10,000 Matching Annotations
  1. Feb 2025
    1. eLife Assessment

      This valuable study presents findings on DNA methylation as an efficient epigenetic transcriptional regulating strategy in bacteria. The authors utilized single-molecule real-time sequencing to profile the DNA methylation landscape across three model pathovars of Pseudomonas syringae, identifying significant epigenetic mechanisms through the Type-I restriction-modification system, which includes a conserved sequence motif associated with N6-methyladenine. The evidence presented is solid and the study provides novel insights into the epigenetic mechanisms of P. syringae, expanding the understanding of bacterial pathogenicity and adaptation.

    2. Reviewer #1 (Public review):

      Summary:

      In this work, Huang et al used SMRT sequencing to identify methylated nucleotides (6mA, 4mC, and 5mC) in Pseudomonas syringae genome. They show that the most abundant modification is 6mA and they identify the enzymes required for this modification as when they mutate HsdMSR they observe a decrease of 6mA. Interestingly, the mutant also displays phenotypes of change in pathogenicity, biofilm formation, and translation activity due to a change in gene expression likely linked to the loss of 6mA.

      Overall, the paper represents an interesting set of new data that can bring forward the field of DNA modification in bacteria.

      Comments on revisions:

      Thank you for the additional work. The authors have now addressed all my concerns.

    3. Reviewer #2 (Public review):

      In the present manuscript, Huang et.al. revealed the significant roles of the DNA methylome in regulating virulence and metabolism within Pseudomonas syringae, with a particular focus on the HsdMSR system in this model strain. The authors used SMRT-seq to profile the DNA methylation patterns (6mA, 5mC, and 4mC) in three P. syringae strains (Psph, Pss, and Psa) and displayed the conservation among them. They further identified the type I restriction-modification system (HsdMSR) in P. syringae, including its specific motif sequence. The HsdMAR participated in the process of metabolism and virulence (T3SS & Biofilm formation), as demonstrated through RNA-seq analyses. Additionally, the authors revealed the mechanisms of the transcriptional regulation by 6mA. Strictly from the point of view of the interest of the question and the work carried out, this is a worthy and timely study that uses third-generation sequencing technology to characterize the DNA methylation in P. syringae. The experimental approaches were solid, and the results obtained were interesting and provided new information on how epigenetics influences the transcription in P. syringae. The conclusions of this paper are mostly well supported by data.

      Comments on revisions:

      The authors have successfully addressed all the comments from the reviewers in their revised manuscript.

    4. Author response:

      The following is the authors’ response to the original reviews.

      Public Reviews:

      Reviewer #1 (Public Review):

      Summary:

      In this work, Huang et al used SMRT sequencing to identify methylated nucleotides (6mA, 4mC, and 5mC) in Pseudomonas syringae genome. They show that the most abundant modification is 6mA and they identify the enzymes required for this modification as when they mutate HsdMSR they observe a decrease of 6mA. Interestingly, the mutant also displays phenotypes of change in pathogenicity, biofilm formation, and translation activity due to a change in gene expression likely linked to the loss of 6mA. Overall, the paper represents an interesting set of new data that can bring forward the field of DNA modification in bacteria.

      Thank you for your valuable feedback on our paper exploring the impact of 6mA modification in P. syringae.

      Major Concerns:

      Most of the authors' data concern Psph pathovar. I am not sure that the authors' conclusions are supported by the two other pathovars they used in the initial 2 figures. If the authors want to broaden their conclusions to Pseudomonas syringe and not restrict it to Psph, the authors should have stronger methylation data using replicates. Additionally, they should discuss why Pss is so different than Pst and Psph. Could they do a blot to confirm it is really the case and not a sequencing artifact? Is the change of methylation during bacterial growth conserved between the pathovar? The authors should obtain mutants in the other pathovar to see if they have the same phenotype. The authors have a nice set of data concerning Psph but the broadening of the results to other pathovar requires further investigation.

      We appreciate the reviewer’s insightful comments. While the majority of our data focuses on the Psph, we recognize the importance of validating these findings in Pss and Pst. To this end, we have performed additional experiments using dot blot and mutant construction to enhance our conclusions in other pathovars.

      We agree that we should discuss why Pss is different from Psph and Pst. We performed a dot blot assay using genome DNA in Pss and Pst, presented in Figure S5A. Meanwhile, we compared the 6mA modification level of Pss and Pst in different growth phases. As shown in Figure S5A, the change of methylation during bacterial growth is conserved in Pst. The change was not obvious in Pss, which might be due to the lack of a type I R-M system.

      “In accordance with previous studies showing that growth conditions affect the bacterial methylation status, we applied dot blot experiments using the same amount of DNA (1 μg) from these three P. syringae strains to detect the 6mA levels during both logarithmic and stationary phases. The results revealed that 6mA levels in the stationary phase were much higher compared to the logarithmic phase in Psph and Pst, but no significant change in Pss. Additionally, we found that during the stationary phase, 6mA methylation levels in Psph and Pst were higher than those in Pss. These findings were consistent with the MTases predication on these three strains, since Pss does not harbor any type I R-M systems, which are important for 6mA medication in bacteria.”

      Please see Figure S5A and Lines 220-228 in the revised manuscript.

      We also tried to construct an HsdM mutant in Pst to explore whether the influence of 6mA methylation was conserved in P. syringae, but it failed after multiple attempts. We did not construct a Pss mutant because no type I R-M system was predicted, and few methylation sites were identified via SMRT-seq in this strain. Therefore, we overexpressed HsdM in Pst instead. We have performed additional experiments in WT and the HsdM overexpression strains, including dot blot and growth curve assays.

      Please see Figures S5B-C and Lines228-232 in the revised manuscript.

      The authors should include proper statistical analysis of their data. A lot of terms are descriptive but not supported by a deeper analysis to sustain the conclusions. For example, in Figure 4E, we do not know if the overlap is significant or not. Are DEGs more overlapping to 6mA sites than non-DEGs? Here is a non-exhaustive list of terms that need to be supported by statistics: different level (L145), greater conservation (L162), significant conservation (L165), considerable similarity (L175), credible motifs (L189), Less strong (L277) and several "lower" and "higher" throughout the text.

      Thank you for the insightful feedback. We have made the following revisions in the manuscript to ensure that the terms are more precise and do not require statistical significance testing.

      (1) Statistical analysis: We have added statistical tests for the overlap between DEGs and 6mA sites in Figure 4E. We performed the Fisher test, and we found the overlap was not significant (p> 0.05). DEGs and non-DEGs were both non-significant overlapped 6mA sites. Please see Figure 4E and Lines 261-262.

      “Less strong” was used to indicate the influence of HsdM on biofilm in Figure 5D. All Figures with “*” labels were analyzed using students' two-tailed t-tests with a significant change (p < 0.05).

      (2) Revised wording: For terms used to describe comparisons, we have revised the wording to be clearer and ensure that the terminology used did not imply the need for statistical significance testing when not required. For example:

      “Different level” has been removed. Please see Lines 143-144.

      “Greater conservation” has been revised to “more conserved functional terms”. Please see Lines 161-162.

      “Significant conservation” has been revised to “notable conservation”. Please see Line 165.

      “Credible motifs” has been revised to “identified motifs”. Please see Line 186.

      The authors performed SMRT sequencing of the delta hsdMSR showing a reduction of 6mA. Could they include a description of their results similar to Figures 1-2. How reduced is the 6mA level? Is it everywhere in the genome? Does it affect other methylation marks? This analysis would strengthen their conclusions.

      Yes, we agree. We have provided additional analysis and descriptions to strengthen the conclusions regarding these valuable comments. We determined three methylation sites in the HsdMSR mutant strain and compared the overlapped genes within these modification patterns. Specifically, we focused on the 6mA sites in Psph WT, HsdMSR mutant, and HsdM motif CAGCN<sub>(6)</sub>CTC. As expected, we found almost all of the reduction 6mA sites in the ΔhsdMSR were from motif CAGCN<sub>(6)</sub>CTC. We also noticed that 5mC and 4mC sites in the mutant were relatively similar to that in WT, and the slight difference might be caused by sequencing errors. Overall, we propose that HsdMSR only catalyze the 6mA located on the motif CAGCN<sub>(6)</sub>CTC, but does not affect other 6mA sites and other modification types.

      Please see Figures S4D-E and Lines 212-218 in the revised manuscript.

      In Figure 6E to conclude that methylation is required on both strands, the authors are missing the control CAGCN6CGC construct otherwise the effect could be linked to the A on the complementary strand.

      Thank you for your valuable suggestions. We have provided the control result on the complementary strand. Please see Figure 6C. The new result evidences the conclusion that 6mA methylation regulates gene transcription based on methylation on both strands.

      Please see Figure 6C and Lines 329-330 in the revised manuscript.

      Reviewer #2 (Public Review):

      In the present manuscript, Huang et.al. revealed the significant roles of the DNA methylome in regulating virulence and metabolism within Pseudomonas syringae, with a particular focus on the HsdMSR system in this model strain. The authors used SMRT-seq to profile the DNA methylation patterns (6mA, 5mC, and 4mC) in three P. syringae strains (Psph, Pss, and Psa) and displayed the conservation among them. They further identified the type I restriction-modification system (HsdMSR) in P. syringae, including its specific motif sequence. The HsdMAR participated in the process of metabolism and virulence (T3SS & Biofilm formation), as demonstrated through RNA-seq analyses. Additionally, the authors revealed the mechanisms of the transcriptional regulation by 6mA. Strictly from the point of view of the interest of the question and the work carried out, this is a worthy and timely study that uses third-generation sequencing technology to characterize the DNA methylation in P. syringae. The experimental approaches were solid, and the results obtained were interesting and provided new information on how epigenetics influences the transcription in P. syringae. The conclusions of this paper are mostly well supported by data, but some aspects of data analysis and discussion need to be clarified and extended.

      Thank you for your positive feedback and recognition of the importance of our study. We appreciate the suggestions for further clarification and extension of some aspects of data analysis and discussion. We added further analysis of the SMRT-seq result of the ΔhsdMSR and overexpressed HsdM in Pst to provide more information on conservation. We added these contents to the discussion in the revised manuscript. Please see Figure 6C and  Figure S5.

      Reviewer #3 (Public Review):

      Summary:

      The article by Huang et.al. presents an in-depth study on the role of DNA methylation in regulating virulence and metabolism in Pseudomonas syringae, a model phytopathogenic bacterium. This comprehensive research utilized single-molecule real-time (SMRT) sequencing to profile the DNA methylation landscape across three model pathovars of P. syringae, identifying significant epigenetic mechanisms through the Type-I restriction-modification system (HsdMSR), which includes a conserved sequence motif associated with N6-methyladenine (6mA). The study provides novel insights into the epigenetic mechanisms of P. syringae, expanding the understanding of bacterial pathogenicity and adaptation. The use of SMRT sequencing for methylome profiling, coupled with transcriptomic analysis and in vivo validation, establishes a robust evidence base for the findings

      Strengths:

      The results are presented clearly, with well-organized figures and tables that effectively illustrate the study's findings.

      Weaknesses:

      It would be helpful to add more details, especially in the methods, which make it easy to evaluate and enhance the manuscript's reproducibility.

      Thank you for the positive evaluation of our study, as well as the constructive feedback provided. We have added more details in methods for RNA-seq analysis and Ribo-seq analysis. Please see Lines 484-515.

      “Briefly, bacteria were cultured to an OD<sub>600</sub> of 0.4, at which point chloramphenicol was added to a final concentration of 100 µg/mL for 2 minutes. Cells were then pelleted and washed with pre-chilled lysis buffer [25 mM Tris-HCl, pH 8.0; 25 mM NH4Cl; 10 mM MgOAc; 0.8% Triton X-100; 100 U/mL RNase-free DNase I; 0.3 U/mL Superase-In; 1.55 mM chloramphenicol; and 17 mM GMPPNP]. The pellet was resuspended in lysis buffer, followed by three freeze-thaw cycles using liquid nitrogen. Sodium deoxycholate was then added to a final concentration of 0.3% before centrifugation. The resulting supernatant was adjusted to 25 A260 units and mixed with 2 mL of 500 mM CaCl<sub>2</sub> and 12 µL MNase, making up a total volume of 200 µL. After the digestion, the reaction was quenched with 2.5 mL of 500 mM EGTA. Monosomes were isolated using Sephacryl S400 MicroSpin columns, and RNA was purified using the miRNeasy Mini Kit (Qiagen). rRNA was removed using the NEBNext rRNA Depletion Kit, and the final library was constructed with the NEBNext Small RNA Library Prep Kit. For each sample, ribosome footprint reads were mapped to the Psph 1448A reference genome, and the translational efficiency was calculated by dividing the normalized Ribo-seq counts by the normalized RNA counts. Two biological replicates were performed for all experiments.”

      Recommendations For The Authors:

      Reviewer #1 (Recommendations For The Authors):

      I would recommend the authors limit their manuscript to Psph pathovar and include statistical analysis supporting their conclusions.

      Thank you for your suggestion.

      Minor

      • L104: "significantly" please add a p-value and explain the analysis.

      Sorry for the confusion. We have added the p-value and explained the analysis in the method section. The p-value used for SMRT-seq was the modification quality value (QV) score, which were used to call the modified bases A (QV=50) and C (QV=100). Please see Lines 452-454.

      • Figures 1B, D, F, and Figure 2A: make the Venn diagram to scale

      Yes, we have revised.

      • L110-111: missing p-value to say that the authors observe a bigger overlap in Pst than Psph as they observed more modified sites in Pst

      Sorry for the confusion. We said it had a bigger overlap in Pst because the number 17.7 was bigger than the number of 15 in Psph. To avoid misunderstanding, we revised the wording to “more genes equipped with all three modification types were detected in Pst than Psph”. Please see Lines 110-111.

      • L112: missing description of their Pss analysis (IDP, sites...)

      We have added the information for Pss in the revised manuscript.

      “Additionally, the methylome atlas of Pss revealed a lower incidence of methylation than those of Psph and Pst, particularly in terms of 6mA modifications, which were only seen in 457 significant 6mA occurrences under the same threshold (IPD > 1.5) and a total of 2,853 and 1,438 methylation sites were detected as 5mC and 4mC, respectively”. Please see Lines 114-116.

      • L118: "modification" to "modified "

      We have revised. Please see Line 119.

      • L120: "modification sites" to "modified nucleotides"

      We have revised. Please see Line 121.

      • L142: correct the title "Methylated genes revealed highly functional conservation among three P. syringae strains" maybe to "Methylated genes are functionally conserved among ..."

      We have revised. Please see Line 142.

      • Figure 2C: not easy to read and interpret

      Sorry for the confusion. Figure 2C revealed the significantly enriched functional pathways in GO and KEGG databases among three P. syringae strains. The specific names of each pathway were listed on the left, and each column with dots indicated the number of genes within one kind of methylation in one of three P. syringae strains. The larger the size, the bigger the number.

      We have revised the legend of Figure 2C. Please see Lines 575-579.

      “The dot plot revealed the significantly enriched functional pathways in GO and KEGG databases among three P. syringae strains. The specific names of each pathway were listed on the left, and each column with dots indicated the number of genes within one kind of methylation in one of three P. syringae strains. The size of the dots indicates the number of related genes.”

      • Figure 6B-C: what is the difference between B 24h and C?

      Figure 6B revealed the expression difference between WT and mutant during 24 hours. Figure 6C only showed a time point in 24 hours. To avoid repetition, we have removed Figure 6C.

      • Figure 6C-D: if the same maybe remove Figure 2C

      We have removed Figure 6D.

      Reviewer #2 (Recommendations For The Authors):

      The manuscript could be improved by addressing the following concerns:

      (1) In line 146: How to understand the percentage conserved in "more than two of the strains"?

      Sorry for the confusion, we planned to indicate the pattern that conserved in two strains and three strains. We have revised it to: “Notable, about 25% to 45% of methylated genes were conserved in two and three strains”. Please see Line 145.

      (2) In line 178: Five conserved sequence motifs should be replaced by "Six conserved sequence motifs".

      We have revised. Please see Line 176.

      (3) In Figure 2B, specify the C1, C2 and C3. "m6A" should be replaced by "6mA".

      Yes, we have revised.

      (4) In Figure S2, "m6A" should be replaced by "6mA".

      Yes, we have revised.

      (5) In line 212, please add references for the previous studies showing that growth conditions affect bacterial methylation status.

      Thank you for your suggestion. We have added the relevant references (Gonzalez and Collier, 2013), (Krebes et al., 2014), (Sanchez-Romero and Casadesus, 2020).

      (6) In line 217, "illustrate" should be "illustrated".

      Yes, we have revised. Please see Line 210.

      (7) There are some genes colored in grey, revealing bigger differences between the two strains than those related to ribosomal protein, T3SS, and alginate synthesis in Fig. 4A. Do they have important functional roles as well?

      Thank you for your suggestion. A total of 116 genes with bigger differences (|Log<sub>2</sub>FC| > 2) except for genes related to ribosomal protein, T3SS, and alginate synthesis. Among these genes, 31 were annotated as hypothetical proteins and 4 as transcription factors with unknown functions, and the remaining genes mostly encoded metabolism-related enzymes. These enzymes might have effects on growth defects in ΔhsdMSR. We added this information in the revised manuscript. Please see Line 249-254.

      (8) The authors should discuss what will be the potential signals or factors that can regulate the activity of HsdMSR. In other words, what can decide the extent of methylation through activating or suppressing the expression of HsdMSR?

      Thank you for your valuable suggestion. We have added this part in the discussion part. Please see Lines 404-415.

      “Apart from the established roles of 6mA and HsdMSR in P. syringae, certain signals or factors may influence HsdMSR expression. For instance, we confirmed that the growth phase affects methylation levels in P. syringae. Previous studies have shown that increased temperatures can reduce methylation levels, as observed in PAO1(Doberenz et al., 2017). These findings suggest that HsdMSR expression may be responsive to both intrinsic cellular states and extrinsic environmental conditions. To further explore potential upstream TFs regulating the expression of HsdMSR, we searched for TF-binding sites in the HsdMSR promoter using our own databases (Fan et al., 2020; Shao et al., 2021; Sun et al., 2024). As a result, we found three candidate TFs (PSPPH_0061, PSPPH_3268, and PSPPH_3504) that might directly bind and regulate HsdMSR expression. Future studies on these TFs and their interactions with the HsdMSR promoter would help clarify the regulatory network governing HsdMSR activity.”

      Reviewer #3 (Recommendations For The Authors):

      (1) Some figures contain dense information, which may be overwhelming for readers. Streamlining the legend for Figure 1 and resizing the Venn diagrams within it could enhance clarity and visual appeal.

      Thank you for your suggestion. We have scaled all the Venn plots in the revised version.

      (2) Incorporating a discussion about the role of the restriction-modification (RM) system in bacterial defense against phage infection into the discussion section could enrich the manuscript's context and relevance.

      Thank you for your valuable suggestion. We have added this part in the Discussion part. Please see Lines 416-427.

      “RM systems are known for their intrinsic role as innate immune systems in anti-phage infection, and present in around 90% of bacterial genomes(Oliveira et al., 2014). RM systems protect bacteria self by recognizing and degrading foreign phage DNA via methylation-specific site and restriction endonucleases (REases) (Loenen et al., 2014). In addition, other phage-resistance systems are similar to RM systems but carry extra genes. One is called the phage growth limitation (Pgl) system, which modifies and cleaves phage DNA. However, the Pgl only modifies the phage DNA in the first infection cycle, and cleaves phage DNA in the subsequent infections, which gives a warn to the neighboring cells(Hampton et al., 2020; Hoskisson et al., 2015). To counteract RM and RM-like systems, phages have evolved strategies, including unusual modifications such as hydroxymethylation, glycosylation, and glucosylation. They can also encode their own MTases to protect their DNA or employ strategies to evade restriction systems and other anti-RM defenses.(Iida et al., 1987; Murphy et al., 2013; Vasu and Nagaraja, 2013).

      (3) In line 152: What is the importance of the mentioned example of Cro/CI family TF?

      Thank you for your comments. The Cro/CI are important TFs present in phages. The interaction between Cro and CI affects bacteria immunity status in Enterohemorrhagic Escherichia coli (EHEC) strains(Jin et al., 2022). RM systems are known as a kind of phage-defense system, and hence we mentioned it here. We have added this description in the revised manuscript. Please see Lines 152-154.

      Reference:

      (1) Doberenz, S., Eckweiler, D., Reichert, O., Jensen, V., Bunk, B., Sproer, C., Kordes, A., Frangipani, E., Luong, K., Korlach, J., et al. (2017). Identification of a Pseudomonas aeruginosa PAO1 DNA Methyltransferase, Its Targets, and Physiological Roles. mBio 8. 10.1128/mBio.02312-16.

      (2) Fan, L., Wang, T., Hua, C., Sun, W., Li, X., Grunwald, L., Liu, J., Wu, N., Shao, X., Yin, Y., et al. (2020). A compendium of DNA-binding specificities of transcription factors in Pseudomonas syringae. Nat Commun 11, 4947. 10.1038/s41467-020-18744-7.

      (3) Gonzalez, D., and Collier, J. (2013). DNA methylation by CcrM activates the transcription of two genes required for the division of Caulobacter crescentus. Mol Microbiol 88, 203-218. 10.1111/mmi.12180.

      (4) Hampton, H.G., Watson, B.N., and Fineran, P.C. (2020). The arms race between bacteria and their phage foes. Nature 577, 327-336.

      (5) Hoskisson, P.A., Sumby, P., and Smith, M.C. (2015). The phage growth limitation system in Streptomyces coelicolor A (3) 2 is a toxin/antitoxin system, comprising enzymes with DNA methyltransferase, protein kinase and ATPase activity. Virology 477, 100-109.

      (6) Iida, S., Streiff, M.B., Bickle, T.A., and Arber, W. (1987). Two DNA antirestriction systems of bacteriophage P1, darA, and darB: characterization of darA− phages. Virology 157, 156-166.

      (7) Jin, M., Chen, J., Zhao, X., Hu, G., Wang, H., Liu, Z., and Chen, W.-H. (2022). An engineered λ phage enables enhanced and strain-specific killing of enterohemorrhagic Escherichia coli. Microbiology Spectrum 10, e01271-01222.

      (8) Krebes, J., Morgan, R.D., Bunk, B., Sproer, C., Luong, K., Parusel, R., Anton, B.P., Konig, C., Josenhans, C., Overmann, J., et al. (2014). The complex methylome of the human gastric pathogen Helicobacter pylori. Nucleic Acids Res 42, 2415-2432. 10.1093/nar/gkt1201.

      (9) Loenen, W.A., Dryden, D.T., Raleigh, E.A., Wilson, G.G., and Murray, N.E. (2014). Highlights of the DNA cutters: a short history of the restriction enzymes. Nucleic Acids Res 42, 3-19.

      (10) Murphy, J., Mahony, J., Ainsworth, S., Nauta, A., and van Sinderen, D. (2013). Bacteriophage orphan DNA methyltransferases: insights from their bacterial origin, function, and occurrence. Appl Environ Microb 79, 7547-7555.

      (11) Oliveira, P.H., Touchon, M., and Rocha, E.P. (2014). The interplay of restriction-modification systems with mobile genetic elements and their prokaryotic hosts. Nucleic Acids Res 42, 10618-10631.

      (12) Sanchez-Romero, M.A., and Casadesus, J. (2020). The bacterial epigenome. Nature reviews. Microbiology 18, 7-20. 10.1038/s41579-019-0286-2.

      (13) Shao, X., Tan, M., Xie, Y., Yao, C., Wang, T., Huang, H., Zhang, Y., Ding, Y., Liu, J., Han, L., et al. (2021). Integrated regulatory network in Pseudomonas syringae reveals dynamics of virulence. Cell Rep 34, 108920. 10.1016/j.celrep.2021.108920.

      (14) Sun, Y., Li, J., Huang, J., Li, S., Li, Y., Lu, B., and Deng, X. (2024). Architecture of genome-wide transcriptional regulatory network reveals dynamic functions and evolutionary trajectories in Pseudomonas syringae. bioRxiv, 2024.2001. 2018.576191.

      (15) Vasu, K., and Nagaraja, V. (2013). Diverse functions of restriction-modification systems in addition to cellular defense. Microbiol Mol Biol Rev 77, 53-72. 10.1128/MMBR.00044-12.

    1. eLife Assessment

      This manuscript offers valuable theoretical predictions on how horizontal gene transfer (HGT) can lead to alternative stable states in microbial communities. Using a modeling framework, solid theoretical evidence is provided to support the claimed role of HGT. However, given that the model has many degrees of freedom, a more comprehensive analysis of the role of different parameters could strengthen the study. Additionally, potential interactions between plasmids that carry out HGT are not discussed in the model. This paper would be of interest to researchers in microbiology, ecology, and evolutionary biology.

    2. Reviewer #2 (Public review):

      Summary:

      In this work, the authors use a theoretical model to study the potential impact of Horizontal Gene Transfer on the number of alternative stable states of microbial communities. For this, they use a modified version of the competitive Lotka Volterra model-which accounts for the effects of pairwise, competitive interactions on species growth-that incorporates terms for the effects of both an added death (dilution) rate acting on all species and the rates of horizontal transfer of mobile genetic elements-which can, in turn, affect species growth rates. The authors analyze the impact of horizontal gene transfer in different scenarios--such as bistability between pairs of species and multistability in communities--over an extended range of parameter values. In almost all these cases, the authors report an increase in either the number of alternative stable states or the parameter region (e.g. growth rate values) in which they occur.

      Understanding the origin of alternative stable states in microbial communities and how often they may occur is an important challenge in microbial ecology and evolution. Shifts between these alternative stable states can drive transitions between e.g. a healthy microbiome and dysbiosis. A better understanding of how horizontal gene transfer can drive multistability could help predict alternative stable states in microbial communities, as well as inspire novel treatments to steer communities towards the most desired (e.g. healthy) stable states. In my opinion, this manuscript is a solid theoretical approach to the subject.

      Strengths:<br /> - Generality of the model: the work is based on a phenomenological model that has been extensively used to predict the dynamics of ecological communities in many different scenarios.<br /> - The question of how horizontal gene transfer can drive alternative stable states in microbial communities is important and there are very few studies addressing it.

      Weaknesses:<br /> - In the revised version of the manuscript, the authors significantly extended the analyzed region of parameter values. Still, the model has many parameters and the analysis is typically done by changing one or two parameters at a time. Thus, the work shows how HGT can indeed promote multistability, but it remains hard to grasp whether it consistently does so across a large region of the parameter values space.

    3. Reviewer #3 (Public review):

      Hong et al. used a model they previously developed to study the impact of plasmid transfer on microbial multispecies communities. They investigated the effect of plasmid transfer on the existence of alternative stable states in a community. The model most closely resembles plasmid conjugation, where the transferred genes confer independent growth-related fitness effects and different plasmids do not affect each other's transfer or growth effects. For this process, the authors find that increasing the rate of plasmid transfer leads to an increasing number of stable states, as long as the model includes a constant death/dilution term.

      This is an interesting and important topic, and I welcome the authors' efforts to explore these topics with mathematical modeling. The addition of sensitivity analyses also strengthens the usefulness for quantitative microbial ecologists. However, the additional sections have made the main text harder to read. Between the effect of the dilution rate, the increase in subpopulations with HGT, and the modulation of interspecies competition, the reviewers have suggested a number of factors that may explain the way plasmid transfer modulates multistability. I think it would be helpful if the authors could summarize some of these effects/interactions between different parameters in their model more. I personally continue to find the model very unintuitive, especially in the way it averages over subpopulations carrying more than one foreign plasmid. Additional sentences that give the reader intuition for the sensitivity analyses and how these interplay with the results would be good.

      Specific points

      (1) The model makes strong assumptions about the biology of HGT, that could be spelled out even more. Since the model is primarily applicable to HGT driven by the exchange of plasmids, I believe the abstract (and perhaps even the title of the paper) should be updated to reflect that.

      (2) I am not surprised that a mechanism that creates diversity will lead to more alternative stable states. Specifically, the null model for the absence of HGT is to set gamma to zero, resulting in pij=0 for all subpopulations (line 454). This means that a model with N^2 classes is effectively reduced to N classes. It seems intuitive that an LV-model with many more species would also allow for more alternative stable states. For a fair comparison one would really want to initialize these subpopulations in the model (with the same growth rates - e.g. mu1(1+lambda2)) but without gene mobility.<br /> [Update:] It is good that it seems that initializing pij with non-zero abundance did not seem to affect the conclusion that higher amounts of HGT increases multi stability. However, rather than listing it as one control for a specific condition, I would argue that this is the appropriate null model across the board (where HGT rate is varied from 0 to a high value), including figures S9 and S10.

      (3) The possibility that the same cell may be counted in different pij runs counter to all intuition that researchers coming from a background of compartmental /epidemiological modeling may have. The associated assumption that plasmids do not affect each other's dynamics or (growth/interaction) effects at all is also a very strong assumption. This should be signaled much earlier in the manuscript, possibly already in line 106 when the model is introduced.

    4. Author response:

      The following is the authors’ response to the original reviews.

      Reviewer #1 (Public review):

      Summary:

      The authors present a modelling study to test the hypothesis that horizontal gene transfer (HGT) can modulate the outcome of interspecies competition in microbiomes, and in particular promote bistability in systems across scales. The premise is a model developed by the same authors in a previous paper where bistability happens because of a balance between growth rates and competition for a mutual resource pool (common carrying capacity). They show that introducing a transferrable element that gives a "growth rate bonus" expands the region of parameter space where bistability happens. The authors then investigate how often (in terms of parameter space) this bistability occurs across different scales of complexity, and finally under selection for the mobile element (framed as ABR selection).

      Strengths:

      The authors tackle an important, yet complex, question: how do different evolutionary processes impact the ecology of microbial ecosystems? They do a nice job at increasing the scales of heterogeneity and asking how these impact their main observable: bistability.

      We appreciate the reviewer for agreeing with the potential value of our analysis. We are also grateful for the constructive comments and suggestions on further analyzing the influence of the model structure and the associated assumptions. We have fully addressed the raised issues in the updated manuscript and below.

      Weaknesses:

      The author's starting point is their interaction LV model and the manuscript then explores how this model behaves under different scenarios. Because the structure of the model and the underlying assumptions essentially dictate these outcomes, I would expect to see much more focus on how these two aspects relate to the specific scenarios that are discussed. For example:

      A key assumption is that the mobile element conveys a multiplicative growth rate benefit (1+lambda). However, the competition between the species is modelled as a factor gamma that modulates the competition for overall resource and thus appears in the saturation term (1+ S1/Nm + gamma2*S2/Nm). This means that gamma changes the perceived abundance of the other species (if gamma > 1, then from the point of view of S1 it looks like there are more S2 than there really are). Most importantly, the relationship between these parameters dictates whether or not there will be bistability (as the authors state).

      This decoupling between the transferred benefit and the competition can have different consequences. One of them is that - from the point of view of the mobile element - the mobile element competes at different strengths within the same population compared to between. To what degree introducing such a mobile element modifies the baseline bistability expectation thus strongly depends on how it modifies gamma and lambda.

      Thus, this structural aspect needs to be much more carefully presented to help the reader follow how much of the results are just trivial given the model assumptions and which have more of an emergent flavour. From my point of view, this has an important impact on helping the reader understand how the model that the authors present can contribute to the understanding of the question "how microbes competing for a limited number of resources stably coexist". I do appreciate that this changes the focus of the manuscript from a presentation of simulation results to more of a discussion of mathematical modelling.

      We thank the reviewer for the insightful suggestions. We agree with the reviewer that the model structure and the underlying assumptions need to be carefully discussed, in order to understand the generality of the theoretical predictions. In particular, the reviewer emphasized that how HGT affects bistability might depend on how mobile genetic elements modified growth rates and competition. In the main text, we have shown that when mobile genes only influence species growth rates, HGT is expected to promote multistability (Fig. 1 and 2). However, when mobile genes modify species interactions, the effect of HGT on multistability is dependent on how mobile genes change competition strength (Fig. 3a to f). When mobile genes increase competition, HGT promotes multistability (Fig. 3c and e). In contrast, when mobile genes relax competition, HGT is expected to reduce multistability (Fig. 3d and f).

      In light of the reviewer’s comments, we have further generalized the model structure, by accounting for the scenario where mobile genes simultaneously modify growth rates and competition. The effect of mobile genes on growth rates is represented by the magnitude of 𝜆’s, and the influence on competition is described by another parameter 𝛿. By varying these two parameters, we can evaluate how the model structure and the underlying assumptions affect the baseline expectation. We performed additional simulations with broad ranges of 𝜆 and 𝛿 values. In particular, we analyzed whether HGT would promote the likelihood of bistability in two-species communities compared with the scenario without gene transfer (Fig. 3g-i). Our results suggested that: (1) With or without HGT, reducing 𝜆 (increasing neutrality) promotes bistability; (2) With HGT, increasing 𝛿 promotes bistability; (2) Compared with the population without HGT, gene transfer promotes bistability when 𝛿 is zero or positive, while reduces bistability when 𝛿 is largely negative. These results agree with the reviewer’s comment that the baseline bistability expectation depends on how HGT modifies gamma and lambda. In the updated manuscript, we have thoroughly discussed how the model structure and the underlying assumptions can influence the predictions (line 238-253). 

      We further expanded our analysis, by calculating how other parameters, including competition strength, growth rate ranges, and death/dilution rate, would affect the multistability of communities undergoing horizontal gene transfer (Fig. S2, S3, S9, S10, S11, S12, S13, S15). Together with the results presented in the first draft, these analysis enables a more comprehensive understanding of how different mechanisms, including but not limited to HGT, collectively shaped community multistability. In the updated manuscript, the reviewer can see the change of focus from exploring the effects of HGT to a more thorough discussion of the mathematical model. The revised texts highlighted in blue and the supplemented figures reflect such a change.

      Reviewer #2 (Public review):

      Summary:

      In this work, the authors use a theoretical model to study the potential impact of Horizontal Gene Transfer on the number of alternative stable states of microbial communities. For this, they use a modified version of the competitive Lotka Volterra model-which accounts for the effects of pairwise, competitive interactions on species growth-that incorporates terms for the effects of both an added death (dilution) rate acting on all species and the rates of horizontal transfer of mobile genetic elements-which can in turn affect species growth rates. The authors analyze the impact of horizontal gene transfer in different scenarios: bistability between pairs of species, multistability in communities, and a modular structure in the interaction matrix to simulate multiple niches. They also incorporate additional elements to the model, such as spatial structure to simulate metacommunities and modification of pairwise interactions by mobile genetic elements. In almost all these cases, the authors report an increase in either the number of alternative stable states or the parameter region (e.g. growth rate values) in which they occur.

      In my opinion, understanding the role of horizontal gene transfer in community multistability is a

      very important subject. This manuscript is a useful approach to the subject, but I'm afraid that a thorough analysis of the role of different parameters under different scenarios is missing in order to support the general claims of the authors. The authors have extended their analysis to increase their biological relevance, but I believe that the analysis still lacks comprehensiveness.

      Understanding the origin of alternative stable states in microbial communities and how often they may occur is an important challenge in microbial ecology and evolution. Shifts between these alternative stable states can drive transitions between e.g. a healthy microbiome and dysbiosis. A better understanding of how horizontal gene transfer can drive multistability could help predict alternative stable states in microbial communities, as well as inspire novel treatments to steer communities towards the most desired (e.g. healthy) stable states.

      Strengths:

      (1) Generality of the model: the work is based on a phenomenological model that has been extensively used to predict the dynamics of ecological communities in many different scenarios.

      (2) The question of how horizontal gene transfer can drive alternative stable states in microbial communities is important and there are very few studies addressing it.

      We thank the reviewer for the positive comments on the potential novelty and conceptual importance of our work. We are also grateful for the constructive suggestions on the generality and comprehensiveness of our analysis. In particular, we agree with the reviewer that a thorough analysis of the role of different parameter could further improve the rigor of this work. We have fully addressed the raised issues in the updated manuscript and below.

      Weaknesses:

      (1) There is a need for a more comprehensive analysis of the relative importance of the different model parameters in driving multistability. For example, there is no analysis of the effects of the added death rate in multistability. This parameter has been shown to determine whether a given pair of interacting species exhibits bistability or not (see e.g. Abreu et al 2019 Nature Communications 10:2120). Similarly, each scenario is analyzed for a unique value of species interspecies interaction strength-with the exception of the case for mobile genetic elements affecting interaction strength, which considers three specific values. Considering heterogeneous interaction strengths (e.g. sampling from a random distribution) could also lead to more realistic scenarios - the authors generally considered that all species pairs interact with the same strength. Analyzing a larger range of growth rates effects of mobile genetic elements would also help generalize the results. In order to achieve a more generic assessment of the impact of horizontal gene transfer in driving multistability, its role should be systematically compared to the effects of the rest of the parameters of the model.

      We appreciate the suggestions. For each of the parameters that the reviewer mentioned, we have performed additional simulations to evaluate its importance in driving multistability. 

      For the added death rate, we have calculated the bistability feasibility of two-species populations under different values of 𝐷. Our results suggested that (1) varying death rate indeed changed the bistability probability of the system; (2) when the death rate was zero, mobile genetic elements that only modify growth rates would have no effects on system’s bistability. These results highlighted the importance of added death rate in driving multistability (Fig. S2, line 136-142). 

      For the interspecies interaction strength, we first extended our analysis on two-species populations. By calculating the bistability probability under different values of 𝛾, we showed that when interspecies interaction strength was smaller than 1, the influence of HGT on population bistability became weak (Fig. S3, line 143-147). We also considered heterogenous interaction strengths in multispecies communities, by randomly sampling 𝛾<sub>ij</sub> values from uniform distributions. While our results suggested the heterogeneous distribution of 𝛾<sub>ij</sub> didn’t fundamentally change the main conclusion, the mean value and variance of 𝛾<sub>ij</sub> affected the influence of HGT on multistability. The effects of HGT on community multistability becomes stronger when the mean value of 𝛾<sub>ij</sub> gets larger than 1 and the variance of 𝛾<sub>ij</sub> is small (Fig. S12, line 190-196).

      We also analyzed different ranges of growth rates effects of mobile genetic elements. In particular, we sampled 𝜆<sub>ij</sub> values from uniform distributions with given widths. Greater width led to larger range of growth rate effects. We used five-species populations as an example and tested different ranges. Our results suggested that multistability was more feasible when the growth rate effects of MGEs were small. The qualitative relationship between HGT and community was not dependent on the range of growth rate effects (Fig. S13, line 197-205).

      (2) The authors previously developed this theoretical model to study the impact of horizontal gene transfer on species coexistence. In this sense, it seems that the authors are exploring a different (stronger interspecies competition) range of parameter values of the same model, which could potentially limit novelty and generality.

      We appreciate the comment. In a previous work (PMID: 38280843), we developed a theoretical model that incorporated horizontal gene transfer process into the classic LV framework. This model provides opportunities to investigate the role of HGT in different open questions of microbial ecology. In the previous work, we considered one fundamental question: how competing microbes coexist stably. In this work, however, we focused on a different problem: how alternative stable states emerge in complex communities. While the basic theoretical tool that we applied in the two works were similar, the scientific questions, application contexts and the implications of our analysis were largely different. The novelty of this work arose from the fact that it revealed the conceptual linkage between alternative stable states and a ubiquitous biological process, horizontal gene transfer. This linkage is largely unknown in previous studies. Exploring such a linkage naturally required us to consider stronger interspecies competitions, which in general would diminish coexistence but give rise to multistability. We believe that the analysis performed in this work provide novel and valuable insights for the field of microbial ecology. 

      With all the supplemented simulations that we carried out in light of the all the reviewer’s comments, we believe the updated manuscript also provide a unified framework to understand how different biological processes collectively shaped the multistability landscape of complex microbiota undergoing horizontal gene transfer. The comprehensive analyses performed and the diverse scenarios considered in this study also contribute to the novelty and generality of this work.  

      (3) The authors analyze several scenarios that, in my opinion, naturally follow from the results and parameter value choices in the first sections, making their analysis not very informative. For example, after showing that horizontal gene transfer can increase multistability both between pairs of species and in a community context, the way they model different niches does not bring significantly new results. Given that the authors showed previously in the manuscript that horizontal gene transfer can impact multistability in a community in which all species interact with each other, one might expect that it will also impact multistability in a larger community made of (sub)communities that are independent of (not interacting with) each-which is the proposed way for modelling niches. A similar argument can be made regarding the analysis of (spatially structured) metacommunities. It is known that, for smaller enough dispersal rates, space can promote regional diversity by enabling each local community to remain in a different stable state. Therefore, in conditions in which the impact of horizontal gene transfer drives multistability, it will also drive regional diversity in a metacommunity.

