "Cell Type Relevance score" regression model to best separate cells into discrete cell types. The authors compare TACIT to select similar algorithms and compare across technologies including proteomic PhenoCycler and transcriptomic MERFISH and Xenium and tissue types including colorectal cancer, healthy intestine, and salivary glands with graft- versus- host disease. Overall, the manuscript was adequately written and clearly described the method and offered a favorable view of the algorithm compared to others. However, there are some major points which hamper enthusiasm for the report. One of the major issues is the benchmarking, which appears to compare algorithms in an unfair manner, especially the Louvain algorithm. Additionally, the quantitative results may need to be refined based on the lack of a ground truth, perhaps using, for example, alternative Silhouette- based measures.  
+
+## Question: How are values chosen for relevance of markers defining cell types? Is this arbitrary?  
+
+Response: The selection of markers for defining cell types is not arbitrary; it follows a systematic process guided by biological knowledge and empirical data. For multiplex IHC, such as PhenoCycler, the panel is designed by selecting markers that capture specific cell types of interest. For example, proteins or genes like CD4 and CD8 are well- established markers for helper and cytotoxic T cells, respectively, making them ideal in immunological studies. For single- cell spatial transcriptomics data, such as MERFISH and Xenium, panel design is often informed by companion scRNA- seq data, where genes with high potential to distinguish cell types are selected as markers. This selection is further validated using domain- specific literature, databases, and scRNA- seq datasets to ensure accuracy. Taking a predefined list of markers for each cell type as input, TACIT computes a linear combination of these markers, assuming equal weight of relevance, to create an aggregated cell type relevance score. This ensures that all signature markers contribute collectively to the final score, minimizing the risk of overemphasizing individual markers. This approach reinforces the overall identity of each cell type and reduces the likelihood of misclassification based on the expression of a single marker.  
+
+Question: TACIT attempts to cluster highly homogenous cell communities by choosing a Louvain resolution to contain small numbers of cells. There are two issues with this assumption. First, Louvain clusters are highly connected which is not necessarily true which led to the development of the Leiden method. Second, if the assumption is that the cells are homogeneous then the cutoff should not be cluster size but rather cell similarity within clusters by using something akin to Silhouette.  
+
+Response: Thank you for your thoughtful feedback. In response to your points:  
+
+Choice of Clustering Algorithm: We have evaluated both the Louvain and Leiden algorithms within TACIT across multiple datasets used in the paper. Our results show that TACIT with the Leiden algorithm achieved very similar accuracy when compared with the Louvain algorithm (Figure R1). Given these results, either clustering method can be considered effective within TACIT, depending on specific requirements and context.
+
+<--- Page Split --->
+![](images/Figure_unknown_0.jpg)
+
+<center>Figure R1: Overall accuracy comparison of the effects of Leiden and Louvain clustering to TACIT performance in PCF-CRC, PCF-HI, and MERFISH. </center>  
+
+We agree that capturing homogeneous cell population should be the goal. Unfortunately, there doesn't exist a parameter directly linked to similarity we can set for current off- the- shelf clustering algorithm. While silhouette score is a great measure for similarity, computing the silhouette score is computationally prohibitive for the large datasets consisting of hundreds of thousands of cells or more, introducing a bottleneck in our computation. Thus, we are trying to leverage the resolution parameter assuming higher resolution leading to intra- cluster higher homogeneity. We have conducted experimented to assess the accuracy as a function of varying resolutions and observed that the accuracy converged as resolution increases to a certain point (Extended Data 2e- g).  
+
+![](images/Figure_unknown_1.jpg)
+  
+
+Extended Data 2 in the original paper: (e- g) Various resolution levels were tested to assess their impact on the performance of TACIT in PCF- CRC. Higher resolution levels, which correspond to an increased number of MicroClusters, showed a positive correlation with recall values, particularly for rare cell types that constitute less than \(1\%\) of the data. This enhancement in recall is crucial for accurately identifying and characterizing rare cell populations.  
+
+Question: TACIT learns a threshold to separate positive signals and background noise - does this hold true for highly homogeneous data? What about heterogeneous data of the same cell type such as a tumor?  
+
+Response: Theoretically, if all signals are highly homogeneous, TACIT may treat all signals as background, or as a single class, since the data would provide no intrinsic distinction between positive signals and noise. It is difficult to call positive signals without a reliable reference.
+
+<--- Page Split --->
+
+Fortunately, in real- world applications, where each slide typically contains around half a million cells, we expect to see a mixture of different cell populations. In these cases, variations in marker intensity will correlate with the positivity of specific cell types, allowing TACIT to effectively differentiate signals as demonstrated in the series of experiments reported in the paper.  
+
+The PCF- CRC data used in the paper represents colon cancer patient with tumor data with varies cancer stages (Figure 2 in the manuscript)1.  
+
+In the case of heterogeneous data from the same cell type, the ability to identify different subtypes depend heavily on the specific cell markers being profiled. As shown across multiple experiments in the paper, TACIT can identify immune subtypes with appropriate markers. Similarly, if tumor subtyping markers are included, TACIT can distinguish between various tumor subtypes.  
+
+Question: For the benchmarking, as there is no real ground truth, the authors should also include an analysis with a measure quantifying feature similarity such as entropy, purity, or Silhouette.  
+
+Response: Thank you for your suggestion. In response, we conducted additional experiments to benchmark TACIT against the Leiden and Louvain clustering algorithms across various resolution settings. We employed the metrics recommended by the reviewer—Purity and Entropy—to assess feature similarity. These analyses were performed on the PCF- CRC. The results are presented in Figure R2 below. While we also considered using the Silhouette score, it was infeasible due to the high memory demands associated with calculating pairwise distances between cells. Our dataset includes between \(\sim 250,000\) and 2,000,000 cells, resulting in a matrix too large to efficiently store and process.  
+
+In Figure R2a, the graph illustrates that both Leiden and Louvain methods demonstrate an increase in the number of clusters at different rates as the resolution increases. The Leiden method consistently identifies more clusters than the Louvain method at the same resolution. Our experiment stops at the resolution when the number of clusters reach over 40 or 50, significantly higher than the number of expected cell populations, which is 18 in this case.  
+
+TACIT demonstrates significantly lower average entropy scores with 18 clusters compared to both Louvain and Leiden across all tested resolutions (Figure R2b). This suggests that TACIT clusters are more homogeneous, while Louvain and Leiden clusters often contain a mixture of cell types. Furthermore, TACIT achieves significantly higher purity scores than Louvain and Leiden (Figure R2c). Louvain and Leiden's purity scores initially increase with fewer clusters but decrease at higher resolutions. This decrease in purity at higher resolutions is likely due to over- fragmentation, where clusters become smaller and more likely to contain a mix of cell types by chance.  
+
+![](images/Figure_unknown_2.jpg)
+
+
+<--- Page Split --->
+
+Figure R2: (a) Number of clusters across different resolutions (0.1–1.9) for the Leiden and Louvain methods. The red dotted line indicates the expected number of cell types. (b) Comparison of average entropy score at varying resolutions using Leiden and Louvain. (c) Comparison of purity score at different resolutions using Leiden and Louvain.  
+
+Question: How sensitive is TACIT to shifting parameters? For example, how much does varying the resolution of the initial Louvain clustering change the results? Any other parameters should be checked as well.  
+
+Response: Thank you for your question. We provide detailed results in Extended Data Figure 2 (Copied below), which includes experimental setups to test the sensitivity of TACIT to the initial resolution parameter: perform multiple resolutions using \(100\%\) of the PCF- CRC dataset.  