      Thanks. Based on the reviewer’s comments, we have move Fig. 3 and 4 to Supplementary Information. In the updated manuscript, we have focused more on analyzing the roles of different parameters in shaping community multistability.

      (4) In some cases, the authors consider that mobile genetic elements can lead to ~50% growth rate differences. In the presence of an added death rate, this can be a relatively strong advantage that makes the fastest grower easily take over their competitors. It would be important to discuss biologically relevant examples in which such growth advantages driven by mobile genetic elements could be expected, and how common such scenarios might be.

      We appreciate the suggestion. Mobile genetic elements can drive large growth rate differences when they encode adaptative traits like antibiotic resistance (line 197-198). 

      We also analyzed different ranges of growth rates effects of mobile genetic elements, by sampling 𝜆<sub>ij</sub> values from uniform distributions with given widths. Our results suggested that multistability was more feasible when the fitness effects of MGEs were small (Fig. S13b). The qualitative relationship between HGT and community was not dependent on the range of growth rate effects (Fig. S13a and b). We discussed these results in line 197-205 of the updated main text.

      Reviewer #3 (Public review):

      Hong et al. used a model they previously developed to study the impact of horizontal gene transfer (HGT) on microbial multispecies communities. They investigated the effect of HGT on the existence of alternative stable states in a community. The model most closely resembles HGT through the conjugation of incompatible plasmids, where the transferred genes confer independent growth-related fitness effects. For this type of HGT, the authors find that increasing the rate of HGT leads to an increasing number of stable states. This effect of HGT persists when the model is extended to include multiple competitive niches (under a shared carrying capacity) or spatially distinct patches (that interact in a grid-like fashion). Instead, if the mobile gene is assumed to reduce between-species competition, increasing HGT leads to a smaller region of multistability and fewer stable states. Similarly, if the mobile gene is deleterious an increase in HGT reduces the parameter region that supports multistability.

      This is an interesting and important topic, and I welcome the authors' efforts to explore these topics with mathematical modeling. The manuscript is well written and the analyses seem appropriate and well-carried out. However, I believe the model is not as general as the authors imply and more discussion of the assumptions would be helpful (both to readers + to promote future theoretical work on this topic). Also, given the model, it is not clear that the conclusions hold quite so generally as the authors claim and for biologically relevant parameters. To address this, I would recommend adding sensitivity analyses to the manuscript.

      We thank the reviewer for the agreeing that our work addressed an important topic and was wellconducted. We are also grateful for the suggestion on sensitivity analysis, which is very helpful to improve the rigor and generality of our conclusion. All the raised issues have been fully addressed in the updated manuscript and below.

      Specific points

      (1) The model makes strong assumptions about the biology of HGT, that are not adequately spelled out in the main text or methods, and will not generally prove true in all biological systems. These include:

      a) The process of HGT can be described by mass action kinetics. This is a common assumption for plasmid conjugation, but for phage transduction and natural transformation, people use other models (e.g. with free phage that adsorp to all populations and transfer in bursts).

      b) A subpopulation will not acquire more than one mobile gene, subpopulations can not transfer multiple genes at a time, and populations do not lose their own mobilizable genes. [this may introduce bias, see below].

      c) The species internal inhibition is independent of the acquired MGE (i.e. for p1 the self-inhibition is by s1).

      These points are in addition to the assumptions explored in the supplementary materials, regarding epistasis, the independence of interspecies competition from the mobile genes, etc. I would appreciate it if the authors could be more explicit in the main text about the range of applicability of their model, and in the methods about the assumptions that are made.

      We are grateful for the reviewer’s suggestions. In main text and methods of the updated manuscript, we have made clear the assumptions underlying our analysis. For point (a), we have clarified that our model primarily focused on plasmid transfer dynamics (line 74, 101, 517). Therefore, the process of HGT can be described by mass action kinetics, which is commonly assumed for plasmid transfer (line 537-538). For point (b), our model allows a cell to acquire more than one mobile genes. Please see our response to point (3) for details. We have also made it clear that we assumed the populations would not lose their own mobile gene completely (line 526-527). For (c), we have also clarified it in the updated manuscript (line 111-112, 527-528). 

      We have also performed a series of additional simulations to show the range of applicability of our model. In particular, we discuss the role of other mechanisms, including interspecies interaction strength, the growth rate effects of MGEs, MGE epistasis and microbial death rates in shaping the multistability of microbial communities undergoing HGT. These results were provided in Fig. S2, S3, S9, S10, S11, S12, S13 and S15.

      (2) I am not surprised that a mechanism that creates diversity will lead to more alternative stable states. Specifically, the null model for the absence of HGT is to set gamma to zero, resulting in pij=0 for all subpopulations (line 454). This means that a model with N^2 classes is effectively reduced to N classes. It seems intuitive that an LV-model with many more species would also allow for more alternative stable states. For a fair comparison, one would really want to initialize these subpopulations in the model (with the same growth rates - e.g. mu1(1+lambda2)) but without gene mobility.

      We appreciate the insightful comments. The reviewer was right that in our model HGT created additional subpopulations in the community. However, with or without HGT, we calculated the species diversity and multistability based on the abundances of the 𝑁 species (s<sub>i</sub> in our model), instead of all the p<sub>ij</sub> subpopulations. Therefore, although there exist more ‘classes’ in the model with HGT, the number of ‘classes’ considered when we calculated community diversity and multistability was equal. In light of the reviewer’s suggestion, we have also performed additional simulations, where we initialized the subpopulations in the model with nonzero abundances. Our results suggested that initializing the p<sub>ij</sub> subpopulations with non-zero abundances didn’t change the main conclusion (Fig. S11, line 188-189).

      (3) I am worried that the absence of double gene acquisitions from the model may unintentionally promote bistability. This assumption is equivalent to an implicit assumption of incompatibility between the genes transferred from different species. A highly abundant species with high HGT rates could fill up the "MGE niche" in a species before any other species have reached appreciable size. This would lead to greater importance of initial conditions and could thus lead to increased multistability.

      This concern also feels reminiscent of the "coexistence for free" literature (first described here http://dx.doi.org/10.1016/j.epidem.2008.07.001 ) which was recently discussed in the context of plasmid conjugation models in the supplementary material (section 3) of https://doi.org/10.1098/rstb.2020.0478 .

      We appreciate the comments. Our model didn’t assume the incompatibility between MGEs transferred from different species. Instead, it allows a cell to acquire more than one MGEs. In our model, p<sub>ij</sub> described the subpopulation in the 𝑖-th species that acquired the MGE from the 𝑗th species. Here, p<sub>ij</sub> can have overlaps with p<sub>ik</sub> (𝑗 ≠ 𝑘). In other words, a cell can belong to p<sub>ij</sub> and p<sub>ik</sub> at the same time. The p<sub>ij</sub> subpopulation is allowed to carry the MGEs from the other species. In the model, we used to describe the influence of the other MGEs on the growth of p<sub>ij</sub>.

      We also thank the reviewer for bringing two papers into our attention. We have cited and discussed these papers in the updated manuscript (line 355-362).

      (4) The parameter values tested seem to focus on very large effects, which are unlikely to occur commonly in nature. If I understand the parameters in Figure 1b correctly for instance, lambda2 leads to a 60% increase in growth rate. Such huge effects of mobile genes (here also assumed independent from genetic background) seem unlikely except for rare cases. To make this figure easier to interpret and relate to real-world systems, it could be worthwhile to plot the axes in terms of the assumed cost/benefit of the mobile genes of each species.

      Thanks for the comments. In the main text, we presented one simulation results that assumed relatively large effects of MGE on species fitness, as the reviewer pointed out. In the updated manuscript, we have supplemented numerical simulations that considered different ranges of fitness effects, including the fitness effect as small as 10% (Fig. S13a). We have also plotted the relationship between community multistability and the assumed fitness effects of MGEs, as the reviewer suggested (Fig. S13b). Our results suggested that multistability was more feasible when the fitness effects of MGEs were small, and changing the range of MGE fitness effects didn’t fundamentally change our main conclusion. These results were discussed in line 197-205 of the updated main text.

      Something similar holds for the HGT rate (eta): given that the population of E. coli or Klebsiella in the gut is probably closer to 10^9 than 10^12 (they make up only a fraction of all cells in the gut), the assumed rates for eta are definitely at the high end of measured plasmid transfer rates (e.g. F plasmid transfers at a rate of 10^-9 mL/CFU h-1, but it is derepressed and considered among the fastest - https://doi.org/10.1016/j.plasmid.2020.102489 ). To adequately assess the impact of the HGT rate on microbial community stability it would need to be scanned on a log (rather than a linear) scale. Considering the meta-analysis by Sheppard et al. it would make sense to scan it from 10^-7 to 1 for a community with a carrying capacity around 10^9.

      We thank the reviewer for the constructive suggestion. We have carried out additional simulations by scanning the 𝜂 value from 10<sup>-7</sup> to 1. The results suggested that increasing HGT rates started to promote multistability when 𝜂 value exceeded 10<sup>-2</sup> per hour (Fig. S9, line 337-346). This corresponds to a conjugation efficiency of 10<sup>-11</sup> cell<sup>-1</sup> ∙ mL<sup>-1</sup>∙ mL when the maximum carrying capacity equals 10<sup>9</sup> cellsmL<sup>-1</sup>, or a conjugation efficiency of 10<sup>-14</sup> cell<sup>-1</sup> ∙ hr<sup>-1</sup>∙ mL when the maximum carrying capacity equals 10<sup>12</sup> cellsmL<sup>-1</sup>.

      (5) It is not clear how sensitive the results (e.g. Figure 2a on the effect of HGT) are to the assumption of the fitness effect distribution of the mobile genes. This is related to the previous point that these fitness effects seem quite large. I think some sensitivity analysis of the results to the other parameters of the simulation (also the assumed interspecies competition varies from figure to figure) would be helpful to put the results into perspective and relate them to real biological systems.

      We appreciate the comments. In light of the reviewer’s suggestion, we have changed the range of the fitness effects and analyzed the sensitivity of our predictions to this range. As shown in Fig. S13, changing the range of MGE fitness effects didn’t alter the qualitative interplay between HGT and community multistability. We have also examined the sensitivity of the results to the strength of interspecies competition strength (Fig. S3, S10, S12). These results suggested that while the strength of interspecies interactions played an important role in shaping community multistability, the relationship between HGT rate and multistability was not fundamentally changed by varying interaction strength. In addition, we examined the role of death rates (Fig. S2). In the updated manuscript, we discussed the sensitivity of our prediction to these parameters in line 136-147, 190205, 335-354.

      Recommendations for the authors:

      Reviewer #2 (Recommendations for the authors):

      Please find below a few suggestions that, in my opinion, could help improve the manuscript.

      TITLE

      It might not be clear what I 'gene exchange communities' are. Perhaps it could be rewritten for more specificity (e.g. '...communities undergoing horizontal gene transfer').

      We have updated the title as the reviewer suggested.

      ABSTRACT

      The abstract could also be edited to improve clarity and specificity. Terms like 'complicating factors' are vague, and enumerating specific factors would be better. The results are largely based on simulations, no analytical results are plotted, so I find that the sentence starting with 'Combining theoretical derivation and numerical simulations' can be a bit misleading.

      We appreciate the suggestions. We have enumerated the specific factors and scenarios in the updated abstract (line 18-26). We have also replaced 'Combining theoretical derivation and numerical simulations' with ‘Combining mathematical modeling and numerical simulations’.

      INTRODUCTION

      -  Line 42, please revise this paragraph. The logical flow is not so clear, it seems a bit like a list of facts, but the main message might not be clear enough. Also, it would be good to define 'hidden' states or just rewrite this sentence.

      We appreciate the suggestion. In the updated manuscript, we have rewritten this paragraph to improve the logical flow and clarity (line 46-52).

      -  Line 54, there is little detail about both theoretical models and HGT in this paragraph, and mixing the two makes the paragraph less focused. I suggest to divide into two paragraphs and expand its content. For example, you could explain a bit some relevant implications of MGE.

      We appreciate the suggestion. In the updated manuscript, we have divided this paragraph into two paragraphs, focusing on theoretical models and HGT, respectively (line 55-71). In particular, we have added explanations on the implications of MGEs (line 66-69), as the reviewer suggested.

      -  Line 72, as mentioned in the abstract, it would be better to explicitly mention which confounding factors are going to be discussed.

      Thanks for the suggestion. We have rewritten this part as “We further extended our analysis to scenarios where HGT changed interspecies interactions, where microbial communities were subjected to strong environmental selections and where microbes lived in metacommunities consisting of multiple local habitats. We also analyzed the role of different mechanisms, including interspecies interaction strength, the growth rate effects of MGEs, MGE epistasis and microbial death rates in shaping the multistability of microbial communities. These results created a comprehensive framework to understand how different dynamic processes, including but not limited to HGT rates, collectively shaped community multistability and diversity” (line 75-82).

      RESULTS

      -  The basic concepts (line 77) should be explained with more detail, keeping the non-familiar reader in mind. The reader might not be familiar with the concept of bistability in terms of species abundance. Also, note that mutual inhibition does not necessarily lead to positive feedback, as an interaction strength between 0 and 1 might still be considered inhibition. In any case, in Figure 1 it is not obvious how the positive feedback is represented, the caption should explain it. Note that neither the main text nor the caption explains the metaphor of the landscape and the marble that you are using in Figure 1a.

      We have rewritten this paragraph to provide more details on the basic concepts (line 86-99). We have removed the statement about ‘mutual inhibition’ to avoid being misleading. We have also updated the caption of Fig. 1a to explain the metaphor of the landscape and the marble (line 389396). 

      -  In the classical LV model, bistability does not depend on growth rates, but only on interaction strength. Therefore, I think that much of the results are significantly influenced by the added death rate. I believe that if the death rate is set to zero, mobile genetic elements that only modify growth rates will have no effect on the system's bistability. Because of this, I think that a thorough analysis of the role of the added death (dilution) rate and the distribution of growth rates is especially needed.

      We are grateful for the reviewer’s insightful comments. In the updated manuscript, we have thoroughly analyzed the role of the added death (dilution) rate on the bistability of communities composed of two species (Fig. S2). Indeed, as the reviewer pointed out, if the death rate equals zero, mobile genetic elements that only modify growth rates will have no effect on the system's bistability. We have discussed the role of death rate in line 136-142 of the updated manuscript.

      We have also expanded our analysis on the distribution of growth rates. In particular, we considered different ranges of growth rates effects of mobile genetic elements, by sampling 𝜆<sub>ij</sub> values from uniform distributions with given widths (Fig. S13). Greater width led to larger range of growth rate effects. We used five-species populations as an example and tested different ranges.

      Our results suggested that multistability was more feasible when the growth rate effects of MGEs were small (Fig. S13b). The qualitative relationship between HGT and community was not dependent on the range of growth rate effects (Fig. S13a). These results are discussed in line 197205 of the updated manuscript.

      -  The analysis uses gamma values that, in the absence of an added death rate, render a species pair bistable. Therefore, multistability would be quite expected for a 5 species community. Note that, multistability is possible in communities of more than 2 species even if all gamma values are smaller than 1. Analyzing a wide range of interaction strength distributions would really inform on the relative role of HGT in multistability across different community scenarios.

      We are grateful for the reviewer’s suggestion. In light of the reviewer’s comments, in the updated manuscript, we have performed additional analysis by focusing on a broader range of interaction strengths (Fig. S3, S10, S12), especially the gamma values below 1 (Fig. S10). Our results agreed with the reviewer’s notion that multistability was possible in communities of more than 2 species even if all gamma values were smaller than 1 (Fig. S10). 

      -  I would recommend the authors extend the analysis of the model used for Figures 1 and 2. Figures 3 and 4 could be moved to the supplement (see my point in the public review), unless the authors extend the analysis to explain some non-intuitive outcomes for niches and metacommunities.

      Thanks. In the updated manuscript we have performed additional simulations to extend the analysis in Figure 1 and 2. These results were presented in Fig. S2, S3, S9, S10, S11, S12, and S13. We have also moved Figure 3 and 4 to SI as the reviewer suggested.

      -  The authors seem to refer to fitness and growth rates as the same thing. This could lead to confusion - the strongest competitor in a species pair could also be interpreted as the fittest species despite being the slowest grower. I think there's no need to use fitness if they refer to growth rates. In any case, they should define fitness if they want to use this concept in the text.

      We are grateful for the insightful suggestion. To avoid confusion, we have used ‘growth rate’ throughout the updated manuscript.

      -  Across the text, the language needs some revision for clarity, specificity, and scientific style. In lines 105 - 109 there are some examples, like the use of 'in a lot of systems', and ' interspecies competitions' (I believe they mean interspecies interaction strengths).

      We appreciate the reviewer for pointing them out. We have thoroughly checked the text and made the revisions whenever applicable to improve the clarity and specificity.

      -  Many plots present the HGT rate on the horizontal axis. Could the authors explain why is it that the rate of HGT is relatively important for the number of alternative stable states? I understand how from zero to a small positive number there is a qualitative change. Beyond that, it shouldn't affect bistability too much, I think. If I am right, then other parameters could be more informative to plot in the horizontal axis. If I am wrong, I think that providing an explanation for this would be valuable.

      Thanks. To address the reviewer’s comment, we have systematically analyzed the effects of HGT on community multistability, by scanning the HGT rate from 10<sup>-7</sup> to 10<sup>0</sup>hr<sup>-1</sup> . In communities of two or multiple species, our simulation results showed that multistability gradually increased with HGT rate when HGT rate exceeded 10<sup>2</sup>hr<sup>-1</sup>. These results, presented in Fig. S9 and discussed in line 337-346, provided a more quantitative relationship between multistability and HGT rate.

      While in this work we showed the potential role of HGT in modulating community multistability, our results didn’t exclude the role of the other parameters. Motivated by the comments raised by the reviewers, in the updated manuscript, we have performed additional simulations to analyze the influence of other parameters in shaping community multistability. These parameters include death or dilution rate (Fig. S2), interaction strength (Fig. S3, S9, S10, S11, S12, S14, S15), 𝜆 range (Fig. S13, S15) and 𝛿 value (Fig. 3g, h, i). In many of the supplemented results (Fig. S2b, S3b, S13b, Fig. 3g, 3h and 3i), we have also plotted the data by using these parameters as the x axis. We believe the updated work now provided a more comprehensive framework to understand how different mechanisms, including but not limited to HGT, might shape the multistability of complex microbiota. These points were discussed in line 136-147, 190-205, 238-253, 334-354 of the updated main text. 

      -  My overall thoughts on the case of antibiotic exposure are similar to those of previous sections. Very few of the different parameters of the model are analyzed and discussed. In this case, the authors increased the interaction strength to ~0.4 times higher compared to previous sections. Was this necessary, and why?

      Thanks for the comments. In the previous draft, the interaction strength 𝛾=1.5 was tested as an example. Motivated by the reviewer’s comments, in the updated manuscript, we have examined different interaction strengths, including the strength ( 𝛾 = 1.1 ) commonly tested in other scenarios. The prediction equally held for different 𝛾 values (Fig. S15). We have also analyzed different 𝜆 ranges (Fig. S15). These results, together with the analyses presented in the earlier version of the manuscript, suggested the potential role of HGT in promoting multistability for communities under strong selection. The supplemented results were presented in Fig. S15 and discussed in line 293-295 of the updated manuscript.

      -  Line 195, if a gene encodes for the production of a public good, why would its HGT reduce interaction strength? I can think of the opposite scenario: the gene is a public good, and without HGT there is only one species that can produce it. Let's imagine that the public good is an enzyme that deactivates an antibiotic that is present in the environment, and then the species that produces has a positive interaction with another species in a pairwise coculture. If HGT happens, the second species becomes a producer and does not need the other one to survive in the presence of antibiotics anymore. The interaction can then become more competitive, as e.g. competition for resources could become the dominant interaction.

      We are grateful for pointing it out. In the updated manuscript, we have removed this statement.

      DISCUSSION

      -  L 267 "by comparison with empirical estimates of plasmid conjugation rates from a previous study [42], the HGT rates in our analysis are biologically relevant in a variety of natural environments". The authors are using a normalized model and the relevance of other parameter values is not discussed. If the authors want to claim that they are using biologically relevant HGT, they should also discuss whether the rest of the parameter values are biologically relevant. I recommend relaxing this statement about HGT rates.

      We appreciate the suggestion. We agree with the reviewer that other parameters including the death/dilution rate, interactions strength and 𝜆 ranges are also important in shaping community multistability. We have performed additional analysis to show the effects of these parameters. In light of the reviewer’s suggestion, we have relaxed this statement and thoroughly discussed the context-dependent effect of HGT as well as the roles of different parameters (line 334-354).

      -  Last sentence: "Therefore, inhibiting the MGE spread using small molecules might offer new opportunities to reshape the stability landscape and narrow down the attraction domains of the disease states". It is not clear what procedure/technique the authors are suggesting. If they want to keep this statement, the authors should give more details on how small molecules can be/are used to inhibit MGE.

      We appreciated the comments. Previous studies have shown some small molecules like unsaturated fatty acids can inhibit the conjugative transfer of plasmids. By binding the type IV secretion traffic ATPase TrwD, these compounds limit the pilus biogenesis and DNA translocation. We have provided more details regarding this statement in the updated manuscripts (line 376-379).

      METHODS

      -  Line 439, mu_i should be presented as the maximum 'per capita' growth rate.

      We have updated the definition of 𝜇i following the suggestion (line 529).

      -  Line 444, this explanation is hard to follow, please expand it to provide more details. You could provide an example, like explaining that all individuals from S1 have the MGE1 and therefore they have mu_1 = mu_01 ... After HGT, their fitness changes if they get the plasmid from S2, so a term lambda2 appears.

      Thanks. In the updated manuscript, we have expanded the explanation by providing an example as the reviewer suggested (line 534-537).

      -  The normalization assumes a common carrying capacity Nm (Eqs 1-4) and then it's normalized (Eqs. 5-8). It would be better to start from a more general scenario in which each species has a different carrying capacity and then proceed with the normalization.

      We appreciate the suggestion. In the updated manuscript, we have started our derivation from the scenario where each species has a different carrying capacity before proceeding with the normalization (section 1 of Methods, line 516-554). The same equations can be obtained after normalization.

      -  I think that the meaning of kappa (the plasmid loss rate) is not explained in the text.

      Thanks for pointing it out. We have explained the meaning of kappa in the updated text (line 108, 154, 539-541, 586-587, 607).

      SUPPLEMENT

      -  Figure S4, what are the different colors in panel b?

      In panel b of Fig. S4, the different colors represent the simulation results repeated with randomized growth rates. We have made it clear in the updated SI.

      Reviewer #3 (Recommendations for the authors):

      (1) Please extend your description of the model, so it is easier to understand for readers who have not read the first paper. Especially the choice to describe the model as species and subpopulations, as opposed to writing it as MGE-carrying and MGE-free populations of each species makes it quite complicated to understand which parameters influence each other.

      Thanks for the suggestion. We have extended the model description in the updated manuscript, which provides a more detailed introduction on model configurations and parameter definitions (line 86-99, 101-113, 151-159). We have also updated the Methods to extend the model description.

      (2) Please define gamma_ji in equation 13 and eta_jki in equation 14 (how to map the indices onto the assumed directionality of the interaction).

      We have defined these two parameters in the updated manuscript (line 584-586, 630-632).

      (3)  Line 511: please add at the beginning of this paragraph that you are assuming a grid-like arrangement of patches which will be captured by dispersal term H.

      We have updated this paragraph to make this assumption clear (line 636-637).

      (4)  Line 540: "used in our model" (missing a word).

      We have corrected it in the updated manuscript.

      (5)  Currently the analyses looking at the types of growth effects HGT brings (Figures 5-7) feel very "tacked on". These are not just "confounding factors", but rather scenarios that are much more biologically realistic than the assumption of independent effects. I would introduce them earlier in the text, as I think many readers may not trust your results until they know this was considered (+ how it changes the conclusions).

      We are grateful for the suggestion. We agree with the reviewer that these biologically realistic scenarios should be introduced earlier in the text. In the updated manuscript, we have moved these analyses forward, as sections 3, 4 and 5. We have also avoided the term “confounding factors”. Instead, in the updated manuscript, we have separated these analyses into different sections, and clearly described each scenario in the section title (line 217-218, 254, 275).

      (6)  In some places the manuscript refers to HGT, in others to MGE presence (e.g. caption of Figure 6). These are not generally the same thing, as HGT could also occur due to extracellular vesicles or natural transformation etc. Please standardize the nomenclature and make it clearer which type of processes the model describes.

      We appreciate the comment. The model in this work primarily focused on the process of plasmid transfer. We have made it clear throughout the main text. 

      (7)  In many figures the y-axis starts at a value other than 0. This is a bit misleading. In addition, I would recommend changing the title "Area of bistability region" to "Area of bistability" or perhaps even "Area of multistability" (since more than two species are considered).

      Thanks for the suggestion. We have updated all the relevant figures to make sure that their y-axes start at 0. We have also changed the title “Area of bistability region” to “Area of multistability”, whenever it is applicable.

      (8)  Figure 7: what are the assumed fitness effects of the mobile genes in the simulation? Which distribution were they drawn from? Please add this info to the figure caption here and elsewhere.

      In Figure 7, we explored an extreme scenario of the fitness effects of the mobile genes, where the population was subjected to strong environmental selection and only cells carrying the mobile gene could grow. Therefore, the carriage of the mobile gene changed the species growth rate from 0 to a positive value µ<sub>i</sub>. When calculating the number of stable states in the communities, we randomly drew the µ<sub>i</sub> values from a uniform distribution between 0.3 and 0.7 hr<sup>-1</sup>. We had added this information in the figure caption (line 505-508) and method (line 615-617) of the updated manuscript.

    1. eLife Assessment

      This important study combines virology experiments and mathematical modeling to determine the nuclear export rate of each of the eight RNA segments of the influenza A virus, leading to the proposal that a specific retention of mRNA within the nucleus delays the expression of antigenic viral proteins. The proposed model for explaining the differential rate of export is compelling, going beyond the state of the art, but the experimental setup is only in partial support and further studies will be needed to confirm the proposed mechanism.

    2. Reviewer #1 (Public review):

      The authors studied why the two more antigenic proteins of the influenza A virus, hemagglutinin (HA) and neuraminidase (NA), are expressed later during the infection. They set an experimental approach consisting of a 2-hour-long infection at a multiplicity of infection of 2 with the viral strain WSN. They used cells from the lung carcinoma cell line A549. They used the FISH technique to detect the mRNAs in situ and developed an imaging-based assay for mathematically modeling and estimating the nuclear export rate of each of the eight viral segments. They propose that the delay in the expression of HA and NA is based on the retention of their mRNA within the nucleus.

      Strength

      The study of an unaddressed mechanism in influenza A virus infectious cycle, as is the late expression of HA and NA, by creating a work flow including mRNA detection (FISH) plus mathematical calculations to arrive at a model, which additionally could be useful for general biological processes where transcription occurs in a burst-like manner.

      Weakness

      The authors built on several assumptions regarding the viral infection to "quantify" the transcript' export rate lacking experimental support. It would greatly improve if more precise experiments could be performed and/or include demonstration of the assumptions made (i.e., empirically demonstrating that cRNA production does not occur within the first 2 hours of infection, and the late expression of HA and NA proteins).

    3. Reviewer #2 (Public review):

      In this study the authors developed a framework to investigate the export rates of Influenza viral RNAs translocating from the nucleus to the cytoplasm. This model suggests that the influenza virus may control gene expression at the RNA export level, namely, the retention of certain transcripts in the nucleus for longer times, allows the generation of other viral encoded proteins that are exported regularly, and only later on do certain mRNAs get exported. These encode proteins that alert the cell to the presence of viral molecules, hence keeping their emergence to very end, might help the virus to avoid detection as late as possible in the infection cycle.

      The study is of limited scope. The notion that some mRNAs are retained in the nucleus after transcription is concluded early on from the FISH data. The model does not contribute much to the understanding and is mostly confirming the FISH data. The export rate is an ambiguous number and this part is not elaborated upon. One is left with more questions since no mechanistic knowledge emerges, and no additional experimentation is attempted to try drive to a deeper understanding.

      Comments on revisions:

      The authors have implemented the comments that required textual rewriting, which does make the paper clearer. On the experimental side, very little was done. It is fine to answer that the suggested experiments are not relevant or feasible for one reason or another, but one would expect to see some effort in providing other experimental sets to address key comments, and not only to modify a sentence in the text. So in my mind this round of revision feels more like some kind of intellectual discussion, which is fine, but I would have expected more, particularly after so much time has passed. I am still not satisfied with the way the analysis is presented in Fig. 2B, and writing a line about what is not analyzed in the legend, does not seem clear enough.

    4. Author response:

      The following is the authors’ response to the original reviews.

      We thank the editors and reviewers for the comments and suggestions on our manuscript.  The main point that we wished to convey in this paper was the concept and the kinetic model that enabled the estimation of nuclear export rate from an image of single mRNAs localised in single cells.  By studying the influenza viral transcripts with this model, we report the variation in the mRNA nuclear export rate of the eight viral segments.  Of note, the hemagglutinin and neuraminidase mRNAs were the slowest among the eight segments in exiting the nucleus.  We agree that the potential mechanism and the biological impact of this observation require further validation, as the reviewers pointed out.  We revised our manuscript to describe these points separately (Lines 21-25, Abstract; Lines 86-91, Introduction; Lines 316-320, Results; Lines 372-381, Discussion).  We also highlight below, the revisions that we made to address the specific points raised by the reviewers.  

      Influenza viral transcription

      The authors used specific settings for their virology experiments and several assumptions regarding their mathematical modelling, so it's extremely important that the reader has the viral life cycle clearly understood before immersing themselves in the results. Thus, a detailed explanation of the viral life cycle, including the kinetics of each step, would be extremely helpful if included in the introduction section.  Reviewer #1

      We have included the molecular composition of influenza vRNP and the mechanism of viral transcription in the revised manuscript (Lines 46-53).  

      Line 45: "Eight viral RNA segments are transcribed by the same set of molecular machinery" (Ref. 7). What's known about the arrival of the viral RNA segments in the nucleus? Is it synchronized? The authors will understand that my concern is related to the fact that a differential arrival would indeed impact the transcription and export processes.  Reviewer #1

      The arrival of eight vRNPs in the nucleus is not synchronised, with each of the eight vRNPs arriving independently (Chou et al. PLOS Pathogens 2013) (Lakadamyali et al, PNAS 2003).  This does not compromise our model, as our model estimates the export rate of each mRNA species individually (also please see our response in Model assumption below).  This is included in the second paragraph of the Discussion section (Lines 390-400).  

      Model assumption

      Even though I do not have the expertise to assess the authors' mathematical model, I do not doubt its robustness. Even so, I find some virological concerns related to the set-up of their experiments. According to what I understand, the authors performed non-synchronized 2 h-long infections with the WSN strain of influenza A virus. They did this to avoid cRNA production (and cross-reaction of the probes), which they claim to occur "much later than mRNA synthesis". Then they omit the degradation of the mRNAs for their model without giving an explanation for having done so. So, taking all these into account, it seems to me that too many assumptions are made without a strong argument. I understand that they are made in order to simplify their model, but I strongly consider that the model would gain strength if some of these events were experimentally considered. Thus, would it be possible to perform synchronized infections? Would it be possible to empirically demonstrate that cRNA production does not occur within the first 2 hours of infection and/or separate transcription and replication? Would it be possible to incorporate a degradation inhibitor of the mRNAs into their infections? If all these could be achieved, then the results coming out of the mathematical model would be enormously reinforced.  Reviewer #1

      * The study lacks experimental data that would help support the conclusions. For instance, perturbations are many times used to prove a point related to gene expression. An example for Fig. 2 for such an experiment could be to treat the cells with transcription inhibitors (e.g. DRB, 5,6-dichloro1-beta-D-ribofuranosylbenzimidazole). Preventing transcription leaves only mature RNAs in the nucleus, and then using this system one can compare the export rate of different RNAs.  Reviewer #2

      We agreed that the primary concern in our model was the assumption that the mRNA degradation could be omitted.  Synchronised infection is not necessary; in fact, non-synchronised infection is preferred, as we explain later in our response.  Additionally, the dominance of mRNA production over the cRNA production has been documented elsewhere.  To address mRNA degradation and validate our model estimation, we performed a time-course measurement using baloxavir.  Baloxavir efficiently blocks the viral transcription by inhibiting the nuclease activity in PA.  DRB, suggested by the reviewer, allows influenza viral transcription and causes viral transcripts to accumulate in the nucleus for unknown mechanisms (Amorim et al. Traffic 2007 and our observation using smFISH, not shown).  The additional experiment, now presented in Fig. 5 in the revised manuscript, indicated that the mRNA degradation is minimal, and the export rate estimated in our model and the time-course experiment agreed well for the HA segment.  The experiment raised the possibility that the time-course measurement underestimates the export rate of transcripts that exit the nucleus rapidly, such as NP.  A real-time imaging of single transcripts would be necessary to directly measure the true nuclear export rate; however, this is beyond the scope of our paper.  The new result is now presented in Fig. 5, Supplementary figures 3 and 4, and in the main text (Lines 322-360).  An alteration was also made in Line 286 to guide to Fig. 5.  The Materials and Methods section was updated (Lines 478-482).  

      We note that our model does not require synchronised infection.  Even under synchronised infection, such as incubating cells with the virus at 4°C to facilitate attachment and subsequently shifting to 37°C to allow viral entry, the inherent heterogeneity in vRNP migration to the nucleus still remains.  This randomness does not compromise our model; rather, our model exploits this random arrival of each vRNP in each cell in the system.  This variation, in turn, generates cells carrying varying amounts of transcripts, enabling the estimation of nuclear export rate.  Importantly, more variation ensures the broader distribution of transcript levels, enabling more precise parameter fitting in our model.  It is also important to note that our model does not require the correlation between segments.  Our model estimates the export rate of each mRNA species individually.  These important points were explained in the Discussion section (Lines 390-400).  

      * There is no concrete value given for the export rates and what they might mean biologically (e.g. time present/stuck in the nucleus) - Fig. 4D. This leaves the reader in the dark.  Reviewer #2

      The export rate lambda (previously denoted as k) in our model (Fig. 4) and the decay constant k in the time-course measurement (Fig. 5) represent the proportion of mRNAs exported from the nucleus in an infinitesimal time, defining the nuclear export rate.  This has been clarified in the revised manuscript (Lines 314-316), with some alterations to make the parameter use more comprehensive.  

      -  The Greek letter k previously used in Fig. 4 and the associated equations was consistently replaced with lambda to avoid the confusion with the parameter k that is subsequently used for the exponent decay in Fig. 5 in the revised manuscript.  

      -  The Greek letter epsilon (previously used to represent export) was replaced with mu, slightly more common for representing the rate of transport.  

      -  The term “velocity” was consistently replaced with “rate” in the context of the nuclear export (Lines 163, 215, 320, 441).  

      -  The phrase “molar concentrations of mRNAs” was corrected for “molecules of mRNAs” (Line 282).

      Also, we have now described our model in two sections: “Conceiving the model” and “Implementing a kinetic model to estimate the nuclear export rate” in the Result.  The first section outlines the conceptual framework of the model, and the second focuses on its implementation and the parameter extraction (Lines 227 and 277).  

      Applicability of the model

      Lines 27-29. "Our framework presented in this study can be widely used for investigating the nuclear retention of nascent transcripts produced in a transcription burst." In my opinion, this is the strongest point of the manuscript: developing a mathematical model to analyze nuclear export retention as a mechanism of protein expression control, which could lay the foundation for further biological processes. The authors revisit this idea in the Discussion section. However, which would be those processes for which the model could be helpful? I consider that a more conspicuous discussion on this topic would broaden the readers scope, a crucial point under the eLife scope.  Reviewer #1

      * Could this framework be used to quantify the nuclear export rate of cellular RNAs? According to the explanation in the Discussion, it would seem that this approach is limited to quantifying the export rate of influenza RNAs.  Reviewer #2

      Our model is not limited to the influenza virus infection.  Our model is applicable for systems where transcription is initiated concurrently, such as when stimuli trigger the activation of a certain set of genes for transcription.  Therefore, this makes it particularly valuable for quantifying the nuclear retention of mRNAs in a transcription burst.  This point is reiterated in Line 383-390.  

      Potential mechanisms for differential nuclear export rate of viral segments

      * There is no mechanistic insight in the study. The idea driven by this study is that gene expression is regulated by the RNA export rate. But how is that explained? Is there any molecular pathway or explanation for this model? If the transcripts are ready for export, why do the mRNAs stay inside the nucleus? One option to consider are the export factors. Viral RNAs are exported by different pathways as mentioned (line 362), or by TREX2 (Bhat P et al Nat Comm 2023). The data shows that there is no difference observed in the export rate of different pathways. How about knocking down an important export factor to show how this affects the export rates. Or the opposite, overexpress a certain factor, would this change the nucleus/cytoplasm distribution of the retained RNAs.  Reviewer #2

      As we discussed in the paper, we are beginning to consider that each viral segment has an intrinsic sequence that determines its nuclear export rate, because previous studies on the export factors does not fully explain the variation in the nuclear export rate observed in our study.  As the reviewer suggested, a recent study (Bhat et al. Nature Communications 2023) exactly pointed out the internal sequence in the HA segment, aligning with our working hypothesis.  This point is discussed and their work (Bhat et al. 2023) has been cited in the Discussion section in the revised manuscript (Lines 446-449).  

      Biological impact of the nuclear retention

      The authors mention several times throughout the manuscript that the virus might use the nuclear retention of mRNA for HA and NA to postpone the expression of these antigenic molecules. At this point, I need to admit that a great question mark appeared in my mind, maybe related to the fact that some knowledge is lacking in my analysis. Lines 328-330: "On the other hand, pushing back the expression of viral antigens HA and NA would be beneficial for the virus to delay the host immune response against the infected cells in which the virus is being replicated." As I tend to understand, the host immune response recognizes HA and NA within the viral particle, if so and independently of the time that HA and Na arrive at the virus assembly step, the progeny' viral particles that are complete and extruded from the cells would be those awakening the host immunity response. If this is right, how would the delayed export of those proteins from the nucleus (and their late expression) be beneficial for delaying the immune response? I would appreciate an explanation for this point, and if I am wrong, then there could exist a relationship between nuclear export rate and the pathogenicity of different strains of influenza A virus. If so, could the authors challenge their model with additional viral strains showing a differential immune response pattern? A deeper analysis in this direction would greatly strengthen the message in their manuscript.  Reviewer #1

      * Is the timing of viral protein appearance in accordance with the time the mRNA is exported to the cytoplasm. It is logical that the first mRNA to go to the cytoplasm would be the first to become a protein. Can the authors show that nuclear retention of mRNA would push back the expression of the viral antigens HA and NA.  Reviewer #2

      Three types of immune reactions are being studied extensively.  The first is the humoral immune response, where antibodies target the viral antigens HA and NA on the viral envelope, coating and inactivating the viral particles.  The second is the cytotoxic T cell response.  There is growing evidence that cytotoxic T cells react against NP, eliciting cross-reaction to broader range of influenza viral strains.  This reaction is not specific to HA and NA, and antigens are processed in the cytoplasm and presented on the MHC.  The third is antibody-dependent cellular cytotoxicity (ADCC), where antibodies recognise the viral proteins on the cellular surface (HA and NA) of infected cells, facilitating their elimination by the NK cells.  Although protein translation may begin as soon as the first mRNA exits the nucleus, the virus may delay the peak of the antigen production and therefore, postpone the NK-mediated ADCC.  This specific point, along with references to ADCC in influenza virus infection, has been clarified in the Discussion section (Lines 377-381).  

      Data analysis and presentation

      Lines 99-101. "Viral mRNAs were detected as single diffraction-limited spots in the three-dimensional image stacks, allowing for absolute mRNA quantification (Fig. 1B)". What do the authors mean to say by "absolute mRNA quantification"? Do they refer to the total spots or the total mRNAs? Is it assumed that one spot corresponds to a single mRNA transcript? This is not clear at all for this reviewer, which could be the situation for a potential reader. Since it's the beginning of the story, this should be clearly stated in the manuscript.  Reviewer #1

      Each spot of fluorescent signal corresponds to a single molecule of viral mRNA.  We quantified the absolute number of transcripts in each cell.  This is clarified in the revised manuscript (Lines 104-106).  

      * Line 151: does the baseline change according to the RNA in question? The authors say that the "baseline is defined by the median of the Z distribution of peripheral mRNAs" - it seems that the number 0.731 refers only to one type of RNA (which is not mentioned at all not in the text and not in the legend). Reviewer #2

      The baseline was set using the NP mRNAs in the cytoplasm because the NP mRNA showed the widest distribution across the cytoplasm (Line 157).  

      * Also, what is all the signal that is seen outside the marked cells in Fig. 2B? There seems to be significant background in the field, does this mean much false-positive in the multiplex FISH? If so, then how do the authors know that the staining inside the cells isn't to some degree non-specific? It would be necessary to back this up with some other type of quantitative assay like qRT-PCR.  Reviewer #2

      The cells were removed from the analysis if the cytoplasmic boundary touched any edge of the field-of-view, while the signals were recovered across the entire field-of-view.  This is clarified in the figure legend (Lines 194-195).  

      Others

      * The meaning and explanation for Figure 1H -are unclear. Rephrase and make the legend more reader friendly.  Reviewer #2

      We made alterations to the legend (Lines 132-134) and the relevant lines in the main text (Lines 148-151).  

      * Fig. 2E: Is this the total transcript count or only in the nucleus? Would it be possible to find some correlation between the segments if a pair-wise analysis is performed according to nuclear-cytoplasm distribution?  Reviewer #2

      The total counts are presented.  This is clarified in the legend (Lines 199-200).  

      * Abstract -"A mathematical modelling indicated that the relationship between the nuclear ratio and the total count of mRNAs in single cells is dictated by a proxy for the nuclear export rate." - this sentence is very unclear.  Reviewer #2

      The sentence was removed in the revised manuscript (Line 21).  This removal did not affect the overall meaning in the abstract.  We also made an alteration to Line 279 that contained a similar phrase.  

      * The use of the word "acutely" (lines 16 and 35) is strange.  Reviewer #2

      They have been removed (now Lines 15, 33).  

      * Line 157 - "This result indicates that the velocity of viral mRNA export from the nucleus varies according to the viral segments." - not velocity, maybe timing.  Reviewer #2

      We consistently replaced “velocity” with “rate” (Lines 163, 215, 320, 441).

      * Reference for line 41.  Reviewer #2

      A reference (Waker et al. Trends Microbiol. 2019) has been cited (Line 39).  

      * Reference for lines 105-106.  Reviewer #2

      The gene length of each segment was indicated in the sentence (Line 137).  

      * Line 264- why here is 0.02 M.O.I used compared to line 97 where 2 is used?  Reviewer #2

      We used M.O.I. of 0.02 to allow for spot quantification over longer periods of observation (Lines 269-270).  

      * NS1 is expressed at late infection times and might alter the nuclear export of viral mRNAs (line 352). Need to show that indeed it is not expressed in the experiments done here.  Reviewer #2

      It is not possible to definitely prove that NS1 is not expressed due to the sensitivity limitations.  However, we minimised the its impact by investigating at the early time point (Lines 415416).  

      * Line 459- 30% formamide? Is this correct or should it be 10%?  Reviewer #2

      This is correct.  The probes used were longer than the others for smFISH.  Therefore, we washed away the probes with the stringent condition.

    1. eLife Assessment

      This study reports a model of 8 somatosensory areas of the rat cortex consisting of 4.2 million morphologically and electrically detailed neurons. The authors carry out simulation experiments aimed at understanding how multiscale organization of the cortical network shapes neural activity. While the reviewers found the results to be solid, they note that they could have likely been obtained using a much smaller portion of the model. Nonetheless, the release of the modeling platform represents a significant contribution to the field by providing a valuable resource for the scientific community.

    2. Reviewer #1 (Public review):

      This paper presents a model of the whole somatosensory non-barrel cortex of the rat, with 4.2 million morphologically and electrically detailed neurons, with many aspects of the model constrained by a variety of data. The paper focuses on simulation experiments, testing a range of observations. These experiments are aimed at understanding how multiscale organization of the cortical network shapes neural activity.

      Strengths

      • The model is very large and detailed. With 4.2 million neurons and 13.2 billion synapses, as well as the level of biophysical realism employed, it is a highly comprehensive computational representation of the cortical network.

      • Large scope of work - the authors cover a variety of properties of the network structure and activity in this paper, from dendritic and synaptic physiology to multi-area neural activity.

      • Direct comparisons with experiments, shown throughout the paper, are laudable.

      • The authors make a number of observations, like describing how high-dimensional connectivity motifs shape patterns of neural activity, which can be useful for thinking about the relations between the structure and the function of the cortical network.

      • Sharing the simulation tools and a "large subvolume of the model" is appreciated.

      Weaknesses

      • A substantial part of this paper - the first few figures - focuses on single-cell and single-synapse properties, with high similarity to what was shown in Markram et al., 2015. Details may differ, but overall it is quite similar.

      • Although the paper is about the model of the whole non-barrel somatosensory cortex, out of all figures, only one deals with simulations of the whole non-barrel somatosensory cortex. Most figures focus on simulations that involve one or a few "microcolumns". Again, it is rather similar to what was done in Markram et al., 2015 and constitutes relatively incremental progress.

      • With a model like this, one has an opportunity to investigate computations and interactions across an extensive cortical network in an in vivo-like context. However, the simulations presented are not addressing realistic specific situations corresponding to animals performing a task or perceiving a relevant somatosensory stimulus. This makes the insights into roles of cell types or connectivity architecture less interesting, as they are presented for relatively abstract situations. It is hard to see their relationship to important questions that the community would be excited about - theoretical concepts like predictive coding, biophysical mechanisms like dendritic nonlinearities, or circuit properties like feedforward, lateral, and feedback processing across interacting cortical areas. In other words, what do we learn from this work conceptually, especially, about the whole non-barrel somatosensory cortex?

      • Most of comparisons with in vivo-like activity are done using experimental data for whisker deflection (plus some from the visual stimulation in V1). But this model is for the non-barrel somatosensory cortex, so exactly the part of the cortex that has less to do with whiskers (or vision). Is it not possible to find any in vivo neural activity data from non-barrel cortex?

      • The authors almost do not show raw spike rasters or firing rates. I am sure most readers would want to decide for themselves whether the model makes sense, and for that the first thing to do is to look at raster plots and distributions of firing rates. Instead, the authors show comparisons with in vivo data using highly processed, normalized metrics.

      • While the authors claim that their model with one set of parameters reproduces many experimentally established metrics, that is not entirely what one finds. Instead, they provide different levels of overall stimulation to their model (adjusting the target "P_FR" parameter, with values from 0 to 1, and other parameters), and that influences results. If I get this right (the figures could really be improved with better organization and labeling), simulations with P_FR closer to 1 provide more realistic firing rate levels for a few different cases, however, P_FR of 0.3 and possibly above tends to cause highly synchronized activity - what the authors call bursting, but which also could be called epileptic-like activity in the network.

      • The authors mention that the model is available online, but the "Resource availability" section does not describe that in substantial detail. As they mention in the Abstract, it is only a subvolume that is available. That might be fine, but more detail in appropriate parts of the paper would be useful.

      Comments on revisions:

      The authors addressed all my comments by revising and adding text as well as revising and adding some figures and videos. The limitations described in my previous review (above) mostly remain, but they are much better acknowledged and described now. These limitations can be addressed in the future work, whereas the current paper represents a step forward relative to the state of the art and provides a useful resource for the community.

      Two minor points about the new additions to the paper:

      (1) Something does not seem right in the sentence, "Unlike the Markram et al. (2015) model, the new model can also be exploited by the community and has already been used in a number of follow up papers studying (Ecker et al., 2024a,b; ...)". Should the authors remove "studying"?

      (2) It is great that the authors added more plots and videos of the firing rates, but most of them show maximum-normalized rates, which sort of defeats the purpose. No scale on the y-axis is shown (it can be useful even for normalized data). And it is impossible to see anything for inhibitory populations.

      These are minor points that may not need to be addressed. Overall, it is a nice study that is certainly useful for the field.

      A great improvement is that the model is made fully available to the public.

    3. Author response:

      The following is the authors’ response to the previous reviews.

      Public Reviews:

      Reviewer #1 (Public review):

      This paper presents a model of the whole somatosensory non-barrel cortex of the rat, with 4.2 million morphologically and electrically detailed neurons, with many aspects of the model constrained by a variety of data. The paper focuses on simulation experiments, testing a range of observations. These experiments are aimed at understanding how the multiscale organization of the cortical network shapes neural activity.

      Strengths:

      (1) The model is very large and detailed. With 4.2 million neurons and 13.2 billion synapses, as well as the level of biophysical realism employed, it is a highly comprehensive computational representation of the cortical network.

      (2) Large scope of work - the authors cover a variety of properties of the network structure and activity in this paper, from dendritic and synaptic physiology to multi-area neural activity.

      (3) Direct comparisons with experiments, shown throughout the paper, are laudable.

      (4) The authors make a number of observations, like describing how high-dimensional connectivity motifs shape patterns of neural activity, which can be useful for thinking about the relations between the structure and the function of the cortical network.

      (5) Sharing the simulation tools and a "large subvolume of the model" is appreciated.

      We thank the reviewer for these comments and are pleased they appreciated these aspects of the work.

      Weaknesses:

      (1) A substantial part of this paper - the first few figures - focuses on single-cell and single-synapse properties, with high similarity to what was shown in Markram et al., 2015. Details may differ, but overall it is quite similar.

      We thank the reviewer for this useful comment and agree that it is important to better highlight the incremental improvements to the model’s low-level physiology. The validity of any model can continuously be improved at all spatial scales and the validity of emergent network activity increases with improved validity at lower levels. For this reason, we felt it was valuable to improve the low-level physiology of the model.

      Regarding neuron physiology, we have added the following in Section 2.1 on page 5:

      “2.1 Improved modeling and validation of neuron physiology

      Similarly to Markram et al. (2015), electrical properties of single neurons were modelled by optimizing ion channel densities in specific compartment-types (soma, axon initial segment (AIS), basal dendrite, and apical dendrite) (Figure 2B) using an evolutionary algorithm (IBEA; Van Geit et al., 2016) so that each neuron recreates electrical features of its corresponding electrical type (e-type) under multiple standardized protocols. Compared to Markram et al. (2015), electrical models were optimized and validated using 1) additional in vitro data, features and protocols, 2) ion channel and electrophysiological data corrected for the liquid junction potential, and 3) stochastic channels (StochKv3) now including inactivation profiles. The methodology and resulting electrical models are described in Reva et al. (2023) (see Methods), and generated quantitatively more accurate electrical activity, including improved attenuation of excitatory postsynaptic potentials (EPSPs) and back-propagating action potentials.”

      And page 8:

      “The new neuron models saw a 5-fold improvement in generalizability compared to Markram et al. (2015) (Reva et al., 2023).”

      We have also made the descriptions of the improvements to synaptic physiology more explicit in Section 2.2 on page 9:

      “2.2 Improved modeling and validation of synaptic physiology

      The biological realism of synaptic physiology was improved relative to Markram et al. (2015) using additional data sources and by extending the stochastic version of the Tsodyks-Markram model (Tsodyks and Markram, 1997; Markram et al., 1998; Fuhrmann et al., 2002; Loebel et al., 2009) to feature multi-vesicular release, which in turn improved the accuracy of the coefficient of variations (CV; std/mean) of postsynaptic potentials (PSPs) as described in Barros-Zulaica et al. (2019) and Ecker et al. (2020). The model assumes a pool of available vesicles that is utilized by a presynaptic action potential, with a release probability dependent on the extracellular calcium concentration ([Ca2+]o; Ohana and Sakmann, 1998; Rozov et al., 2001; Borst, 2010). Additionally, single vesicles spontaneously release as an additional source of variability with a low frequency (with improved calibration relative to Markram et al. (2015)). The utilization of vesicles leads to a postsynaptic conductance with bi-exponential kinetics. Short-term plasticity (STP) dynamics in response to sustained presynaptic activation are either facilitating (E1/I1), depressing (E2/I2), or pseudo-linear (I3). E synaptic currents consist of both AMPA and NMDA components, whilst I currents consist of a single GABAA component, except for neurogliaform cells, whose synapses also feature a slow GABAB component. The NMDA component of E synaptic currents depends on the state of the Mg2+ block (Jahr and Stevens, 1990), with the improved fitting of parameters to cortical recordings from Vargas-Caballero and Robinson (2003) by Chindemi et al. (2022).”

      (2) Although the paper is about the model of the whole non-barrel somatosensory cortex, out of all figures, only one deals with simulations of the whole non-barrel somatosensory cortex. Most figures focus on simulations that involve one or a few "microcolumns". Again, it is rather similar to what was done by Markram et al., 2015 and constitutes relatively incremental progress.

      We thank the reviewer for this comment and have added the following text to the Discussion on page 33 to explain our rationale:

      “In keeping with the philosophy of compartmentalization of parameters and continuous model refinement (see Introduction), it was essential to improve validity at the columnar scale (relative to Markram et al. (2015)) as part of demonstrating validity of the full nbS1. Indeed, improved parametrization and validation at smaller scales was essential to parameterizing background input which generated robust nbS1 activity within realistic [Ca<sup>2+</sup>]<sub>o</sub> and firing rate ranges. We view this as a major achievement, as it was unknown whether the model would achieve a stable and meaningful regime at the start of our investigation. Whilst we would have liked to go further, our primary goal was to publish a well characterized model as an open resource that others could use to undertake further in-depth studies. In this regard, we are pleased that the parametrization of the nbS1 model has already been used to study EEG signals (Tharayil et al., 2024), as well as propagation of activity between two subregions (Bolaños-Puchet and Reimann, 2024).”

      We also make it clearer in the Introduction on page 4 that the improved validation of the emergent columnar regime was essential to stable activity at the larger scale:

      “These initial validations demonstrated that the model was in a more accurate regime compared to Markram et al. (2015) – an essential step before testing more complex or larger-scale validations. For example, under the same parameterization we then observed selective propagation of stimulus-evoked activity to downstream areas, and…”

      (3) With a model like this, one has an opportunity to investigate computations and interactions across an extensive cortical network in an in vivo-like context. However, the simulations presented are not addressing realistic specific situations corresponding to animals performing a task or perceiving a relevant somatosensory stimulus. This makes the insights into the roles of cell types or connectivity architecture less interesting, as they are presented for relatively abstract situations. It is hard to see their relationship to important questions that the community would be excited about - theoretical concepts like predictive coding, biophysical mechanisms like dendritic nonlinearities, or circuit properties like feedforward, lateral, and feedback processing across interacting cortical areas. In other words, what do we learn from this work conceptually, especially, about the whole non-barrel somatosensory cortex?

      We thank the reviewer for this comment and agree that it would be very interesting to explore such topics. In the Introduction on page 4, we have updated the list of papers which have so far used the model for more in depth studies:

      “…propagation of activity between cortical areas (Bolaños-Puchet and Reimann, 2024) the role of non-random connectivity motifs on network activity (Pokorny et al., 2024) and reliability (Egas Santander et al., 2024), the composition of high-level electrical signals such as the EEG (Tharayil et al., 2024), and how spike sorting biases population codes (Laquitaine et al., 2024).”

      In the Discussion on page 33 we also add our additional thoughts on this topic:

      “Whilst we would have liked to go further, our primary goal was to publish a well characterized model as an open resource that others could use to undertake further in-depth studies. In this regard, we are pleased that the parametrization of the nbS1 model has already been used to study EEG signals (Tharayil et al., 2024), as well as propagation of activity between two subregions (Bolaños-Puchet and Reimann, 2024). Investigation, improvement and validation must be continued at all spatial scales in follow up papers with detailed description, figures and analysis, which cannot be covered in this manuscript. Each new study increases the scope and validity of future investigations. In this way, this model and paper act as a stepping stone towards more complex questions of interest to the community such as perception, task performance, predictive coding and dendritic processing. This was similar for Markram et al. (2015) where the initial paper was followed by more detailed studies. Unlike the Markram et al. (2015) model, the new model can also be exploited by the community and has already been used in a number of follow up papers studying (Ecker et al., 2024a,b; Bolaños-Puchet and Reimann, 2024; Pokorny et al., 2024; Egas Santander et al., 2024; Tharayil et al., 2024; Laquitaine et al., 2024). We believe that the number of use cases for such a general model is vast, and is made larger by the increased size of the model.”

      (4) Most comparisons with in vivo-like activity are done using experimental data for whisker deflection (plus some from the visual stimulation in V1). But this model is for the non-barrel somatosensory cortex, so exactly the part of the cortex that has less to do with whiskers (or vision). Is it not possible to find any in vivo neural activity data from the non-barrel cortex?

      We agree with the reviewer that this is a weakness. We have expanded our discussion of the need to mix data sources to also consider our view for network level activity:

      “This paper and its companion paper serve to present a methodology for modeling micro- and mesoscale anatomy and physiology, which can be applied for other cortical regions and species. With the rapid increase in openly available data, efforts are already in progress to build models of mouse brain regions with reduced reliance on data mixing thanks to much larger quantities of available atlas-based data. This also includes data for the validation of emergent network level activity. Here we chose to compare network-level activity to data mostly from the barrel cortex, as well as a single study from primary visual cortex. Whilst a lot of the data used to build the model was from the barrel cortex, the barrel cortex also represents a very well characterized model of cortical processing for simple and controlled sensory stimuli. The initial comparison of population-wise responses in response to accurate thalamic input for single whisker deflections was essential to demonstrating that the model was closer to in vivo, and we were unaware of similar data for nonbarrel somatosensory regions. Moreover, our optogenetic & lesion study demonstrated the capacity to compare and extend studies of canonical cortical processing in the whisker system.”

      (5) The authors almost do not show raw spike rasters or firing rates. I am sure most readers would want to decide for themselves whether the model makes sense, and for that, the first thing to do is to look at raster plots and distributions of firing rates. Instead, the authors show comparisons with in vivo data using highly processed, normalized metrics.

      We thank the reviewer for this comment and agree that better visualizations of the network activity under different conditions is essential for helping the reader assess the work. In addition to raster plots in Video 1, Video 3, Fig 6, Fig 5C, Fig S9a, S16a, we have additionally:

      a) Changed the histograms of spontaneous activity in Fig 4G on page 13 to raster plots for the seven column subvolume for two contrasting meta-parameter regimes.

      b) Added 4 new videos (Video 6a,b and 8a,b) showing all spontaneous and evoked meta-parameter combinations in hex0 and hex39 of the nbS1:

      We have added improved plots showing the distributions of firing rates in the seven column subvolume on page 74:

      With more detailed consideration in the Results on page 15:

      “Long-tailed population firing rate distributions with means ∼ 1Hz

      To study the firing rate distributions of different subpopulations and m-types, we ran 50s simulations for the meta-parameter combinations: [Ca<sup>2+</sup>]<sub>o</sub>: 1.05mM, R<sub>OU</sub>: 0.4,P<sub>FR</sub>: 0.3, 0.7 (Figure S4). Different subpopulations showed different sparsity levels (proportion of neurons spiking at least once) ranging from 6.6 to 42.5%. Wohrer et al. (2013) considered in detail the biases and challenges in obtaining ground truth firing rate distributions in vivo, and discuss the wide heterogeneity of reports in different modalities using different recording techniques. They conclude that most evidence points towards longtailed distributions with peaks just below 1Hz. We confirmed that spontaneous firing rate distributions were long-tailed (approximately lognormally distributed) with means on the order of 1Hz for most subpopulations. Importantly the layer-wise means were just below 1Hz in all layers for the P<sub>FR</sub> = 0.3 meta-parameter combination. Moreover, our recent work applying spike sorting to extracellular activity using this meta-parameter combination found spike sorted firing rate distributions to be lognormally distributed and very similar to in vivo distributions obtained using the same probe geometry and spike sorter (Laquitaine et al., 2024).

      (6) While the authors claim that their model with one set of parameters reproduces many experimentally established metrics, that is not entirely what one finds. Instead, they provide different levels of overall stimulation to their model (adjusting the target "P_FR" parameter, with values from 0 to 1, and other parameters), and that influences results. If I get this right (the figures could really be improved with better organization and labeling), simulations withP<sub>FR</sub> closer to 1 provide more realistic firing rate levels for a few different cases, however, P<sub>FR</sub> of 0.3 and possibly above tends to cause highly synchronized activity - what the authors call bursting, but which also could be called epileptic-like activity in the network.

      We thank the reviewer for this comment. We can now see that the motivation for P<sub>FR</sub> parameter was introduced very briefly in the results and that the results of the calibration and analysis of the spontaneous activity regime are not interpreted in relation to this parameter.

      To address this, we have given more detail where it is first introduced in the Results on page 12:

      “to account for uncertainty in the firing rate bias during spontaneous activity from extracellular spike sorted recordings…”

      We then reconsider that it represents an unknown bias when interpreting the calibration and spontaneous activity results on page 15:

      “We reemphasize that the [Ca<sup>2+</sup>]<sub>o</sub>, R<sub>OU</sub> and P<sub>FR</sub> meta-parameters account for uncertainty of in vivo extracellular calcium concentration, the nature of inputs from other brain regions and the bias of extracellularly recorded firing rates. Whilst estimates for [Ca<sup>2+</sup>]<sub>o</sub> are between 1.0 - 1.1mM (Jones and Keep, 1988; Massimini and Amzica, 2001; Amzica et al., 2002; Gonzalez et al., 2022) and estimates for PFR are in the range of 0.1 - 0.3 (Olshausen and Field, 2006), combinations of these parameters supporting in vivo-like stimulus responses in later sections will offer a prediction for the true values of these parameters. Both these later results and our recent analysis of spike sorting bias using this model (Laquitaine et al., 2024) predict a spike sorting bias corresponding to P<sub>FR</sub> ∼ 0.3, confirming the prediction of Olshausen and Field (2006).”

      And in relation to the stimulus evoked responses on page 17:

      “Specifically, simulations with PFR from 0.1 to 0.5 robustly support realistic stimulus responses, with the middle of this range (0.3) corresponding with estimates of in vivo recording bias; both the previous estimates of Olshausen and Field (2006) and from a spike sorting study using this model (Laquitaine et al., 2024).”

      Following these considerations, the remainder of the experiments using the seven column subvolume only use a single meta-parameter on page 19.

      For the full nbS1 we further discuss the importance of a P_FR value between 0.1 and 0.3 in the Results on page 26:

      “Stable spontaneous activity only emerges in nbS1 at predicted in vivo firing rates

      After calibrating the model of extrinsic synaptic input for the seven column subvolume, we tested to what degree the calibration generalizes to the entire nbS1. Notably, this included the addition of mid-range connectivity (Reimann et al., 2024). The total number of local and mid-range synapses in the model was 9138 billion and 4075 billion, i.e., on average full model simulations increased the number of intrinsic synapses onto a neuron by 45%. Particularly, we ran simulations for P<sub>FR</sub></i ∈ [0.1, 0.15, ..., 0.3] using the OU parameters calibrated for the seven column subvolume for [Ca<sup>2+</sup>]<sub>o</sub> = 1.05mM and R<sub>OU</sub> = 0.4. Each of these full nbS1 simulations produced stable non-bursting activity (Figure 8A), except for the simulation for P<sub>FR</sub></i = 0.3, which produced network-wide bursting activity (Video 6). Activity levels in the simulations of spontaneous activity were heterogeneous (Figure 8B, Video 7). In some areas, firing rates were equal to the target P<sub>FR</sub>, whilst in others they increased above the target (Figure 8C). In the more active regions, mean firing rates (averaged over layers) were on the order of 30-35% of the in vivo references for the maximum non-bursting P<sub>FR</sub> simulation (target P<sub>FR</sub> : 0.25). This range of firing rates again fits with the estimate of firing rate bias from our paper studying spike sorting bias (Laquitaine et al., 2024) and the meta-parameter range supporting realistic stimulus responses in the seven column subvolume. This also predicts that the nbS1 cannot sustain higher firing rates without entering a bursting regime.

      Finally, we also added to our discussion of biases in extracellular firing rates in the Discussion on page 32:

      “This is also inline with our recent work using the model, which estimated a spike sorting bias corresponding to PFR = 0.3 using virtual extracellular electrodes (Laquitaine et al., 2024).”

      We also thank the reviewer for pointing out that we did not define the term “bursting” in the main text. We have added the following definition and discussion in the Results on page 15:

      “Note that the most correlated meta-parameter combination [Ca<sup>2+</sup>]<sub>o</sub>: 1.1mM, R<sub>OU</sub>: 0.2, P<sub>FR</sub>: 1.0 produced network-wide “bursting” activity, which we define as highly synchronous all or nothing events (Video 1). Such activity, which may be characteristic of epileptic activity, can be studied with the model but is not the focus of this study.”

      (7) The authors mention that the model is available online, but the "Resource availability" section does not describe that in substantial detail. As they mention in the Abstract, it is only a subvolume that is available. That might be fine, but more detail in appropriate parts of the paper would be useful.

      Firstly, we are pleased to say that the full nbS1 model is now available to download, in addition to the seven hexagon subvolume. In the manuscript, we have:

      a) Added to the Introduction at the bottom of page 4:

      “To provide a framework for further studies and integration of experimental data, the full model is made available with simulation tools, as well as a smaller subvolume with the optional new connectome capturing inhibitory targeting rules from electron microscopy”.

      b) Updated the open source panel of Figure 1:

      Secondly, we thank the reviewer for noticing that the description of the available model is not well described in the “Resource availability” statement and have addressed this by:

      a) Adding the following to the “Resource availability” statement on page 36:

      “Both the full nbS1 model and smaller seven hexagon subvolume are available on Harvard Dataverse and Zenodo respectively in SONATA format (Dai et al., 2020) with simulation code. DOIs are listed under the heading ``Final simulatable models'' in the Key resources table. An additional link is provided to the SM-Connectome with instructions on how to use it with the seven hexagon subvolume model.”

      b) Creating a new subheading in the “Key resources table” titled: “Final simulatable models” to make it clearer which links refer to the final models.

      Reviewer #2 (Public review):

      Summary:

      This paper is a companion to Reimann et al. (2022), presenting a large-scale, data-driven, biophysically detailed model of the non-barrel primary somatosensory cortex (nbS1). To achieve this unprecedented scale of a bottom-up model, approximately 140 times larger than the previous model (Markram et al., 2015), they developed new methods to account for inputs from missing brain areas, among other improvements. Isbister et al. focus on detailing these methodological advancements and describing the model's ability to reproduce in vivo-like spontaneous, stimulus-evoked, and optogenetically modified activity.

      Strengths:

      The model generated a series of predictions that are currently impossible in vivo, as summarized in Table S1. Additionally, the tools used in this study are made available online, fostering community-based exploration. Together with the companion paper, this study makes significant contributions by detailing the model's constraints, validations, and potential caveats, which are likely to serve as a basis for advancing further research in this area.

      We thank the reviewer for these comments, and are pleased they appreciate these aspects of the work.

      Weaknesses:

      That said, I have several suggestions to improve clarity and strengthen the validation of the model's in vivo relevance.

      Major:

      (1) For the stimulus-response simulations, the authors should also reference, analyze, and compare data from O'Connor et al. (2010; https://pubmed.ncbi.nlm.nih.gov/20869600/) and Yu et al .(2016; https://pubmed.ncbi.nlm.nih.gov/27749825/) in addition to Yu et al. 2019, which is the only data source the authors consider for an awake response. The authors mentioned bias in spike rate measurements, but O'Connor et al. used cell-attached recordings, which do not suffer from activity-based selection bias (in addition, they also performed Ca2+ imaging of L2/3). This was done in the exact same task as Yu et al., 2019, and they recorded from over 100 neurons across layers. Combining this data with Yu et al., 2019 would provide a comprehensive view of activity across layers and inhibitory cell types. Additionally, Yu et al. (2016) recorded VPM neurons in the same task, alongside whole-cell recordings in L4, showing that L4 PV neurons filter movement-related signals encoded in thalamocortical inputs during active touch. This dataset is more suitable for extracting VPM activity, as it was collected under the same behavior and from the same species (Unlike Diamond et al., 1992, which used anesthetized rats). Furthermore, this filtering is an interesting computation performed by the network the authors modeled. The validation would be significantly strengthened and more biologically interesting if the authors could also reproduce the filtering properties, membrane potential dynamics, and variability in the encoding of touch across neurons, not just the latency (which is likely largely determined by the distance and number of synapses).

      We thank the reviewer for pointing out these very useful studies. We have taken on board this suggestion for a future model of the mouse barrel cortex.

      (2) The authors mention that in the model, the response of the main activated downstream area was confined to L6. Is this consistent with in vivo observations? Additionally, is there any in vivo characterization of the distance dependence of spiking correlation to validate Figure 8I?

      We are not aware of data confirming the propagation of activity to downstream areas being confined to layer 6 but have considered the connectivity further between these two regions on page 27, as well as studying this further in follow up work:

      “Stable propagation of evoked activity through mid-range connectivity only emerges in nbS1 at predicted in vivo firing rates

      We repeated the previous single whisker deflection evoked activity experiment in the full model, providing a synchronous thalamic input into the forelimb sub-region (S1FL; Figure 8E; Video 8 & 9). Responses in S1FL were remarkably similar to the ones in the seven column subvolume, including the delays and decays of activity (Figure 8F). However, in addition to a localized primary response in S1FL within 350μm of the stimulus, we found several secondary responses at distal locations (Figure 8E; Video 9), which was suggestive of selective propagation of the stimulus-evoked signal to downstream areas efferently connected by mid-range connectivity. The response of the main activated downstream area (visible in Figure 8E) was confined to L6 (Figure 8G). In a follow up study using the model to explore the propagation of activity between cortical regions (Bolaños-Puchet and Reimann, 2024), it is described how the model contains both a feedforward projection pattern, which projects to principally to synapses in L1 & L23, and a feedback type pattern, which principally projects to synapses in L1 & L6. On visualizing the innervation profile from the stimulated hexagon to the downstream hexagon we can see that we have stimulated a feedback pathway (Figure S16)”

      With referenced Figure S16 on page 85:

      We did find in vivo evidence of similar layer-wise and distance dependence of correlations in the somatosensory cortex discussed on page 27 of the Results:

      “The distance dependence of correlations followed a similar profile to that observed in a dataset characterizing spontaneous activity in the somatosensory cortex (Reyes-Puerta et al., 2015a) (compare red line in Figure 8I with Figure S16). In the in vivo dataset spiking correlation was also low but highest in lower layers, with short “up-states” in spiking activity constrained to L5 & 6 (see Figure 1E,F in (Reyes-Puerta et al., 2015a)). In the model, they are constrained to L6.”

      With Figure S16a on page 85 showing the distance dependence of correlations in the anaesthetized barrel cortex during spontaneous activity (digitization from the reference paper):

      (3) Across the figures, activity is averaged across neurons within layers and E or I cell types, with a limited description of single-cell type and single-cell responses. Were there any predictions regarding the responses of particular cell types that significantly differ from others in the same layer? Such predictions could be valuable for future investigations and could showcase the advantages of a data-driven, biophysically detailed model.

      We thank the review for this comment. In addition to new analyses at higher granularity addressed in other comments, we have added the following comparison of stimulus-evoked membrane potential dynamics in different subpopulations for the original connectome and SM-connectome in Figure 7 on page 24.

      This gave interesting results discussed in a new subsection on page 26:

      “EM targeting trends hyperpolarize Sst+ and HT3aR+ late response, and disinhibit L5/6 E

      Studying somatic membrane potentials for different subpopulations in response to whisker deflections shows that PV+, L23E and L4E subpopulations are largely unaffected in the SM-connectome (Figure 7E). Interestingly, Sst+ and 5HT3aR+ subpopulations show a strong hyperpolarization in the late response that isn’t present in the original connectome. Interestingly, this corresponds with a stronger late response in L5/6 E populations, which could be caused by disinhibition due to the Sst+ and 5HT3aR+ hyperpolarization. This could be explored further in follow up studies using our connectome manipulator tool (Pokorny et al., 2024).”

      (4) 2.4: Are there caveats to assuming the OU process as a model for missing inputs? Inputs to the cortex are usually correlated and low-dimensional (i.e., communication subspace between cortical regions), but the OU process assumes independent conductance injection. Can (weakly) correlated inputs give rise to different activity regimes in the model? Can you add a discussion on this?

      We agree with the reviewer that there are caveats to assuming an OU process for the model of missing inputs and have added the following to the Discussion on page 31:

      “The calibration framework could optimize per population parameters for other compensation methods, whilst still offering an interpretable spectrum of firing rate regimes at different levels of P<sub>FR</sub>. For example, more realistic compensation schemes could be explored which introduce a) correlations between the inputs received by different neurons and b) compensation distributed across dendrites, as well as at the soma. We predict that such changes would make spontaneous activity more correlated at the lower spontaneous firing rates which supported in vivo like responses (P<sub>FR</sub> : 0.1 − 0.5), which would in turn make stimulus-responses more noise correlated.”

      (5) 2.6: The network structure is well characterized in the companion paper, where the authors report that correlations in higher dimensions were driven by a small number of neurons with high participation ratios. It would be interesting to identify which cell types exhibit high node participation in high-dimensional simplices and examine the spiking activity of cells within these motifs. This could generate testable predictions and inform theoretical cell-type-specific point neuron models for excitatory/inhibitory balanced networks and cortical processing.

      We thank the reviewer for this suggestion. We have added two supplementary figures to address this suggestion, which are discussed in the Results on Page 16:

      “Additionally, we studied the structural effect on the firing rate (here measured as the inverse of the inter-spike interval, ISI, which can be thought of as a proxy of non-zero firing rate). We found that for the connected circuit, the firing rate increases with simplex dimension; in contrast with the disconnected circuit, where this relationship remains flat (see Figure S6 red vs. blue curves and Methods).

      This also demonstrates high variability between neurons, in line with biology, both structurally (Towlson et al., 2013; Nigam et al., 2016) and functionally (Wohrer et al., 2013; Buzs´aki and Mizuseki, 2014). We next identified the cell types that are overexpressed in the group of neurons that have the 5% highest values of node participation across dimensions (Figure S7). This could inform theoretical point neuron models with cell-type specificity, for example. We found that while in dimension one (i.e., node degree) this consists mostly of inhibitory cells, in higher dimensions the cell types concentrate in layers 4, 5 and 6, especially for TPC neurons. This is in line with our structural layer-wise findings in Figure 8B in Reimann et al. (2024).”

      Which reference new Figures S6 and S7:

      With the methodology for S6 described on page 49 of the Methods:

      “For any numeric property of neurons, e.g., firing rate, we evaluate the effect of dimension on it by taking weighted averages across dimensions. That is for each dimension k, we take the weighted average of the property across neurons where the weights are given by node participation on dimension k. More precisely, let N be the number of neurons and −→V ∈ RN, be a vector of a property on all the neurons e.g., the vector of firing rates. Then in each dimension k we compute

      Where is the vector of node participation on dimension k for all neurons and ・ is the dot product.

      To measure the over and underexpression of the different m-types among those with the highest 5% of values of node participation, we used the hypergeometric distribution to determine the expected distribution of m-types in a random sample of the same size. More precisely, for each dimension k and m-type m, let N<sub>total</sub> be the total number of neurons in the circuit, Nm be the number of neurons of m-type m in the circuit, Ctop be the number of neurons with the highest 5% values of node participation in dimension k, Cm the number of neurons of mtype m among these, and let P = hypergeom(N<sub>total</sub<,N<sub>m</sub>,C<sub>top</sub>) be the hypergeometric distribution.

      By definition, P(x) describes the probability of sampling x neurons of m-type m in a random sample of size C<sub>top</sub>. Therefore, using the cumulative distribution F(x) = P(Counts ≤ x), we can compute the p-values as follows:

      Small values indicate under and over representation respectively….”

      Minor:

      (1) Since the previous model was published in 2015, the neuroscience field has seen significant advancements in single-cell and single-nucleus sequencing, leading to the clustering of transcriptomic cell types in the entire mouse brain. For instance, the Allen Institute has identified ~10 distinct glutamatergic cell types in layer 5, which exceeds the number incorporated into the current model. Could you discuss 1) the relationship between the modeled me-types and these transcriptomic cell types, and 2) how future models will evolve to integrate this new information? If there are gaps in knowledge in order to incorporate some transcriptome cell types into your model, it would be helpful to highlight them so that efforts can be directed toward addressing these areas.

      We thank the reviewer for this suggestion, particularly the idea to describe what types of data would be valuable towards improving the model in future. We have added the following to the Discussion on page 33:

      “In our previous work (Roussel et al., 2023) we linked mouse inhibitory me-models to transcriptomic types (t-types) in a whole mouse cortex transcriptomic dataset (Gouwens et al., 2019). This can provide a direct correspondence in future large-scale mouse models. As we model only a single electrical type for pyramidal cells there is no one-to-one correspondence between our me-models and the 10 different pyramidal cell types identified there. We are not currently aware of any method which can recreate the electrical features of different types of pyramidal cells using only generic ion channel models. To achieve the firing pattern behavior of more specific electrical types, usually ion channel kinetics are tweaked, and this would violate the compartmentalization of parameters. In future we hope to build morpho-electric-transcriptomic type (met-type) models by selecting gene-specific ion channel models (Ranjan et al., 2019, 2024) based on the met-type’s gene expression. Data specific to different neuron sections (i.e. soma, AIS, apical/basel dendrites) of different met-types, such as gene expression, distribution of ion channels, and voltage recordings under standard single cell protocols would be particularly useful.”

      (2) For the optogenetic manipulation, it would be interesting if the model could reproduce the paradoxical effects (for example, Mahrach et al. reported paradoxical effects caused by PV manipulation in S1; https://pubmed.ncbi.nlm.nih.gov/31951197/). This seems a more relevant and non-trivial network phenomenon than the V1 manipulation the authors attempted to replicate.

      We thank the reviewer for this valuable idea. Indeed, our model is able to reproduce paradoxical effects under certain conditions. We added the following new supplementary Figure S12 demonstrating this finding (black arrows).

      Which we discuss in the Results on page 22:

      “However, at high contrasts, we observed a paradoxical effect of the optogenetic stimulation on L6 PV+ neurons, reducing their activity with increasing stimulation strength (Figure S12B; cf. Mahrach et al. (2020)). This effect did not occur under grey screen conditions (i.e., at contrast 0.0) with a constant background firing rate of 0.2 Hz or 5 Hz respectively (not shown). The individual…”

      and added to the Discussion on page 32:

      “Also, we predicted a paradoxical effect of optogenetic stimulation on L6 PV+ interneurons, namely a decrease in firing with increased stimulus strength. This is reminiscent of the paradoxical responses found by Mahrach et al. (2020) in the mouse anterior lateral motor cortex (in L5, but not in L2/3) and barrel cortex (no layer distinction) respectively. While Mahrach et al. (2020) conducted their recordings in awake mice not engaged in any behavior, we observed this effect only when drifting grating patterns with high contrast were presented. Nevertheless, consistent with their findings, we found the effect only in deep but not in superficial layers, and only for PV+ interneurons but not for PCs. Our model could therefore be used to improve the understanding of this paradoxical effect in follow up studies. These examples demonstrate that the approach of modeling entire brain regions can be used to further probe the topics of the original articles and cortical processing.”

      Recommendations for the authors:

      Reviewer #1 (Recommendations for the authors):

      My specific comments are in the Public Review. The summarizing point is that this is a sprawling paper, and it is easy for readers to get confused. Focusing on specific connections between known functional properties and findings in this model, especially for the full-scale model, will be helpful.

      We thank the reviewer for this comment and for their related recommendation (4) below, and have added subheadings through-out the results.

      Reviewer #2 (Recommendations for the authors):

      (1) P4. What are the 10 free parameters?

      We thank the reviewer for pointing out that it would be useful to summarize the 10 parameters at this stage of the text, and have adjusted the sentence to:

      “As a result, the emerging in-vivo like activity is the consequence of only 10 free parameters representing the strength of extrinsic input from other brain regions into 9 layer-specific excitatory and inhibitory populations, and a parameter controlling the noise structure of this extrinsic input.”

      (2) Table 1 and S1 are extremely useful. Could you provide a table summarizing the major assumptions or gaps in the model, their potential influence on the results, and possible ways to collect data that could support or challenge these assumptions? Currently, this information is scattered throughout the manuscript.

      We thank the reviewer for this very useful suggestion and have added a Table S8 on page 68:

      (3) Figure 4F is important, but the legend is unclear. What is the unit on the x-axis? The values seem too large to represent per-neuron measurements.