+
+Using the full dataset ( \(100\%\) ) at varying resolutions, we found that higher resolutions to a certain extent enhanced the accuracy for both rare and abundant cell types. The finer clustering allowed better detection of rare cell types, leading to improved recall. As we continued to increase the resolution, the accuracy in terms of the recall, precision and F1 score stabilized as the resolution of Louvain clustering reaches 50 (Extended Data Figures 2e- g), suggesting an optimal resolution to balanced accuracy and model complexity.  
+
+![](images/Figure_unknown_3.jpg)
+  
+
+Extended Data 2: (e- g) Various resolution levels were tested to assess their impact on the performance of TACIT. Higher resolution levels, which correspond to an increased number of MicroClusters, showed a positive correlation with recall values, particularly for rare cell types that constitute less than \(1\%\) of the data when resolution is less than 50. This enhancement in recall is crucial for accurately identifying and characterizing rare cell populations.  
+
+Question: CELESTA and SCINA were not included in the PCF- HI dataset comparison due to too many "Other" annotations. However, algorithms have customizable parameters to account for this for example, SCINA has an "allow_unknown" parameter that can be set to 0 if this is too much of an issue to limit comparison.  
+
+Response: Thank you for your question. Indeed, SCINA provides an "allow_unknown" parameter that can be set to exclude uncertain cell classifications. This is particularly useful when there is a need to minimize "Other" annotations in datasets with a significant number of unclassifiable cells. The authors of SCINA recommend enabling this setting when rare cell types are not a primary focus<sup>2</sup>.  
+
+In response to the reviewer's advice, we ran SCINA with the "allow_unknown" parameter set to "False" to limit the uncertain classifications. The result in comparison to other methods is shown in Figure R3. SCINA achieved a very low accuracy of 0.41 with a predominance of the
+
+<--- Page Split --->
+
+'Enterocyte' label and notably missing several important cell types such as "NK", "Paneth", "Goblet", "Neuroendocrine", "M2 Macrophage", and "CD7+ Immune". This suggests that SCINA has challenges in identifying diverse and rare cell types even with known markers.  
+
+For CELESTA, there doesn't exist a simple approach/parameter one can fine tune to lower the amount of unassigned population. Theoretically, CELESTA may be able to lower the number of unknown cells by adjusting a large number of parameters ( \(>100\) parameters with 25 cell types). This is because there are four thresholds needed to be specified for each cell type:  
+
+high_expression_threshold_anchor, low_expression_threshold_anchor, high_expression_threshold_index, low_expression_threshold_index.  
+
+However, there is no guidance available to tune these parameters. Given that the PCF- HI dataset contains 25 cell types, this involves setting 100 parameters which is infeasible to tune. Due to these limitations, we opted not to include CELESTA in the comparison for this dataset.  
+
+![](images/Figure_unknown_4.jpg)
+
+<center>Figure R3: (a) Proportion of cell type predictions between TACIT and SCINA (do not allow unknown cells). (b) Overall accuracy comparison between SCINA (do not allow unknown cells) and TACIT in the PCF-HI datasets. </center>  
+
+Question: The authors mention significantly higher recall, precision, and F1 scores, but there is no listed p-value at that point in the text. A solution to this would be to split the data by sample and have a score for each sample to compare.  
+
+Response: Thank you for pointing out the absence of listed p- values for the reported recall, precision, and F1 scores. In our analysis presented in Figure R4, we split the data by sample to calculate these scores separately for each sample. This allowed us to use statistical tests, specifically the Wilcoxon signed- rank test, to compare TACIT against existing methods for each sample. The results shown in Figure below confirmed that TACIT consistently achieved significantly higher scores (p- value \(< 0.0005\) ).
+
+<--- Page Split --->
+![](images/Figure_unknown_5.jpg)
+
+<center>Figure R4: Recall, Precision, and F1 scores by sample, comparing the PCF-CRC (140 samples) and PCF-HI (64 samples) datasets across four methods: TACIT, CELESTA, Louvain, and SCINA. </center>  
+
+Question: Why was TACIT always compared with Louvain following the benchmark? Especially when it seems like CELESTA was better performing than their method for Louvain.  
+
+Response: Thank you for your question. We chose to compare TACIT with the Louvain method because Louvain is a representative of the category of unsupervised clustering approach that is widely adopted for analyzing both spatial proteomics and spatial transcriptomics data. It is embedded in the popularly adopted Seurat and Scanpy pipelines as well as the commercial pipelines as the go- to algorithm for cell annotation. However, CELESTA is specifically tailored for spatial proteomics due to its reliance on continuous data inputs and its use of a Gaussian mixture model for normal distribution fitting. Spatial transcriptomics data, being count- based, doesn't align as well with CELESTA's underlying assumptions and methodologies. In our paper, we included evaluations of CELESTA on both proteomics datasets—colorectal cancer and human intestine, as shown in Figure 2 and Extended Figure 1. Due to these considerations, for a broader and more consistent benchmark across different types of spatial data, Louvain was chosen as the primary comparison point for TACIT.  
+
+Question: Minor: Figure 1e "i" layout is confusing and should be clearer in the figure or have a better notation.  
+
+Response: Thank you. We have made adjustments based on your comments  
+
+There are typos in the manuscript (e.g. pg. 8: "transcriptomics, providing") Page 9: "release, highlighting" -> "release, highlighting"  
+
+Response: Thank you. We have made adjustments based on your comments  
+
+Question: The authors compared Louvain with their algorithm based on the UMAP embedding, which is highly controversial and not recommended as UMAP is known to destroy distances between observations. Furthermore, they most likely did not use a resolution on par with their own algorithm so the comparison may not be fair. Lastly, they only looked at the top 5 markers instead of all the a priori known markers to define each cluster, unlike their own algorithm. In all, I am not entirely convinced that the Louvain method they compared to was fair, and if corrected, how much TACIT would add to this existing pipeline.
+
+<--- Page Split --->
+
+Response: Thank you for your valuable insights. To clarify, in our benchmarking, Louvain clustering was performed on the PCA embeddings of the original data. UMAP was only used for visualization purposes. Thus, the core of our analysis remains independent of UMAP's limitations.  
+
+Both TACIT and Louvain clustering were provided with the same set of markers to identify cell types. While TACIT directly assigns cell identities using the given markers, Louvain simply groups cells without assigning labels. To determine cell types in Louvain clusters, researchers must manually analyze the genes expressed in each group by first identifying a ranked list of differentially expressed genes in each cluster and then match these genes to the known cell type markers using the top ranked genes, which is not limited to the top 5 especially when there is no match of marker genes with the top 5.  
+
+Regarding resolution, TACIT starts with a high- resolution Louvain clustering to generate a large number of microclusters, but it ultimately assigns cells into a predefined set of cell populations including rare populations. Using the same high resolution is unrealistic for Louvain clustering because it is infeasible to manually annotate hundreds of clusters manually. Thus, comparing TACIT and Louvain with the same resolution is inappropriate.  
+
+The challenge with Louvain clustering or any unsupervised clustering is that often times there does not exist a magic clustering resolution or number of clusters when the variety of cell populations can be easily identified as expected, especially when there is a mixture of rare and dominant cell populations. The cell populations are often dictated by how good the clustering algorithm is. Often times, users have to try different resolutions to approach an acceptable clustering result. They may have to subset some large clusters and further divide them by using clustering algorithm. Our experiment with experienced bioinformatician with lots of known biological knowledge shows that this step often takes at least 12 hours to annotate one dataset.  
+
+Thus, unlike clustering based approach, TACIT is fully automatic and scalable. It takes the predefined set of cell type markers as input and generate the cell populations with high enrichment of these markers. There is no need for human intervention at any step. It is designed to work seamlessly with both spatial proteomics and transcriptomics datasets, making it highly adaptable for various spatial omics applications. Its key strength lies in its versatility in handling both data types, offering scalability and high accuracy in cell type annotation. This accuracy is particularly notable for identifying both rare and abundant cell types, consistently outperforming existing cell annotation method even with scRNA- seq transfer methods. TACIT operates without the need for training data or labeled scRNA- seq data, significantly reducing manual intervention. Its automated workflow speeds up the analysis process, enabling fast and efficient cell type annotation without sacrificing precision. These combined features make TACIT a robust and efficient tool for spatial omics, with clear advantages over traditional methods like Louvain. For example, analyzing a typical PCF dataset with a 48- marker panel across four tissues from the intestine would typically take at least 25 hours of manual work for clustering, merging, and cell type assignment<sup>3</sup>. In contrast, TACIT's scalability allows it to handle much larger datasets, such as the 64 tissues in the PCF- HI dataset within one hour, with greater efficiency and accuracy, while also reducing manual effort.  
+
+Reviewer #1 (Remarks on code availability): The code is deposited on Code Ocean, although I could not review the GitHub repository as it was private. The authors present the code on Code Ocean as a large monolithic file and not all functions are documented, this could be improved. Furthermore, there appears to be no README, tutorial, or vignette, hampering transparency, reproducibility, and ease
+
+<--- Page Split --->
+
+of- use. These would absolutely need to be added before moving forward with any publication. Running the code with a reproducible run did not appear to successfully output anything, perhaps better documentation for this run would help as well.  
+
+Response: Thank you for your feedback regarding the availability of our code. We apologize for the inconvenience caused by the private GitHub repository. We plan to make the GitHub repository publicly accessible upon acceptance of our paper. The repository will include a README, tutorial, and vignette to guide users on how to utilize the functions. Additionally, we included the TACIT vignette (using PCF- CRC) as an example in TACIT_Vignette.pdf files.
+
+<--- Page Split --->
+
+## Reviewer #2 (Remarks to the Author):  
+
+Comments to NCOMMS- 24- 38058- T.  
+
+The author developed TACIT, an unsupervised algorithm for cell annotation in spatial transcriptomics and proteomics. Benchmark analysis showed that TACIT outperformed existing methods in accuracy and scalability when tested on five datasets. The authors further revealed new phenotypes in inflammatory gland diseases and identified under- and overrepresented immune cell types through the integration of spatial transcriptomics and proteomics. However, substantial revisions are still required to enhance the clarity, robustness, and impact of your algorithm and findings.  
+
+Question: The TACIT method appears to be heavily constrained by the signature. Is there a metric to evaluate the efficiency of markers in distinguishing cell types, thereby assisting in the design of a marker panel? What if the markers are insufficient for differentiation?  
+
+Response: Thank you for your question. In spatial proteomics studies, researchers have a set of markers that are well- documented in literature and have been validated in previous studies. Markers in spatial transcriptomics data such as Xenium, MERFISH and CosMx are often guided by the associated scRNA- seq datasets and validated by well- documented literature, which have much higher resolution in cell type identification. From there, subsets of genes with appropriate expression levels that can uniquely identify each cell type of interest are carefully selected during the design of the panel. Cytomarker (https://cytomarker.ai/) is a recently developed tool that can assist in the identification of efficient markers for panel design. In the case when markers are insufficient for differentiation, alternative markers will need to be identified to include in future panels.  
+
+Question: In scenarios where signatures are subject to noise, an investigation into the robustness of the TACIT method is warranted, particularly considering the addition or subtraction of markers. Moreover, determining the minimum number of markers required for accurate cell type recognition would be beneficial.  
+
+Response: This is a great question. To evaluate the robustness of the TACIT method in the presence of noise, we tested it using the MERFISH spatial transcriptomics dataset, which contains (300 genes with 9 cell types and 5.6 markers per cell type). We conducted simulations by either randomly adding 1, 2, or 5 genes to the cell type signatures or randomly subtracting 1, 2, and 5 genes from each marker (Figure R5a). Each simulation was repeated 5 times. We found that adding 1 or 2 irrelevant genes had a negligible impact on accuracy compared to the baseline, but adding 5 irrelevant genes led to a noticeable drop in accuracy, with the mean score decreasing to 0.8. Conversely, subtracting 1 gene did not significantly affect accuracy, but subtracting 2 or 5 genes resulted in a substantial drop, with accuracy falling to around 0.4 (Figure R5b).  
+
+The drop in accuracy can be attributed to the distribution of cell type signatures in the dataset. For instance, the signatures for Astrocytes and Inhibitory cells each have only 2 markers but are critical since these cell types collectively make up approximately \(50\%\) of the dataset (Astrocyte: \(12.41\%\) , Inhibitory: \(37.06\%\) ). If we subtract 2 markers from the signature list, these cell types are effectively excluded, leading to inaccuracies in identifying or labeling them. With only 2 markers available, removing 2 results in no markers left for these cell types, making them unidentifiable. Similarly, subtracting 5 markers further exacerbates this issue. For other cell types such as Endothelial (6 markers), Ependymal (8 markers), and OD Mature (9 markers), the impact is less
+
+<--- Page Split --->
+
+severe. Therefore, maintaining an adequate number of markers is essential for accurate cell type recognition.  
+
+Our results on real datasets suggest that at least three markers per cell type are required for spatial proteomics, and at least five markers per cell type for spatial transcriptomics, to ensure accurate identification and minimize the risk of significant performance degradation caused by marker variability.  
+
+![](images/Figure_unknown_6.jpg)
+
+<center>Figure R5: (a) Boxplot showing the number of genes per cell type for each simulation. (b) Boxplot of accuracy scores (5 bootstraps) for each simulation. </center>  
+
+Question: The performance of the Louvain method critically depends on clustering resolution, and misclassifications could potentially stem from clustering inaccuracies. When evaluating, would changes in the resolution of the Louvain method help improve performance? Also, it would be insightful to evaluate the outcomes when utilizing preexisting cluster assignments (if available) directly.  
+
+Response: Thank you for your question. We conducted additional experiment running both Louvain and Leiden under different resolutions using PCF- CRC dataset with existing annotation for each cell. We used both entropy and purity score to evaluate the homogeneity of the clusters. In Figure R2a, the graph shows that both the Leiden and Louvain methods increase the number of clusters at different rates as the resolution is raised. TACIT achieves significantly lower average entropy scores compared to both Louvain and Leiden across all tested resolutions (Figure R2b). High entropy values for Louvain and Leiden suggest that their clusters contain a high mixture of cells from different populations, indicating reduced homogeneity. These findings are further supported by the purity scores in Figure R2c, where TACIT achieves significantly higher purity compared to Louvain and Leiden.
+
+<--- Page Split --->
+![](images/Figure_unknown_7.jpg)
+
+<center>Figure R2: (a) Number of clusters across different resolutions (0.1-1.9) for the Leiden and Louvain methods. The red dotted line indicates the expected number of cell types. (b) Comparison of average entropy score at varying resolutions using Leiden and Louvain. (c) Comparison of purity score at different resolutions using Leiden and Louvain. </center>  
+
+Question: It is imperative to conduct a comparative analysis of TACIT against contemporary methodologies, notably those employing deep learning frameworks such as SPACEL, and emergent single-cell RNA sequencing (scRNA- seq) techniques exemplified by ACT.  
+
+Response: To address this question, we reviewed recent publications on cell type annotation methods for spatial proteomics and transcriptomics. We now included accuracy scores from the TYPEx paper for spatial proteomics (PCF- CRC and PCF- HI) and the Spatial- ID paper for spatial transcriptomics (MERFISH)4,5. Using the same datasets and cell IDs as in the published papers, we compared TACIT with existing models incorporating machine learning, statistical methods (e.g., Gaussian mixture model, Markov chain), deep learning, and scRNA- seq transfer across shared datasets like PCF- CRC, PCF- HI, and MERFISH. We excluded SPACEL from the comparison, as its focus is on identifying tissue structures rather than cell types6.  