      Thank you to the reviewer for raising this. Indeed the values are estimated mean numbers of missing number synapses per neuron by population. Such numbers are difficult to estimate but we have further discussed our rationale, justification and consideration of whether these numbers are accurate in the Results, as follows:

      “Heterogeneity in synaptic density within and across neuron classes and sections makes estimating the number of missing synapses challenging (DeFelipe and Fariñas, 1992). Changing the assumed synaptic density value of 1.1 synapses/μm would only change the slope of the relationship, however. Estimates of mean number of existing and missing synapses per population were within reasonable ranges; even the larger estimate for L5 E (due to higher dendritic length; Figure S3) was within biological estimates of 13,000 ± 3,500 total afferent synapses (DeFelipe and Fariñas, 1992).”

      This text references the new supplementary Figure S3:

      Moreover, these numbers represent the number of synapses, rather than the number of connections. The number of connections is usually used for quantifications such as indegree, and are usually much lower.

      We have also updated the caption and axis labels of the original figure:

      (4) Including additional subsections or improving the indexing in the Results section could be beneficial. In its current format, it's difficult to distinguish where the model description ends and where the validation begins. Some readers may want to focus more on the validation than other parts, so clearer segmentation would improve readability.

      We have addressed this comment with the opening comment in the authors “Recommendations for authors”.

      (5) P4. 2nd paragraph. Original vs rewired connectome. The term "rewired connectome" may give the impression that it refers to an artificial manipulation rather than a modification based on the latest data. It might be helpful to use a different term (e.g., SM-connectome as described later in the paper?).

      We have adjusted the text in the introduction:

      “Additionally, we generated a new connectome which captured recently characterized spatially-specific targeting rules for different inhibitory neuron types (Schneider-Mizell et al., 2023) in the MICrONS electron microscopy dataset (MICrONS-Consortium et al., 2021), such as increased perisomatic targeting by PV+ neurons, and increased targeting of inhibitory populations by VIP+ neurons. Comparing activity to the original connectome gave predictions about the role of these additional targeting rules.”

      (6) Figures 7 B, C, D: what is v1/v2? Original vs SM-Connectome?

      We thank the reviewer for noticing this and have corrected the figure to use “Orig” and “SM” consistent with the rest of the figure.

      (7) Page 23, 2.10: what is phi?

      We thank the reviewer for noticing this inconsistency with the earlier text, and have updated the text to read: “Particularly, we ran simulations for PF R ∈ [0.1, 0.15, ..., 0.3] using the OU para-maters calibrated for the seven column subvolume for [Ca<sup>2+</sup>] = 1.05 mM and R<sub>OU</sub> = 0.4.”

    1. eLife Assessment

      This important study investigates the implications of human endogenous retrovirus (HERV) activity in myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) and fibromyalgia (FM). These findings indicate significant associations that coincide with previous literature, which has suggested roles for differential HERV activity in degenerative, inflammatory, and aging-related pathologies of the central nervous system (CNS), as well as neurotropic infections. These seminal studies can be strengthened with minor improvements to the methodologies of characterizing differential HERV activity, further characterizing downstream mechanisms by which HERV activity impacts disease and by an expansion of the datasets utilized to include additional cohorts. These compelling findings are of immediate importance to clinicians, policymakers, and researchers interested in the underlying etiology of human health and disease.

    2. Reviewer #1 (Public review):

      Summary:

      Giménez-Orenga et al. investigate the origin and pathophysiology of myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) and fibromyalgia (FM). Using RNA microarrays, the authors compare the expression profiles and evaluate the biomarker potential of human endogenous retroviruses (HERV) in these two conditions. Altogether, the authors show that HERV expression is distinct between ME/CFS and FM patients, and HERV dysregulation is associated with higher symptom intensity in ME/CFS. HERV expression in ME/CFS patients is associated with impaired immune function and higher estimated levels of plasma cells and resting CD4 memory T cells. This work provides interesting insights into the pathophysiology of ME/CFS and FM, creating opportunities for several follow-up studies.

      Strengths:

      (1) Overall, the data is convincing and supports the authors' claims. The manuscript is clear and easy to understand, and the methods are generally well-detailed. It was quite enjoyable to read.

      (2) The authors combined several unbiased approaches to analyse HERV expression in ME/CFS and FM. The tools, thresholds, and statistical models used all seem appropriate to answer their biological questions.

      (3) The authors propose an interesting alternative to diagnosing these two conditions. Transcriptomic analysis of blood samples using an RNA microarray could allow a minimally invasive and reproducible way of diagnosing ME/CFS and FM.

      Weaknesses:

      (1) The cohort analysed in this study was phenotyped by a single clinician. As ME/CFS and FM are diagnosed based on unspecific symptoms and are frequently misdiagnosed, this raises the question of whether the results can be generalised to external cohorts.

      (2) The analyses performed to unravel the causes and effects of HERV expression in ME/CFS and FM are solely based on sequencing data. Experimental approaches could be used to validate some of the transcriptomic observations.

    3. Reviewer #2 (Public review):

      Summary:

      Giménez-Orenga carried out this study to assess whether human endogenous retroviruses (HERVs) could be used to improve the diagnosis of Myalgic Encephalomyelitis/Chronic Fatigue Syndrome (ME/CFS) and Fibromyalgia (FM). To this end, they used the HERV-V3 array developed previously, to characterize the genome-wide changes in the expression of HERVs in patients suffering from ME/CFS, FM, or both, compared to controls. In turn, they present a useful repertoire of HERVs that might characterize ME/CFS and FM. For the most part, the paper is written in a manner that allows a natural understanding of the workflow and analyses carried out, making it compelling. The figures and additional tables present solid support for the findings. However, some statements made by the authors seem incomplete and would benefit from a more thorough literature review. Overall, this work will be of interest to the medical community seeking in better understanding of the co-occurrence of these pathologies, hinting at a novel angle by integrating HERVs, which are often overlooked, into their assessment.

      Strengths:

      (1) The work is well-presented, allowing the reader to understand the overall workflow and how the specific aims contribute to filling the knowledge gap in the field.

      (2) The analyses carried out to understand the potential impact on gene expression mediated by HERVs are in line with previous works, making it solid and robust in the context of this study.

      Weaknesses:

      (1) The authors claim to obtain genome-wide HERV expression profiles. However, the array used was developed using hg19, while the genomic analysis of this work are carried out using a liftover to hg38. It would improve the statement and findings to include a comparison of the differences in HERVs available in hg38, and how this could impact the "genome-wide" findings.

      (2) The authors in some points are not thorough with the cited literature. Two examples are:<br /> a) Lines 396-397 the authors say "the MLT1, usually found enriched near DE genes (Bogdan et al., 2020)". I checked the work by Bogdan, and they studied bacterial infection. A single work in a specific topic is not sufficient to support the statement that MLT1 is "usually" in close vicinity to differentially expressed genes. More works are needed to support this.<br /> b) After the previous statement, the authors go on to mention "contributing to the coding of conserved lncRNAs (Ramsay et al., 2017)". First, lnc = long non-coding, so this doesn't make sense. Second, in the work by Ramsay they mention "that contributed a significant amount of sequence to primate lncRNAs whose expression was conserved", which is different from what the authors in this study are trying to convey. Again, additional work and a rephrasing might help to support this idea.

      (3) When presenting the clusters, the authors overlook the fact that cluster 4 is clearly control-specific, and fail to discuss what this means. Could this subset of HERV be used as bona fide markers of healthy individuals in the context of these diseases? Are they associated with DE genes? What could be the impact of such associations?

      Appraisals on aims:

      The authors set specific questions and presented the results to successfully answer them. The evidence is solid, with some weaknesses discussed above that will methodologically strengthen the work.

      Likely impact of work on the field:

      This work will be of interest to the medical community looking for novel ways to improve clinical diagnosis. Although future works with a greater population size, and more robust techniques such as RNA-Seq, are needed, this is the first step in presenting a novel way to distinguish these pathologies.

      It would be of great benefit to the community to provide a table/spreadsheet indicating the specific genomic locations of the HERVs specific to each condition. This will allow proper provenance for future researchers interested in expanding on this knowledge, as these genomic coordinates will be independent of the technique used (as was the array used here).

    4. Reviewer #3 (Public review):

      The authors find that HERV expression patterns can be used as new criteria for differential diagnosis of FM and ME/CFS and patient subtyping. The data are based on transcriptome analysis by microarray for HERVs using patient blood samples, followed by differential expression of ERVs and bioinformatic analyses. This is a standard and solid data processing pipeline, and the results are well presented and support the authors' claim.

    5. Author response:

      Public Reviews:

      Reviewer #1 (Public review):

      Summary:

      Giménez-Orenga et al. investigate the origin and pathophysiology of myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) and fibromyalgia (FM). Using RNA microarrays, the authors compare the expression profiles and evaluate the biomarker potential of human endogenous retroviruses (HERV) in these two conditions. Altogether, the authors show that HERV expression is distinct between ME/CFS and FM patients, and HERV dysregulation is associated with higher symptom intensity in ME/CFS. HERV expression in ME/CFS patients is associated with impaired immune function and higher estimated levels of plasma cells and resting CD4 memory T cells. This work provides interesting insights into the pathophysiology of ME/CFS and FM, creating opportunities for several follow-up studies.

      Strengths:

      (1) Overall, the data is convincing and supports the authors' claims. The manuscript is clear and easy to understand, and the methods are generally well-detailed. It was quite enjoyable to read.

      (2) The authors combined several unbiased approaches to analyse HERV expression in ME/CFS and FM. The tools, thresholds, and statistical models used all seem appropriate to answer their biological questions.

      (3) The authors propose an interesting alternative to diagnosing these two conditions. Transcriptomic analysis of blood samples using an RNA microarray could allow a minimally invasive and reproducible way of diagnosing ME/CFS and FM.

      Weaknesses:

      (1) The cohort analysed in this study was phenotyped by a single clinician. As ME/CFS and FM are diagnosed based on unspecific symptoms and are frequently misdiagnosed, this raises the question of whether the results can be generalised to external cohorts.

      Thank you for your comment. Surely the study of larger cohorts will determine the external validity of these results in a clinical scenario. However, this pilot study, first of its kind, was designed to maximize homogeneity across participants which seemed primarily ensured by inclusion of females only diagnosed by a single experienced observer.

      (2) The analyses performed to unravel the causes and effects of HERV expression in ME/CFS and FM are solely based on sequencing data. Experimental approaches could be used to validate some of the transcriptomic observations.

      Certainly, experimental approaches may add robustness to our findings. We in fact consider taking this avenue to deepen in the observations presented here. However, the limited knowledge of HERV-mediated physiological functions may hinder the task of revealing causes and effects of HERV expression in ME/CFS and FM in the short term.

      Reviewer #2 (Public review):

      Summary:

      Giménez-Orenga carried out this study to assess whether human endogenous retroviruses (HERVs) could be used to improve the diagnosis of Myalgic Encephalomyelitis/Chronic Fatigue Syndrome (ME/CFS) and Fibromyalgia (FM). To this end, they used the HERV-V3 array developed previously, to characterize the genome-wide changes in the expression of HERVs in patients suffering from ME/CFS, FM, or both, compared to controls. In turn, they present a useful repertoire of HERVs that might characterize ME/CFS and FM. For the most part, the paper is written in a manner that allows a natural understanding of the workflow and analyses carried out, making it compelling. The figures and additional tables present solid support for the findings. However, some statements made by the authors seem incomplete and would benefit from a more thorough literature review. Overall, this work will be of interest to the medical community seeking in better understanding of the co-occurrence of these pathologies, hinting at a novel angle by integrating HERVs, which are often overlooked, into their assessment.

      Strengths:

      (1) The work is well-presented, allowing the reader to understand the overall workflow and how the specific aims contribute to filling the knowledge gap in the field.

      (2) The analyses carried out to understand the potential impact on gene expression mediated by HERVs are in line with previous works, making it solid and robust in the context of this study.

      Weaknesses:

      (1) The authors claim to obtain genome-wide HERV expression profiles. However, the array used was developed using hg19, while the genomic analysis of this work are carried out using a liftover to hg38. It would improve the statement and findings to include a comparison of the differences in HERVs available in hg38, and how this could impact the "genome-wide" findings.

      This is an important point. However, the low number of probes that were excluded from our analysis by lack of correspondence with hg38, less than 100 among the 1,290,800 probesets, was interpreted as insignificant for "genome-wide" claims. An aspect that will be detailed in the revised version of this manuscript.

      (2) The authors in some points are not thorough with the cited literature. Two examples are:

      a) Lines 396-397 the authors say "the MLT1, usually found enriched near DE genes (Bogdan et al., 2020)". I checked the work by Bogdan, and they studied bacterial infection. A single work in a specific topic is not sufficient to support the statement that MLT1 is "usually" in close vicinity to differentially expressed genes. More works are needed to support this.

      b) After the previous statement, the authors go on to mention "contributing to the coding of conserved lncRNAs (Ramsay et al., 2017)". First, lnc = long non-coding, so this doesn't make sense. Second, in the work by Ramsay they mention "that contributed a significant amount of sequence to primate lncRNAs whose expression was conserved", which is different from what the authors in this study are trying to convey. Again, additional work and a rephrasing might help to support this idea.

      Certainly, these two sentences need rephrasing to better adjust statements to current evidence and will be replaced in the revised version of this manuscript.

      (3) When presenting the clusters, the authors overlook the fact that cluster 4 is clearly control-specific, and fail to discuss what this means. Could this subset of HERV be used as bona fide markers of healthy individuals in the context of these diseases? Are they associated with DE genes? What could be the impact of such associations?

      Using control DE HERV as bona fide markers of healthy individuals seems like an interesting possibility worth exploring. Control DE HERVs (cluster 4) are indeed associated with DE genes involved in apoptosis, T cell activation and cell-cell adhesion (modules 1 and 6) (Figure 3A). The impact of which deserves further study.

      Appraisals on aims:

      The authors set specific questions and presented the results to successfully answer them. The evidence is solid, with some weaknesses discussed above that will methodologically strengthen the work.

      Likely impact of work on the field:

      This work will be of interest to the medical community looking for novel ways to improve clinical diagnosis. Although future works with a greater population size, and more robust techniques such as RNA-Seq, are needed, this is the first step in presenting a novel way to distinguish these pathologies.

      It would be of great benefit to the community to provide a table/spreadsheet indicating the specific genomic locations of the HERVs specific to each condition. This will allow proper provenance for future researchers interested in expanding on this knowledge, as these genomic coordinates will be independent of the technique used (as was the array used here).

      We agree with the reviewer that sharing genomic locations of DE HERVs in these pathologies would contribute to further development of our findings. Unfortunately, we do not hold the rights to share probe coordinates from this custom HERV-V3 microarray which we used under MTA agreement with its developer.

      Reviewer #3 (Public review):

      The authors find that HERV expression patterns can be used as new criteria for differential diagnosis of FM and ME/CFS and patient subtyping. The data are based on transcriptome analysis by microarray for HERVs using patient blood samples, followed by differential expression of ERVs and bioinformatic analyses. This is a standard and solid data processing pipeline, and the results are well presented and support the authors' claim.

    1. eLife Assessment

      This study investigated the influence of genomic information and timing of vaccine strain selection on the accuracy of influenza A/H3N2 forecasting. The authors utilised appropriate statistical methods and have provided solid evidence that is an important contribution to the evidence base. While the study addresses a key aspect of public health, the impact is rather limited by its exclusive reliance on predictive methods using genomic information, without incorporating phenotypic data.

    2. Reviewer #1 (Public review):

      Summary:

      In the paper, the authors investigate how the availability of genomic information and the timing of vaccine strain selection influence the accuracy of influenza A/H3N2 forecasting. The manuscript presents three key findings:

      (1) Using real and simulated data, the authors demonstrate that shortening the forecasting horizon and reducing submission delays for sharing genomic data improve the accuracy of virus forecasting.

      (2) Reducing submission delays also enhances estimates of current clade frequencies.

      (3) Shorter forecasting horizons, for example, allowed by the proposed use of "faster" vaccine platforms such as mRNA, resulting in the most significant improvements in forecasting accuracy.

      Strengths:

      The authors present a robust analysis, using statistical methods based on previously published genetic-based techniques to forecast influenza evolution. Optimizing prediction methods is crucial from both scientific and public health perspectives. The use of simulated as well as real genetic data (collected between April 1, 2005, and October 1, 2019) to assess the effects of shorter forecasting horizons and reduced submission delays is valuable and provides a comprehensive dataset. Moreover, the accompanying code is openly available on GitHub and is well-documented.

      Weaknesses:

      While the study addresses a critical public health issue related to vaccine strain selection and explores potential improvements, its impact is somewhat constrained by its exclusive reliance on predictive methods using genomic information, without incorporating phenotypic data. The analysis remains at a high level, lacking a detailed exploration of factors such as the genetic distance of antigenic sites.

      Another limitation is the subsampling of the available dataset, which reduces several tens of thousands of sequences to just 90 sequences per month with even sampling across regions. This approach, possibly due to computational constraints, might overlook potential effects of regional biases in clade distribution that could be significant. The effect of dataset sampling on presented findings remains unexplored. Although the authors acknowledge limitations in their discussion section, the depth of the analysis could be improved to provide a more comprehensive understanding of the underlying dynamics and their effects.

    3. Reviewer #2 (Public review):

      Summary:

      The authors have examined the effects of two parameters that could improve their clade forecasting predictions for A(H3N2) seasonal influenza viruses based solely on analysis of haemagglutinin gene sequences deposited on the GISAID Epiflu database. Sequences were analysed from viruses collected between April 1, 2005 and October 1, 2019. The parameters they investigated were various lag periods (0, 1, 3 months) for sequences to be deposited in GISAID from the time the viruses were sequenced. The second parameter was the time the forecast was accurate over projecting forward (for 3,6,9,12 months). Their conclusion (not surprisingly) was that "the single most valuable intervention we could make to improve forecast accuracy would be to reduce the forecast horizon to 6 months or less through more rapid vaccine development". This is not practical using conventional influenza vaccine production and regulatory procedures. Nevertheless, this study does identify some practical steps that could improve the accuracy and utility of forecasting such as a few suggested modifications by the authors such as "..... changing the start and end times of our long-term forecasts. We could change our forecasting target from the middle of the next season to the beginning of the season, reducing the forecast horizon from 12 to 9 months.'

      Strengths:

      The authors are very familiar with the type of forecasting tools used in this analysis (LBI and mutational load models) and the processes used currently for influenza vaccine virus selection by the WHO committees having participated in a number of WHO Influenza Vaccine Consultation meetings for both the Southern and Northern Hemispheres.

      Weaknesses:

      The conclusion of limiting the forecasting to 6 months would only be achievable from the current influenza vaccine production platforms with mRNA. However, there are no currently approved mRNA influenza vaccines, and mRNA influenza vaccines have also yet to demonstrate their real-world efficacy, longevity, and cost-effectiveness and therefore are only a potential platform for a future influenza vaccine. Hence other avenues to improve the forecasting should be investigated.

      While it is inevitable that more influenza HA sequences will become available over time a better understanding of where new influenza variants emerge would enable a higher weighting to be used for those countries rather than giving an equal weighting to all HA sequences.

      Also, other groups are considering neuraminidase sequences and how these contribute to the emergence of new or potentially predominant clades.

    4. Author response:

      Thank you to the reviewers and editors for their positive and constructive comments. Based on this feedback, we can see that we need to clarify that the primary goal of this paper is a test of potential changes in public health policy rather than a test of technical improvements to forecasting models. We briefly summarize the primary goal below to address these public reviews and list our proposed revisions to the manuscript based on reviewer feedback.

      All real-time forecasting models contend with 2 major constraints:

      (1) How far into the future they have to predict

      (2) How rapidly the data used for predictions become available in real time

      In the case of evolutionary influenza forecasts, the current values of these constraints are 1) 12 months into the future and 2) an average lag of ~3 months for hemagglutinin (HA) sequences to become available after sample collection. Regardless of the predictors we use in these models (genetic or phenotypic), our units of prediction always depend on HA: the HA protein is the primary target of our immunity, HA is the only gene whose composition is determined by the vaccine selection process, and influenza diversity is historically defined by clades in HA phylogenies.

      Our primary goal of this study was to understand the relative effect sizes of these two common constraints on forecasting while holding all other variables as constant as possible. With this understanding, we hoped to better inform public health priorities and set realistic expectations for current and future forecasting efforts regardless of the technical specifications of each forecasting model. In other words, the goal of this study was not to optimize prediction methods but to estimate the effects of potential policy changes on forecast accuracy.

      We found that reducing how far into the future we need to predict consistently reduced our forecasting error in simulated populations (where we knew the true fitness of each virus) and in natural populations (where we either estimated fitness from genetic predictors or we knew the true fitness of each virus based on its future success). Figure 6 and its first supplemental figure show these effect sizes for natural and simulated populations, respectively, when the future fitness of each virus is known at the time of prediction. By definition, we cannot hope to improve our estimates of viral fitness for these forecasts by using other genetic or phenotypic information.

      Figure 6 shows that reducing how far into the future we need to predict from 12 to 6 months improves our forecasting accuracy 3 times as much as reducing the lag between sample collection and HA sequence submission to public databases. The impact of this finding is the confirmation that a faster vaccine development process would improve our forecast accuracy substantially more than faster turnaround between sample collection and sequence submission. If our public health goal is to make better predictions of future influenza populations, then this result indicates that our main priority is to speed up the vaccine development process.

      If our public health goal is to better understand the composition of currently circulating influenza populations (the units of our forecasts), then Figure 3 shows that reducing the lag between sample collection and HA sequence submission from ~3 months on average to 1 month on average reduces our uncertainty in current clade frequency estimates by half. This impact is also independent of the predictors we use in our forecasting models and is not lessened by the lack of other genetic or phenotypic information in our analyses.

      We realize that neither a 6-month vaccine development process nor a 1-month average sequence submission lag exist yet, but we believe that these are realistic and achievable goals for scientific and public health communities. We also realize that these public health goals are not mutually exclusive. By measuring the effects of these realistic changes to current policy through our forecasting experiments, we hope to inspire and motivate researchers and decision-makers who are empowered to make both of these goals a reality.

      Finally, we want to emphasize that the use of phenotypic data in forecasts introduces additional delays caused by the lag between when genetic sequences become available and when serological experiments can be performed. Most WHO influenza collaborating centers use a "sequence-first" approach where they characterize the genetic sequence and use available sequences to prioritize phenotypic experiments with serology. This additional lag in availability of phenotypic data means that a forecasting model based on genetic and phenotypic data will necessarily have a greater lag in data availability than a model based on genetic data only. This lag is important for practical forecasts, too, but because the lag reflects specific characteristics of each collaborating center and not a global policy change, we believe this topic falls outside of the scope of this study.

      Based on these public reviews and the private recommendations from reviewers, we plan to make the following revisions to this manuscript.

      ● Clarify the introduction, discussion, and abstract to emphasize the primary goal for this study to test effects of realistic changes to public health policy and note that this study does not cover improvements to forecasting models. As part of these changes, we will include a rationale for our choice of a genetic-information-only approach rather than a model that integrates phenotypic data. We will also refine Figure 1 to more clearly communicate the two factors we tested in this study.

      ● Provide a clearer explanation for the subsampling approach we use, include supplemental materials to communicate the geographic and temporal biases that exist in available HA sequence data, and discuss potential effects of different subsampling strategies.

      ● Evaluate the robustness of our results to different randomly subsampled data. We will perform additional technical replicates of our analysis workflow for natural populations, and summarize the effects of realistic interventions across replicates in a supplemental figure and the main text of the results.

      ● Investigate time-dependent effects of forecast horizons and submission lags on model accuracy to identify any potential biases in accuracy during specific historical epochs or any seasonal trends in accuracy associated with predicting future populations for the Northern or Southern Hemispheres.

      ● In the discussion, clarify how reducing submission lags would practically improve the WHO's ability to select vaccine candidate viruses and minimize jargon that currently makes the discussion less accessible to the average reader.

      ● Investigate how changes in forecast horizons and submission lags change the distance between predicted and observed future populations at antigenic positions (i.e., "epitope" positions) to understand whether we see the same effects with that subset of positions as we see across all HA positions.

    1. Author Response:

      We greatly appreciate the feedback provided by reviewers on this manuscript. One of our key objectives was to provide a comprehensive, detailed resource for researchers using single-cell transcriptomics to study arthritis, especially immune cells like macrophages. We strived to perform thorough, wide-ranging analyses that are both accessible and useful to other scientists in the field, and that we hope will serve as the basis for many future avenues of study. As such, we acknowledge that this work is a “first step”, providing a strong descriptive foundation with some mechanistic insight that we and others will continue pursuing. Preliminary studies in our laboratory seeking to dissect signaling mechanisms associated with the M-CSF pathway have illuminated how complex and context-dependent this signaling is, which is an important consideration for future in vivo investigations. Further, it is indeed true that attempting to harmonize transcriptomic data across studies, models, laboratories, and dissection/processing methods is fraught with difficulty and prone to misinterpretation – and we made an effort to highlight this in our manuscript, particularly with respect to where synovial immune cells were recovered from, and how. We encourage healthy discussion within the field for developing shared, unified protocols for harvests and processing upstream of transcriptomic experiments.

    1. eLife Assessment

      The authors report how a previously published method, ReplicaDock, can be used to improve predictions from AlphaFold-multimer (AFm) for protein docking studies. The level of improvement is modest for cases where AFm is successful; for cases where AFm is not as successful, the improvement is more significant, although the accuracy of prediction is also notably lower. The evidence for the ReplicaDock approach being more predictive than AFm is particularly convincing for the antibody-antigen test case. Overall, the study makes a valuable contribution by combining data- and physics-driven approaches.

    2. Reviewer #1 (Public review):

      Summary:

      The authors wanted to use AlphaFold-multimer (AFm) predictions to reduce the challenge of physics-based protein-protein docking.

      Strengths:

      They found two features of AFm predictions that are very useful. 1) pLLDT is predictive of flexible residues, which they could target for conformational sampling during docking; 2) the interface-pLLDT score is predictive of the quality of AFm predictions, which allows the authors to decide whether to do local or global docking.

      Weaknesses:

      (1) As admitted by the authors, the AFm predictions for the main dataset are undoubtedly biased because these structures were used for AFm training. Could the authors find a way to assess the extent of this bias?<br /> (2) For the CASP15 targets where this bias is absent, the presentation was very brief. In particular, I'm interested in seeing how AFm helped with the docking? They may even want to do a direct comparison with docking results w/o the help of AFm.

      Comments on revisions:

      This revision has addressed my previous comments.

    3. Reviewer #2 (Public review):

      Summary:

      In short, this paper uses a previously published method, ReplicaDock to improve predictions from AlphaFold-multimer. The method generated about 25% more acceptable predictions than AFm, but more important is improving an Antibody-antigen set, where more than 50% of the models become improved.

      When looking at the results in more detail, it is clear that for the models where the AFm models are good, the improvement is modest (or not at all). See, for instance, the blue dots in Fig 6. However, in the cases where AFm fails, the improvement is substantial (red dots in Fig 6), but no models reach a very high accuracy (Fnat ~0.5 compared to 0.8 for the good AFm models). So the paper could be summarized by claiming, "We apply ReplicaDock when AFm fails", instead of trying to sell the paper as an utterly novel pipeline. I must also say that I am surprised by the excellent performance of ReplicaDock - it seems to be a significant step ahead of other (not AlphaFold) docking methods, and from reading the original paper, that was unclear. Having a better benchmark of it alone (without AFm) would be very interesting.

      These results also highlight several questions I try to describe in the weakness section below. In short, they boil down to the fact that the authors must show how good/bad ReplicaDock is at all targets (not only the ones where AFm fails. In addition, I have several more technical comments.

      Strengths:

      Impressive increase in performance on AB-AG set (although a small set and no proteins ).

      Weaknesses:

      The presentation is a bit hard to follow. The authors mix several measures (Fnat, iRMS, RMSDbound, etc). In addition, it is not always clear what is shown. For instance, in Fig 1, is the RMSD calculated for a single chain or the entire protein? I would suggest that the author replace all these measures with two: TM-score when evaluating the quality of a single chain and DockQ when evaluating the results for docking. This would provide a clearer picture of the performance. This applies to most figures and tables. For instance, Fig 9 could be shown as a distribution of DockQ scores.

      The improvements on the models where AFm is good are minimal (if at all), and it is unclear how global docking would perform on these targets, nor exactly why the plDDT<0.85 cutoff was chosen. To better understand the performance of ReplicaDock, the authors should therefore (i) run global and local docking on all targets and report the results, (ii) report the results if AlphaFold (not multimer) models of the chains were used as input to ReplicaDock (I would assume it is similar). These models can be downloaded from AlphaFoldDB.

      Further, it would be interesting to see if ReplicaDock could be combined with AFsample (or any other model to generate structural diversity) to improve performance further.

      The estimates of computing costs for the AFsample are incorrect (check what is presented in their paper). What are the computational costs for RepliaDock global docking?

      It is unclear strictly what sequences were used as input to the modelling. The authors should use full-length UniProt sequences if that were not done.

      The antibody-antigen dataset is small. It could easily be expanded to thousands of proteins. It would be interesting to know the performance of ReplicaDock on a more extensive set of Antibodies and nanobodies.

      Using pLDDT on the interface region to identify good/bas models is likely suboptimal. It was acceptable (as a part of the score) for AlphaFold-2.0 (monomer), but AFm behaves differently. Here, AFm provides a direct score to evaluate the quality of the interaction (ipTM or Ranking Confidence). The authors should use these to separate good/bad models (for global/local docking), or at least show that these scores are less good than the one they used.

      Comments on revisions:

      The inclusion of the DockQ improved the paper. No further comments.

    4. Author response:

      The following is the authors’ response to the original reviews.

      Reviewer #1 (Public Review)

      Summary:

      The authors wanted to use AlphaFold-multimer (AFm) predictions to reduce the challenge of physics-based protein-protein docking.

      Strengths:

      They found that two features of AFm predictions are very useful. 1) pLLDT is predictive of flexible residues, which they could target for conformational sampling during docking; 2) the interface-pLLDT score is predictive of the quality of AFm predictions, which allows the authors to decide whether to do local or global docking.

      Weaknesses:

      (1) As admitted by the authors, the AFm predictions for the main dataset are undoubtedly biased because these structures were used for AFm training. Could the authors find a way to assess the extent of this bias?

      Indeed, the AFm training included most of the structures in the DB5 benchmark for its training as many structures (either unbound or bound) were deposited before the training cut-off period. One of the challenges of estimating this bias is the availability of new structures - both bound and unbound deposited after the training cut-off. Estimating the extent of training bias is therefore conditional on these factors and difficult. A few studies have attempted to address this bias (Yin et al, 2022, https://doi.org/10.1002/pro.4379).

      In our study, we assess this bias by comparing the AFm structures to the bound and unbound forms and calculating their Ca RMSDs and TM-scores (new addition). We now elaborate in the Results:Dataset curation section and we have added a figure comparing the TM-scores in the supplement.

      We added a clarifying text and a note about the TM-score calculation in the manuscript as follows:

      “Since most of the benchmark targets in DB5.5 were included in AlphaFold training, there would be training bias associated with their predictions (i.e. our measured success rates are an upper bound).”

      “We also calculated the TM-scores of the AFm predicted complex structures with respect to the bound and the unbound crystal structures (Supplementary Figure S2). As TM-scores reflect a global comparison between structures and are less sensitive to local structural deviations, no strong conclusions could be derived. This is in agreement with our intuition that since both unbound and bound states of proteins will share a similar fold, and AlphaFold can predict structures with high TM-scores in most cases, gauging the conformational deviations with TM-scores would be inconclusive.”

      (2) For the CASP15 targets where this bias is absent, the presentation was very brief. In particular, it would be interesting to see how AFm helped with the docking. The authors may even want to do a direct comparison with docking results without the help of AFm.

      Unfortunately since this was a CASP-CAPRI round, the structure of the unbound Antigen or the nanobodies was unavailable. Thus we cannot perform a comparison without using AF2 at all since we need a structure prediction tool to produce the unbound nanobody and the nanobody-antigen complex template structure to dock. This has been clarified in the main text for better understanding for the readers.

      “Since the nanobody-antigen complexes were CASP targets, we did not have unbound structures, rather only the sequences of individual chains. Therefore, for each target, we employed the AlphaRED strategy as described in Fig 7.”

      Reviewer #1 (Recommendations For The Authors):

      For suggestions for major improvements, see comments under weaknesses. One additional suggestion: the authors found that pLLDT is predictive of flexible residues. Can they try to find AFm features that are predictive of the interface site? Such information may guide their docking to a local site.

      This is a great idea that we and others have been thinking about considerably. Prior work by Burke et al. (Towards a structurally resolved human protein interaction network) examines AlphaFold’s ability to predict PPIs. For high-confidence predicted models of interacting protein complexes, the authors showed that pDockQ correlated reasonably well with correct protein interactions.

      That being said, binding site identification, particularly in a partner-agnostic fashion, i.e. determining binding patches on a given protein, is an area of on-going research . We hope a future study examines AlphaFold3 or ESM3 specifically for this task.

      “Further, we tested multiple thresholds to estimate the optimum cut-off for distinguishing near-native structures (defined as an interface-RMSD < 4 Å) from the predictions. Figure 3.B summarizes the performance with a confusion matrix for the chosen interface-pLDDT cutoff of 85. 79 % of the targets are classified accurately with a precision of 75%, thereby validating the utility of interface-pLDDT as a discriminating metric to rank the docking quality of the AFm complex structure predictions. With AlphaFold3 and ESM3 being released, investigating features that could predict flexible residues or interface site would be valuable, as this information may guide local docking.”

      Minor:

      Page 3, lines 73-77, state how many targets were curated from DB5.5.

      We have now clarified this in the manuscript. All 254 targets curated from DB5.5 at the time of this benchmark study.

      “For each protein target, we extracted the amino acid sequences from the bound structure and predicted a corresponding three-dimensional complex structure with the ColabFold implementation of the AlphaFold multimer v2.3.0 (released in March 2023) for the 254 benchmark targets from DB5.5.”

      In Figure 1, the color used for medium is too difficult to distinguish from the grey color used for rigid.

      We thank you for this suggestion. We have updated the color to olive. Further, based on Reviewer 2’s suggestions, we have moved this plot to the Supplementary.

      Reviewer #2 (Public Review):

      Summary:

      In short, this paper uses a previously published method, ReplicaDock, to improve predictions from AlphaFold-multimer. The method generated about 25% more acceptable predictions than AFm, but more important is improving an Antibody-antigen set, where more than 50% of the models become improved.

      When looking at the results in more detail, it is clear that for the models where the AFm models are good, the improvement is modest (or not at all). See, for instance, the blue dots in Figure 6. However, in the cases where AFm fails, the improvement is substantial (red dots in Figure 6), but no models reach a very high accuracy (Fnat ~0.5 compared to 0.8 for the good AFm models). So the paper could be summarized by claiming, "We apply ReplicaDock when AFm fails", instead of trying to sell the paper as an utterly novel pipeline. I must also say that I am surprised by the excellent performance of ReplicaDock - it seems to be a significant step ahead of other (not AlphaFold) docking methods, and from reading the original paper, that was unclear. Having a better benchmark of it alone (without AFm) would be very interesting.

      We thank the reviewer for highlighting the performance of ReplicaDock. ReplicaDock alone is benchmarked in the original paper (10.1371/journal.pcbi.1010124), with full details on the 2022 version of DB5.5 in the supplement. Indeed ReplicaDock2 achieves the highest reported success rates on flexible docking targets reported in the literature (until this AlphaRED paper!).

      Regarding this statement about “the paper could be summarized…” it might be helpful to give more context. ReplicaDock is a replica exchange Monte Carlo sampling approach for protein docking that incorporates flexibility in an induced-fit fashion. However, the choice of which backbone residues to move is solely dependent on contacts made during each docking trajectory. In the last section of the ReplicaDock paper, we introduced “Directed Induced-fit” where we biased the backbone sampling only towards those residues where we knew the backbone is flexible (this information is obtained because for the benchmark set, we had both unbound and bound structures and hence could cherry-pick the specific residues which are mobile). We agree with the reviewers that AlphaRED is essentially a derivative of ReplicaDock, however, the two major claims that we make in this paper are:

      (1) AlphaFold pLDDT is an effective predictor of backbone flexibility for practical use in docking.

      (2) We can automate the Directed InducedFit approach within ReplicaDock by utilizing this pLDDT information per residue for conformational sampling in protein docking; and in doing so, create a pipeline that would allow us to go from sequence-to-structure-to-complex, specifically capturing conformational changes.

      To conclude these claims, we pose the following questions in the Introduction:

      “(1) Do the residue-specific estimates from AF/AFm relate to potential metrics demonstrating conformational flexibility?

      (2) Can AF/AFm metrics deduce information about docking accuracy?

      (3) Can we create a docking pipeline for in-silico complex structure prediction incorporating AFm to convert sequence-to-structure-to-docked complexes?”

      This work requires a pipeline, the center of which lies in ReplicaDock as a docking method, but has functionalities that were absent in prior work. The goal is also to develop a one-stop shop without manual intervention (a prerequisite for biasing backbone sampling in ReplicaDock) that could be utilized by structural biologists efficiently.

      We clarify this points in the abstract and main text as follows:

      Abstract: “In this work, we combine AlphaFold as a structural template generator with a physics-based replica exchange docking algorithm \add{to better sample conformational changes.”

      Introduction:

      “The overarching goal is to create a one-stop, fully-automated pipeline for simple, reproducible, and accurate modeling of protein complexes. We investigate the aforementioned questions and create a protocol to resolve AFm failures and capture binding-induced conformational changes. We first assess the utility of AFm confidence metrics to detect conformational flexibility and binding site confidence.”

      These results also highlight several questions I try to describe in the weakness section below. In short, they boil down to the fact that the authors must show how good/bad ReplicaDock is at all targets (not only the ones where AFm fails. In addition, I have several more technical comments.

      Strengths:

      Impressive increase in performance on AB-AG set (although a small set and no proteins).

      We thank the reviewer for their comments.

      Weaknesses:

      The presentation is a bit hard to follow. The authors mix several measures (Fnat, iRMS, RMSDbound, etc). In addition, it is not always clear what is shown. For instance, in Figure 1, is the RMSD calculated for a single chain or the entire protein? I would suggest that the author replace all these measures with two: TM-score when evaluating the quality of a single chain and DockQ when evaluating the results for docking. This would provide a clearer picture of the performance. This applies to most figures and tables.

      We apologize for the lack of clarity owing to different metrics. Irms and fnat are standard performance metrics in the docking field, but we agree that DockQ would be simpler when the detail of the other metrics are not required. We have updated the figures Figure 5 and Figure 8 to also show DockQ comparisons.

      Regarding Figure 1, as highlighted in Line 90 of the main-text, “Figure 1 shows the Ca-RMSD of all protein partners of the AFm predicted complex structures with respect to the bound and the unbound.” As suggested by the reviewer in their further comments, we have moved this FIgure to the Supplementary. We have also included TM-score comparison in the Supplementary ( SupFig S2) and included clarifying statements in the main text:

      “We also tested TM-scores to measure the structural deviations of the AFm predicted complex structures with respect to the bound and unbound structures (Supplementary Figure S2). However, this metric is not sensitive enough to detect the subtle, local conformational changes upon binding.”

      For instance, Figure 9 could be shown as a distribution of DockQ scores.

      We have now updated Figure 5 to include DockQ scores in Panel D. Since DockQ is a function of iRMSD, fnat and L-RMSD, it shows cumulative improvement in performance. Some of the nuanced details, such as, the protocol improves i-RMSD considerably but fnat improvement is lacking, and can highlight whether backbone sampling is the challenge or is it sidechain refinement.Therefore, we need to retain the iRMSD and fnat metrics in panel A-C . But We have incorporated this in the main text as follows:

      “Finally, to evaluate docking success rates, we calculate DockQ for top predictions from AFm and AlphaRED respectively (Figure 5D). AlphaRED demonstrates a success rate (DockQ>0.23) for 63% of the benchmark targets. Particularly for Ab-Ag complexes, AFm predicted acceptable or better quality docked structures in only 20% of the 67 targets. In contrast, the AlphaRED pipeline succeeds in 43% of the targets, a significant improvement.”

      Further, we have reevaluated success rates in Figure 8 (previously Figure 9) and have updated the manuscript to report these updated success rates.

      “By utilizing the AlphaRED strategy, we show that failure cases in AFm predicted models are improved for all targets (lower Irms for 97 of 254 failed targets) with CAPRI acceptable-quality or better models generated for 62% of targets overall (Fig 8)”.

      The improvements on the models where AFm is good are minimal (if at all), and it is unclear how global docking would perform on these targets, nor exactly why the plDDT<0.85 cutoff was chosen.

      We agree with the reviewers that the improvement on the models with good AFm predictions is minimal. We acknowledge this in the text now as follows:

      “Most of the improvements in the success rates are for cases where AFm predictions are worse. For targets with good AFm predictions, AlphaRED refinement results in minimal improvements in docking accuracy.”

      The choice of pLDDT cutoff = 85 is elaborated in the “Interface-pLDDT correlates with DockQ and discriminates poorly docked structures” section, paragraph 3. Briefly, we tested multiple metrics and the interface pLDDT had the highest AUC, indicating that it is the best metric for this task. For interface-pLDDT we tested multiple thresholds, and the cutoff of 85 resulted in the highest percentage of true-positive and true-negative rates. This is illustrated with the confusion matrix in Figure 3.B with the precision scores. We now clarify this in the text as follows:

      “With interface-pLDDT as a discriminating metric, we tested multiple thresholds to estimate the optimum cut-off for distinguishing near-native structures (defined as an interface-RMSD < 4 Å) from the predictions. Figure 3B summarizes the performance with a confusion matrix for the chosen interface-pLDDT cutoff of 85. 79% of the targets are classified accurately with a precision of 75%, thereby validating the utility of interface-pLDDT as a discriminating metric to rank the docking quality of the AFm complex structure predictions.”

      To better understand the performance of ReplicaDock, the authors should therefore (i) run global and local docking on all targets and report the results, (ii) report the results if AlphaFold (not multimer) models of the chains were used as input to ReplicaDock (I would assume it is similar). These models can be downloaded from AlphaFoldDB.

      The performance of ReplicaDock on DB5.5 is tabulated in our prior work (https://doi.org/10.1371/journal.pcbi.1010124) and we direct the reviewers there for the detailed performance and results. In our opinion, the benchmark suggested by the reviewer would be redundant and not worth the computational expense.

      The scope of this paper is to highlight a structure prediction + physics-based modeling pipeline for docking to adapt to the accuracy of up-and-coming structure prediction tools.

      Using AlphaFold monomer chains as input and benchmarking on that, albeit interesting scientifically, will not be useful for either the pipeline or biologists who would want a complex structure prediction. We thank the authors for their comments but want to reemphasize that the end goal of this work is to increase the accuracy of complex structure predictions and PPIs obtained from computational tools.

      Further, it would be interesting to see if ReplicaDock could be combined with AFsample (or any other model to generate structural diversity) to improve performance further.

      We would like to highlight that ReplicaDock is a stand-alone tool for protein docking and here we demonstrate the ability of adapting it with metrics derived from AlphaFold or other structure prediction tools (say ESMFold) such as pLDDT for conformational sampling and improving docking accuracy. We definitely agree that adapting it to use with tools such as AFSample will be interesting but it is out of scope of this work.

      The estimates of computing costs for the AFsample are incorrect (check what is presented in their paper). What are the computational costs for RepliaDock global docking?

      The authors of the AFSample paper report that “AFsample requires more computational time than AF2, as it generates 240 models, and including the extra recycles, the overall timing is 1000 more costly than the baseline.” We have reported these exact numbers in our manuscript.

      The computational costs of ReplicaDock are 8-72 CPU hours on a single node with 24 processors as reported in our prior work.

      For AlphaRED, the costs are slightly higher owing to the structure prediction module in the beginning and are up to 100 CPU hrs for our largest (max Nres) target.

      It is unclear strictly what sequences were used as input to the modelling. The authors should use full-length UniProt sequences if they were not done.

      We report this in the methods section of the manuscript as well as in Figure 5. Full length complex sequences were used for the models that we extracted from DB5.5.

      “As illustrated in Fig. 5, given a sequence of a protein complex, we use the ColabFold implementation of AF2-multimer to obtain a predictive template.”

      We clarify this in the methods section as:

      “For each target in the DB5.5 dataset, we first extracted the corresponding FASTA sequence for the bound complex and then obtained AlphaFold predicted models with the ColabFold v1.5.2 implementation of AlphaFold and AlphaFold-multimer (v.2.3.0).”

      The antibody-antigen dataset is small. It could easily be expanded to thousands of proteins. It would be interesting to know the performance of ReplicaDock on a more extensive set of Antibodies and nanobodies.

      This work demonstrates the performance on the docking benchmark, i.e. given unbound structure can you predict the bound complexes. With this regard, our analysis has been focussed on targets where both the unbound and bound structures are available so that we could evaluate the ability of AlphaRED on modeling protein flexibility and docking accuracy. For antibody-antigen complexes, there are only 67 structures with both unbound and bound complexes available and they constituted our dataset. Benchmarking AlphaRED on all antibody-antigen targets can give biased results as most Ab-Ag complexes are in AlphaFold training set. Further, our work is more aimed towards predicting conformational flexibility in docking and not rigid-body docked complexes, so benchmarking on existing bound Ab-Ag structures is out of scope for this work.

      Using pLDDT on the interface region to identify good/bas models is likely suboptimal. It was acceptable (as a part of the score) for AlphaFold-2.0 (monomer), but AFm behaves differently. Here, AFm provides a direct score to evaluate the quality of the interaction (ipTM or Ranking Confidence). The authors should use these to separate good/bad models (for global/local docking), or at least show that these scores are less good than the one they used.

      We thank the reviewers for this suggestion.

      Reviewer #2 (Recommendations For The Authors):

      Some Figures could be skipped/improved

      Fig 1: Use TM-score instead a much better measure (and the figure is not necessary).

      Figure 1 compares the bias of AlphaFold towards unbound or bound forms of the proteins. We believe that this figure highlights the slight inherent bias of AlphaFold towards bound structures over unbound.

      As the reviewers have suggested we have included a plot comparing the TM-scores for the structures. Further, we have moved this figure to the Supplementary.

      Fig 2. Skip B (why compare RMSD with pLDDT?). Add a figure to see how this correlates over all targets not just two.

      RMSD and LDDT both represent metrics to evaluate conformational variability between two structures, such as the bound and unbound forms of the same protein structure. On one hand where RMSD measures overall deviation of residues, LDDT allows the estimation of relative domain orientations and concerted proteins. We have elaborated this in Methods as well as in the Results section titled “AlphaFold pLDDT provides a predictive confidence measure for backbone flexibility”.

      The data for the benchmark targets is now included in the Supplementary (Supplementary Figures S3-S4).

      Fig 3. Color the different chains of a protein differently. Thereby the Receptor/Ligand/Bound labels can be omitted.

      We thank the reviewers for this suggestion. However, the color scheme is chosen to highlight (1) the relative orientation of protein partners relative to each other. We have ensured that the alignment is over one partner (Receptor) so that you could see the relative orientation of the other partner (Ligand) in the modeled protein over the bound structure (in one color). (2) The coloring of the receptor and ligand chain is by pLDDT (from red to blue) to highlight that for decoys with incorrectly predicted interfaces, the pLDDT scores of the interface residues are indeed lower and can be a discriminating metric. We elaborate this in the caption of Figure 3 as well as in the section “Interface-pLDDT correlates with DockQ and discriminates poorly docked structures”. Coloring the chains of a protein differently will obfuscate the point that we are aiming to make and will be inconclusive for the readers as they would need to rely only on quantitative metrics (Irms and DockQ) reported but won’t be able to visualize the interface pLDDT of the incorrectly bound structures. We hope that this justifies the choice of our color scheme.

      Fig 4. Include RankConf, ipTM, pDockQ, and other measures in the plos (they are likely better). Include DockQ for the top targets. It is difficult to estimate for multi chain complexes.

      We thank the reviewer for this suggestion. We have now included the DockQ performances for all targets in Figure 5 (previously Figure 6) as well as re-evaluated our final success rates based on the DockQ calculations in Figure 8 (previously Figure 9).

      Fig 5. use a better measure to split (see above).

      We have elaborated on the choice of the split for the comments above and the interface pLDDT threshold of 85 is a decision made post observation on the docking benchmark. We do want to highlight that the cut-off is arbitrary and in our online server (ROSIE) as well as in custom scripts, this cut-off can be tuned by the user as required. We would suggest a cut-off of 85 based on our observations but the users are welcome to tune this as per their needs.

      Fig 6. Replace lrms/fnat with DockQ.

      We have now included DockQ scores in our manuscript.

      Fig 7. Color the different chains of a protein differently.

      We have colored the protein chains differently. AlphaFold models are in Orange, Bound complexes are in Gray, and predicted proteins from AlphaRED are in Blue-Green indicating the two partners. All models are aligned over the receptor so relative orientations of the ligand protein can be observed.

      Fig 8 Color the different chains of a protein differently.

      The chains are colored differently. We would like the reviewer to elaborate more on what they would like to observe as we believe our color scheme makes intuitive sense for readers.

      Fig 9. Use DockQ instead of CAPRI criteria.

      The figure has been updated based on DockQ. To elaborate, the CAPRI criteria is set based on DockQ scores as elaborated in the figure caption.

    1. eLife Assessment

      This manuscript reports important findings that the methyltransferase METTL3 is involved in the repair of abasic sites and uracil in DNA, mediating resistance to floxuridine-driven cytotoxicity. Convincing evidence shows the involvement of m6A in DNA based on single cell imaging and mass spec data. The authors present evidence that the m6A signal does not result from bacterial contamination or RNA, but the text does not make this overly clear.

    2. Reviewer #1 (Public review):

      Summary:

      The authors sought to identify unknown factors involved in the repair of uracil in DNA through a CRISPR knockout screen.

      Strengths:

      The screen identified both known and unknown proteins involved in DNA repair resulting from uracil or modified uracil base incorporation into DNA. The conclusion is that the protein activity of METTL3, which converts A nucleotides to 6mA nucleotides, plays a role in the DNA damage/repair response. The importance of METTL3 in DNA repair, and its colocalization with a known DNA repair enzyme, UNG2, is well characterized.

      Weaknesses:

      This reviewer identified no major weaknesses in this study. The manuscript could be improved by tightening the text throughout, and more accurate and consistent word choice around the origin of U and 6mA in DNA. The dUTP nucleotide is misincorporated into DNA, and 6mA is formed by methylation of the A base present in DNA. Using words like 6mA "deposition in DNA" seems to imply it results from incorporation of a methylated dATP nucleotide during DNA synthesis.

    3. Reviewer #2 (Public review):

      Summary:

      In this work, the authors performed a CRISPR knockout screen in the presence of floxuridine, a chemotherapeutic agent that incorporates uracil and fluoro-uracil into DNA, and identified unexpected factors, such as the RNA m6A methyltransferase METTL3, as required to overcome floxuridine-driven cytotoxicity in mammalian cells. Interestingly, the observed N6-methyladenosine was embedded in DNA, which has been reported as DNA 6mA in mammalian genomes and is currently confirmed with mass spectrometry in this model. Therefore, this work consolidated the functional role of mammalian genomic DNA 6mA, and supported with solid evidence to uncover the METTL3-6mA-UNG2 axis in response to DNA base damage.

      Strengths:

      In this work, the authors took an unbiased, genome-wide CRISPR approach to identify novel factors involved in uracil repair with potential clinical interest.

      The authors designed elegant experiments to confirm the METTL3 works through genomic DNA, adding the methylation into DNA (6mA) but not the RNA (m6A), in this base damage repair context. The authors employ different enzymes, such as RNase A, RNase H, DNase, and liquid chromatography coupled to tandem mass spectrometry to validate that METTL3 deposits 6mA in DNA in response to agents that increase genomic uracil.

      They also have the Mettl3-KO and the METTL3 inhibition results to support their conclusion.

      Weaknesses:

      Although this study demonstrates that METTL3-dependent 6mA deposition in DNA is functionally relevant to DNA damage repair in mammalian cells, there are still several concerns and issues that need to be improved to strengthen this research.

      First, in the whole paper, the authors never claim or mention the mammalian cell lines contamination testing result, which is the fundamental assay that has to be done for the mammalian cell lines DNA 6mA study.

      Second, in the whole work, the authors have not supplied any genomic sequencing data to support their conclusions. Although the sequencing of DNA 6mA in mammalian models is challenging, recent breakthroughs in sequencing techniques, such as DR-Seq or NT/NAME-seq, have lowered the bar and improved a lot in the 6mA sequencing assay. Therefore, the authors should consider employing the sequencing methods to further confirm the functional role of 6mA in base repair.

      Third, the authors used the METTL3 inhibitor and Mettl3-KO to validate the METTL3-6mA-UNG2 functional roles. However, the catalytic mutant and rescue of Mettl3 may be the further experiments to confirm the conclusion.

    4. Reviewer #3 (Public review):

      Summary:

      The authors are showing evidence that they claim establishes the controversial epigenetic mark, DNA 6mA, as promoting genome stability.

      Strengths:

      The identification of a poorly understood protein, METTL3, and its subsequent characterization in DDR is of high quality and interesting.

      Weaknesses:

      (1) The very presence of 6mA (DNA) in mammalian DNA is still highly controversial and numerous studies have been conclusively shown to have reported the presence of 6mA due to technical artifacts and bacterial contamination. Thus, to my knowledge there is no clear evidence for 6mA as an epigenetic mark in mammals, and consequently, no evidence of writers and readers of 6mA. None of this is mentioned in the introduction. Much of the introduction can be reduced, but a paragraph clearly stating the controversy and lack of evidence for 6mA in mammals needs to be added, otherwise, the reader is given an entirely distorted view of the field.

      These concerns must also be clearly in the limitations section and even in the results section which fails to nuance the authors' findings.

      (2) What is the motivation for using HT-29 cells? Moreover, the materials and methods do not state how the authors controlled for bacterial contamination, which has been the most common cause of erroneous 6mA signals to date. Did the authors routinely check for mycoplasma?

      (3) The single cell imaging of 6mA in various cells is nice. The results are confirmed by mass spec as an orthogonal approach. Another orthogonal and quantitative approach to assessing 6mA levels would be PacBio. Similarly, it is unclear why the authors have not performed dot-blots of 6mA for genomic DNA from the given cell lines.

      (4) The results of Figure 3 need further investigation and validation. If the results are correct the authors are suggesting that the majority of 6mA in their cell lines is present in the DNA, and not the RNA, which is completely contrary to every other study of 6mA in mammalian cells that I am aware of. This could suggest that the antibody is not, in fact, binding to 6mA, but to unmodified adenine, which would explain why the signal disappears after DNAse treatment. Indeed, binding of 6mA to unmethylated DNA is a commonly known problem with most 6mA antibodies and is well described elsewhere.

      (5) Given the lack of orthologous validation of the observed DNA 6mA and the lack of evidence supporting the presence of 6mA in mammalian DNA and consequently any functional role for 6mA in mammalian biology, the manuscript's conclusions need to be toned down significantly, and the inherent difficulty in assessing 6mA accurately in mammals acknowledged throughout.

    5. Author response:

      eLife Assessment <br /> This manuscript reports important findings that the methyltransferase METTL3 is involved in the repair of abasic sites and uracil in DNA, mediating resistance to floxuridine-driven cytotoxicity. The presented evidence for the involvement of m6A in DNA is incomplete and requires further validation with orthogonal approaches to conclusively show the presence of 6mA in the DNA and exclude that the source is RNA or bacterial contamination. 

      We thank the editors for recognizing the importance of our work and the relevance of METTL3 in DNA repair. However, we wholly disagree with the second sentence in the eLife assessment, and we want to clarify why our evidence for the involvement of 6mA in DNA is complete.  

      The identification of 6mA in DNA, upon DNA damage, is based first on immunofluorescence observations using an anti-m6A antibody. In this setting, removal of RNA with RNase treatment fails to reduce the 6mA signal, excluding the possibility that the source of signal is RNA. In contrast, removal of DNA with DNase treatment removes all 6mA signal, strongly suggesting that the species carrying the N6-methyladenosine modification is DNA (Figure 3D, E). Importantly, in Figure 3F, we provide orthogonal, quantitative mass spectrometry data that independently confirm this finding. Mass spectrometry-liquid chromatography of DNA analytes, conclusively shows the presence of 6mA in DNA upon treatment with DNA damaging agents and excludes that the source is RNA, based on exact mass. Reviewer #2 recognized the strengths of this approach to generate solid evidence for 6mA in DNA.

      Cells only show the 6mA signal when treated with DNA damaging agents, and the 6mA is absent from untreated cells (Figure 3D, E, F). This provides strong evidence that the 6mA signal is not a result of bacterial contamination in our cell lines. Moreover, our cell lines are routinely tested for mycoplasma contamination. It could be possible that stock solutions of DNA damaging agents may be contaminated, but this would need to be true for all individual drugs and stocks tested. The data showing 6mA signal is not significantly different from untreated cells when a DNA damaging agent is combined with a METTL3 inhibitor (Figure 3G, H) provides strong evidence against bacterial contamination in our stocks.  

      In summary, we provide conclusive evidence, based on orthogonal methods, that the METTL3-dependent N6-methyladenosine modification is deposited in DNA, not RNA, in response to DNA damage. 

      Public Reviews: <br /> Reviewer #1 (Public review): <br /> Summary: 

      The authors sought to identify unknown factors involved in the repair of uracil in DNA through a CRISPER knockout screen. 

      Typo above: “CRISPER” should be “CRISPR”.

      Strengths: 

      The screen identified both known and unknown proteins involved in DNA repair resulting from uracil or modified uracil base incorporation into DNA. The conclusion is that the protein activity of METTL3, which converts A nucleotides to 5mA nucleotides, plays a role in the DNA damage/repair response. The importance of METTL3 in DNA repair, and its colocalization with a known DNA repair enzyme, UNG2, is well characterized. 

      Typo above: “5mA” should be “6mA”.

      Weaknesses: <br /> This reviewer identified no major weaknesses in this study. The manuscript could be improved by tightening the text throughout, and more accurate and consistent word choice around the origin of U and 6mA in DNA. The dUTP nucleotide is misincorporated into DNA, and 6mA is formed by methylation of the A base present in DNA. Using words like 6mA "deposition in DNA" seems to imply it results from incorporation of a methylated dATP nucleotide during DNA synthesis.

      The increased presence of 6mA during DNA damage could result from methylation at the A base itself (within DNA) or from incorporation of pre-modified 6mA during DNA synthesis. Our data do not directly discriminate between these two mechanisms, and we will clarify this point in the discussion.

      Reviewer #2 (Public review): <br /> Summary: <br /> In this work, the authors performed a CRISPR knockout screen in the presence of floxuridine, a chemotherapeutic agent that incorporates uracil and fluoro-uracil into DNA, and identified unexpected factors, such as the RNA m6A methyltransferase METTL3, as required to overcome floxuridine-driven cytotoxicity in mammalian cells. Interestingly, the observed N6-methyladenosine was embedded in DNA, which has been reported as DNA 6mA in mammalian genomes and is currently confirmed with mass spectrometry in this model. Therefore, this work consolidated the functional role of mammalian genomic DNA 6mA, and supported with solid evidence to uncover the METTL3-6mA-UNG2 axis in response to DNA base damage. <br /> Strengths: <br /> In this work, the authors took an unbiased, genome-wide CRISPR approach to identify novel factors involved in uracil repair with potential clinical interest. 

      The authors designed elegant experiments to confirm the METTL3 works through genomic DNA, adding the methylation into DNA (6mA) but not the RNA (m6A), in this base damage repair context. The authors employ different enzymes, such as RNase A, RNase H, DNase, and liquid chromatography coupled to tandem mass spectrometry to validate that METTL3 deposits 6mA in DNA in response to agents that increase genomic uracil. <br /> They also have the Mettl3-KO and the METTL3 inhibition results to support their conclusion. <br /> Weaknesses:<br /> Although this study demonstrates that METTL3-dependent 6mA deposition in DNA is functionally relevant to DNA damage repair in mammalian cells, there are still several concerns and issues that need to be improved to strengthen this research.

      First, in the whole paper, the authors never claim or mention the mammalian cell lines contamination testing result, which is the fundamental assay that has to be done for the mammalian cell lines DNA 6mA study.

      Our cell lines are routinely tested for bacterial contamination, specifically mycoplasma, and we plan to state this information in a revised version of the manuscript.

      Importantly, we do not observe 6mA in untreated cells, strongly suggesting that the 6mA signal observed is dependent on the presence of DNA damage and not caused by contamination in the cell lines (Figure 3D, E, F). While it could be possible that stock solutions of DNA damaging agents may be contaminated, this would need to be the case for all individual drugs and stocks tested that induce 6mA, which seems very unlikely. Finally, the data showing 6mA signal is not significantly different from untreated cells when a DNA damaging agent is combined with a METTL3 inhibitor (Figure 3 G, H) provides strong evidence against bacterial contamination in our drug stocks.

      Second, in the whole work, the authors have not supplied any genomic sequencing data to support their conclusions. Although the sequencing of DNA 6mA in mammalian models is challenging, recent breakthroughs in sequencing techniques, such as DR-Seq or NT/NAME-seq, have lowered the bar and improved a lot in the 6mA sequencing assay. Therefore, the authors should consider employing the sequencing methods to further confirm the functional role of 6mA in base repair.

      While we agree that it could be important to understand the precise genomic location of 6mA in relation to DNA damage, this is outside the scope of the current study. Moreover, this exercise may prove unproductive. If 6mA is enriched in DNA at damage sites or as DNA is replicated, the genomic mapping of 6mA is likely to be stochastic. If stochastic, it would be impossible to obtain the read depth necessary to map 6mA accurately.

      Third, the authors used the METTL3 inhibitor and Mettl3-KO to validate the METTL3-6mA-UNG2 functional roles. However, the catalytic mutant and rescue of Mettl3 may be the further experiments to confirm the conclusion. 

      We believe this to be an excellent suggestion from Reviewer #2 but we are unable to perform the proposed experiment at this time. We encourage future studies to explore the rescue experiment.

      Reviewer #3 (Public review):

      Summary:

      The authors are showing evidence that they claim establishes the controversial epigenetic mark, DNA 6mA, as promoting genome stability.

      Strengths:

      The identification of a poorly understood protein, METTL3, and its subsequent characterization in DDR is of high quality and interesting.

      Weaknesses:

      (1) The very presence of 6mA (DNA) in mammalian DNA is still highly controversial and numerous studies have been conclusively shown to have reported the presence of 6mA due to technical artifacts and bacterial contamination. Thus, to my knowledge there is no clear evidence for 6mA as an epigenetic mark in mammals, and consequently, no evidence of writers and readers of 6mA. None of this is mentioned in the introduction. Much of the introduction can be reduced, but a paragraph clearly stating the controversy and lack of evidence for 6mA in mammals needs to be added, otherwise, the reader is given an entirely distorted view of the field.

      These concerns must also be clearly in the limitations section and even in the results section which fails to nuance the authors' findings.

      We agree with the reviewer that the presence and potential function of 6mA in mammalian DNA has been debated. Importantly, the debate regarding the presence and quantity of 6mA in DNA has been previously restricted to undamaged, baseline conditions. In complete agreement with this notion, we do not detect appreciable levels of 6mA in untreated cells. We will revise the introduction to introduce the debate about 6mA in DNA. We, however, want to highlight that our study provides for the first time, convincing evidence (based on orthogonal methods) that 6mA is present in DNA in response to a stimulus, DNA damage.

      (2) What is the motivation for using HT-29 cells? Moreover, the materials and methods do not state how the authors controlled for bacterial contamination, which has been the most common cause of erroneous 6mA signals to date. Did the authors routinely check for mycoplasma?

      HT-29 is a cell line of colorectal origin and chemotherapeutic agents that introduce uracil and uracil derivatives in DNA, as those used in this study, are relevant for the treatment of colorectal cancer. As indicated above, we do not observe 6mA in untreated cells, strongly suggesting that the 6mA signal observed is dependent on DNA damage and not caused by a potential bacterial contamination (Figure 3D, E, F). Additionally, our cell lines are routinely tested for bacterial contamination, specifically mycoplasma.

      (3) The single-cell imaging of 6mA in various cells is nice but must be confirmed by orthogonal approaches. PacBio would provide an alternative and quantitative approach to assessing 6mA levels. Similarly, it is unclear why the authors have not performed dot-blots of 6mA for genomic DNA from the given cell lines.

      We are confused by this point since an orthogonal approach to detect 6mA, mass spectrometry-liquid chromatography, was employed. This method does not use an antibody and confirms the increase of 6mA in DNA when cells were treated with DNA damaging agents. This data is presented in Figure 3F.

      It is sensible to hypothesize that the localization of 6mA is consistent with DNA replication (like uracil deposition). In this event, the genomic mapping of 6mA is likely to be stochastic. This would make quantification with PacBio sequencing difficult because it would be very challenging to achieve the appropriate read depth to call a modified base.

      Dot blots rely on an antibody and thus are not truly orthogonal to our immunofluorescence-based measurements. We preferred the mass spectrometry-liquid chromatography approach we took as a true orthogonal approach.

      (4) The results of Figure 3 need further investigation and validation. If the results are correct the authors are suggesting that the majority of 6mA in their cell lines is present in the DNA, and not the RNA, which is completely contrary to every other study of 6mA in mammalian cells that I am aware of. This could suggest that the antibody is not, in fact, binding to 6mA, but to unmodified adenine, which would explain why the signal disappears after DNAse treatment. Indeed, binding of 6mA to unmethylated DNA is a commonly known problem with most 6mA antibodies and is well described elsewhere.

      Based on this and the following comment, we are convinced that Reviewer #3 has overlooked two critical elements of our study:

      First, the immunofluorescence work presented in Figure 3, showing 6mA signal in response to DNA damage, uses cells that were pre-extracted to remove excess cytoplasmic RNA. This method is often used in immunofluorescence experiments of this kind. The pre-extraction method removes most of the cytoplasmic content, and the majority of the cytoplasmic m6A RNA signal. Supplementary Figure 3D shows cells that have not been pre-extracted prior to staining. These images show the cytoplasmic m6A signal is abundant if we do not perform the pre-extraction step.

      If the antibody used to label 6mA significantly reacted with unmodified adenine, we would expect a large signal in untreated or untreated and denatured conditions. In contrast, an increase in 6mA is not observed in either case.

      Second, the orthogonal approach we employed, mass spectrometry coupled with liquid chromatography, measures 6mA DNA analytes specifically by exact mass. This approach does not depend on an antibody and yields results consistent with those from the immunofluorescence experiments.

      (5) Given the lack of orthologous validation of the observed DNA 6mA and the lack of evidence supporting the presence of 6mA in mammalian DNA and consequently any functional role for 6mA in mammalian biology, the manuscript's conclusions need to be toned down significantly, and the inherent difficultly in assessing 6mA accurately in mammals acknowledged throughout.

      Typo above: “difficultly” should be “difficulty”.

      As discussed in response to prior comments, Figure 3 does provide two independent and orthologous methods that demonstrate 6mA presence in DNA specifically, and not RNA, in response to DNA damage. Complementary and orthogonal datasets are presented using either immunofluorescence microscopy or mass spectrometry-liquid chromatography of extracted DNA. The latter method does not rely on an antibody and can discriminate 6mA DNA versus RNA based on exact mass. We will revise the text to clarify that Figure 3F is a completely orthogonal approach.

    1. eLife Assessment

      This valuable study marks a significant advancement in brain aging research by centering on Asian populations (Chinese, Malay, and Indian Singaporeans), a group frequently underrepresented in such studies. It unveils solid evidence for anatomical differences in brain aging predictors between the young and old age groups. Overall, this study broadens our understanding of brain aging across diverse ethnicities.

    2. Joint Public Review:

      Summary:

      The authors of the study investigated the generalization capabilities of a deep learning brain age model across different age groups within the Singaporean population, encompassing both elderly individuals aged 55 to 88 years and children aged 4 to 11 years. The model, originally trained on a dataset primarily consisting of Caucasian adults, demonstrated a varying degree of adaptability across these age groups. For the elderly, the authors observed that the model could be applied with minimal modifications, whereas for children, significant fine-tuning was necessary to achieve accurate predictions. Through their analysis, the authors established a correlation between changes in the brain age gap and future executive function performance across both demographics. Additionally, they identified distinct neuroanatomical predictors for brain age in each group: lateral ventricles and frontal areas were key in elderly participants, while white matter and posterior brain regions played a crucial role in children. These findings underscore the authors' conclusion that brain age models hold the potential for generalization across diverse populations, further emphasizing the significance of brain age progression as an indicator of cognitive development and aging processes.

      Strengths:

      (1) The study tackles a crucial research gap by exploring the adaptability of a brain age model across Asian demographics (Chinese, Malay, and Indian Singaporeans), enriching our knowledge of brain aging beyond Western populations.<br /> (2) It uncovers distinct anatomical predictors of brain aging between elderly and younger individuals, highlighting a significant finding in the understanding of age-related changes and ethnic differences.

      In summary, this paper underscores the critical need to include diverse ethnicities in model testing and estimation.

      Comments on revisions:

      The previously mentioned weaknesses were addressed in the revision process. As stated earlier the paper tackles a crucial research gap by exploring the adaptability of a brain-age model across Asian demographics (Chinese, Malay, and Indian Singaporeans), enriching our knowledge of brain aging beyond Western populations.

    1. eLife Assessment

      This valuable study examines the variability in spacing and direction of entorhinal grid cells, providing convincing evidence that such variability helps disambiguate locations within an environment. This study will be of interest to neuroscientists working on spatial navigation and, more broadly, on neural coding.

    2. Reviewer #1 (Public review):

      Summary:

      The present paper by Redman et al. investigated the variability of grid cell properties in the MEC by analyzing publicly available large-scale neural recording data. Although previous studies have proposed that grid spacing and orientation are homogeneous within the same grid module, the authors found a small but robust variability in grid spacing and orientation across grid cells in the same module. The authors also showed, through model simulations, that such variability is useful for decoding spatial position.

      Strengths:

      The results of this study provide novel and intriguing insights into how grid cells compose the cognitive map in the axis of the entorhinal cortex and hippocampus. This study analyzes large data sets in an appropriate manner and the results are convincing.

      Comments on revisions:

      In the revised version of the manuscript, the authors have addressed all the concerns I raised.

    3. Reviewer #2 (Public review):

      Summary:

      This paper presents an interesting and useful analysis of grid cell heterogeneity, showing that the experimentally observed heterogeneity of spacing and orientation within a grid cell module can allow more accurate decoding of location from a single module.

      Strengths:

      (1) I found the statistical analysis of the grid cell variability to be very systematic and convincing. I also found the evidence for enhanced decoding of location based on between cell variability within a module to be convincing and important, supporting their conclusions.

      (2) Theoreticians have developed models that focus on the use of grid cells that are highly regular in their parameters, and usually vary only in the spatial phase of cells within modules and the spacing and orientation between modules. This focus on consistency is partly to obtain the generalization of the grid cell code to a broad range of previously unvisited locations. In contrast, most experimentalists working with grid cells know that many if not most grid cells show high variability of firing fields, as demonstrated in the figures in experimental papers. The authors of this current paper have highlighted this discrepancy, and shown that the variability shown in the data could actually enhance decoding of location.

    4. Reviewer #3 (Public review):

      Summary:

      Redman and colleagues analyze grid cell data obtained from public databases. They show that there is significant variability in spacing and orientation within a module. They show that the difference in spacing and orientation for a pair of cells is larger than the one obtained for two independent maps of the same cell. They speculate that this variability could be useful to disambiguate the rat position if only information from a single module is used by a decoder.

      Strengths:

      The strengths of this work lie in its conciseness, clarity, and the potential significance of its findings for the grid cell community, which has largely overlooked this issue for the past two decades. Their hypothesis is well stated and the analyses are solid.

      Weaknesses:

      Major weaknesses identified in the original version have been addressed.

      The authors have addressed all of our concerns, providing control analyses that strengthen their claim.

    5. Author response:

      The following is the authors’ response to the original reviews.

      We thank the reviewers for their time and thoughtful comments. We believe that the further analyses suggested have made the results clearer and more robust. Below, we briefly highlight the key points addressed in the revision and the new evidence supporting them. Then, we address each reviewer’s critiques point-by-point.

      - Changes in variability with respect to time/experience

      Both reviewers #1 and #3 asked whether the variability in grid properties observed was dependent on time or experience. This is an important point, given that such a dependence on time could lead to interesting hypotheses about the underlying dynamics of the grid code. However, in the new analyses we performed, we do not observe changes in grid variability within a session (Fig S5 of the revised manuscript), suggesting that the grid variability seen is constant within the timescale of the data set.

      - The assumption of constant grid parameters in the literature

      Reviewer #2 pointed out that it had been appreciated by experimentalists that grid properties are variable within a module. We agree that we may have overstated the universality of this assumption in the original manuscript, and we have toned down the language in the revision. However, we note that many previous theoretical studies assumed these properties to be constant, within a given module. We provide some examples below, and have added evidence of this assertion, with citations to the theoretical literature, to the revised manuscript .

      - Additional sources of variability

      Reviewer #3 pointed out additional sources that might explain the variability observed in the paper (beyond time and experience). These sources include: field width, border location, and the impact of conjunctive cells. We have run additional analyses and have found no significant impact on the observed variability from any of these factors. We believe that these are important controls, and have added them to the manuscript (Fig S4-S7 of the revised manuscript)

      - Analysis of computational models

      Reviewer #3 noted that our results could be strengthened by performing similar analyses on the output of computational models of grid cells. This is a good idea. We have now measured the variability of grid properties in a recent normative recurrent neural network (RNN) model that develops grid cells when trained to perform path integration (Sorscher et al., 2019). This model has been shown to develop signatures of a 2D toroidal attractor (Sorscher et al., 2023) and achieves a high accuracy on a simple path integration task. Interestingly, the units with the greatest grid scores also exhibit a range of grid spacings and grid orientations (Fig S8 of the revised manuscript). Furthermore, by decreasing the amount of sparsity (through decreasing the weight decay regularization), we found an increase in the variability of the grid properties. This analysis demonstrates a heretofore unknown similarity between the RNN models trained to perform path integration and recorded grid cells from MEC. It additionally provides a framework for computational analysis of the emergence of grid property variability.

      Reviewer #1:

      (1) Is the variability in grid spacing and orientation that the authors found intrinsically organized or is it shaped by experience? Previous research has shown that grid representations can be modified through experience (e.g., Boccara et al., Science 2019). To understand the dynamics of the network, it would be important to investigate whether robust variability exists from the beginning of the task period (recording period) or whether variability emerges in an experience-dependent manner within a session.

      This is an interesting question that was not addressed in the paper. To test this, we performed additional analysis to resolve whether the variability changes across a session.

      Using a sliding window, we have measured changes in variability with respect to recording time (Fig S5A). To this end, we compute grid orientation and spacing over a time-window whose length is half the total length of the recording. From the population distribution of orientation and spacing values, we compute the standard deviation as a measure of variability. We repeat the same procedure, sliding the window forward until the variability for the second half of the recording is computed.

      We applied this approach to recording ID R12 (the same as in Figs 2-4) given that this recording session was significantly longer than the rest (nearly two hours). Results are shown in Fig S5B-C. For both orientation and spacing, no changes of variability with respect to time can be observed. Similar results were found for other modules (see caption of Fig S5 for statistics).

      We also note that the rats were already familiarized with the environment for 10-20 sessions prior to the recordings, so there may not be further learning during the period of the grid cell recordings. No changes in variability can be seen in Rat R across days (e.g., in Fig 5B R12 and R22 have similar distributions of variability). However, we note that it may be possible that there are changes in grid properties at time-scales greater than the recordings.

      (2) It is important to consider the optimal variability size. The larger the variability, the better it is for decoding. On the other hand, as the authors state in the

      Discussion, it is assumed that variability does not exist in the continuous attractor model. Although this study describes that it does not address how such variability fits the attractor theory, it would be better if more detailed ideas and suggestions were provided as to what direction the study could take to clarify the optimal size of variability.

      We appreciate this suggestion and agree that more discussion is warranted on how our results can be reconciled with previously observed attractor dynamics. To explore this, we studied the recurrent neural network (RNN) model from Sorscher et al. (2019), which develops grid responses when trained on path integration. This network has previously been found to develop signatures of toroidal topology (Sorscher et al., 2023), yet we find its grid responses also contain heterogeneity in grid properties (Fig S8). By decreasing the strength of the weight decay regularization (which leads to denser connectivity in the recurrent layer), we find an increase in the grid property variability. Interestingly, decreasing the weight decay regularization has been previously found to lead to weaker grid responses and worse ability of the RNN to perform path integration on environments larger than it was trained on. This approach not only provides preliminary evidence to our claim that too much variability can lead to weaker continuous attractor structure, but also provides a modeling framework with which future work can explore this question in more detail. We have added discussion of this issue to the manuscript text (Discussion).

      Reviewer #2:

      (1) Even though theoreticians might have gotten the mistaken impression that grid cells are highly regular, this might be due to an overemphasis on regularity in a subset of papers. Most experimentalists working with grid cells know that many if not most grid cells show high variability of firing fields within a single neuron, though this analysis focuses on between neurons. In response to this comment, the reviewers should tone down and modify their statements about what are the current assumptions of the field (and if possible provide a short supplemental section with direct quotes from various papers that have made these assumptions).

      We agree that some experimentalists are aware of variability in the recorded grid response patterns and that this work may not come as a complete surprise to them. We have toned down our language in the Introduction, changing “our results challenge a long-held assumption” to “our results challenge a frequently made assumption in the theoretical literature”. Additionally, we have added a caveat that “experimentalists have been aware” of the observed variability in grid properties.

      We would like to emphasize that the lack of work carefully examining the robustness of this variability has prevented a firm understanding of whether this is an inherent property of grid cells or due to measurement noise. The impact of this can be seen in theoretical neuroscience work where a considerable number of articles (including recent publications) start with the assumption that all grid cells within a module have identical properties, with the exception of phase shift and noise. We have now cited a number of these papers in the Introduction, to provide specific references. To further illustrate the pervasiveness of this assumption being explicitly made in theoretical neuroscience, below we provide quotes from a few important papers:

      “Cells with a common spatial period also share a common grid orientation; their responses differ only by spatial translations, or different preferred firing phases, with respect to their common response period” (Sreenivasan and Fiete, 2011)”

      “Grid cells are organized into discrete modules; within each module, the spatial scale and orientation of the grid lattice are the same, but the lattice for different cells is shifted in space.” (Stemmler et al., 2015)”

      “Recently, it was shown that grid cells are organized in discrete modules within which cells share the same orientation and periodicity but vary randomly in phase” (Wei et al., 2015)”

      “...cells within one module have receptive fields that are translated versions of one another, and different modules have firing lattices of different scales and orientations” (Dorrell et al., 2023)”

      In these works, this assumption is used to derive properties relating to the computational properties of grid cells (e.g., error correction, optimal scaling between grid spacings in different modules).

      In addition, since grid cells are assumed to be identical in the computational neuroscience community, there has been little work on quantifying how much variability a given model produces. This makes it challenging to understand how consistent different models are with our observations. This is illustrated in our analysis of a recent recurrent neural network (RNN) model of grid cells (Fig S8), which does exhibit variability.

      (2) The authors state that "no characterization of the degree and robustness of variability in grid properties within individual modules has been performed." It is always dangerous to speak in absolute terms about what has been done in scientific studies. It is true that few studies have had the number of grid cells necessary to make comparisons within and between modules, but many studies have clearly shown the distribution of spacing in neuronal data (e.g. Hafting et al., 2005; Barry et al., 2007; Stensola et al., 2012; Hardcastle et al., 2015) so the variability has been visible in the data presentations. Also, most researchers in the field are well aware that highly consistent grid cells are much rarer than messy grid cells that have unevenly spaced firing fields. This doesn't hurt the importance of the paper, but they need to tone down their statements about the lack of previous awareness of variability (specific locations are noted in the specific comments).

      We have toned down our language in the Introduction. However, we note that our point that no detailed analysis had been done on measuring the robustness of this variability stands. Thus, for the general community, it has not been clear whether this previously observed variability is noise or a real feature of the grid code.

      (3) The methods section needs to have a separate subheading entitled: How grid cells were assigned to modules" that clearly describes how the grid cells were assigned to a module (i.e. was this done by Gardner et al., or done as part of this paper's post-processing?

      We thank the reviewer for pointing out this missing information. We have added a new subsection in the Materials and Methods section, entitled “Grid module classification” to clarify how the grid cells are assigned to modules. In short, this was done by Gardner et al. (2022) using an unsupervised clustering approach that was viewed as enabling a less biased identification of modules. We did not perform any additional processing steps on module identity.