+
+From spatial proteomics comparisons (Figure R6a), TACIT consistently achieved high F1 scores without requiring training data or reference scRNA- seq datasets. In human intestine and colorectal cancer datasets, TACIT outperformed models like TYPEx (statistical model)4, CELESTA (statistical model)7, Astir (deep neural network)8, and STELLAR (graph convolutional neural network)3, with TACIT achieving an F1 score of 0.75 compared to their scores of 0.45 to 0.6. Although STELLAR achieved an F1 score of 0.8 on the PCF- HI dataset, this was obtained through cross- validation, where the model was trained on three patients and tested on four, highlighting its reliance on patient- specific training data, in contrast to TACIT's unsupervised approach (Figure R6a).  
+
+From spatial transcriptomics comparisons on MERFISH dataset(Figure R6b), we included deep learning models such as Tangram9 and Spatial- ID5 which are trained on scRNA- seq data and transfer them to spatial transcriptomics as well as cell2location10 uses Negative Binomial regression. Additionally, we also include machine learning models such as ScNym11 and SciBet12. Our experiment shows that all methods do not achieve the same level of accuracy as TACIT (Figure R7b). Additionally, all methods except TACIT in this comparison require extensive computational resources and some requires extensive training data such as well classified scRNA- seq data.
+
+<--- Page Split --->
+
+Additionally, we evaluated the ACT (Annotation of Cell Types) method, which showed an accuracy of \(65.58\%\) using MERFISH data (see ACT_celltype.csv and ACT_Annotation_results_top1.txt), compared to TACIT's accuracy of \(87\%\) . It is worth noting that ACT required selecting more than the top 10 genes as specified in the original paper, so our comparison was limited to spatial transcriptomics data.  
+
+In summary, TACIT stands out by achieving high accuracy without the need for training data or reference scRNA- seq datasets, making it highly effective in scenarios where such data is unavailable. The accuracy metrics for other methods were gathered from their respective publications with best reported performance, highlighting TACIT's performance advantages.  
+
+![](images/Figure_unknown_8.jpg)
+
+<center>Figure R6: (a) Bar plot comparing F1 scores of recent cell type annotation methods for spatial proteomics, including TACIT, Astrir, CELESTA (from the Spatial-ID paper), STELLAR, and TYPEx, across the PCF-HI and PCF-CRC datasets. (b) Bar plot comparing F1 scores of TACIT with recently published cell type annotation methods for spatial transcriptomics data using MERFISH dataset. </center>  
+
+Question: Certain cell types had very low F1 scores. A detailed inquiry is warranted to elucidate which cell type exhibits the peak predictive accuracy and conversely, which demonstrates the nadir, along with an explanation grounded in the underlying characteristics or limitations of the dataset or methodology.  
+
+Response: Cell types with low F1 scores (below \(20\%\) ) in the CRC dataset include immune cells ( \(\mathrm{F1} = 0.002\) , \(\mathrm{n} = 3127\) cells, \(1.3\%\) ), CD11c+ DCs ( \(\mathrm{F1} = 0.12\) , \(\mathrm{n} = 400\) cells, \(0.2\%\) ), and NK cells ( \(\mathrm{F1} = 0.19\) , \(\mathrm{n} = 323\) cells, \(0.1\%\) ). These rare cell types, due to their small population sizes, pose challenges for manual annotation. The immune cell category was broadly labeled as "other immune," encompassing cells expressing only CD45, without further specificity in the original ground truth.  
+
+Figure R7 presents heatmaps depicting the mean expression levels of protein markers across TACIT- predicted cell types using PCF- CRC (Figure R7a) and PCF- HI (Figure R7b). Rows represent distinct cell types, while columns represent various protein markers. The heatmaps are divided into two sections: matched cells (left) and unmatched cells (right) compared to the reference. Notably, unmatched cells display similar marker signatures to matched cells, suggesting that TACIT effectively captures their unique expression patterns. For instance, CD11c+ DCs in the unmatched group still exhibit high CD11c expression, while NK cells show light red CD56 expression. However, in the unmatched CD11c+ DCs group, there is also Granzyme B expression, which is not a typical marker for CD11c+ DCs. This combination of CD11c and Granzyme B suggests possible segmentation noise in neighboring cells. Improvement
+
+<--- Page Split --->
+
+on cell segmentation and Integrating cell morphological information could provide additional context and improve accuracy for these and other cell types.  
+
+![](images/Figure_unknown_9.jpg)
+
+<center>Figure R7: (a) Heatmap of the mean expression of protein markers in TACIT predictions for the PCF-CRC dataset, displaying cells matched with the reference (left) and cells not matched with the reference (right). (b) Heatmap of the mean expression of protein markers in TACIT predictions for the PCF-HI dataset, showing cells matched with the reference (left) and cells not matched with the reference (right). </center>  
+
+## Question: When performing MC clustering, how does one determine the relationship between resolution and the average number of cells per cluster (0.1% to 0.5%)?  
+
+Response: To determine the relationship between resolution and the average number of cells per cluster in terms of the proportion of total cells (0.1% to 0.5%), we conducted clustering at varying resolutions and calculated the proportion of cells per cluster relative to the total number of cells for different datasets. The results showed a clear trend: as the resolution increases, the average number of cells per cluster decreases (Figure R8). Specifically, the proportion of cells per cluster ranged from approximately 0.93% at a resolution of 0.5 to about 0.18% at a resolution of 10.0.  
+
+We usually start with the default resolution of 5 for clustering. At this resolution, the average number of cells per cluster is approximately 0.27% of the total cell population, which falls comfortably within the desired range of 0.1% to 0.5%. The resolution will be adjusted accordingly if the average number of cells per cluster falls out of the expected range.
+
+<--- Page Split --->
+![](images/Figure_unknown_10.jpg)
+
+<center>Figure R8: Average proportion of cells per cluster at each resolution. </center>  
+
+Question: Results > Section 3: An investigation into the disparity of outcomes when leveraging the top 5 markers as signatures versus established markers is essential. This comparison will shed light on the potential advantages or shortcomings of each strategy.  
+
+Response: To clarify, the top 5 most prominently expressed markers derived from scRNA- seq datasets in used in combination with established markers for the cell type identification process in this experiment. Because transcriptomic markers can be tissue- specific, location- specific and even disease- specific, existing knowledge regarding established markers on a specific tissue type is very much incomplete, diminishing the accuracy of cell type annotation. Thus, leveraging the highly enriched genes revealed in scRNA- seq in the same tissue condition allows us to use more relevant but previously unknown biomarkers to improve cell type annotation.  
+
+Question: Results > Section 3: While scRNA- seq has successfully identified cell subtypes, subsequent application of Seurat's transfer method fails to recapitulate these distinctions. An analysis is required to uncover the cause of this discrepancy.  
+
+Response: The discrepancy arises primarily from the fundamental differences between scRNA- seq and spatial data modalities. scRNA- seq typically involves around 20,000 genes, providing a rich dataset that captures extensive biological information, making it easier to distinguish and cluster cell subtypes effectively. This is evident in the UMAP plot from scRNA- seq data, where distinct clusters are clearly separated, as shown by the distinct groups, such as the acinar cells highlighted in blue within the red circle, which are differentiated into three distinct clusters (Figure R9 left).  