      Reviewer #3:

      (1) One possible explanation of the dispersion in lambda (not in theta) could be variability in the typical width of the field. For a fixed spacing, wider fields might push the six fields around the center of the autocorrelogram toward the outside, depending on the details of how exactly the position of these fields is calculated. We recommend authors show that lambda does not correlate with field width, or at least that the variability explained by field width is smaller than the overall lambda variability.

      We agree that this option had not been carefully ruled out by our previous analyses. To tackle this question, we compute the field width of a given cell using the value at the minima of its spatial autocorrelogram (Fig S4A-B). For all cells in recording ID R12, there is a non-significant negative linear correlation between grid field width and between-cell variability (Fig S4C) . The variability explained by the width of the field is 4% of the variability, as indicated by the R<sup>2</sup> value of the linear fit. Similar results were found for all other modules (see caption of Fig S4C for statistics). Therefore, we do not think that grid field width explains spacing variability.

      (2) An alternative explanation could be related to what happens at the borders. The authors tackle this issue in Figure S2 but introduce a different way of measuring lambda based on three fields, which in our view is not optimal. We recommend showing that the dispersions in lambda and theta remain invariant as one removes the border-most part of the maps but estimating lambda through the autocorrelogram of the remaining part of the map. Of course, there is a limit to how much can be removed before measures of lambda and theta become very noisy.

      We have performed additional analysis to explore the role of borders in grid property variability. To do so, we have followed the suggestion by the reviewer and have re-analyzed grid properties from the autocorrelogram when the border-most part of the maps are removed (Fig S6A-B). For all modules, we do not see any changes in variability (computed as the standard deviation of the population distribution) for either orientation or spacing. As predicted by the reviewer, after removing about 25% of the border-most part of the environment we start seeing changes in variability, as measures of theta and lambda become noisy and computed over a smaller spatial range. This result holds for all other modules (Fig S6C-D).

      (3) A third possibility is slightly more tricky. Some works (for example Kropff et al, 2015) have shown that fields anticipate the rat position, so every time the rat traverses them they appear slightly displaced opposite to the direction of movement. The amount of displacement depends on the velocity. Maps that we construct out of a whole session should be deformed in a perfectly symmetric way if rats traverse fields in all directions and speeds. However, if the cell is conjunctive, we would expect a deformation mainly along the cell's preferred head direction. Since conjunctive cells have all possible preferred directions, and many grid cells are not conjunctive at all, this phenomenon could create variability in theta and lambda that is not a legitimate one but rather associated with the way we pool data to construct maps. To rule away this possibility, we recommend the authors study the variability in theta and lambda of conjunctive vs non-conjunctive grid cells. If the authors suspect that this phenomenon could explain part of their results, they should also take into account the findings of Gerlei and colleagues (2020) from the Nolan lab, that add complexity to this issue.

      We appreciate the reviewer pointing out the possible role conjunctive cells may play. To investigate how conjunctive cells may affect the observed grid property variability, we have performed additional analyses taking into account if the grid cells included in the study are conjunctive. Comparing within- and between-cell variability of conjunctive vs. non-conjunctive cells in recording R12, we do not see any qualitative differences for either orientation or spacing (Fig S7A-B). When excluding conjunctive cells from the between-variability comparison, we do not see any significant difference compared to when these cells are included (Fig S7C-D). As such, it does not appear that conjunctive cells are the source of variability in the population.

      We further note that the number of putative conjunctive cells varied across modules and recordings. For instance, in recording Q1 and Q2, Gardner et al. (2022) reported 3 (out of 97) and 1 (out of 66) conjunctive cells, respectively. Given that we see variability robustly across recordings (Fig 5), we do not believe that conjunctive cells can explain the presence of variability we observe.

      (4) The results in Figure 6 are correct, but we are not convinced by the argument. The fact that grid cells fire in the same way in different parts of the environment and in different environments is what gives them their appeal as a platform for path integration since displacement can be calculated independently of the location of the animal. Losing this universal platform is, in our view, too much of a price to pay when the only gain is the possibility of decoding position from a single module (or non-adjacent modules) which, as the authors discuss, is probably never the case. Besides, similar disambiguation of positions within the environment would come for free by adding to the decoding algorithm spatial cells (non-hexagonal but spatially stable), which are ubiquitous across the entorhinal cortex. Thus, it seems to us that - at least along this line of argumentation - with variability the network is losing a lot but not gaining much.

      We agree that losing the continuous attractor network (CAN) structure and the ability to path integrate would be a very large loss. However, we do not believe that the variability we observe necessarily destroys either the CAN or path integration. We argue this for two reasons. First, the data we analyzed [from Gardner et al. (2022)] is exactly the data set that was found to have toroidal topology and therefore viewed to be consistent with a major prediction of CANs. Thus, the amount of variability in grid properties does not rule out the underlying presence of a continuous attractor. Second, path integration may still be possible with grid cells that have variable properties. To illustrate this, we analyzed data from Sorscher et al. (2019) recurrent neural network model (RNN) that was trained explicitly on path integration, and found that the grid representations that emerged had variability in spacing and orientation (see point #6 below).

      (5) In Figure 4 one axis has markedly lower variability. Is this always the same axis? Can the authors comment more on this finding?

      We agree that in Fig 4 the first axis has lower variability. We believe that this is specific to the module R12 and does not reflect any differences in axis or bias in the methods used to compute the axis metrics. To test this, we have performed the same analyses for other modules, finding that other recordings do not exhibit the same bias. Results for the modules with the most cells are shown below (Author response image 1).

      Author response image 1.

      Grid propertied along Axis 1 are not less variable for many recorded grid modules. Same as Fig.4C-D, but for four other recorded modules. Note that the variability along each axis is similar.

      (6) The paper would gain in depth if maps coming out of different computational models could be analyzed in the same way.

      We agree with the reviewer that examining computational models using the same approach would strengthen our results and we appreciate the suggestion. To address this, we have analyzed the results from a previous normative model for grid cells [Sorscher et al., (2019)] that trained a recurrent neural network (RNN) model to perform path integration and found that units developed grid cell like responses. These models have been found to exhibit signatures of toroidal attractor dynamics [Sorscher et al. (2023)] and exhibit a diversity of responses beyond pure grid cells, making them a good starting point for understanding whether models of MEC may contain uncharacterized variability in grid properties.

      We find that RNN units in these normative models exhibit similar amounts of variability in grid spacing and orientation as observed in the real grid cell recordings (Fig S8A-D). This provides additional evidence that this variability may be expected from a normative framework, and that the variability does not destroy the ability to path integrate (which the RNN is explicitly trained to perform).

      The RNN model offers possibilities to assess what might cause this variability. While we leave a detailed investigation of this to future work, we varied the weight decay regularization hyper-parameter. This value controls how sparse the weights in the hidden recurrent layer are. Large weight decay regularization strength encourages sparser connectivity, while small weight decay regularization strength allows for denser connectivity. We find that increasing this penalty (and enforcing sparser connectivity) decreases the variability of grid properties (Fig S8E-F). This suggests that the observed variability in the Gardner et al. (2022) data set could be due to the fact that grid cells are synaptically connected to other, non-grid cells in MEC.

      (7) Similarly, it would be very interesting to expand the study with some other data to understand if between-cell delta_theta and delta_lambda are invariant across environments. In a related matter, is there a correlation between delta_theta (delta_lambda) for the first vs for the second half of the session? We expect there should be a significant correlation, it would be nice to show it.

      We agree this would be interesting to examine. For this analysis, it is essential to have a large number of grid cells, and we are not aware of other published data sets with comparable cell numbers using different environments.

      Using a sliding window analysis, we have characterized changes in variability with respect to the recording time (Figure S5A). To do so, we compute grid orientation and spacing over a time-window whose length is half of the total length of the recording. From the population distribution of orientation and spacing values, we compute the standard deviation as a measure of between-cell variability. We repeat the same procedure, sliding the window forward until the variability for the second half of the recording is computed.

      We applied this approach to recording ID R12 (the same as in Figs 2-4) given that this recording session was significantly longer than the rest (almost two hours). Results are shown in Fig S5 B-C. For both orientation and spacing, no systematic changes of variability with respect to time were observed. Similar results were found for other modules (see caption of Fig S5 for statistics).

      We also note that the rats were already familiarized with the environment for 10-20 sessions prior to the recordings, so there may not be further learning during the period of the grid cell recordings. No changes in variability can be seen in Rat R across days (e.g., in Fig 5B R12 and R22 have similar distributions of variability). However, we note that it may be possible that there are changes in grid properties at time-scales greater than the recordings.

    1. eLife Assessment

      This important study reports a detailed quantification of the population dynamics of Salmonella enterica serovar Typhimurium in mice. Bacterial burden and founding population sizes across various organs were quantified, revealing pathways of dissemination and reseeding of the gastrointestinal tract from systemic organs. Using various techniques, including genetic distance measurements, the authors present compelling evidence to support their conclusions, thus presenting new knowledge that will be of broad interest to scientists focusing on infectious diseases.

    2. Reviewer #1 (Public review):

      Hotinger et al. explore the population dynamics of Salmonella enterica serovar Typhimurium in mice using genetically tagged bacteria. In addition to physiological observations, pathology assessments, and CFU measurements, the study emphasizes quantifying host bottleneck sizes that limit Salmonella colonization and dissemination. The authors also investigate the genetic distances between bacterial populations at various infection sites within the host.

      Initially, the study confirms that pretreatment with the antibiotic streptomycin before inoculation via orogastric gavage increases the bacterial burden in the gastrointestinal (GI) tract, leading to more severe symptoms and heightened fecal shedding of bacteria. This pretreatment also significantly reduces between-animal variation in bacterial burden and fecal shedding. The authors then calculate founding population sizes across different organs, discovering a severe bottleneck in the intestine, with founding populations reduced by approximately 10^6-fold compared to the inoculum size. Streptomycin pretreatment increases the founding population size and bacterial replication in the GI tract. Moreover, by calculating genetic distances between populations, the authors demonstrate that, in untreated mice, Salmonella populations within the GI tract are genetically dissimilar, suggesting limited exchange between colonization sites. In contrast, streptomycin pretreatment reduces genetic distances, indicating increased exchange.

      In extraintestinal organs, the bacterial burden is generally not substantially increased by streptomycin pretreatment, with significant differences observed only in the mesenteric lymph nodes and bile. However, the founding population sizes in these organs are increased. By comparing genetic distances between organs, the authors provide evidence that subpopulations colonizing extraintestinal organs diverge early after infection from those in the GI tract. This hypothesis is further tested by measuring bacterial burden and founding population sizes in the liver and GI tract at 5 and 120 hours post-infection. Additionally, they compare orogastric gavage infection with the less injurious method of infection via drinking, finding similar results for CFUs, founding populations, and genetic distances. These results argue against injuries during gavage as a route of direct infection.

      To bypass bottlenecks associated with the GI tract, the authors compare intravenous (IV) and intraperitoneal (IP) routes of infection. They find approximately a 10-fold increase in bacterial burden and founding population size in immune-rich organs with IV/IP routes compared to orogastric gavage in streptomycin-pretreated animals. This difference is interpreted as a result of "extra steps required to reach systemic organs."

      While IP and IV routes yield similar results in immune-rich organs, IP infections lead to higher bacterial burdens in nearby sites, such as the pancreas, adipose tissue, and intraperitoneal wash, as well as somewhat increased founding population sizes. The authors correlate these findings with the presence of white lesions in adipose tissue. Genetic distance comparisons reveal that, apart from the spleen and liver, IP infections lead to genetically distinct populations in infected organs, whereas IV infections generally result in higher genetic similarity.

      Finally, the authors investigate GI tract reseeding, identifying two distinct routes. They observe that the GI tracts of IP/IV-infected mice are colonized either by a clonal or a diversely tagged bacterial population. In clonally reseeded animals, the genetic distance within the GI tract is very low (often zero) compared to the bile population, which is predominantly clonal or pauciclonal. These animals also display pathological signs, such as cloudy/hardened bile and increased bacterial burden, leading the authors to conclude that the GI tract was reseeded by bacteria from the gallbladder bile. In contrast, animals reseeded by more complex bacterial populations show that bile contributes only a minor fraction of the tags. Given the large founding population size in these animals' GI tracts, which is larger than in orogastrically infected animals, the authors suggest a highly permissive second reseeding route, largely independent of bile. They speculate that this route may involve a reversal of known mechanisms that the pathogen uses to escape from the intestine.

      The manuscript presents a substantial body of work that offers a meticulously detailed understanding of the population dynamics of S. Typhimurium in mice. It quantifies the processes shaping the within-host dynamics of this pathogen and provides new insights into its spread, including previously unrecognized dissemination routes. The methodology is appropriate and carefully executed, and the manuscript is well-written, clearly presented, and concise. The authors' conclusions are well-supported by experimental results and thoroughly discussed. This work underscores the power of using highly diverse barcoded pathogens to uncover the within-host population dynamics of infections and will likely inspire further investigations into the molecular mechanisms underlying the bottlenecks and dissemination routes described here.

    3. Reviewer #2 (Public review):

      In this paper, Hotinger et. al. propose an improved barcoded library system, called STAMPR, to study Salmonella population dynamics during infection. Using this system, the authors demonstrate significant diversity in the colonization of different Salmonella clones (defined by the presence of different barcodes) not only across different organs (liver, spleen, adipose tissues, pancreas and gall bladder) but also within different compartments of the same gastrointestinal tissue. Additionally, this system revealed that microbiota competition is the major bottleneck in Salmonella intestinal colonization, which can be mitigated by streptomycin treatment. However, this has been demonstrated previously in numerous publications. They also show that there was minimal sharing between populations found in the intestine and those in the other organs. Upon IV and IP infection to bypass the intestinal bottleneck, they were able to demonstrate, using this library, that Salmonella can renter the intestine through two possible routes. One route is essentially the reverse path used to escape the gut, leading to a diverse intestinal population; while the other, through the bile, typically results in a clonal population.

      Comments on latest version:

      The authors have addressed my concerns.

    4. Author response:

      The following is the authors’ response to the original reviews.

      Public Reviews:

      Reviewer #1 (Public review): 

      Hotinger et al. explore the population dynamics of Salmonella enterica serovar Typhimurium in mice using genetically tagged bacteria. In addition to physiological observations, pathology assessments, and CFU measurements, the study emphasizes quantifying host bottleneck sizes that limit Salmonella colonization and dissemination. The authors also investigate the genetic distances between bacterial populations at various infection sites within the host.

      Initially, the study confirms that pretreatment with the antibiotic streptomycin before inoculation via orogastric gavage increases the bacterial burden in the gastrointestinal (GI) tract, leading to more severe symptoms and heightened fecal shedding of bacteria. This pretreatment also significantly reduces between-animal variation in bacterial burden and fecal shedding. The authors then calculate founding population sizes across different organs, discovering a severe bottleneck in the intestine, with founding populations reduced by approximately 10^6-fold compared to the inoculum size. Streptomycin pretreatment increases the founding population size and bacterial replication in the GI tract. Moreover, by calculating genetic distances between populations, the authors demonstrate that, in untreated mice, Salmonella populations within the GI tract are genetically dissimilar, suggesting limited exchange between colonization sites. In contrast, streptomycin pretreatment reduces genetic distances, indicating increased exchange.

      In extraintestinal organs, the bacterial burden is generally not substantially increased by streptomycin pretreatment, with significant differences observed only in the mesenteric lymph nodes and bile. However, the founding population sizes in these organs are increased. By comparing genetic distances between organs, the authors provide evidence that subpopulations colonizing extraintestinal organs diverge early after infection from those in the GI tract. This hypothesis is further tested by measuring bacterial burden and founding population sizes in the liver and GI tract at 5 and 120 hours post-infection. Additionally, they compare orogastric gavage infection with the less injurious method of infection via drinking, finding similar results for CFUs, founding populations, and genetic distances. These results argue against injuries during gavage as a route of direct infection. 

      To bypass bottlenecks associated with the GI tract, the authors compare intravenous (IV) and intraperitoneal (IP) routes of infection. They find approximately a 10-fold increase in bacterial burden and founding population size in immune-rich organs with IV/IP routes compared to orogastric gavage in streptomycin-pretreated animals. This difference is interpreted as a result of "extra steps required to reach systemic organs."

      While IP and IV routes yield similar results in immune-rich organs, IP infections lead to higher bacterial burdens in nearby sites, such as the pancreas, adipose tissue, and intraperitoneal wash, as well as somewhat increased founding population sizes. The authors correlate these findings with the presence of white lesions in adipose tissue. Genetic distance comparisons reveal that, apart from the spleen and liver, IP infections lead to genetically distinct populations in infected organs, whereas IV infections generally result in higher genetic similarity. 

      Finally, the authors investigate GI tract reseeding, identifying two distinct routes. They observe that the GI tracts of IP/IV-infected mice are colonized either by a clonal or a diversely tagged bacterial population. In clonally reseeded animals, the genetic distance within the GI tract is very low (often zero) compared to the bile population, which is predominantly clonal or pauciclonal. These animals also display pathological signs, such as cloudy/hardened bile and increased bacterial burden, leading the authors to conclude that the GI tract was reseeded by bacteria from the gallbladder bile. In contrast, animals reseeded by more complex bacterial populations show that bile contributes only a minor fraction of the tags. Given the large founding population size in these animals' GI tracts, which is larger than in orogastrically infected animals, the authors suggest a highly permissive second reseeding route, largely independent of bile. They speculate that this route may involve a reversal of known mechanisms that the pathogen uses to escape from the intestine. 

      The manuscript presents a substantial body of work that offers a meticulously detailed understanding of the population dynamics of S. Typhimurium in mice. It quantifies the processes shaping the within-host dynamics of this pathogen and provides new insights into its spread, including previously unrecognized dissemination routes. The methodology is appropriate and carefully executed, and the manuscript is well-written, clearly presented, and concise. The authors' conclusions are well-supported by experimental results and thoroughly discussed. This work underscores the power of using highly diverse barcoded pathogens to uncover the within-host population dynamics of infections and will likely inspire further investigations into the molecular mechanisms underlying the bottlenecks and dissemination routes described here.

      Major point:

      Substantial conclusions in the manuscript rely on genetic distance measurements using the Cavalli-Sforza chord distance. However, it is unclear whether these genetic distance measurements are independent of the founding population size. I would anticipate that in populations with larger founding population sizes, where the relative tag frequencies are closer to those in the inoculum, the genetic distances would appear smaller compared to populations with smaller founding sizes independent of their actual relatedness. This potential dependency could have implications for the interpretation of findings, such as those in Figures 2B and 2D, where antibiotic-pretreated animals consistently exhibit higher founding population sizes and smaller genetic distances compared to untreated animals.

      Thank you for raising this important point regarding reliance on cord distances for gauging genetic distance in barcoded populations. The reviewer is correct that samples with more founders will be more similar to the inoculum and thus inherently more similar to other samples that also have more founders. However, creation of libraries containing very large numbers of unique barcodes can often circumvent this issue. In this case, the effect size of chance-based similarity is not large enough to change the interpretation of the data in Figures 2B and 2D. In our case, the library has ~6x10<sup>4</sup> barcodes, and the founding populations in Figure 2B are ~10<sup>3</sup>. Randomly resampling to create two populations of 10<sup>3</sup> cells from an initial population with 6x10<sup>4</sup> barcodes is expected to yield largely distinct populations with very little similarity. Thus, the similarity between streptomycin-treated populations in Figure 2D is likely the result of biology rather than chance.  

      Reviewer #2 (Public review):

      In this paper, Hotinger et. al. propose an improved barcoded library system, called STAMPR, to study Salmonella population dynamics during infection. Using this system, the authors demonstrate significant diversity in the colonization of different Salmonella clones (defined by the presence of different barcodes) not only across different organs (liver, spleen, adipose tissues, pancreas, and gall bladder) but also within different compartments of the same gastrointestinal tissue. Additionally, this system revealed that microbiota competition is the major bottleneck in Salmonella intestinal colonization, which can be mitigated by streptomycin treatment. However, this has been demonstrated previously in numerous publications. They also show that there was minimal sharing between populations found in the intestine and those in the other organs. Upon IV and IP infection to bypass the intestinal bottleneck, they were able to demonstrate, using this library, that Salmonella can renter the intestine through two possible routes. One route is essentially the reverse path used to escape the gut, leading to a diverse intestinal population; while the other, through the bile, typically results in a clonal population. Although the authors showed that the STAMPR pipeline improved the ability to identify founder populations and their diversity within the same animal during infections, some of the conclusions appear speculative and not fully supported.

      (1) It's particularly interesting how the authors, using this system, demonstrate the dominant role of the microbiota bottleneck in Salmonella colonization and how it is widened by antibiotic treatment (Figure 1). Additionally, the ability to track Salmonella reseeding of the gut from other organs starting with IV and IP injections of the pathogen provides a new tool to study population dynamics (Figure 5). However, I don't think it is possible to argue that the proximal and distal small intestine, Peyer's patches (PPs), cecum, colon, and feces have different founder populations for reasons other than stochastic variations. All the barcoded Salmonella clones have the same fitness and the fact that some are found or expanded in one region of the gastrointestinal tract rather than another likely results from random chance - such as being forced in a specific region of the gut for physical or spatial reasons-and subsequent expansion, rather than any inherent biological cause. For example, some bacteria may randomly adhere to the mucus, some may swim toward the epithelial layer, while others remain in the lumen; all will proliferate in those respective sites. In this way, different founder populations arise based on random localization during movement through the gastrointestinal tract, which is an observation, but it doesn't significantly contribute to understanding pathogen colonization dynamics or pathogenesis. Therefore, I would suggest placing less emphasis on describing these differences or better discussing this aspect, especially in the context of the gastrointestinal tract.

      Thank you for helping us identify this area for further clarification. We agree with the reviewer’s interpretation that seeding of proximal and distal small intestine, Peyer's patches (PPs), cecum, colon, and feces with different founder populations is likely caused by stochastic variations, consistent with separate stochastic bottlenecks to establishing these separate niches. To clarify this point we have modified the text in the results section, “Streptomycin treatment decreases compartmentalization of S. Typhimurium populations within the intestine”.

      Change to text:

      “Except for the cecum and colon, in untreated animals the S. Typhimurium populations in different regions of the intestine were dissimilar (Avg. GD ranged from 0.369 to 0.729, 2D left); i.e., there is little sharing between populations in the intestine. These data suggest that there are separate bottlenecks in different regions of the intestine that cause stochastic differences in the identity of the founders. Interestingly, when these founders replicate, they do not mix, remaining compartmentalized with little sharing between populations throughout the intestinal tract (i.e., barcodes found in one region are not in other regions, Figure S3). This was surprising as the luminal contents, an environment presumably conducive to bacterial movement, were not removed from these samples.”

      In this section we are interested in the underlying biology that occurs after the initial bottleneck to preserve this compartmentalization during outgrowth of the intestinal population. In other words, what prevents these separate populations from merging (e.g., what prevents the bacteria replicating in the proximal small intestine from traveling through the intestine and establishing a niche in the distal small intestine)? While we do not explore the mechanisms of compartmentalization, we observe that it is disrupted by streptomycin pretreatment, suggesting a microbiota-dependent biological cause. 

      (2) I do think that STAMPR is useful for studying the dynamics of pathogen spread to organs where Salmonella likely resides intracellularly (Figure 3). The observation that the liver is colonized by an early intestinal population, which continues to proliferate at a steady rate throughout the infection, is very interesting and may be due to the unique nature of the organ compared to the mucosal environment. What is the biological relevance during infection? Do the authors observe the same pattern (Figures 3C and G) when normalizing the population data for the spleen and mesenteric lymph nodes (mLN)? If not, what do the authors think is driving this different distribution?

      Thank you for raising this interesting point. These data indicate that the liver is seeded from the intestine early during infection. The timing and source of dissemination have relevance for understanding how host and pathogen variables control the spread of bacteria to systemic sites. For example, our conclusion (early dissemination) indicates that the immune state of a host at the time of exposure to a pathogen, and for a short period thereafter, are what primarily influence the process of dissemination, not the later response to an active infection. 

      We observe that the liver and mucosal environments within the intestine have similar colonization behaviors. Both niches are seeded early during infection, followed by steady pathogen proliferation and compartmentalization that apparently inhibits further seeding. This results in the identity of barcodes in the liver population remaining distinct from the intestinal populations, and the intestinal populations remaining distinct from each other.

      We observe a similar pattern to the liver in the spleen and MLN (the barcodes in the spleen and MLN are dissimilar to the population in the intestine). To clarify this point, we have modified the text (below) and added this analysis as a supplemental figure (S4).

      Change to text:

      Genetic distance comparison of liver samples to other sites revealed that, regardless of streptomycin treatment, there was very little sharing of barcodes between the intestine and extraintestinal sites (Avg. GD >0.75, Figure 3C). Furthermore, the MLN and spleen populations also lacked similarity with the intestine (Figure S4). These analyses strongly support the idea that S. Typhimurium disseminates to extraintestinal organs relatively early following inoculation, before it establishes a replicative niche in the intestine.

      (3) Figure 6: Could the bile pathology be due to increased general bacterial translocation rather than Salmonella colonization specifically? Did the authors check for the presence of other bacteria (potentially also proliferating) in the bile? Do the authors know whether Salmonella's metabolic activity in the bile could be responsible for gallbladder pathology?

      The reviewer raises interesting points for future work. We did not check whether other bacterial species are translocating during S. Typhimurium infection. The relevance of Salmonella’s metabolic activity is also very interesting, and we hope these questions will be answered by future studies.

      Recommendations for the authors:

      Reviewer #1 (Recommendations for the authors):

      Minor points:

      (1) P. 9/10 "... the marked delay in shedding after IP and IV relative to orogastric inoculation suggest that the S. Typhimurium population encounters substantial bottleneck(s) on the route(s) from extraintestinal sites back to the intestine.": Can you conclude that from the data? It could also be possible that there is a biological mechanism (other than chance events) that delays the re-entry to the intestine.

      We propose that the delay in shedding indicates additional obstacles that bacteria face when re-entering the intestine, and that there are likely biological mechanisms that cause this delay. However, these unknown mechanisms effectively act as additional bottlenecks by causing a stochastic loss of population diversity. 

      (2) P. 11 "...both organs would likely contain all 10 barcodes. In contrast, a library with 10,000 barcodes can be used to distinguish between a bottleneck resulting in Ns = 1,000 and Ns = 10,000, since these bottlenecks result in a different number of barcodes in output samples. Furthermore, high diversity libraries reduce the likelihood that two tissue samples share the same barcode(s) due to random chance, enabling more accurate quantification of bacterial dissemination.": I agree with the general analysis, but I find it misleading to talk about the presence of barcodes when the analyses in this manuscript are based on the much more powerful comparison of relative abundance of individual tags instead of their presence or absence.

      The reviewer raises an excellent point, and the distinction between relative abundance versus presence/absence is discussed extensively in the original STAMPR manuscript. Although relative abundance is powerful, the primary metric used in this study (Ns) is calculated principally from the number of barcodes, corrected (via simulations) for the probability of observing the same barcode across distinct founders. Although this correction procedure does rely on barcode abundance, the primary driver of founding population quantification is the number of barcodes.

      (3) P.14 "the library in LB supplemented with SM was not significantly different than the parent strain" and Figure 2C: How was significance tested? How many times were the growth curves recorded? On my print-out, the red color has different shades for different growth curves.

      Significance was tested with a Mann-Whitney and growth curves were performed 5 times. Growth curves are displayed with 50% opacity, and as a result multiple curves directly on top of each other appear darker. The legend to S2 has been modified accordingly.

      (4) P.16: close bracket in the equation for FRD calculation.

      Done

      (5) Figure 2C "Average CFU per founder": I found the wording confusing at first as I thought you divided the average bacterial burden per organ by Ns, instead of averaging the CFU/Ns calculated for each mouse.

      The wording has been clarified. 

      (6) Figure 3B: It would be helpful to include expected genetic distances in the schematic as it is difficult to infer the genetic distance when only two of three, respectively, different "barcode colors" are used. While I find the explanation in the main text intuitive, a graphical representation would have helped me.

      Thank you for the suggestion. Unfortunately, using colors to represent barcodes is imperfect and limits the diversity that can be depicted. We have modified Figure 3B to further clarify. 

      (7) Figure 3C: Why do you compare the genetic distance to the liver, when you discuss the genetic distance of the intestinal population? Is it not possible that the intestinal populations are similar to the extraintestinal organs except the liver?

      For clarity, we chose to highlight exclusively the liver. However, we observed a similar pattern to the liver in other extraintestinal organs. To clarify the generalizability of this point we have added a supplemental figure with comparisons to MLN and Spleen (Supplemental figure S4) as well as further text.

      (8) Figure 3C & S5A: I found "+SM" and "+SM, Drinking" confusing and would have preferred "+SM, Gavage" and "+SM, Drinking" for clarity.

      Done, thank you for the suggestion.

      (9) Figure 3G&H: I find it worthy of discussion that the bacterial burden increases over time, while the founding population decreases. Does that not indicate that replication only occurs at specific sites leading to the amplification of only a few barcodes and thereby a larger change of the relative barcode abundance compared to the inoculum?

      From 5h to 120h the size of the founding population decreases in multiple intestinal sites. This likely indicates that the impact of the initial bottleneck is still ongoing at 5h, although further temporal analysis would be required to define the exact timing of the bottleneck. Notably, the passage time through the mouse intestine is ~5h. Many of the founders observed at 5h could be a population that will never establish a replicative niche, and failing to colonize be shed in the feces, bottlenecking the population between 5h and 120h. To clarify this point we have added the following text:

      Section “S. Typhimurium disseminates out of the intestine before establishing an intestinal replicative niche”.

      “In contrast to the liver, there were more founders present in samples from the intestine (particularly in the colon) at 5 hours versus 120 hours (Figure 3H). These data likely indicate that many of the founders observed in the intestine at 5 hours are shed in the feces prior to establishing a replicative niche, and demonstrates that the forces restricting the S. Typhimurium population in the intestine act over a period of > 5 hours.”  

      (10) Figure S2A: I do not understand this figure. Why are there more than 70.000 tags listed? I was under the impression the barcode library in S. Typhimurium had 55.000 tags while only the plasmid pSM1 had more than 70.000 (but the plasmid should not be relevant here). Why are there distinct lines at approximately 10^-5 and a bit lower? I would have expected continuously distributed barcode frequencies.

      During barcode analysis, each library is mapped to the total barcode list in the barcode donor pSM1, which contains ~70,000 barcodes. This enables consistent analysis across different bacterial libraries. The designation “barcode number” refers to the barcode number in pSM1, meaning many of the barcodes in the Salmonella library are at zero reads. This graph type was chosen to show there was no bias toward a particular barcode, however there is significant overlap of the points, making individual barcode frequencies difficult to see. We have changed the x-axis to state “pSM1 Barcode Number” and clarified in the figure legend.

      Since the y-axes on these graphs is on a log10 scale, the lines represent barcodes with 1 read, 2 reads, 3 reads, etc. As the number of reads per barcode increases linearly, the space between them decreases on logarithmic axes.

      (11) There are a few typos in the figure legends of the supplementary material. For example Figure S2: S. Typhimurium not italicized, ~7x105 no superscript. Fig. S4&5 ", Open circles" is "O" is capitalized.

      Typos have been corrected.

    1. eLife Assessment

      The current human tissue-based study provides compelling evidence correlating hippocampal expressions of RNA guanine-rich G-quadruplexes with aging and with Alzheimer's Disease presence and severity. The results are fundamental and will rejuvenate our understanding of aging and AD's pathogenesis.

      [Editors' note: this paper was reviewed by Review Commons.]

    2. Reviewer #1 (Public review):

      This is an interesting manuscript where the authors systematically measure rG4 levels in brain samples at different ages of patients affected by AD. To the best of my knowledge this is the first time that BG4 staining is used in this context and the authors provide compelling evidence to show an association with BG4 staining and age or AD progression, which interestingly indicates that such RNA structure might play a role in regulating protein homeostasis as previously speculated. The methods used and the results reported seems robust and reproducible.

    3. Reviewer #2 (Public review):

      RNA guanine-rich G-quadruplexes (rG4s) are non-canonical higher order nucleic acid structures that can form under physiological conditions. Interestingly, cellular stress is positively correlated with rG4 induction.

      In this study, the authors examined human hippocampal postmortem tissue for the formation ofrG4s in aging and Alzheimer Disease (AD). rG4 immunostaining strongly increased in the hippocampus with both age and with AD severity. 21 cases were used in this study (age range 30-92).

      This immunostaining co-localized with hyper-phosphorylated tau immunostaining in neurons. The BG4 staining levels were also impacted by APOE status. rG4 structure was previously found to drive tau aggregation. Based on these observations, the authors propose a model of neurodegeneration in which chronic rG4 formation drives proteostasis collapse.

      This model is interesting, and would explain different observations (e.g., RNA is present in AD aggregates and rG4s can enhance protein oligomerization and tau aggregation).

    4. Author response:

      The following is the authors’ response to the original reviews.

      Public Reviews:

      Reviewer #1 (Public review):

      This is an interesting manuscript where the authors systematically measure rG4 levels in brain samples at different ages of patients affected by AD. To the best of my knowledge this is the first time that BG4 staining is used in this context and the authors provide compelling evidence to show an association with BG4 staining and age or AD progression, which interestingly indicates that such RNA structure might play a role in regulating protein homeostasis as previously speculated. The methods used and the results reported seems robust and reproducible. There were two main things that needed addressing:

      (1) Usually in BG4 staining experiments to ensure that the signal detected is genuinely due to rG4 an RNase treatment experiment is performed. This does not have to be extended to all the samples presented but having a couple of controls where the authors observe loss of staining upon RNase treatment will be key to ensure with confidence that rG4s are detected under the experimental conditions. This is particularly relevant for this brain tissue samples where BG4 staining has never been performed before.

      (2) The authors have an association between rG4-formation and age/disease progression. They also observe distribution dependency of this, which is great. However, this is still an association which does not allow the model to be supported. This is not something that can be fixed with an easy experiment and it is what it is, but my point is that the narrative of the manuscript should be more fair and reflect the fact that, although interesting, what the authors are observing is a simple correlation. They should still go ahead and propose a model for it, but they should be more balanced in the conclusion and do not imply that this evidence is sufficient to demonstrate the proposed model. It is absolutely fine to refer to the literature and comment on the fact that similar observations have been reported and this is in line with those, but still this is not an ultimate demonstration.

      Comments on current version:

      The authors have now addressed my concerns.

      We thank the reviewer for their support!

      Reviewer #2 (Public review):

      RNA guanine-rich G-quadruplexes (rG4s) are non-canonical higher order nucleic acid structures that can form under physiological conditions. Interestingly, cellular stress is positively correlated with rG4 induction.

      In this study, the authors examined human hippocampal postmortem tissue for the formation ofrG4s in aging and Alzheimer Disease (AD). rG4 immunostaining strongly increased in the hippocampus with both age and with AD severity. 21 cases were used in this study (age range 30-92).

      This immunostaining co-localized with hyper-phosphorylated tau immunostaining in neurons. The BG4 staining levels were also impacted by APOE status. rG4 structure was previously found to drive tau aggregation. Based on these observations, the authors propose a model of neurodegeneration in which chronic rG4 formation drives proteostasis collapse.

      This model is interesting, and would explain different observations (e.g., RNA is present in AD aggregates and rG4s can enhance protein oligomerization and tau aggregation).

      Main issue from the previous round of review:

      There is indeed a positive correlation between Braak stage severity and BG4 staining, but this correlation is relatively weak and borderline significant ((R = 0.52, p value = 0.028). This is probably the main limitation of this study, which should be clearly acknowledged (together with a reminder that "correlation is not causality"). Related to this, here is no clear justification to exclude the four individuals in Fig 1d (without them R increases to 0.78). Please remove this statement. On the other hand, the difference based on APOE status is more striking.

      Comments on current version:

      The authors have made laudable efforts to address the criticisms I made in my evaluation of the original manuscript.

      We thank the reviewer for their support!

      Recommendations for the authors:

      Reviewing Editor:

      I would suggest two minor edits:

      - The findings are correlative and descriptive, but the title implies functionality (A New Role for RNA G-quadruplexes in Aging and Alzheimer′s Disease). I would suggest toning down this title).

      - While I understand the limitations in performing additional biochemical experiments to validate the immunofluorescence study, I think this is worth mentioning as a limitation in the text.

      We have made these two changes as requested, altering the title to remove the word Role that may imply more meaning than intended, and adding a line to the discussion on the need for future additional biochemical experiments.

      Reviewer #1 (Recommendations for the authors):

      Thanks for addressing the concerns raised.

      We thank the reviewer for their support!

      Reviewer #2 (Recommendations for the authors):

      Minor point:

      Related to the "correlation is not causality" remark I made in my evaluation of the original manuscript: the authors' answer is reasonable. Still, I would suggest to modify the abstract: "we propose a model of neurodegeneration in which chronic rG4 formation drives proteostasis collapse" => "we propose a model of neurodegeneration in which chronic rG4 formation is linked to proteostasis collapse"

      All other remarks I made have been answered properly.

      We thank the reviewer for their support! We have made the change exactly as requested by the reviewer.

    1. eLife Assessment

      This important study provides information on the TMEM16 family of membrane proteins, which play roles in lipid scrambling and ion transport. By simulating 27 structures representing five distinct family members, the authors captured hundreds of lipid scrambling events, offering insights into the mechanisms of lipid translocation and the specific protein regions involved in these processes. However, while the data on groove dilation is compelling, the evidence for outside-the-groove scramblase activity without experimental validation is inadequate and is based on a limited set of observed events.

    2. Reviewer #1 (Public review):

      Summary:

      The manuscript investigates lipid scrambling mechanisms across TMEM16 family members using coarse-grained molecular dynamics (MD) simulations. While the study presents a statistically rigorous analysis of lipid scrambling events across multiple structures and conformations, several critical issues undermine its novelty, impact, and alignment with experimental observations.

      Critical issues:

      (1) Lack of Novelty:<br /> The phenomenon of lipid scrambling via an open hydrophilic groove is already well-established in the literature, including through atomistic MD simulations. The authors themselves acknowledge this fact in their introduction and discussion. By employing coarse-grained simulations, the study essentially reiterates previously known findings with limited additional mechanistic insight. The repeated observation of scrambling occurring predominantly via the groove does not offer significant advancement beyond prior work.

      (2) Redundancy Across Systems:<br /> The manuscript explores multiple TMEM16 family members in activating and non-activating conformations, but the conclusions remain largely confirmatory. The extensive dataset generated through coarse-grained MD simulations primarily reinforces established mechanistic models rather than uncovering fundamentally new insights. The effort, while statistically robust, feels excessive given the incremental nature of the findings.

      (3) Discrepancy with Experimental Observations:<br /> The use of coarse-grained simulations introduces inherent limitations in accurately representing lipid scrambling dynamics at the atomistic level. Experimental studies have highlighted nuances in lipid permeation that are not fully captured by coarse-grained models. This discrepancy raises questions about the biological relevance of the reported scrambling events, especially those occurring outside the canonical groove.

      (4) Alternative Scrambling Sites:<br /> The manuscript reports scrambling events at the dimer-dimer interface as a novel mechanism. While this observation is intriguing, it is not explored in sufficient detail to establish its functional significance. Furthermore, the low frequency of these events (relative to groove-mediated scrambling) suggests they may be artifacts of the simulation model rather than biologically meaningful pathways.

      Conclusion:

      Overall, while the study is technically sound and presents a large dataset of lipid scrambling events across multiple TMEM16 structures, it falls short in terms of novelty and mechanistic advancement. The findings are largely confirmatory and do not bridge the gap between coarse-grained simulations and experimental observations. Future efforts should focus on resolving these limitations, possibly through atomistic simulations or experimental validation of the alternative scrambling pathways.

    3. Reviewer #2 (Public review):

      Summary:

      Stephens et al. present a comprehensive study of TMEM16-members via coarse-grained MD simulations (CGMD). They particularly focus on the scramblase ability of these proteins and aim to characterize the "energetics of scrambling". Through their simulations, the authors interestingly relate protein conformational states to the membrane's thickness and link those to the scrambling ability of TMEM members, measured as the trespassing tendency of lipids across leaflets. They validate their simulation with a direct qualitative comparison with Cryo-EM maps.

      Strengths:

      The study demonstrates an efficient use of CGMD simulations to explore lipid scrambling across various TMEM16 family members. By leveraging this approach, the authors are able to bypass some of the sampling limitations inherent in all-atom simulations, providing a more comprehensive and high-throughput analysis of lipid scrambling. Their comparison of different protein conformations, including open and closed groove states, presents a detailed exploration of how structural features influence scrambling activity, adding significant value to the field. A key contribution of this study is the finding that groove dilation plays a central role in lipid scrambling. The authors observe that for scrambling-competent TMEM16 structures, there is substantial membrane thinning and groove widening. The open Ca2+-bound nhTMEM16 structure (PDB ID 4WIS) was identified as the fastest scrambler in their simulations, with scrambling rates as high as 24.4 {plus minus} 5.2 events per μs. This structure also shows significant membrane thinning (up to 18 Å), which supports the hypothesis that groove dilation lowers the energetic barrier for lipid translocation, facilitating scrambling.

      The study also establishes a correlation between structural features and scrambling competence, though analyses often lack statistical robustness and quantitative comparisons. The simulations differentiate between open and closed conformations of TMEM16 structures, with open-groove structures exhibiting increased scrambling activity, while closed-groove structures do not. This finding aligns with previous research suggesting that the structural dynamics of the groove are critical for scrambling. Furthermore, the authors explore how the physical dimensions of the groove qualitatively correlate with observed scrambling rates. For example, TMEM16K induces increased membrane thinning in its open form, suggesting that membrane properties, along with structural features, play a role in modulating scrambling activity.

      Another significant finding is the concept of "out-of-the-groove" scrambling, where lipid translocation occurs outside the protein's groove. This observation introduces the possibility of alternate scrambling mechanisms that do not follow the traditional "credit-card model" of groove-mediated lipid scrambling. In their simulations, the authors note that these out-of-the-groove events predominantly occur at the dimer interface between TM3 and TM10, especially in mammalian TMEM16 structures. While these events were not observed in fungal TMEM16s, they may provide insight into Ca2+-independent scrambling mechanisms, as they do not require groove opening.

      Weaknesses:

      A significant challenge of the study is the discrepancy between the scrambling rates observed in CGMD simulations and those reported experimentally. Despite the authors' claim that the rates are in line experimentally, the observed differences can mean large energetic discrepancies in describing scrambling (larger than 1kT barrier in reality). For instance, the authors report scrambling rates of 10.7 events per μs for TMEM16F and 24.4 events per μs for nhTMEM16, which are several orders of magnitude faster than experimental rates. While the authors suggest that this discrepancy could be due to the Martini 3 force field's faster diffusion dynamics, this explanation does not fully account for the large difference in rates. A more thorough discussion on how the choice of force field and simulation parameters influence the results, and how these discrepancies can be reconciled with experimental data, would strengthen the conclusions. Likewise, rate calculations in the study are based on 10 μs simulations, while experimental scrambling rates occur over seconds. This timescale discrepancy limits the study's accuracy, as the simulations may not capture rare or slow scrambling events that are observed experimentally and therefore might underestimate the kinetics of scrambling. It's however important to recognize that it's hard (borderline unachievable) to pinpoint reasonable kinetics for systems like this using the currently available computational power and force field accuracy. The faster diffusion in simulations may lead to overestimated scrambling rates, making the simulation results less comparable to real-world observations. Thus, I would therefore read the findings qualitatively rather than quantitatively. An interesting observation is the asymmetry observed in the scrambling rates of the two monomers. Since MARTINI is known to be limited in correctly sampling protein dynamics, the authors - in order to preserve the fold - have applied a strong (500 kJ mol-1 nm-2) elastic network. However, I am wondering how the ENM applies across the dimer and if any asymmetry can be noticed in the application of restraints for each monomer and at the dimer interface. How can this have potentially biased the asymmetry in the scrambling rates observed between the monomers? Is this artificially obtained from restraining the initial structure, or is the asymmetry somehow gatekeeping the scrambling mechanism to occur majorly across a single monomer? Answering this question would have far-reaching implications to better describe the mechanism of scrambling.

      Notably, the manuscript does not explore the impact of membrane composition on scrambling rates. While the authors use a specific lipid composition (DOPC) in their simulations, they acknowledge that membrane composition can influence scrambling activity. However, the study does not explore how different lipids or membrane environments or varying membrane curvature and tension, could alter scrambling behaviour. I appreciate that this might have been beyond the scope of this particular paper and the authors plan to further chase these questions, as this work sets a strong protocol for this study. Contextualizing scrambling in the context of membrane composition is particularly relevant since the authors note that TMEM16K's scrambling rate increases tenfold in thinner membranes, suggesting that lipid-specific or membrane-thickness-dependent effects could play a role.

    4. Reviewer #3 (Public review):

      Summary:

      The paper investigates the TMEM16 family of membrane proteins, which play roles in lipid scrambling and ion transport. A total of 27 experimental structures from five TMEM16 family members were analyzed, including mammalian and fungal homologs (e.g., TMEM16A, TMEM16F, TMEM16K, nhTMEM16, afTMEM16). The identified structures were in both Ca²⁺-bound (open) and Ca²⁺-free (closed) states to compare conformations and were preprocessed (e.g., modeling missing loops) and equilibrated. Coarse-grain simulations were performed in DOPC membranes for 10 microseconds to capture the scrambling events. These events were identified by tracking lipids transitioning between the two membrane leaflets and they analysed the correlation between scrambling rates, in addition, structural properties such as groove dilation and membrane thinning were calculated. They report 700 scrambling events across structures and Figure 2 elaborates on how open structures show higher activity, also as expected. The authors also address how structures may require open grooves, this and other mechanisms around scrambling are a bit controversial in the field.

      Strengths:

      The strength of this study emerges from a comparative analysis of multiple structural starting points and understanding global/local motions of the protein with respect to lipid movement. Although the protein is well-studied, both experimentally and computationally, the understanding of conformational events in different family members, especially membrane thickness less compared to fungal scramblases offers good insights.

      Weaknesses:

      The weakness of the work is to fully reconcile with experimental evidence of Ca²⁺-independent scrambling rates observed in prior studies, but this part is also challenging using coarse-grain molecular simulations. Previous reports have identified lipid crossing, packing defects, and other associated events, so it is difficult to place this paper in that context. However, the absence of validation leaves certain claims, like alternative scrambling pathways, speculative.

    5. Author response:

      Reviewer #1 (Public review):

      Summary:

      The manuscript investigates lipid scrambling mechanisms across TMEM16 family members using coarse-grained molecular dynamics (MD) simulations. While the study presents a statistically rigorous analysis of lipid scrambling events across multiple structures and conformations, several critical issues undermine its novelty, impact, and alignment with experimental observations.

      Critical issues:

      (1) Lack of Novelty:

      The phenomenon of lipid scrambling via an open hydrophilic groove is already well-established in the literature, including through atomistic MD simulations. The authors themselves acknowledge this fact in their introduction and discussion. By employing coarse-grained simulations, the study essentially reiterates previously known findings with limited additional mechanistic insight. The repeated observation of scrambling occurring predominantly via the groove does not offer significant advancement beyond prior work.

      We agree with the reviewer’s statement regarding the lack of novelty when it comes to our observations of scrambling in the groove of open Ca<sup>2+</sup>-bound TMEM16 structures. However, we feel that the inclusion of closed structures in this study, which attempts to address the yet unanswered question of how scrambling by TMEM16s occurs in the absence of Ca<sup>2+</sup>, offers new observations for the field. In our study we specifically address to what extent the induced membrane deformation, which has been theorized to aid lipids cross the bilayer especially in the absence of Ca<sup>2+</sup>, contributes to the rate of scrambling (see references 36, 59, and 66). There are also several TMEM16F structures solved under activating conditions (bound to Ca<sup>2+</sup> and in the presence of PIP2) which feature structural rearrangements to TM6 that may be indicative of an open state (PDB 6P48) and had not been tested in simulations. We show that these structures do not scramble and thereby present evidence against an out-of-the-groove scrambling mechanism for these states. Although we find a handful of examples of lipids being scrambled by Ca<sup>2+</sup>-free structures of TMEM16 scramblases, none of our simulations suggest that these events are related to the degree of deformation.

      (2) Redundancy Across Systems:

      The manuscript explores multiple TMEM16 family members in activating and non-activating conformations, but the conclusions remain largely confirmatory. The extensive dataset generated through coarse-grained MD simulations primarily reinforces established mechanistic models rather than uncovering fundamentally new insights. The effort, while statistically robust, feels excessive given the incremental nature of the findings.

      Again, we agree with the reviewer’s statement that our results largely confirm those published by other groups and our own. We think there is however value in comparing the scrambling competence of these TMEM16 structures in a consistent manner in a single study to reduce inconsistencies that may be introduced by different simulation methods, parameters, environmental variables such as lipid composition as used in other published works of single family members. The consistency across our simulations and high number of observed scrambling events have allowed us to confirm that the mechanism of scrambling is shared by multiple family members and relies most obviously on groove dilation.

      (3) Discrepancy with Experimental Observations:

      The use of coarse-grained simulations introduces inherent limitations in accurately representing lipid scrambling dynamics at the atomistic level. Experimental studies have highlighted nuances in lipid permeation that are not fully captured by coarse-grained models. This discrepancy raises questions about the biological relevance of the reported scrambling events, especially those occurring outside the canonical groove.

      We thank the reviewer for bringing up the possible inaccuracies introduced by coarse graining our simulations. This is also a concern for us, and we address this issue extensively in our discussion. As the reviewer pointed out above, our CG simulations have largely confirmed existing evidence in the field which we think speaks well to the transferability of observations from atomistic simulations to the coarse-grained level of detail. We have made both qualitative and quantitative comparisons between atomistic and coarse-grained simulations of nhTMEM16 and TMEM16F (Figure 1, Figure 4-figure supplement 1, Figure 4-figure supplement 5) showing the two methods give similar answers for where lipids interact with the protein, including outside of the canonical groove. We do not dispute the possible discrepancy between our simulations and experiment, but our goal is to share new nuanced ideas for the predicted TMEM16 scrambling mechanism that we hope will be tested by future experimental studies.

      (4) Alternative Scrambling Sites:

      The manuscript reports scrambling events at the dimer-dimer interface as a novel mechanism. While this observation is intriguing, it is not explored in sufficient detail to establish its functional significance. Furthermore, the low frequency of these events (relative to groove-mediated scrambling) suggests they may be artifacts of the simulation model rather than biologically meaningful pathways.

      We agree with the reviewer that our observed number of scrambling events in the dimer interface is too low to present it as strong evidence for it being the alternative mechanism for Ca<sup>2+</sup>-independent scrambling. This will require additional experiments and computational studies which we plan to do in future research. However, we are less certain that these are artifacts of the coarse-grained simulation system as we observed a similar event in an atomistic simulation of TMEM16F.

      Conclusion:

      Overall, while the study is technically sound and presents a large dataset of lipid scrambling events across multiple TMEM16 structures, it falls short in terms of novelty and mechanistic advancement. The findings are largely confirmatory and do not bridge the gap between coarse-grained simulations and experimental observations. Future efforts should focus on resolving these limitations, possibly through atomistic simulations or experimental validation of the alternative scrambling pathways.

      Reviewer #2 (Public review):

      Summary:

      Stephens et al. present a comprehensive study of TMEM16-members via coarse-grained MD simulations (CGMD). They particularly focus on the scramblase ability of these proteins and aim to characterize the "energetics of scrambling". Through their simulations, the authors interestingly relate protein conformational states to the membrane's thickness and link those to the scrambling ability of TMEM members, measured as the trespassing tendency of lipids across leaflets. They validate their simulation with a direct qualitative comparison with Cryo-EM maps.

      Strengths:

      The study demonstrates an efficient use of CGMD simulations to explore lipid scrambling across various TMEM16 family members. By leveraging this approach, the authors are able to bypass some of the sampling limitations inherent in all-atom simulations, providing a more comprehensive and high-throughput analysis of lipid scrambling. Their comparison of different protein conformations, including open and closed groove states, presents a detailed exploration of how structural features influence scrambling activity, adding significant value to the field. A key contribution of this study is the finding that groove dilation plays a central role in lipid scrambling. The authors observe that for scrambling-competent TMEM16 structures, there is substantial membrane thinning and groove widening. The open Ca<sup>2+</sup>-bound nhTMEM16 structure (PDB ID 4WIS) was identified as the fastest scrambler in their simulations, with scrambling rates as high as 24.4 {plus minus} 5.2 events per μs. This structure also shows significant membrane thinning (up to 18 Å), which supports the hypothesis that groove dilation lowers the energetic barrier for lipid translocation, facilitating scrambling.

      The study also establishes a correlation between structural features and scrambling competence, though analyses often lack statistical robustness and quantitative comparisons. The simulations differentiate between open and closed conformations of TMEM16 structures, with open-groove structures exhibiting increased scrambling activity, while closed-groove structures do not. This finding aligns with previous research suggesting that the structural dynamics of the groove are critical for scrambling. Furthermore, the authors explore how the physical dimensions of the groove qualitatively correlate with observed scrambling rates. For example, TMEM16K induces increased membrane thinning in its open form, suggesting that membrane properties, along with structural features, play a role in modulating scrambling activity.

      Another significant finding is the concept of "out-of-the-groove" scrambling, where lipid translocation occurs outside the protein's groove. This observation introduces the possibility of alternate scrambling mechanisms that do not follow the traditional "credit-card model" of groove-mediated lipid scrambling. In their simulations, the authors note that these out-of-the-groove events predominantly occur at the dimer interface between TM3 and TM10, especially in mammalian TMEM16 structures. While these events were not observed in fungal TMEM16s, they may provide insight into Ca<sup>2+</sup>-independent scrambling mechanisms, as they do not require groove opening.

      Weaknesses:

      A significant challenge of the study is the discrepancy between the scrambling rates observed in CGMD simulations and those reported experimentally. Despite the authors' claim that the rates are in line experimentally, the observed differences can mean large energetic discrepancies in describing scrambling (larger than 1kT barrier in reality). For instance, the authors report scrambling rates of 10.7 events per μs for TMEM16F and 24.4 events per μs for nhTMEM16, which are several orders of magnitude faster than experimental rates. While the authors suggest that this discrepancy could be due to the Martini 3 force field's faster diffusion dynamics, this explanation does not fully account for the large difference in rates. A more thorough discussion on how the choice of force field and simulation parameters influence the results, and how these discrepancies can be reconciled with experimental data, would strengthen the conclusions. Likewise, rate calculations in the study are based on 10 μs simulations, while experimental scrambling rates occur over seconds. This timescale discrepancy limits the study's accuracy, as the simulations may not capture rare or slow scrambling events that are observed experimentally and therefore might underestimate the kinetics of scrambling. It's however important to recognize that it's hard (borderline unachievable) to pinpoint reasonable kinetics for systems like this using the currently available computational power and force field accuracy. The faster diffusion in simulations may lead to overestimated scrambling rates, making the simulation results less comparable to real-world observations. Thus, I would therefore read the findings qualitatively rather than quantitatively. An interesting observation is the asymmetry observed in the scrambling rates of the two monomers. Since MARTINI is known to be limited in correctly sampling protein dynamics, the authors - in order to preserve the fold - have applied a strong (500 kJ mol-1 nm-2) elastic network. However, I am wondering how the ENM applies across the dimer and if any asymmetry can be noticed in the application of restraints for each monomer and at the dimer interface. How can this have potentially biased the asymmetry in the scrambling rates observed between the monomers? Is this artificially obtained from restraining the initial structure, or is the asymmetry somehow gatekeeping the scrambling mechanism to occur majorly across a single monomer? Answering this question would have far-reaching implications to better describe the mechanism of scrambling.

      The main aim of our computational survey was to directly compare all relevant published TMEM16 structures in both open and closed states using the Martini 3 CGMD force field. Our standardized simulation and analysis protocol allowed us to quantitatively compare scrambling rates across the TMEM16 family, something that has never been done before. We do acknowledge that direct comparison between simulated versus experimental scrambling rates is complicated and is best to be interpreted qualitatively. In line with other reports (e.g., Li et al, PNAS 2024), lipid scrambling in CGMD is 2-3 orders of magnitude faster than typical experimental findings. In the CG simulation field, these increased dynamics due to the smoother energy landscape are a well known phenomenon. In our view, this is a valuable trade-off for being able to capture statistically robust scrambling dynamics and gain mechanistic understanding in the first place, since these are currently challenging to obtain otherwise. For example, with all-atom MD it would have been near-impossible to conclude that groove openness and high scrambling rates are closely related, simply because one would only measure a handful of scrambling events in (at most) a handful of structures.

      Considering the elastic network: the reviewer is correct in that the elastic network restrains the overall structure to the experimental conformation. This is necessary because the Martini 3 force field does not accurately model changes in secondary (and tertiary) structure. In fact, by retaining the structural information from the experimental structures, we argue that the elastic network helped us arrive at the conclusion that groove openness is the major contributing factor in determining a protein’s scrambling rate. This is best exemplified by the asymmetric X-ray structure of TMEM16K (5OC9), in which the groove of one subunit is more dilated than the other. In our simulation, this information was stored in the elastic network, yielding a 4x higher rate in the open groove than in the closed groove, within the same trajectory.

      Notably, the manuscript does not explore the impact of membrane composition on scrambling rates. While the authors use a specific lipid composition (DOPC) in their simulations, they acknowledge that membrane composition can influence scrambling activity. However, the study does not explore how different lipids or membrane environments or varying membrane curvature and tension, could alter scrambling behaviour. I appreciate that this might have been beyond the scope of this particular paper and the authors plan to further chase these questions, as this work sets a strong protocol for this study. Contextualizing scrambling in the context of membrane composition is particularly relevant since the authors note that TMEM16K's scrambling rate increases tenfold in thinner membranes, suggesting that lipid-specific or membrane-thickness-dependent effects could play a role.

      Considering different membrane compositions: for this study, we chose to keep the membranes as simple as possible. We opted for pure DOPC membranes, because it has (1) negligible intrinsic curvature, (2) forms fluid membranes, and (3) was used previously by others (Li et al, PNAS 2024). As mentioned by the reviewer, we believe our current study defines a good standardized protocol and solid baseline for future efforts looking into the additional effects of membrane composition, tension, and curvature that could all affect TMEM16-mediated lipid scrambling.

      Reviewer #3 (Public review):

      Strengths:

      The strength of this study emerges from a comparative analysis of multiple structural starting points and understanding global/local motions of the protein with respect to lipid movement. Although the protein is well-studied, both experimentally and computationally, the understanding of conformational events in different family members, especially membrane thickness less compared to fungal scramblases offers good insights.

      We appreciate the reviewer recognizing the value of the comparative study. In addition to valuable insights from previous experimental and computational work, we hope to put forward a unifying framework that highlights various TMEM16 structural features and membrane properties that underlie scrambling function.

      Weaknesses:

      The weakness of the work is to fully reconcile with experimental evidence of Ca²⁺-independent scrambling rates observed in prior studies, but this part is also challenging using coarse-grain molecular simulations. Previous reports have identified lipid crossing, packing defects, and other associated events, so it is difficult to place this paper in that context. However, the absence of validation leaves certain claims, like alternative scrambling pathways, speculative.

      It is generally difficult to quantitatively compare bulk measurements of scrambling phenomena with simulation results. The advantage of simulations is to directly observe the transient scrambling events at a spatial and temporal resolution that is currently unattainable for experiments. The current experimental evidence for the precise mechanism of Ca<sup>2+</sup>-independent scrambling is still under debate. We therefore hope to leverage the strength of MD and statistical rigor of coarse-grained simulations to generate testable hypotheses for further structural, biochemical, and computational studies.

    1. eLife Assessment

      This study presents valuable data on the increase in individual differences in functional connectivity with the auditory cortex in individuals with congenital/early-onset hearing loss compared to individuals with normal hearing. The evidence supporting the study's claims is convincing, although additional work using resting-state functional connectivity in the future could further strengthen the results. The work will be of interest to neuroscientists working on brain plasticity and may have implications for the design of interventions and compensatory strategies.

    2. Reviewer #1 (Public review):

      This experiment sought to determine what effect congenital/early-onset hearing loss (and associated delay in language onset) has on the degree of inter-individual variability in functional connectivity to the auditory cortex. Looking at differences in variability rather than group differences in mean connectivity itself represents an interesting addition to the existing literature. The sample of deaf individuals was large, and quite homogeneous in terms of age of hearing loss onset, which are considerable strengths of the work. The experiment appears well conducted and the results are certainly of interest.

      Comment from Reviewing Editor: In the revised manuscript, the authors have addressed all concerns previously identified by reviewer 1.

    3. Reviewer #3 (Public review):

      Summary:

      This study focuses on changes in brain organization associated with congenital deafness. The authors investigate differences in functional connectivity (FC) and differences in the variability of FC. By comparing congenitally deaf individuals to individuals with normal hearing, and by further separating congenitally deaf individuals into groups of early and late signers, the authors can distinguish between changes in FC due to auditory deprivation and changes in FC due to late language acquisition. They find larger FC variability in deaf than normal-hearing individuals in temporal, frontal, parietal, and midline brain structures, and that FC variability is largely driven by auditory deprivation. They suggest that the regions that show a greater FC difference between groups also show greater FC variability.

      Strengths:

      The manuscript is well-written, and the methods are clearly described and appropriate. Including the three different groups enables the critical contrasts distinguishing between different causes of FC variability changes. The results are interesting and novel.

      Weaknesses:

      Analyses were conducted for task-based data rather than resting-state data. The authors report behavioral differences between groups and include behavioral performance as a nuisance regressor in their analysis. This is a good approach to account for behavioral task differences, given the data. Nevertheless, additional work using resting-state functional connectivity could remove the potential confound fully.

      Comment from Reviewing Editor: In the revised manuscript, the authors have addressed all concerns previously identified by reviewer 3, and the eLife assessment statement reflects the point by reviewer 3 that using resting-state functional connectivity in the future could further strengthen the results.

    4. Author response:

      The following is the authors’ response to the previous reviews.

      Public Reviews: 

      Reviewer #1 (Public review):

      This experiment sought to determine what effect congenital/early-onset hearing loss (and associated delay in language onset) has on the degree of inter-individual variability in functional connectivity to the auditory cortex. Looking at differences in variability rather than group differences in mean connectivity itself represents an interesting addition to the existing literature. The sample of deaf individuals was large, and quite homogeneous in terms of age of hearing loss onset, which are considerable strengths of the work. The experiment appears well conducted and the results are certainly of interest. R: Thank you for your positive and thoughtful feedback.

      Reviewer #3 (Public review):

      Summary:

      This study focuses on changes in brain organization associated with congenital deafness. The authors investigate differences in functional connectivity (FC) and differences in the variability of FC. By comparing congenitally deaf individuals to individuals with normal hearing, and by further separating congenitally deaf individuals into groups of early and late signers, the authors can distinguish between changes in FC due to auditory deprivation and changes in FC due to late language acquisition. They find larger FC variability in deaf than normal-hearing individuals in temporal, frontal, parietal, and midline brain structures, and that FC variability is largely driven by auditory deprivation. They suggest that the regions that show a greater FC difference between groups also show greater FC variability.

      Strengths:

      The manuscript is well-written, and the methods are clearly described and appropriate. Including the three different groups enables the critical contrasts distinguishing between different causes of FC variability changes. The results are interesting and novel.

      Weaknesses:

      Analyses were conducted for task-based data rather than resting-state data. The authors report behavioral differences between groups and include behavioral performance as a nuisance regressor in their analysis. This is a good approach to account for behavioral task differences, given the data. Nevertheless, additional work using resting-state functional connectivity could remove the potential confound fully.

      The authors have addressed my concerns well.

      Thank you for your thoughtful feedback. We appreciate your acknowledgment of the strengths of our study and the approaches taken to address potential confounds. As noted, we discuss the limitation of not including resting-state data in the manuscript, and we agree that this represents an important avenue for future research. We hope to address this question in future studies.

    1. eLife Assessment

      This fundamental study provides a critical challenge to a great many studies of the neural correlates of consciousness that were based on post hoc sorting of reported awareness experience. The evidence supporting this criticism is compelling, based on simulations and decoding analysis of EEG data. The results will be of interest not only to psychologists and neuroscientists but also to philosophers who work on addressing mind-body relationships.

    2. Reviewer #1 (Public review):

      The study aimed to investigate the significant impact of criterion placement on the validity of neural measures of consciousness, examining how different standards for classifying a stimulus as 'seen' or 'unseen' can influence the interpretation of neural data. They conducted simulations and EEG experiments to demonstrate that the Perceptual Awareness Scale, a widely used tool in consciousness research, may not effectively mitigate criterion-related confounds, suggesting that even with the PAS, neural measures can be compromised by how criteria are set. Their study challenged existing paradigms by showing that the construct validity of neural measures of conscious and unconscious processing is threatened by criterion placement, and they provided practical recommendations for improving experimental designs in the field. The authors' work contributes to a deeper understanding of the nature of conscious and unconscious processing and addresses methodological concerns by exploring the pervasive influence of criterion placement on neural measures of consciousness and discussing alternative paradigms that might offer solutions to the criterion problem.

      The study effectively demonstrates that the placement of criteria for determining whether a stimulus is 'seen' or 'unseen' significantly impacts the validity of neural measures of consciousness. The authors found that conservative criteria tend to inflate effect sizes, while liberal criteria reduce them, leading to potentially misleading conclusions about conscious and unconscious processing. The authors employed robust simulations and EEG experiments to demonstrate the effects of criterion placement, ensuring that the findings are well-supported by empirical evidence. The results from both experiments confirm the predicted confounding effects of criterion placement on neural measures of unconscious and conscious processing.

      The results are consistent with their hypotheses and contribute meaningfully to the field of consciousness research.

    3. Reviewer #2 (Public review):

      Summary:

      The study investigates the potential influence of the response criterion on neural decoding accuracy in consciousness and unconsciousness, utilizing either simulated data or reanalyzing experimental data with post-hoc sorting data.

      Strengths:

      When comparing the neural decoding performance of Target versus NonTarget with or without post-hoc sorting based on subject reports, it is evident that response criterion can influence the results. This was observed in simulated data as well as in two experiments that manipulated subject response criterion to be either more liberal or more conservative. One experiment involved a two-level response (seen vs unseen), while the other included a more detailed four-level response (ranging from 0 for no experience to 3 for a clear experience). The findings consistently indicated that adopting a more conservative response criterion could enhance neural decoding performance, whether in conscious or unconscious states, depending on the sensitivity or overall response threshold.

      Weaknesses:

      (1) In the realm of research methodology, conducting post-hoc sorting based on subject reports raises an issue. This operation leads to an imbalance in the number of trials between the two conditions (Target and NonTarget) during the decoding process. Such trial number disparity introduces bias during decoding, likely contributing to fluctuations in neural decoding performance. This potential confounding factor significantly impacts the interpretation of research findings. The trial number imbalance may cause models to exhibit a bias towards the category with more trials during the learning process, leading to misjudgments of neural signal differences between the two conditions and failing to accurately reflect the distinctions in brain neural activity between target and non-target states. Therefore, it is recommended that the authors extensively discuss this confounding factor in their paper. They should analyze in detail how this factor could influence the interpretation of results, such as potentially exaggerating or diminishing certain effects, and whether measures are necessary to correct the bias induced by this imbalance to ensure the reliability and validity of the research conclusions.

    4. Reviewer #3 (Public review):

      Summary:

      Fahrenfort et al. investigate how liberal or conservative criterion placement in a detection task affects the construct validity of neural measures of unconscious cognition and conscious processing. Participants identified instances of "seen" or "unseen" in a detection task, a method known as post hoc sorting. Simulation data convincingly demonstrate that, counterintuitively, a conservative criterion inflates effect sizes of neural measures compared to a liberal criterion. While the impact of criterion shifts on effect size is suggested by signal detection theory, this study is the first to address this explicitly within the consciousness literature. Decoding analysis of data from two EEG experiments further shows that different criteria lead to differential effects on classifier performance in post hoc sorting. The findings underscore the pervasive influence of experimental design and participant reports on neural measures of consciousness, revealing that criterion placement poses a critical challenge for researchers.

      Strengths and Weaknesses

      One of the strengths of this study is the inclusion of the Perceptual Awareness Scale (PAS), which allows participants to provide more nuanced responses regarding their perceptual experiences. This approach ensures that responses at the lowest awareness level (selection 0) are made only when trials are genuinely unseen. This methodological choice is important as it helps prevent the overestimation of unconscious processing, enhancing the validity of the findings.<br /> The authors also do a commendable job in the discussion by addressing alternative paradigms, such as wagering paradigms, as a possible remedy to the criterion problem (Peters & Lau, 2015; Dienes & Seth, 2010). Their consideration of these alternatives provides a balanced view and strengthens the overall discussion.

      Our initial review identified a lack of measures of variance as one potential weakness of this work. However we agree with the authors' response that plotting individual datapoints for each condition is indeed a good visualization of variance within a dataset.

      Impact of the Work:

      This study effectively demonstrates a phenomenon that, while understood within the context of signal detection theory, has been largely unexplored within the consciousness literature. Subjective measures may not reliably capture the construct they aim to measure due to criterion confounds. Future research on neural measures of consciousness should account for this issue, and no-report measures may be necessary until the criterion problem is resolved.

    5. Author response:

      The following is the authors’ response to the original reviews.

      Reviewer #1 (Public review):

      Summary:

      The paper proposes that the placement of criteria for determining whether a stimulus is 'seen' or 'unseen' can significantly impact the validity of neural measures of consciousness. The authors found that conservative criteria, which require stronger evidence to classify a stimulus as 'seen,' tend to inflate effect sizes in neural measures, making conscious processing appear more pronounced than it is. Conversely, liberal criteria, which require less evidence, reduce these effect sizes, potentially underestimating conscious processing. This variability in effect sizes due to criterion placement can lead to misleading conclusions about the nature of conscious and unconscious processing.

      Furthermore, the study highlights that the Perceptual Awareness Scale (PAS), a commonly used tool in consciousness research, does not effectively mitigate these criterion-related confounds. This means that even with PAS, the validity of neural measures can still be compromised by how criteria are set. The authors emphasize the need for careful consideration and standardization of criterion placement in experimental designs to ensure that neural measures accurately reflect the underlying cognitive processes. By addressing this issue, the paper aims to improve the reliability and validity of findings in the field of consciousness research.

      Strengths:

      (1) This research provides a fresh perspective on how criterion placement can significantly impact the validity of neural measures in consciousness research.

      (2) The study employs robust simulations and EEG experiments to demonstrate the effects of criterion placement, ensuring that the findings are well-supported by empirical evidence.

      (3) By highlighting the limitations of the PAS and the impact of criterion placement, the study offers practical recommendations for improving experimental designs in consciousness research.

      Weaknesses:

      The primary focused criterion of PAS is a commonly used tool, but there are other measures of consciousness that were not evaluated, which might also be subject to similar or different criterion limitations. A simulation could applied to these metrics to show how generalizable the conclusion of the study is.

      We would like to thank reviewer 1 for their positive words and for taking the time to evaluate our manuscript. We agree that it would be important to gauge generalization to other metrics of consciousness. Note however, that the most commonly used alternative methods are postdecision wagering and confidence, both of which are known to behave quite similarly to the PAS (Sandberg, Timmermans , Overgaard & Cleeremans, 2010). Indeed, we have confirmed in other work that confidence is also sensitive to criterion shifts (see https://osf.io/preprints/psyarxiv/xa4fj). Although it has been claimed that confidence-derived aggregate metrics like meta-d’ or metacognitive efficiency may overcome criterion shifts, it would require empirical data rather than simulation to settle whether this is true or not (also see the discussion in https://osf.io/preprints/psyarxiv/xa4fj). Furthermore, out of these metrics, the PAS seems to be the preferred one amongst consciouness researchers (see figure 4 in Francken, Beerendonk, Molenaar, Fahrenfort, Kiverstein, Seth, Gaal S van, 2022; as well as https://osf.io/preprints/psyarxiv/bkxzh). Thus, given the fact that other metrics are either expected to behave in similar ways and/or because it would require more empirical work to determine along which dimension(s) criterion shifts would operate in alternative metrics, we see no clear path to implement the suggested simulations. We anticipate that aiming to do this would require a considerable amount of additional work, figuring out many things which we believe would better suit a future project. We would of course be open to doing this if the reviewer would have more specific suggestions for how to go about the proposed simulations.

      Reviewer #2 (Public review):

      Summary:

      The study investigates the potential influence of the response criterion on neural decoding accuracy in consciousness and unconsciousness, utilizing either simulated data or reanalyzing experimental data with post-hoc sorting data.

      Strengths:

      When comparing the neural decoding performance of Target versus NonTarget with or without post-hoc sorting based on subject reports, it is evident that response criterion can influence the results. This was observed in simulated data as well as in two experiments that manipulated the subject response criterion to be either more liberal or more conservative. One experiment involved a two-level response (seen vs unseen), while the other included a more detailed four-level response (ranging from 0 for no experience to 3 for a clear experience). The findings consistently indicated that adopting a more conservative response criterion could enhance neural decoding performance, whether in conscious or unconscious states, depending on the sensitivity or overall response threshold.

      Weaknesses:

      (1) The response criterion plays a crucial role in influencing neural decoding because a subject's report may not always align with the actual stimulus presented. This discrepancy can occur in cases of false alarms, where a subject reports seeing a target that was not actually there, or in cases where a target is present but not reported. Some may argue that only using data from consistent trials (those with correct responses) would not be affected by the response criterion. However, the authors' analysis suggests that a conservative response criterion not only reduces false alarms but also impacts hit rates. It is important for the authors to further investigate how the response criterion affects neural decoding even when considering only correct trials.

      We would like to thank reviewer 2 for taking the time to evaluate our manuscript. We appreciate the suggestion to investigate neural decoding on only correct trials. What we in fact did is consider target trials that are 'correct' (hits = seen target present trials) and 'incorrect' (misses = unseen target present trials) separately, see figure 4A and figure 4B. This shows that the response criterion also affects the neural measure of consciousness when only considering correct target present trials. Note however, that one cannot decode 'unseen' (target present) trials if one only aims to decode 'correct' trials, because those are all incorrect by definition. We did not analyze false alarms (these would be the 'seen' trials on the noise distribution of Figure 1A), as there were not enough trials of those, especially in the conservative condition (see Figure 2C and 2D), making comparisons between conservative and liberal impossible. However, the predictions for false alarms are pretty straightforward, and follow directly from the framework in Figure 1.

      (2) The author has utilized decoding target vs. nontarget as the neural measures of unconscious and/or conscious processing. However, it is important to note that this is just one of the many neural measures used in the field. There are an increasing number of studies that focus on decoding the conscious content, such as target location or target category. If the author were to include results on decoding target orientation and how it may be influenced by response criterion, the field would greatly benefit from this paper.

      We thank the reviewer for the suggestion to decode orientation of the target. In our experiments, the target itself does not have an orientation, but the texture of which it is composed does. We used four orientations, which were balanced out within and across conditions such that presence-absence decoding is never driven by orientation, but rather by texture based figure-ground segregation (for similar logic, see for example Fahrenfort et al, 2007; 2008 etc). There are a couple of things to consider when wanting to apply a decoding analysis on the orientation of these textures:

      (1) Our behavioral task was only on the presence or absence of the target, not on the orientation of the textures. This makes it impossible to draw any conclusions about the visibility of the orientation of the textures. Put differently: based on behavior there is no way of identifying seen or unseen orientations, correctly or incorrectly identified orientations etc. For examply, it is easy to envision that an observer detects a target without knowing the orientation that defines it, or vice versa a situation in which an observer does not detect the target while still being aware of the orientation of a texture in the image (either of the figure, or of the background). The fact that we have no behavioral response to the orientation of the textures severely limits the usefulness of a hypothetical decoding effect on these orientations, as such results would be uninterpretable with respect to the relevant dimension in this experiment, which is visibility.

      (2) This problem is further excarbated by the fact that the orientation of the background is always orthogonal to the orientation of the target. Therefore, one would not only be decoding the orientation of the texture that constitutes the target itself, but also the texture that constitutes the background. Given that we also have no behavioral metric of how/whether the orientation of the background is perceived, it is similarly unclear how one would interpret any observed effect.

      (3) Finally, it is important to note that – even when categorization/content is sometimes used as an auxiliary measure in consciousness research (often as a way to assay objective performance) - consciousness is most commonly conceptualized on the presence-absence dimension. A clear illustration of this is the concept of blindsight. Blindsight is the ability of observers to discriminate stimuli (i.e. identify content) without being able to detect them. Blindsight is often considered the bedrock of the cognitive neuroscience of consciousness as it acts as proof that one can dissociate between unconscious processing (the categorization of a stimulus, i.e. the content) and conscious processing of that stimulus (i.e. the ability to detect it).

      Given the above, we do not see how the suggested analysis could contribute to the conclusions that the manuscript already establishes. We hope that – given the above - the reviewer agrees with this assessment.

      Reviewer #3 (Public review):

      Summary:

      Fahrenfort et al. investigate how liberal or conservative criterion placement in a detection task affects the construct validity of neural measures of unconscious cognition and conscious processing. Participants identified instances of "seen" or "unseen" in a detection task, a method known as post hoc sorting. Simulation data convincingly demonstrate that, counterintuitively, a conservative criterion inflates effect sizes of neural measures compared to a liberal criterion. While the impact of criterion shifts on effect size is suggested by signal detection theory, this study is the first to address this explicitly within the consciousness literature. Decoding analysis of data from two EEG experiments further shows that different criteria lead to differential effects on classifier performance in post hoc sorting. The findings underscore the pervasive influence of experimental design and participants report on neural measures of consciousness, revealing that criterion placement poses a critical challenge for researchers.

      Strengths and Weaknesses:

      One of the strengths of this study is the inclusion of the Perceptual Awareness Scale (PAS), which allows participants to provide more nuanced responses regarding their perceptual experiences. This approach ensures that responses at the lowest awareness level (selection 0) are made only when trials are genuinely unseen. This methodological choice is important as it helps prevent the overestimation of unconscious processing, enhancing the validity of the findings.

      A potential area for improvement in this study is the use of single time-points from peak decoding accuracy to generate current source density topography maps. While we recognize that the decoding analysis employed here differs from traditional ERP approaches, the robustness of the findings could be enhanced by exploring current source density over relevant time windows. Event-related peaks, both in terms of timing and amplitude, can sometimes be influenced by noise or variability in trial-averaged EEG data, and a time-window analysis might provide a more comprehensive and stable representation of the underlying neural dynamics.

      We thank reviewer 3 for their positive words and for taking the time to evaluate our manuscript. If we understand the reviewer correctly, he/she suggests that the signal-to-noise ratio could be improved by averaging over time windows rather than taking the values at singular peaks in time. Before addressing this suggestion, we would like to point out that we plotted the relevant effects across time in Supplementary Figure S1A and S1B. These show that the observed effects were not somehow limited in time, i.e. only occuring around the peaks, but that they consistenly occured throughout the time course of the trial. In line with this observation one might argue that the results could be improved further by averaging across windows of interest rather than taking the peak moments alone, as the reviewer suggests. Although this might be true, there are many analysis choices that one can make, each of which could have a positive (or negative) effect on the signal to noise ratio. For example, when taking a window of interest, one is faced with a new choice to make, this time regarding the number of consecutive samples to average across (i.e. the size of the window), etc. More generally there is a long list of choices that may affect the precise outcome of analyses, either positively or negatively. Having analyzed the data in one way, the problem with adding new analysis approaches is that there is no objective criterion for deciding which analysis would be ‘best’, other than looking at the outcome of the statistical analyses themselves. Doing this would constitute an explorative double-dipping-like approach to analyzing the results, which – aside from potentially increasing the signal to noise ratio – is likely to also result in an increase of the type I error rate. In the past, when the first author of this manuscript has attempted to minimize the number of statistical tests, he has lowered the number of EEG time points by simply taking the peaks (for example see https://doi.org/10.1073/pnas.1617268114), and that is the approach that was taken here as well. Given the above, we prefer not to further ‘try out’ additional analytical approaches on this dataset, simply to improve the results. We hope the reviewer sympathizes with our position that it is methodologically most sound to stick to the analyses we have already performed and reported, without further exploration.

      It is helpful that the authors show the standard error of the mean for the classifier performance over time. A similar indication of a measure of variance in other figures could improve clarity and transparency.

      That said, the paper appears solid regarding technical issues overall. The authors also do a commendable job in the discussion by addressing alternative paradigms, such as wagering paradigms, as a possible remedy to the criterion problem (Peters & Lau, 2015; Dienes & Seth, 2010). Their consideration of these alternatives provides a balanced view and strengthens the overall discussion.