+
+In contrast, spatial data such as spatial proteomics and spatial transcriptomics (e.g., MERFISH or Xenium) are inherently sparse and have significantly fewer markers—around 56 for proteomics and approximately 300 for spatial transcriptomics. This reduced marker set limits the amount of information that can be captured, making clustering more challenging. The sparsity and lower
+
+<--- Page Split --->
+
+feature count in spatial data often result in a less distinct separation of cell subtypes, leading to a tendency for clusters to merge or overlap, as observed in the UMAP plot for Xenium data. As shown in the UMAP plot in Figure R9 right, the same cell types that were distinct in scRNA- seq are more intermingled, reflecting the limitations of clustering when using spatial omics data.  
+
+These inherent differences underscore the challenges of directly transferring clustering results from scRNA- seq to spatial datasets. The scRNA- seq data provides a more granular resolution of cell types, while spatial datasets, due to their lower dimensionality and sparsity, often fail to reproduce these fine distinctions without additional analytical adjustments.  
+
+![](images/Figure_unknown_11.jpg)
+
+<center>Figure R9: UMAP of scRNA-seq with cell type expert manual annotations (left) and Xenium (spatial transcriptomics) with Seurat transfer annotations (right). </center>  
+
+Question: The analytical focus on cell states appears cursory and lacks specificity. Moreover, the identification of cell states is not inherently a feature of the TACIT methodology. In light of these considerations, it may be prudent to reassess the inclusion of 'cell state' in the title to accurately reflect the scope and findings of the study.  
+
+Response: Thank you for your insightful critique. We would like to clarify the role of cell states within our TACIT framework and why its inclusion in the title is integral to the methodology and the study's findings. As we said before, TACIT works by first defining a signature set of markers for each cell type, which is then used to calculate a cell type score through a linear combination of those markers. This method assumes equal weight and relevance across markers, ensuring that no single marker dominates the classification and resulting in a more accurate representation of the cell type. In this way, TACIT minimizes the risk of misclassification while reinforcing the overall identity of the cell type. Regarding cell states, TACIT offers the flexibility to include specific cell states of interest (e.g., exhausted T cell subsets) as part of the overall cell type signature. This allows the user to define relevant cell types and modifiable cell states, adding them to the signature if desired. This flexibility is a key feature of TACIT, as the user has control over the level of annotation—whether it's identifying distinct cell types or incorporating more nuanced states and spatial information, such as intratumoral or peritumoral immune infiltrates. The iterative nature of
+
+<--- Page Split --->
+
+TACIT, where the human- in- the- loop refines biological identities, states, and spatial relevance in rounds, is what makes the inclusion of 'cell state' in the title both appropriate and reflective of the study's scope. This process is crucial for testing and refining hypotheses, making it an essential part of the analysis rather than a cursory addition. We hope this addresses your concerns and demonstrates why cell states are a critical aspect of our methodology and our study's scope.  
+
+Question: Results > Section 6: The consistency between the results of spatial proteomics and transcriptomics data is not high. What are the reasons for poor consistency? Which one should be trusted more, or is it necessary to use matched markers and integrate multi- omics data? Due to poor consistency, it is recommended to visualize markers and directly compare the visualization results of markers.  
+
+Response: Thank you for raising this important question about the consistency between spatial proteomics and transcriptomics data. It is well understood that protein and transcript expression levels often do not correlate perfectly, and this is a common phenomenon observed in biological systems, including human cells<sup>14,15</sup>. This discrepancy can be attributed to post- transcriptional regulation, protein degradation rates, and other cellular processes that control protein abundance independently of mRNA levels. Additionally, the differences in methodology between spatial proteomics and spatial transcriptomics further contribute to this inconsistency. Spatial proteomics platforms like PhenoCycler typically measure fewer markers (around 56 proteins), which limits the resolution at which cell types and functional states can be identified. On the other hand, spatial transcriptomics technologies capture a much broader range of genetic material, from 300 to over 18,000 genes, offering a deeper and more comprehensive view of the transcriptome. This broader coverage can reveal nuances in cell type and state that may be missed by proteomics alone. However, rather than viewing these discrepancies as limitations, it is important to recognize that proteomics and transcriptomics provide complementary information. Proteins are the functional molecules of the cell, while transcripts provide insight into the gene expression landscape. Both datasets should ideally be integrated to create a more complete picture of cellular function. To address the issue of poor consistency, we agree that using matched markers and visualizing the results of both modalities side by side is a valuable strategy (Figure 6d in manuscript). This allows us to directly compare the expression patterns of proteins and transcripts, helping to identify where discrepancies occur and potentially providing insights into regulatory mechanisms at play. We believe integrating multi- omics data, including proteomics and transcriptomics, offers the best path forward to achieve a comprehensive understanding of cell types and states.  
+
+## Minor concerns:  
+
+There are minor errors present throughout the text. Please thoroughly review the entire document to correct these and similar mistakes.  
+
+Section 3: "These findings highlight the efficacy of TACIT in spatial transcriptomics. " Methods, Segmented Regression Model: "This was determined by the lowest Akaike Information Criterion (AIC) score achieved among the three models the three models" Methods, Clinical Protocol University of Sao Paulo: Is there a missing word? Figure 6i: "B Cells::Cell State \((+)\) " Figure 5: The text does not match the figure, the caption does not correspond to the figure, and the captions are incomplete. Figure 1f: Note the parentheses that distinguish 'Clean cells' from 'Mixed cells', and there is an issue with the representation of 'Mixed cells' (One "Clean cell' among them).
+
+<--- Page Split --->
+
+Section 4, Figs. 4c- e only represents the results of one dataset, not corresponding to the two datasets mentioned in the text.  
+
+Response: Thank you, we have made adjustments based on your comments.  
+
+## Question: How is the Bootstrap performed?  
+
+Response: Thank you for your inquiry regarding the bootstrap methodology used in our study, as detailed in Supplement Figure 2. To evaluate the robustness of the TACIT algorithm, we implemented the following bootstrap procedure: In each iteration, we randomly selected \(80\%\) of the data from the PCF- CRC dataset without replacement. This process was repeated independently 10 times to simulate the effect of varying sample sizes and to test the stability of TACIT under different subsets of data. For each iteration, we calculated precision, recall, and F1 scores to assess the performance of the algorithm. These metrics were chosen to evaluate the accuracy and consistency of the cell type identification provided by TACIT. We then aggregated the results from all iterations to analyze the stability of these performance metrics, which are visualized in Supplement Figures 2b to 2d. This approach allowed us to demonstrate that TACIT maintains consistent performance across different data samplings, underscoring its reliability and effectiveness for practical applications in spatial omics analysis.  
+
+Question: The main text sections do not mention Astrograph, yet it is involved in the abstract. Please revise the relevant content accordingly.  
+
+Response: Thank you for pointing out the discrepancy. We have revised the abstract to remove any reference to Astrograph, ensuring it accurately reflects the content and focus described in the main text. This adjustment maintains the integrity and coherence of our manuscript.  
+
+## Question: Reviewer #2 (Remarks on code availability):  
+
+Invalid link.  
+
+Response: Thank you for your feedback regarding the availability of our code. We apologize for the inconvenience caused by the private GitHub repository. We plan to make the GitHub repository publicly accessible upon acceptance of our paper. The repository will include a README, tutorial, and vignette to guide users on how to utilize the functions. Additionally, we included the TACIT vignette (using PCF- CRC) as an example in TACIT_Vignette.pdf files.
+
+<--- Page Split --->
+
+Reviewer #3 (Remarks to the Author):  
+
+Question: The TACIT's performance heavily depended on the quality and comprehensiveness of the marker panels used for different cell types. However, the author did not seem to explain the detailed process of selecting markers. In practice, if there is no expert knowledge especially for some rare cell types, the author should explain how to ensure markers are accurate. Alternatively, if some low- quality markers were included in TACIT's signature matrix, did TACIT's performance remain robust?  
+
+Response: Thank you for your question. Researchers typically begin with well- documented markers from the literature that have been validated in previous studies. These markers are tested and, if necessary, optimized based on experimental outcomes. If the initial panel does not sufficiently distinguish between closely related or phenotypically similar cell types, expanding the marker panel can significantly enhance the method's ability to differentiate between cell types in spatial proteomics or transcriptomics experiments. In our simulations, where we added or subtracted markers from TACIT's signature matrix, we found that TACIT maintained accuracy even with the inclusion of two noisy markers per cell type.  
+
+Additionally, when expert knowledge on rare cell types is limited, we employ an unsupervised approach alongside the literature- based selection. We analyze scRNA- seq data to identify the top 5 genes or proteins most prominently expressed in each cluster (as demonstrated in Results Figure 3). This data- driven approach helps capture potentially overlooked markers that are highly relevant to the specific cellular contexts under study. We further validate these findings by cross- referencing with existing single- cell RNA databases to ensure marker accuracy and relevance. This comprehensive strategy ensures that TACIT remains robust, even when lower- quality markers are present in the initial panel.  
+
+There are also community efforts in developing better marker sets for robust panel design. https://cytomarker.ai/ is a recently developed web interface that support the selection of high quality markers based on large scale scRNA- seq data.  
+
+Question: In TACIT's algorithm, the effectiveness of using KNN for the deconvolution should depend on the number of "clean cells" from the previous "cell type categorization" step, which were used as anchors. How can we ensure there are enough "clean cells" when TACIT is applied to other organs and species? If there are too few "clean cells", the cell type annotation will definitely perform poorly.  
+
+Response: Thank you for your question regarding the reliance of TACIT's deconvolution process on the number of clean cells from the cell type categorization step.  
+
+TACIT's algorithm dynamically establishes clean cells based on thresholds derived from the specific dataset being analyzed, rather than relying on a pre- trained model. This allows TACIT to adapt to the characteristics of each dataset, ensuring it can identify relevant clean cells across various organs and species. This flexibility is crucial for achieving accurate cell type annotation in diverse contexts.  
+
+Our evaluations across multiple datasets, including three public datasets (PCF- CRC, PCF- HI, and MERFISH) and one in- house dataset (Xenium- SjD), have shown TACIT's ability to reliably identify all expected cell types, indicating that enough clean cells were present for annotation in these cases.
+
+<--- Page Split --->
+
+However, in instances where the number of clean cells for a particular cell type is extremely low (fewer than 10), there are two possible explanations: 1) the population of that cell type is genuinely small, or 2) the marker signature for that cell type is not powerful enough to distinguish cells from other cell types. In the first scenario, there is little that can be done to increase the clean cell count. In the second scenario, there are strategies to address this challenge, such as: 1) adding additional markers to better distinguish the mixed population; 2) Improving cell segmentation accuracy will alleviate the issue of mixing cell marker signals from neighboring cells, as seen in both PCF and Xenium data.  
+
+Question: Although the authors claimed TACIT had a promising result in integrating multimodal data, its performance was only demonstrated using transcriptomics and proteomics. Newly emerging modalities like spatial epigenomics and metabolomics were not thoroughly tested in the paper. Further validation on a broader range of spatial omics technologies would strengthen its applicability and increase the impact of TACIT.  
+
+Response: Currently, probe- based spatial omics platforms like Xenium, CosMx, and MERSCOPE primarily focus on spatial transcriptomics, enabling high- resolution mapping of gene expression within tissues. However, to date, the technology for single cell spatial epigenomics and spatial metabolomics is not available yet and there are no publicly available datasets in this area.  
+
+For spatial epigenomics, which involves mapping DNA modifications (such as methylation) or chromatin accessibility across tissue sections, several approaches are being explored, but these are not yet fully realized in commercial probe- based platforms like single cell spatial transcriptomics data. Some approaches involve adapting existing technologies like ATAC- seq or DNA methylation profiling to spatial contexts, but these often require sophisticated custom setups and are not yet as accessible or standardized.  
+
+Some research groups are developing methods for targeted spatial chromatin profiling, using antibodies or probes to mark specific histone modifications or open chromatin regions in tissue sections. However, these techniques are still in the experimental stage and not yet available as commercial platforms. The closest is AtlasXomics, but this is not probe- based, more akin to spatial Visium for epigenome (https://www.atlasxomics.com/)  
+
+Spatial metabolomics such as MALDI- MSI, which involves mapping the distribution of metabolites within tissues, is also in a more nascent stage compared to spatial transcriptomics.  
+
+Currently, there isn't a direct equivalent of Xenium, CosMx, or MERSCOPE specifically for spatial epigenomics or spatial metabolomics that operates on a probe- based platform. These fields are rapidly advancing, and it's likely that more specialized and accessible technologies will emerge in the near future as research progresses and the demand for spatial multimomics increases. TACIT will be used on these technologies as they emerge.  
+
+Question: Although the author discussed TACIT's memory usage, they should also mention TACIT's computation time and compare it with other benchmarking methods on real or simulated data to demonstrate TACIT's practicality.  
+
+Response: Thank you for suggesting a comparison of TACIT's computation time with other benchmarking methods. We have now included a detailed comparison of TACIT, SCINA, and CELESTA across several real datasets. For the PCF- CRC dataset ( \(\sim 240k\) cells), TACIT took 11.25 minutes, SCINA took 15.21 minutes, and CELESTA completed in 4.32 minutes. For the
+
+<--- Page Split --->
+
+MERFISH dataset ( \(\sim 900k\) cells), TACIT required 50.98 minutes, while SCINA and CELESTA were not run since those methods are primarily used for PCF datasets. For the PCF- HI dataset ( \(\sim 2.6M\) cells), TACIT took 105.28 minutes, SCINA took 102.74 minutes, and CELESTA completed in 65.77 minutes (Figure R10). It's important to note that for CELESTA, we used only the default parameters, without accounting for the time required to tune four parameters per cell type during the parameter selection process.  
+
+Additionally, unsupervised clustering methods such as Leiden and Louvain clustering require further steps to convert clusters into cell types, which can be time- consuming. Recent reports from Nature Methods indicate that processing four tissues ( \(\sim 160,000\) cells) in a typical PCF dataset with a 48- marker panel can take over 25 hours for clustering, merging, re- clustering, sub- clustering, and cell type assignment based on marker expressions and spatial locations within PCF images<sup>3</sup>.  
+
+For context, all analyses were performed on a system with an Intel(R) Core(TM) Ultra 9 185H 2.30 GHz processor and 64 GB of RAM. This hardware provided the necessary computational resources to handle large datasets, ensuring the reliability of the benchmarking results.  
+
+![PLACEHOLDER_27_0]
+
+<center>Figure R10: Run time comparison. All analyses were performed on a system with an Intel(R) Core(TM) Ultra 9 185H 2.30 GHz processor and 64 GB of RAM. Note unsupervised clustering methods such as Leiden and Louvain clustering require additional manual annotation steps to convert clusters into cell types, which can be time-consuming. Recent reports from Brbic, M. et al. 2022, Nature Methods reported that processing four tissues ( \(\sim 160,000\) cells) in a typical PCF dataset with a 48-marker panel can take over 25 hours for clustering, merging, re-clustering, sub-clustering, and cell type assignment based on marker expressions and spatial locations within PCF images. </center>  
+
+## Question:  
+
+There are some writing and formatting issues.  
+
+a) In Figure 2 legend, this description should be changed to (b, h) "(e,k) UMAP representations with cell type delineations, showing the clustering of cells in a two dimensional space ... Identifying cell types"
+
+<--- Page Split --->
+
+b) In Figure 2 legend, (f,i) should be (c,i) "(f,i) Heatmaps comparing the mean marker values for each cell type identified by TACIT and other existing methods. TACIT's heatmaps exhibit distinct and clear unique marker expressions for each cell type, with a diagonal pattern that highlights its precise cell type identification capabilities."  
+
+c) The resolution of Figure 3c is very low, making it difficult to read the names of the markers.  
+
+Response: Thank you, we have made adjustments based on your comments.
+
+<--- Page Split --->
+
+## Reference  
+
+1. Schurch, C. Coordinated cellular neighborhoods orchestrate antitumoral immunity at the colorectal cancer invasive front. [object Object] https://doi.org/10.17632/MPJZBTFGFR.1 (2020).2. Zhang, Z. jcao89757/SCINA. (2024).3. Brbic, M. et al. Annotation of spatially resolved single-cell data with STELLAR. Nat. Methods 19, 1411–1418 (2022).4. Magness, A. et al. Deep cell phenotyping and spatial analysis of multiplexed imaging with TRACERx-PHLEX. Nat. Commun. 15, 5135 (2024).5. Shen, R. et al. Spatial-ID: a cell typing method for spatially resolved transcriptomics via transfer learning and spatial embedding. Nat. Commun. 13, 7640 (2022).6. Xu, H. et al. SPACEL: deep learning-based characterization of spatial transcriptome architectures. Nat. Commun. 14, 7603 (2023).7. Zhang, W. et al. Identification of cell types in multiplexed in situ images by combining protein expression and spatial information using CELESTA. Nat. Methods 19, 759–769 (2022).8. Geuenich, M. J. et al. Automated assignment of cell identity from single-cell multiplexed imaging and proteomic data. Cell Syst. 12, 1173-1186. e5 (2021).9. Biancalani, T. et al. Deep learning and alignment of spatially resolved single-cell transcriptomes with Tangram. Nat. Methods 18, 1352–1362 (2021).10. Kleshchevnikov, V. et al. Cell2location maps fine-grained cell types in spatial transcriptomics. Nat. Biotechnol. 40, 661–671 (2022).11. Kimmel, J. C. & Kelley, D. R. Semisupervised adversarial neural networks for single-cell classification. Genome Res. 31, 1781–1793 (2021).12. Li, C. et al. SciBet as a portable and fast single cell type identifier. Nat. Commun. 11, 1818 (2020).
+
+<--- Page Split --->
+
+13. Quan, F. et al. Annotation of cell types (ACT): a convenient web server for cell type annotation. Genome Med. 15, 91 (2023).  
+
+14. Li, J., Zhang, Y., Yang, C. & Rong, R. Discrepant mRNA and Protein Expression in Immune Cells. Curr. Genomics 21, 560-563 (2020).  
+
+15. Bauernfeind, A. L. & Babbitt, C. C. The predictive nature of transcript expression levels on protein expression in adult human brain. BMC Genomics 18, 322 (2017).
+
+<--- Page Split --->
+
+## Contents of point-by-point responses to review comments for NCOMMS-24-38058-T  
+
+Reviewer #1: Page 2  Reviewer #2: Page 4  Reviewer #3: Page 6  
+
+We thank the reviewers for their additional round of reviews and valuable comments on our manuscript. We have addressed their feedback and made significant revisions to enhance the clarity and rigor of our work. Below, we outline our responses to the reviewers' comments and the revisions made.  
+
+Below is a summary of major newly added results in the manuscript:  
+
+- Adding Figure 3: Comparative Analysis of TACIT, Louvain, and Leiden Across Resolutions and Benchmarking Against Recent Methods.- Adding Extended Data 8: Evaluating Segmentation Impact in TACIT Annotations on PCF Datasets.- TACIT is publicly available on GitHub at: https://github.com/huyhnkl953/TACIT
+
+<--- Page Split --->
+
+## Reviewer #1 (Remarks to the Author):  
+
+I thank the authors for their effort and thorough response to my concerns.  
+
+"The main concern I have is still the benchmarking, as while the authors did implement other metrics such as purity and entropy, they did so in a way that may be misleading. For example, in Figure R2a increasing resolution resulted in more clusters which is contrary to the idea of resolution where an increase should have less clusters. In addition, in Figure R2a,b they show a range of resolution values for Leiden and Louvain but not for TACIT. This is odd, especially when in Extended Data 2 they have a range for TACIT. In addition, that range was from 0.005 to 150, but we do not know what TACIT used in this benchmark, it should be the same as Louvain at least as it uses Louvain as part of the algorithm. I understand the authors' point that it would be unfeasible to annotate manually this many clusters, but entropy purity does not require a cluster annotation so that is not the intended purpose of this benchmark."  
+
+Response: Thank you for your comments. As noted in the original Leiden paper [PMID: 30914743], both Louvain and Leiden clustering methods exhibit an increase in the number of communities (clusters) at higher resolutions. A higher resolution means the algorithm detects more detailed substructures within the data, leading to a larger number of smaller clusters.  
+
+To address the reviewer's comment, we compared the Louvain and Leiden methods together with TACIT using a resolution ranging from 0.005 to 50. Please note that, unlike Louvain and Leiden clustering, resolution is not an explicit parameter in TACIT and is determined by TACIT itself to generate a sufficiently large number of microclusters for TACIT to work with. However, to address the reviewer's comment, we included the result of TACIT when changing the underlying resolution parameters ranging from 0.005 to 50. Beyond resolution 50, the number of clusters for Louvain and Leiden methods exceeds 1000, which is undesirable as we prefer fewer clusters. As shown in the figure R1, TACIT consistently output the same number of cell populations with the highest accuracy in cell type annotation result in terms of entropy, and purity in comparison with Louvain and Leiden.  
+
+In the PCF- CRC datasets, TACIT consistently shows lower entropy values (calculated using the DescTools package in R version 0.99.57), ranging from 4.877 at a resolution of 0.005 to 5.108 at 50, indicating stable clustering as the resolution increases. With purity values (using the funtimes package in R version 9.1) starting at 0.631 and stabilizing around 0.646, TACIT generates clusters with a significant proportion of similar true class labels, reflecting high clustering quality. In contrast, Leiden exhibits higher entropy, starting at 4.885 and peaking at 11.787, indicating increasing disorder and less effective clustering at higher resolutions. Its purity declines from 0.349 to 0.013, suggesting many clusters contain mixed class labels. Louvain displays similar trends, with entropy increasing from 4.989 to 12.058 and purity values starting at 0.375 but dropping to 0.012 at higher resolutions, indicative of overall poor clustering performance. Additionally, in the MERFISH datasets, TACIT also demonstrates lower entropy compared to both Louvain and Leiden across resolutions, as well as higher purity scores.
+
+<--- Page Split --->
+![PLACEHOLDER_33_0]
+
+<center>Figure R1: Comparison of the number of clusters, entropy score, and purity score, among Louvain, Leiden, and TACIT methods across resolutions ranging from 0.005 to 50 using PCFCRC datasets (R1a-c) and MERFISH datasets (R1d-f). </center>  
+
+"Lastly, I noticed some typos in the newly included text, some proofreading may be required before publication."  
+
+Response: Thank you for your comments. We have already performed a thorough proofreading of the newly included text to ensure it is free of typos and errors before publication.  
+
+Reviewer #1 (Remarks on code availability):  
+
+"As the GitHub repository is not public, I am unable to review that repository's code quality, vignettes, README, etc.I still did not see any of that documentation in the Code Ocean repository as well."  
+
+Response: Thank you for your comments. The GitHub repository has been made public with vignettes on how to use the TACIT function.
+
+<--- Page Split --->
+
+Reviewer #2 (Remarks to the Author):  
+
+"Most of my concerns have been resolved. However, three minor issues still need to be addressed.  
+
+Minor concerns:  
+
+1. Results, section: Multimodal Cell Identification with TACIT:  
+
+"... capturing both antibody intensities from PCF and count values from Xenium (Figs. 6b, c)."  
+
+Should it be "Figs. 6b, 6c" for consistency?  
+
+2. Figure 6i, Legend:  
+
+"B Cells::Cell State \((+)\) ".  
+
+There seems to be a duplicated colon (":"). This appears to be the same issue as in the legend of Extended Data 9b.  
+
+3. Methods, Segmented Regression Model:  
+
+"This was determined by the lowest Akaike Information Criterion (AIC) score achieved among the three models the three models...".  
+
+The phrase "the three models" is repeated. Please revise for clarity."  
+
+Response: Thank you for your comments. We revised in the manuscript  
+
+Reviewer #2 (Remarks on code availability):  
+
+The code appears to be well- written and meets the standards for reproducibility and usability. Below are the detailed points of evaluation:  
+
+## Reproducibility:  
+
+The results of the paper are largely reproducible using the provided code. The implementation is consistent with the methodology described in the paper, and the code includes all necessary scripts to reproduce the key findings.  
+
+## Usability:  
+
+The repository includes a comprehensive README file that provides clear instructions for installation and running the application. It outlines the required dependencies, setup steps, and example commands for running experiments, which is helpful.  
+
+The repository structure is logical and user- friendly, allowing users to locate files and resources without confusion.  
+
+## Installation and Execution:  
+
+The installation process is straightforward, and all dependencies are either listed in a requirements file or documented explicitly. I was able to install the necessary packages and execute the code without encountering significant issues.  
+
+Sample datasets and configurations are provided, enabling users to quickly test the implementation without needing to prepare custom inputs initially.
+
+<--- Page Split --->
+
+## Community Usability:  
+
+The code serves as a valuable resource for the community, particularly due to its clear structure and adherence to reproducibility practices.  
+
+Overall, the code is functional, well- documented, and reproducible, making it a reliable and usable resource for researchers and practitioners in the field.  
+
+Response: Thank you for your comments.
+
+<--- Page Split --->
+
+Reviewer #3 (Remarks to the Author):  
+
+"We appreciate the authors' detailed and thoughtful responses, which have addressed most of the primary concerns raised in our initial review. The comprehensive explanation of TACIT's marker panel selection pipeline, the adaptive approach to clean cell identification effectively clarify its strengths and address key questions about its applicability and reliability. Additionally, the clarification of TACIT's performance across diverse datasets and its computational scalability underscore its promise as a versatile tool for spatial omics analyses.  
+
+While the authors' responses have resolved many issues, there remain two areas where further clarification or exploration would enhance the robustness and generalizability of TACIT:  
+
+Benchmarking Against Additional Methods: While TACIT has been compared to Louvain clustering, additional benchmarking against other widely used or emerging tools in spatial omics, such as Spatial- ID or Cell2location, would provide a broader context for its strengths and limitations. We noticed that the authors have provided additional comparative analyses in their replies to Reviewer #1. However, the results of these comparisons are crucial to fully appreciating TACIT's performance relative to existing tools. As such, the authors should consider providing these benchmarking results directly in the main text of the manuscript rather than relegating them to supplementary figures.  
+
+Segmentation Dependencies: As the authors mentioned in the replies, the reliance on accurate segmentation masks remains a critical dependency for TACIT's performance. Although the authors propose strategies to mitigate errors, such as using a human- in- the- loop Cellpose3 model, further validation is needed to evaluate how segmentation inaccuracies affect cell type annotation, for example in highly complex or crowded regions like tertiary lymphoid structures (TLS). This dependency may limit TACIT's applicability in cases where segmentation tools underperform or lack sufficient training for the tissue type in question.  
+
+Addressing these points would not only strengthen the manuscript but also provide the community with greater confidence in TACIT's utility across a wide range of spatial omics applications. Thank you for your continued efforts in refining and improving this promising tool."  
+
+Response: We thank the reviewer for raising this issue.  
+
+To address this, we have added a cell type annotation comparison performance under different segmentation tools: including one of each of the segmentation types, such as, cellpose 3 (Human- in- the- loop), stardist (Pre- trained model) and watershed (Cell Expansion) using a sample from a patient with Graft- versus- host disease (GVHD), the same disease used as a proof of concept in this paper.
+
+<--- Page Split --->
+
+We compared the segmentation performance with the annotations made by one of our co- first author who is a pathologist. The evaluation of these methods showed that the human- in- the- loop aligns more consistently with the pathophysiology structure of GVHD. Additionally, both Cellpose 3 and stardist identified cell types that are correlated with the disease progression, and consequently more relevant to GVHD and exhibited fewer mixed cell types than the Watershed method.  
+
+Please note that segmentation is an independent step prior to the cell annotation step, affecting all cell type annotation methods. When testing TACIT under different conditions, we observed that variability in cell morphology significantly impacted cell annotation and cell state analysis. To address this, we conducted multiple analyses to identify the best segmentation methods for whole- slide annotation in cases where diverse cell morphologies coexist on a single slide. In the dataset used for this study, we encountered secretory cells (with large cytoplasm) and small, round immune cells, where human- in- the- loop segmentation methods outperformed automated methods in accuracy compared to histopathological evaluations.  
+
+This challenge is also evident in cancer datasets that feature dysplastic cells with diverse morphologies. This issue motivated our team to pursue a related study on the impact of segmentation on cell type annotation, which we aim to publish in the near future.  
+
+![PLACEHOLDER_37_0]
+
+<center>Figure R2: Evaluating Segmentation Impact in TACIT Annotations on PCF Datasets. (a) Following TACIT annotations using three different methods—Watershed, stardist, and Cellpose version 3—we used Voronoi plots to reconstruct the slides and analyze the variations in reconstructions attributable to different segmentation methods. The Watershed model resulted in a reconstruction with a dense population of ductal cells, while the stardist and Cellpose version 3 models produced reconstructions that more closely represented the architecture of acinar/myoepithelial cells and ducts within the salivary lobules. (b) Each model identified varying </center>
+
+<--- Page Split --->
+
+proportions of cells, with the human in the loop (HITL) model detecting significantly more acinar cells than the Watershed method. In contrast, the Watershed model identified a greater number of ductal cells compared to the other models. Additionally, the HITL model identified fewer mixed cell types than the Watershed method. (c) The diversity of cell types detected increased sequentially from the Watershed model to the stardist and then to the Cellpose version 3 models, showing that HITL methods can find more rare cell types. (d) UMAPs (Uniform Manifold Approximation and Projection) were created using spatial single- cell data from each segmentation method, showcasing differences in the number and types of cells identified by these approaches.  
+
+Reviewer #3 (Remarks on code availability):  
+
+"Codes are not fully released".  
+
+Response: Thank you for your comments. The GitHub repository has been made public with vignettes on how to use the TACIT function.
+
+<--- Page Split --->
+
+# Contents of point-by-point responses to review comments for NCOMMS-24-38058C  
+
+Reviewer #1: Page 1  Reviewer #2: Page 1  Reviewer #3: Page 1  
+
+## REVIEWERS' COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+I thank the authors for their additional comparisons and believe they have fully addressed my concerns.  
+
+Response: Thank you for your comments.  
+
+Reviewer #1 (Remarks on code availability):  
+
+The code is monolithic and filled with global variables which may cause issues with future maintenance, but does install and run.  
+
+Response: Thank you for your comments.  
+
+Reviewer #3 (Remarks to the Author):  
+
+The authors answered all my concerns and I have no more questions.  
+
+Response: Thank you for your comments.
+
+<--- Page Split --->