      We thank the reviewer for this suggestion. Note that we already have a measure of variance in the other figures too, namely showing the connected data points of individual participants. Indvidual data points as a visualization of variance is preferred by many journals (e.g., see https://www.nature.com/documents/cr-gta.pdf), and also shows the spread of relevant differences when paired points are connected. For example, in Figure 2, 3 and 4, the relevant difference is between the liberal and conservative condition. When wanting to show the spread of the differences between these conditions, one option would be to first subtract the two measures in a pairwise fashion (e.g., liberal-conservative), and then plot the spread of those differences using some metric (e.g. standard error/CI of the mean difference). However, this has the disadvantage of no longer separately showing the raw scores on the conditions that are being compared. Showing conditions separately provides clarity to the reader about what is being compared to what. The most common approach to visualizing the variance of the relevant difference in such cases, is to plot the connected individual data points of all participants in the same plot. The uniformity of the slope of these lines in such a visualization provides direct insight into the spread of the relevant difference. Plotting the standard error of the mean on the raw scores of the conditions in these plots would not help, because this would not visualize the spread of the relevant difference (liberal-conservative). We therefore opted in the manuscript to show the mean scores on the conditions that we compare, while also showing the connected raw data points of individual participants in the same plot. One might argue that we should then use that same visualization in figure 3A, but note that this figure is merely intended to identify the peaks, i.e. it does not compare liberal to conservative. Furthermore, plotting the decoding time lines of individual participants would greatly diminish the clarity of this figure. Given our explanation, we hope the reviewer agrees with the approach that we chose, although we are of course open to modifying the figures if the reviewer has a suggestion for doing so while taking into account the points we raise here in our response.

      Impact of the Work:

      This study effectively demonstrates a phenomenon that has been largely unexplored within the consciousness literature. Subjective measures may not reliably capture the construct they aim to measure due to criterion confounds. Future research on neural measures of consciousness should account for this issue, and no-report measures may be necessary until the criterion problem is resolved.

      Recommendations for the authors:

      Reviewer #2 (Recommendations for the authors):

      The authors could further elaborate on the results of the PAS to provide a clearer insight into the impact of response criteria, which is notably more complex than in other experiments. Specifically, the results demonstrate that conservative response criterion condition displays a considerably higher sensitivity compared to those with a liberal response criterion. It would be interesting to explore whether this shift in sensitivity suggests a correlation between changes in response criteria and conscious experiences, and how the interaction between sensitivity and response criteria can affect the neural measure of consciousness.

      We thank the reviewer for this suggestion. Note that the change in sensitivity that we observed is minor compared to the change we observed in response criterion (hedges g criterion in exp 2 = 2.02, compared to hedges g sensitivity/d’ in exp 2 = 0.42). However, we do investigate the effect of sensitivity (disregarding response criterion) on decoding accuracy. To this end we devised Figure 3C (for the full decoding time course see Supplementary Figure S1B). These figures show that the small behavioral sensitivity effects observed in both experiments (hedges g sensitivity in exp 1 = 0.30, exp 2 = 0.42) did not translate into significant decoding differences between conservative and liberal in either experiment. This comes as no surprise given the small corresponding behavioral effects. Note that small sensitivity differences between liberal and conservative conditions are commonplace, plausibly driven by the fact that being liberal also involves being more noisy in one’s response tendencies (i.e. sometimes randomly indicating presence). Further, the reviewer suggests that we might correlate changes in response criteria to changes in conscious experience. The only relevant metric of conscious experience for which we have data in this manuscript is the Perceptual Awareness Scale (PAS), so we assume the reviewer asks for a correlation between experimentally induced changes in response criterion with the equivalent changes in d’. To this end we computed the difference in the PAS-based d’ metric between conservative and liberal, as well as the difference in the PAS-based criterion metric between conservative and liberal, and correlated these across subjects (N=26) using a Spearman rank correlation. The result shows that these metrics do not correlate r(24)=0.04, p=0.85. Note however that small-N correlations like these are only somewhat reliable for large effect sizes. An N of 26 and a mere power of 80% requires an effect size of at least r=0.5 to be detectable, so even if a correlation were to exist we may not have had enough power to detect it. Due to these caveats we opted to not report this null-correlation in the manuscript, but we are of course willing to do so if the reviewer and/or editor disagrees with this assessment.

    1. eLife Assessment

      The authors investigated the mechanisms underlying the pause in striatal cholinergic interneurons (SCINs) induced by thalamic input, identifying that Kv1 channels play a key role in this burst-dependent pause. The valuable study provides mechanistic insights into how burst activity in SCINs leads to a subsequent pause, highlighting the involvement of D1/D5 receptors. The experimental evidence is solid; however, the reviewers suggest further clarifying the mechanism by which clozapine reduces D5R ligand-independent activity in the L-DOPA-off state.

    2. Reviewer #1 (Public review):

      Summary:

      Tubert C. et al. investigated the role of dopamine D5 receptors (D5R) and their downstream potassium channel, Kv1, in the striatal cholinergic neuron pause response induced by thalamic excitatory input. Using slice electrophysiological analysis combined with pharmacological approaches, the authors tested which receptors and channels contribute to the cholinergic interneuron pause response in both control and dyskinetic mice (in the L-DOPA off state). They found that activation of Kv1 was necessary for the pause response, while activation of D5R blocked the pause response in control mice. Furthermore, in the L-DOPA off state of dyskinetic mice, the absence of the pause response was restored by the application of clozapine. The authors claimed that 1) the D5R-Kv1 pathway contributes to the cholinergic interneuron pause response in a phasic dopamine concentration-dependent manner, and 2) clozapine inhibits D5R in the L-DOPA off state, which restores the pause response.

      Strengths

      The electrophysiological and pharmacological approaches used in this study are powerful tools for testing channel properties and functions. The authors' group has well-established these methodologies and analysis pipelines. Indeed, the data presented were robust and reliable.

      Weaknesses:

      Although the paper has strengths in its methodological approaches, there is a significant gap between the presented data and the authors' claims.

      The authors answered the most of concerns I raised. However, the critical issue remains unresolved.

      I am still not convinced by the results presented in Fig. 6 and their interpretation. Since Clozapine acts as an agonist in the absence of an endogenous agonist, it may stimulate the D5R-cAMP-Kv1 pathway. Stimulation of this pathway should abolish the pause response mediated by thalamic stimulation in SCINs, rather than restoring the pause response. Clarification is needed regarding how Clozapine reduces D5R-ligand-independent activity in the absence of dopamine (the endogenous agonist). In addition, the author's argued that D5R antagonist does not work in the absence of dopamine, therefore solely D5R antagonist didn't restore the pause response. However, if D5R-cAMP-Kv1 pathway is already active in L-DOPA off state, why D5R antagonist didn't contribute to inhibition of D5R pathway?<br /> Since Clozapine is not D5 specific and Clozapine experiments were not concrete, I recommend testing whether other receptors, such as the D2 receptor, contribute to the Clozapine-induced pause response in the L-DOPA-off state.

    3. Reviewer #2 (Public review):

      Summary:

      This manuscript by Tubert et al. presents the role of D5 receptors (D5R) in regulating the striatal cholinergic interneuron (CIN) pause response through D5R-cAMP-Kv1 inhibitory signaling. Their findings provide a compelling model explaining the "on/off" switch of the CIN pause, driven by the distinct dopamine affinities of D2R and D5R. This mechanism, coupled with varying dopamine states, is likely critical for modulating synaptic plasticity in cortico-striatal circuits during motor learning and execution. Furthermore, the study bridges their previous finding of CIN hyperexcitability (Paz et al., Movement Disorder 2022) with the loss of the pause response in LID mice and demonstrates the restore of the pause through D1/D5 inverse agonism.

      Strengths:

      The study presents solid findings, and the writing is logically structured and easy to follow. The experiments are well-designed, properly combining ex vivo electrophysiology recording, optogenetics, and pharmacological treatment to dissect / rule out most, if not all, alternative mechanisms in their model.

      Weaknesses:

      While the manuscript is overall satisfying, one conceptual gap needs to be further addressed or discussed: the potential "imbalance" between D2R and D5R signaling due to the ligand-independent activity of D5R in LID. Given that D2R and D5R oppositely regulate CIN pause responses through cAMP signaling, investigating the role of D2R under LID off L-DOPA (e.g., by applying D2 agonists or antagonists, even together with intracellular cAMP analogs or inhibitors) could provide critical insights. Addressing this aspect would strengthen the manuscript in understanding CIN pause loss under pathological conditions.

    4. Reviewer #3 (Public review):

      Summary:

      Tubert et al. investigate the mechanisms underlying the pause response in striatal cholinergic interneurons (SCINs). The authors demonstrate that optogenetic activation of thalamic axons in the striatum induces burst activity in SCINs, followed by a brief pause in firing. They show that the duration of this pause correlates with the number of elicited action potentials, suggesting a burst-dependent pause mechanism. The authors demonstrated this burst-dependent pause relied on Kv1 channels. The pause is blocked by a SKF81297 and partially by sulpiride and mecamylamine, implicating D1/D5 receptor involvement. The study also shows that the ZD7288 does not reduce the duration of the pause, and that lesioning dopamine neurons abolishes this response, which can be restored by clozapine.

      Weaknesses:

      While this study presents an interesting mechanism for SCIN pausing after burst activity, there are several major concerns that should be addressed:

      (1) Scope of the Mechanism: It is important to clarify that the proposed mechanism may apply specifically to the pause in SCINs following burst activity. The manuscript does not provide clear evidence that this mechanism contributes to the pause response observed in behavioral animals. While the thalamus is crucial for SCIN pauses in behavioral contexts, the exact mechanism remains unclear. Activating thalamic input triggers burst activity in SCINs, leading to a subsequent pause, but this mechanism may not be generalizable across different scenarios. For instance, approximately half of TANs do not exhibit initial excitation but still pause during behavior, suggesting that the burst-dependent pause mechanism is unlikely to explain this phenomenon. Furthermore, in behavioral animals, the duration of the pause seems consistent, whereas the proposed mechanism suggests it depends on the prior burst, which is not aligned with in vivo observations. Additionally, many in vivo recordings show that the pause response is a reduction in firing rate, not complete silence, which the mechanism described here does not explain. Please address these in the manuscript.

      (2) Terminology: The use of "pause response" throughout the manuscript is misleading. The pause induced by thalamic input in brain slices is distinct from the pause observed in behavioral animals. Given the lack of a clear link between these two phenomena in the manuscript, it is essential to use more precise terminology throughout, including in the title, bullet points, and body of the manuscript.

      (3) Kv1 Blocker Specificity: It is unclear how the authors ruled out the possibility that the Kv1 blocker did not act directly on SCINs. Could there be an indirect effect contributing to the burst-dependent pause? Clarification on this point would strengthen the interpretation of the results.

      (4) Role of D1 Receptors: While it is well-established that activating thalamic input to SCINs triggers dopamine release, contributing to SCIN pausing (as shown in Figure 3), it would be helpful to assess the extent to which D1 receptors contribute to this burst-dependent pause. This could be achieved by applying the D1 agonist SKF81297 after blocking nAChRs and D2 receptors.

      (5) Clozapine's Mechanism of Action: The restoration of the burst-dependent pause by clozapine following dopamine neuron lesioning is interesting, but clozapine acts on multiple receptors beyond D1 and D5. Although it may be challenging to find a specific D5 antagonist or inverse agonist, it would be more accurate to state that clozapine restores the burst-dependent pause without conclusively attributing this effect to D5 receptors.

      Comments on revisions:

      The authors have addressed many of my concerns. However, I remain unconvinced that adding an 'ex vivo' experiment fully resolves the fundamental differences between the burst-dependent pause observed in slices - defined by the duration of a single AHP - and the pause response in CHINs observed in vivo, which may involve contributions from more than one prolonged AHP. In vivo, neurons can still fire action potentials during the pause, albeit at a lower frequency. Moreover, in behaving animals, pause duration does not vary with or without initial excitation. The mechanism proposed demonstrates that the pause duration, defined by the length of a single AHP, is positively correlated with preceding burst activity.

      To improve clarity, I recommend using the term 'SCIN pause' to describe the ex vivo findings, distinguishing them more explicitly from the 'pause response' observed in behaving animals. This distinction would help contextualize the ex vivo findings as potentially contributing to, but not fully representing, the pause response in vivo.

      Again, it would be helpful to present raw data for pause durations rather than relying solely on ratios. This approach would provide the audience with a clearer understanding of the absolute duration of the burst-dependent pause and allow for better comparison to the ~200 ms pause observed in behaving animals.

    5. Author response:

      The following is the authors’ response to the original reviews.

      Public Reviews:

      Reviewer #1 (Public review):

      Summary:

      Tubert C. et al. investigated the role of dopamine D5 receptors (D5R) and their downstream potassium channel, Kv1, in the striatal cholinergic neuron pause response induced by thalamic excitatory input. Using slice electrophysiological analysis combined with pharmacological approaches, the authors tested which receptors and channels contribute to the cholinergic interneuron pause response in both control and dyskinetic mice (in the LDOPA off state). They found that activation of Kv1 was necessary for the pause response, while activation of D5R blocked the pause response in control mice. Furthermore, in the LDOPA off-state of dyskinetic mice, the absence of the pause response was restored by the application of clozapine. The authors claimed that (1) the D5R-Kv1 pathway contributes to the cholinergic interneuron pause response in a phasic dopamine concentration-dependent manner, and (2) clozapine inhibits D5R in the L-DOPA off state, which restores the pause response.

      Strengths:

      The electrophysiological and pharmacological approaches used in this study are powerful tools for testing channel properties and functions. The authors' group has well-established these methodologies and analysis pipelines. Indeed, the data presented were robust and reliable.

      Thank you for your comments.

      Weaknesses:

      Although the paper has strengths in its methodological approaches, there is a significant gap between the presented data and the authors' claims.

      There was no direct demonstration that the D5R-Kv1 pathway is dominant when dopamine levels are high. The term 'high' is ambiguous, and it raises the question of whether the authors believe that dopamine levels do not reach the threshold required to activate D5R under physiological conditions.

      We acknowledge that further work is necessary to clarify the role of the D5R in physiological conditions. While we haven’t found effects of the D1/D5 receptor antagonist SCH23390 on the pause response in control animals (Fig. 3), it is still possible that dopamine levels reach the threshold to stimulate D5R when burst firing of dopaminergic neurons contributes to dopamine release. We believe the pause response depends, among other factors, on the relative stimulation levels of SCIN D2 and D5 receptors, which is likely not an all-or-nothing phenomenon. To reduce ambiguity, we have eliminated the labels referring to dopamine levels in Figure 6F.

      Furthermore, the data presented in Figure 6 are confusing. If clozapine inhibits active D5R and restores the pause response, the D5R antagonist SCH23390 should have the same effect. The data suggest that clozapine-induced restoration of the pause response might be mediated by other receptors, rather than D5R alone.

      Thank you for letting us clarify this issue. Please note that the levels of endogenous dopamine 24 h after the last L-DOPA challenge in severe parkinsonian mice are expected to be very low. In the absence of an agonist, a pure D1/D5 antagonist would not exert an effect, as demonstrated with SCH23390 alone, which did not have an impact on the SCIN response to thalamic stimulation (Fig. 6). While clozapine can also act as a D1/D5 receptor antagonist, its D1/D5 effects in absence of an agonist are attributed to its inverse agonist properties (PMID: 24931197). Notably, SCH23390 prevented the effect of clozapine, allowing us to conclude that ligand-independent D1/D5 receptor-mediated mechanisms are involved in suppressing the pause response in dyskinetic mice. We now made it clearer in the third paragraph of the Discussion.

      Reviewer #2 (Public review):

      Summary:

      This manuscript by Tubert et al presents the role of the D5 receptor in modulating the striatal cholinergic interneuron (CIN) pause response through D5R-cAMP-Kv1 inhibitory signaling. Their model elucidates the on / off switch of CIN pause, likely due to the different DA affinity between D2R and D5R. This machinery may be crucial in modulating synaptic plasticity in cortical-striatal circuits during motor learning and execution. Furthermore, the study bridges their previous finding of CIN hyperexcitability (Paz et al., Movement Disorder 2022) with the loss of pause response in LID mice.

      Strengths:

      The study had solid findings, and the writing was logically structured and easy to follow. The experiments are well-designed, and they properly combined electrophysiology recording, optogenetics, and pharmacological treatment to dissect/rule out most, if not all, possible mechanisms in their model.

      Thank you for your comments.

      Weaknesses:

      The manuscript is overall satisfying with only some minor concerns that need to be addressed. Manipulation of intracellular cAMP (e.g. using pharmacological analogs or inhibitors) can add additional evidence to strengthen the conclusion.

      Thank you for the suggestion. While we acknowledge that we are not providing direct evidence of the role of cAMP, we chose not to conduct these experiments because cAMP levels influence several intrinsic and synaptic currents beyond Kv1, significantly affecting  membrane oscillations and spontaneous firing, as shown in Paz et al. 2021. However, we are modifying the fourth paragraph of the Discussion so there is no misinterpretation about our findings in the current work.

      Reviewer #3 (Public review):

      Summary:

      Tubert et al. investigate the mechanisms underlying the pause response in striatal cholinergic interneurons (SCINs). The authors demonstrate that optogenetic activation of thalamic axons in the striatum induces burst activity in SCINs, followed by a brief pause in firing. They show that the duration of this pause correlates with the number of elicited action potentials, suggesting a burst-dependent pause mechanism. The authors demonstrated this burst-dependent pause relied on Kv1 channels. The pause is blocked by an SKF81297 and partially by sulpiride and mecamylamine, implicating D1/D5 receptor involvement. The study also shows that the ZD7288 does not reduce the duration of the pause and that lesioning dopamine neurons abolishes this response, which can be restored by clozapine.

      Weaknesses:

      While this study presents an interesting mechanism for SCIN pausing after burst activity, there are several major concerns that should be addressed:

      (1) Scope of the Mechanism:

      It is important to clarify that the proposed mechanism may apply specifically to the pause in SCINs following burst activity. The manuscript does not provide clear evidence that this mechanism contributes to the pause response observed in behavioral animals. While the thalamus is crucial for SCIN pauses in behavioral contexts, the exact mechanism remains unclear. Activating thalamic input triggers burst activity in SCINs, leading to a subsequent pause, but this mechanism may not be generalizable across different scenarios. For instance, approximately half of TANs do not exhibit initial excitation but still pause during behavior, suggesting that the burst-dependent pause mechanism is unlikely to explain this phenomenon. Furthermore, in behavioral animals, the duration of the pause seems consistent, whereas the proposed mechanism suggests it depends on the prior burst, which is not aligned with in vivo observations. Additionally, many in vivo recordings show that the pause response is a reduction in firing rate, not complete silence, which the mechanism described here does not explain. Please address these in the manuscript.

      Thank you for your valuable feedback. While the absence of an initial burst in some TANs in vivo may suggest the involvement of alternative or additional mechanisms, this does not exclude a participation of Kv1 currents. We have seen that subthreshold depolarizations induced by thalamic inputs are sufficient to produce an afterhyperpolarization (AHP) mediated by Kv1 channels (see Tubert et al., 2016, PMID: 27568555). Although such subthreshold depolarizations are not captured in current recordings from behaving animals, intracellular in vivo recordings have demonstrated an intrinsically generated AHP after subthreshold depolarization of SCIN caused by stimulation of excitatory afferents (PMID: 15525771). Additionally, when pause duration is plotted against the number of spikes elicited by thalamic input (Fig. 1G), we found that one elicited spike is followed by an interspike interval 1.4 times longer than the average spontaneous interspike interval. We acknowledge the potential involvement of additional factors, including a decrease of excitatory thalamic input coinciding with the pause, followed by a second volley of thalamic inputs (Fig. 1J-K, after observations by Matsumoto et al., 2001- PMID: 11160526), as well as the timing of elicited spikes relative to ongoing spontaneous firing (Fig. 1D-E). Dopaminergic modulation (Fig. 3) and regional differences among striatal regions (PMID: 24559678) may also contribute to the complexity of these dynamics. 

      (2) Terminology:

      The use of "pause response" throughout the manuscript is misleading. The pause induced by thalamic input in brain slices is distinct from the pause observed in behavioral animals. Given the lack of a clear link between these two phenomena in the manuscript, it is essential to use more precise terminology throughout, including in the title, bullet points, and body of the manuscript.

      While we acknowledge that our study does not include in vivo evidence, we believe ex vivo preparations have been instrumental in elucidating the mechanisms underlying the responses observed in vivo. We also agree with previous ex vivo studies in using consistent terminology. However, we will clarify the ex vivo nature of our work in the abstract and bullet points for greater transparency.

      (3) Kv1 Blocker Specificity:

      It is unclear how the authors ruled out the possibility that the Kv1 blocker did not act directly on SCINs. Could there be an indirect effect contributing to the burst-dependent pause? Clarification on this point would strengthen the interpretation of the results.

      Thank you for letting us clarify this issue. In our previous work (Tubert et al., 2016) we showed that the Kv1.3 and Kv1.1 subunits are selectively expressed in SCIN throughout the striatum. Moreover, gabaergic transmission is blocked in our preparations. We are including a phrase to make it clearer in the manuscript (Results section, subheading “The pause response to thalamic stimulation requires activation of Kv1 channels”).

      (4) Role of D1 Receptors:

      While it is well-established that activating thalamic input to SCINs triggers dopamine release, contributing to SCIN pausing (as shown in Figure 3), it would be helpful to assess the extent to which D1 receptors contribute to this burst-dependent pause. This could be achieved by applying the D1 agonist SKF81297 after blocking nAChRs and D2 receptors.

      Thank you for letting us clarify this point. We show that blocking D2R or nAChR reduces the pause only for strong thalamic stimulation eliciting 4 SCIN spikes (Figure 3G), whereas the D1/D5 agonist SKF81297 is able to reduce the pause induced by weaker stimulation as well (Figure 3C). In addition, the D1/D5 receptor antagonist SCH23390 does not modify the pause response (Figure 3C). This may indicate that nAChR-mediated dopamine release induced by thalamic-induced bursts more efficiently activates D2R compared to D5R. We speculate that, in this context, lack of D5R activation may be necessary to keep normal levels of Kv1.3 currents necessary for SCIN pauses.

      (5) Clozapine's Mechanism of Action:

      The restoration of the burst-dependent pause by clozapine following dopamine neuron lesioning is interesting, but clozapine acts on multiple receptors beyond D1 and D5.

      Although it may be challenging to find a specific D5 antagonist or inverse agonist, it would be more accurate to state that clozapine restores the burst-dependent pause without conclusively attributing this effect to D5 receptors.

      Thank you for your insightful observation. We acknowledge the difficulty of targeting dopamine receptors pharmacologically due to the lack of highly selective D1/D5 inverse agonists. We used SCH23390, which is a highly selective D1/D5 receptor antagonist devoid of inverse agonist effects, to block clozapine’s ability to restore SCIN pauses (Figure 6C). This indicates that the restoration of SCIN pauses by clozapine depends on D1/D5 receptors. Furthermore, in a previous study, we demonstrated that clozapine’s effect on restoring SCIN excitability in dyskinetic mice (a phenomenon mediated by Kv1 channels in SCIN; Tubert et al., 2016) was not due to its action on serotonin receptors (Paz, Stahl et al., 2022). While our data do not rule out the potential contribution of other receptors, such as muscarinic acetylcholine receptors, we believe they strongly support the role of D1/D5 receptors. To reflect this, we added a statement discussing the potential contribution of receptors beyond D1/D5 in the last paragraph of the Discussion.

      Recommendations for the authors:

      Reviewer #1 (Recommendations for the authors):

      (1) The effect of MgTx was not consistent with the previous study (Tubert, 2016). I expected MgTx to increase the basal firing rate of cholinergic interneurons.

      Thank you for highlighting this. In our previous study we used ACSF in the recording pipette, instead of the intracellular solution -higher in potassium- used in the present study. This is likely related to the higher spontaneous firing rates of SCIN observed in the present study, which made the SCIN response stand out. In addition, our previous study analyzed the effect of MgTx on spontaneous firing frequency of SCIN isolated from major circuit regulation by adding CNQX and picrotoxin to the bath, while in this study we needed to preserve the thalamic input and only picrotoxin in the bath was used. Given these differences, the two conditions are not strictly comparable but rather give complementary information.

      (2) In the text, the authors claim that "SCINs recorded in the parkinsonian OFF-L-DOPA condition show an increase in membrane excitability that mimics changes acutely induced by SKF81297 in SCINs from control mice." However, the data for SKF81297 do not support this claim.

      We modified the text to make it clearer that the cited phrase refers to a previous publication (PMID: 35535012) in which SCIN intrinsic excitability was characterized via analysis of responses to somatic current injection in whole-cell recordings. In the present study Fig. 3D shows SKF81297 effects on interspike intervals during spontaneous activity with a trend towards increased firing, and Fig. 4E a lack of effect on “burst duration” for responses with different numbers of spikes elicited by thalamic afferent stimulation. 

      (3) I recommend testing whether other receptors, such as D2R, contribute to the clozapineinduced pause response in the L-DOPA off state.

      Thank you for your suggestion. We acknowledge that studying the role of D2R is important. However, our preliminary data suggest that a comprehensive follow up study, which is beyond the scope of this manuscript, is necessary to understand their role. 

      Reviewer #2 (Recommendations for the authors):

      (1) For Figure 1D-E, I understand that the authors are trying to state that the previous spontaneous spike contributes to a hyperpolarized window that induces a delay in the evoked spikes. However, it is almost impossible to discriminate between spontaneous and evoked spikes in this experiment. Furthermore, considering the tonic firing property, I highly suspect that even a sham control design (no optogenetic light) will give you a similar distribution as in Figure 1E (the longer IN X1, the shorter in X2).

      We agree that some spikes following stimulus onset may have occurred independently of the light stimulus, as it is also possible during behavioral tasks. We used the baseline recordings to estimate the effects of a sham stimulus as requested and included the data in Fig. 1E-F. As expected, the sham stimulation data showed a similar inverse relationship with the time elapsed from the preceding spike, but latencies were longer than with the stimulus (except for times close to the average ISI), suggesting that the optical stimulation increased the probability of evoking a spike (Fig. 1F). Remarkably, the pause following this threshold stimulation was significantly longer than the baseline ISI, as reported in the main text (Results section, last sentence of first paragraph).

      (2) The authors used optogenetics to induce thalamic inputs to induce the pause after bursts. Considering CINs also receive inputs from different brain regions, e.g. cortex, does this phenomena/pause after bursts also exist following cortical inputs?

      We did not study the SCIN response to cortical inputs, but both thalamic and cortical inputs seem to drive SCIN pause-responses as observed by others (PMID: 24553950).  

      (3) The effect of the D5R inverse agonism, and the further combined with D5R agonist and antagonist, faithfully reveal/confirm the increase of ligand-independent activity of D5R in LID reported previously. It would be ideal to also directly modulate intracellular cAMP (as in the 2022 paper) to confirm the rescue effects on the CIN pause response.

      Please, see our response in the public review.

      (4) In healthy conditions, the balance between D2R and D5R signaling (shown in Figure 6F left) switches the pause and no pause modes which potentially contributes to cortical-striatal plasticity. How about in LID off L-DOPA condition? Is it possible to rescue/modulate the pause on/off mode by D2R agonism in LID?

      We haven’t tested the effect of D2 agonists yet, but this is scheduled for follow up studies. 

      Reviewer #3 (Recommendations for the authors):

      (1) The authors use the ratio of pause duration to baseline ISI to describe the pause, which is useful for detecting significant differences. However, it would be beneficial to also report the actual duration of the burst-dependent pause to provide readers with a clearer understanding of the variation in pauses.

      In all figures we report the average baseline ISI duration for each experiment / experimental condition, allowing readers to estimate actual pause durations. We added in the main text actual average pause durations corresponding to Fig. 1H, which are representative of those observed along the study.

      (2) In Figure 3D, a more detailed comparison would be helpful, as there appears to be a significant difference between the SKF81297 group and others.

      We acknowledge that there might be a trend towards reduced ISIs, however, it was statistically non-significant (see legend of figure 3). In addition, the effect of SKF81297 seems unrelated to this trend, as its effect is also seen under the effect of ZD7288, which substantially prolongs the baseline ISI (Fig. 4E-F).

    1. Author response:

      The following is the authors’ response to the current reviews.

      Comments on revisions:

      I thank the authors for addressing my comments.

      - I believe that additional in vivo experiments, or the inclusion of controls for the specificity of the inhibitor, which the authors argue are beyond the scope of the current study, are essential to address the weaknesses and limitations stated in my current evaluation.

      We respectfully acknowledge the reviewer's concern but would like to reiterate that demonstrating the specificity of the inhibitor is beyond the scope of this study. Alpelisib (BYL-719) is a clinically approved drug widely recognized as a specific inhibitor of p110α, primarily used in the treatment of breast cancer. Its selectivity for the p110α isoform has been extensively validated in the literature.

      In our study, we used Alpelisib to assess whether pharmacological inhibition of p110α would produce effects similar to those observed in our genetic model, which is particularly relevant for the potential translational implications of our findings. Given the well-documented specificity of this inhibitor, we believe that additional controls to confirm its selectivity are unnecessary within the context of this study. Instead, our focus has been to investigate the functional role of p110α activity in macrophage-driven inflammation using the models described.

      We appreciate the reviewer’s insight and hope this clarification addresses their concern.

      - While the neutrophil depletion suggests neutrophils are not required for the phenotype, there are multiple other myeloid cells, in addition to macrophages, that could be contributing or accounting for the in vivo phenotype observed in the mutant strain (not macrophage specific).

      We appreciate the reviewer's observation regarding the potential involvement of other myeloid cells. However, it is important to highlight that the inflammatory process follows a well-characterized sequential pattern. Our data clearly demonstrate that in the paw inflammation model:

      ·       Neutrophils are effectively recruited, as evidenced by the inflammatory abscess filled with polymorphonuclear cells.

      ·       However, macrophages fail to be recruited in the RBD model.

      Given that this critical step is disrupted, it is reasonable to expect that any subsequent steps in the inflammatory cascade would also be affected. A precise dissection of the role of other myeloid populations would require additional lineage-specific models to selectively target each subset, which, as we have previously stated, would be the focus of an independent study.

      While we cannot entirely exclude the contribution of other myeloid cells, our data strongly support the conclusion that macrophages are, at the very least, a key component of the observed phenotype. We explicitly address this point in the Discussion section, where we acknowledge the potential involvement of other myeloid populations.

      - Inclusion of absolute cell numbers (in addition to the %) is essential. I do not understand why the authors are not including these data. Have they not counted the cells?

      We appreciate the reviewer’s concern regarding the inclusion of absolute cell numbers. However, as stated in the Materials and Methods section, we analyzed 50,000 cells per sample, and the percentages reported in the manuscript are directly derived from this standardized count.

      Our decision to present the data as percentages follows standard practices in flow cytometry-based analyses, as it allows for a clearer and more biologically relevant comparison of relative changes between conditions. This approach ensures consistency across samples and facilitates the interpretation of population dynamics during inflammation.

      We would also like to clarify that all data are based on actual counts, and rigorous controls were implemented throughout the study to ensure accuracy and reproducibility. We hope this explanation addresses the reviewer’s concern and provides further clarity on our approach.

      - Lastly, inclusion of representatives staining and gating strategies for all immune profiling measurements carried out by flow cytometry is important. This point has not been addressed, not even in writing.

      We appreciate the reviewer’s concern regarding the inclusion of absolute cell numbers. However, as stated in the Materials and Methods section, we analyzed 50,000 cells per sample, and the percentages reported in the manuscript are directly derived from this standardized count.

      Our decision to present the data as percentages follows standard practices in flow cytometry-based analyses, as it allows for a clearer and more biologically relevant comparison of relative changes between conditions. This approach ensures consistency across samples and facilitates the interpretation of population dynamics during inflammation.

      We would also like to clarify that all data are based on actual counts, and rigorous controls were implemented throughout the study to ensure accuracy and reproducibility. We hope this explanation addresses the reviewer’s concern and provides further clarity on our approach.


      The following is the authors’ response to the original reviews.

      Public Reviews:

      Reviewer #1 (Public review):

      This study by Alejandro Rosell et al. reveals the immunoregulatory role of the RAS-p110α pathway in macrophages, specifically in regulating monocyte extravasation and lysosomal digestion during inflammation. Disrupting this pathway, through genetic tools or pharmacological intervention in mice, impairs the inflammatory response, leading to delayed resolution and more severe acute inflammation. The authors suggest that activating p110α with small molecules could be a potential therapeutic strategy for treating chronic inflammation. These findings provide important insights into the mechanisms by which p110α regulates macrophage function and the overall inflammatory response.

      The updates made by the authors in the revised version have addressed the main points raised in the initial review, further improving the strength of their findings.

      Reviewer #2 (Public review):

      Summary:

      Cell intrinsic signaling pathways controlling the function of macrophages in inflammatory processes, including in response to infection, injury or in the resolution of inflammation are incompletely understood. In this study, Rosell et al. investigate the contribution of RAS-p110α signaling to macrophage activity. p110α is a ubiquitously expressed catalytic subunit of PI3K with previously described roles in multiple biological processes including in epithelial cell growth and survival, and carcinogenesis. While previous studies have already suggested a role for RAS-p110α signaling in macrophage function, the cell intrinsic impact of disrupting the interaction between RAS and p110α in this central myeloid cell subset is not known.

      Strengths:

      Exploiting a sound previously described genetically engineered mouse model that allows tamoxifen-inducible disruption of the RAS-p110α pathway and using different readouts of macrophage activity in vitro and in vivo, the authors provide data consistent with their conclusion that alteration in RAS-p110α signaling impairs various but selective aspects of macrophage function in a cell-intrinsic manner.

      Weaknesses:

      My main concern is that for various readouts, the difference between wild-type and mutant macrophages in vitro or between wild-type and Pik3caRBD mice in vivo is modest, even if statistically significant. To further substantiate the extent of macrophage function alteration upon disruption of RAS-p110α signaling and its impact on the initiation and resolution of inflammatory responses, the manuscript would benefit from a more extensive assessment of macrophage activity and inflammatory responses in vivo.

      Thank you for raising this point. We understand the reviewer’s concern regarding the modest yet statistically significant differences observed between wild-type and mutant macrophages in vitro, as well as between wild-type and Pik3ca<sup>RBD</sup> mice in vivo. Our current study aimed to provide a foundational exploration of the role of RAS-p110α signaling in macrophage function and inflammatory response, focusing on a set of core readouts that demonstrate the physiological relevance of this pathway. While a more extensive in vivo assessment could offer additional insights into macrophage activity and the nuanced effects of RAS-p110α disruption, it would require an array of new experiments that are beyond the current scope.

      However, we believe that the current data provide significant insights into the pathway’s role, highlighting important alterations in macrophage function and inflammatory processes due to RAS-p110α disruption. These findings lay the groundwork for future studies that can build upon our results with a more comprehensive analysis of macrophage activity in various inflammatory contexts.

      In the in vivo model, all cells have disrupted RAS-p100α signaling, not only macrophages. Given that other myeloid cells besides macrophages contribute to the orchestration of inflammatory responses, it remains unclear whether the phenotype described in vivo results from impaired RAS-p100α signaling within macrophages or from defects in other haematopoietic cells such as neutrophils, dendritic cells, etc.

      Thank you for raising this point. To address this, we have added a paragraph in the Discussion section acknowledging that RAS-p110α signaling disruption affects all hematopoietic cells (lines 461-470 in the discussion). However, we also provide several lines of evidence that support macrophages as the primary cell type involved in the observed phenotype. Specifically, we note that neutrophil depletion in chimera mice did not alter transendothelial extravasation, and that macrophages were the primary cell type showing significant functional defects in the paw edema model. These findings, combined with specific deficiencies in myeloid populations, suggest a predominant role of macrophages in the impaired inflammatory response, though we acknowledge the potential contributions of other myeloid cells.

      Inclusion of information on the absolute number of macrophages, and total immune cells (e.g. for the spleen analysis) would help determine if the reduced frequency of macrophages represents an actual difference in their total number or rather reflects a relative decrease due to an increase in the number of other/s immune cell/s.

      Thank you for this suggestion. We understand the value of presenting actual measurements; however, we opted to display normalized data to provide a clearer comparison between WT and RBD mice, as this approach highlights the relative differences in immune populations between the two groups. Normalizing data helps to focus on the specific impact of the RAS-p110α disruption by minimizing inter-sample variability that can obscure these differences.

      To further address the reviewer’s concern regarding the interpretation of macrophage frequencies, we have included a pie chart that represents the relative proportions of the various immune cell populations studied within our dataset. Author response image 1 provides a visual overview of the immune cell distribution, enabling a clearer understanding of whether the observed decrease in macrophage frequency represents an actual reduction in total macrophage numbers or a shift in their relative abundance due to changes in other immune populations.

      We hope this approach satisfactorily addresses reviewer’s concerns by providing both a normalized dataset for clearer interpretation of genotype-specific effects and an overall immune profile that contextualizes macrophage frequency within the broader immune cell landscape.

      Author response image 1.

      Recommendations for the authors:

      Reviewer #2 (Recommendations for the authors):

      As proof of concept data that activation of RAS-p110α signaling constitutes indeed a putative approach for treating chronic inflammation is not included in the manuscript, I suggest removing this implication from the abstract.

      Thank you for this suggestion. We have now removed this implication from the abstract to maintain clarity and to better reflect the scope of the data presented in the manuscript.

      Inclusion of a control in which RBD/- cells are also treated with BYL719, across experiments in which the inhibitor is used, would be important to determine, among other things, the specificity of the inhibitor.

      We appreciate the reviewer’s suggestion to include RBD/- cells treated with BYL719 as an additional control. However, we would like to clarify that this approach would raise a different biological question, as treating RBD mice with BYL719 would not only address the specificity of the inhibitor but also examine the combined effects of genetic and pharmacologic disruptions on PI3K pathway signaling. Investigating this dual disruption falls outside the scope of our current study, which is focused specifically on the effects of RAS-p110α disruption.

      It is also important to note that our RBD mouse model selectively disrupts RAS-mediated activation of p110α, while PI3K activation can still occur through other pathways, such as receptor tyrosine kinases (RTKs) and G protein-coupled receptors (GPCRs). Thus, inhibiting p110α with BYL719 would produce broader effects beyond the inhibition of RAS-PI3K signaling, impacting PI3K activation regardless of its upstream source.

      In addition, incorporating this control would require us to repeat nearly all experiments in the manuscript, as it would necessitate generating and analyzing new samples for each experimental condition. Given the scope and resources involved, we believe this approach is unfeasible at this stage of the revision process.

      We hope this explanation is satisfactory and that the current data in the manuscript provide a rigorous assessment of the RAS-p110α signaling pathway within the defined experimental scope.

      Figure 3I is missing the statistical analysis (this is mentioned in the legend though).

      Thank you for pointing this out. We apologize for the oversight. The statistical analysis for Figure 3I has now been added.

      Gating strategies and representative staining should be included more generally across the manuscript.

      Thank you for this suggestion. To address this, we have added a new supplementary figure (Figure 2-Supplement Figure 2) that illustrates the gating strategy along with a representative dataset. Additionally, a brief summary of the gating strategy has been included in the main text to further clarify the methodology.

      It is recommended that authors show actual measurements rather than only data normalized to the control (or arbitrary units).

      Thank you for this suggestion. We understand the value of presenting actual measurements; however, we opted to display normalized data to provide a clearer comparison between WT and RBD mice, as this approach highlights the relative differences in immune populations between the two groups. Normalizing data helps to focus on the specific impact of the RAS-p110α disruption by minimizing inter-sample variability that can obscure these differences.

      To further address the reviewer’s concern regarding the interpretation of macrophage frequencies, we have included a pie chart that represents the relative proportions of the various immune cell populations studied within our dataset. Author response image 1 provides a visual overview of the immune cell distribution, enabling a clearer understanding of whether the observed decrease in macrophage frequency represents an actual reduction in total macrophage numbers or a shift in their relative abundance due to changes in other immune populations.

      We hope this approach satisfactorily addresses reviewer’s concerns by providing both a normalized dataset for clearer interpretation of genotype-specific effects and an overall immune profile that contextualizes macrophage frequency within the broader immune cell landscape.

    2. eLife Assessment

      This useful study investigates the impact of disrupting the interaction of RAS with the PI3K subunit p110α in macrophage function in vitro and inflammatory responses in vivo. Solid data overall supports a role for RAS-p110α signalling in regulating macrophage activity and so inflammation, however for many of the readouts presented the magnitude of the phenotype is not particularly pronounced. Further analysis would be required to substantiate the claims that RAS-p110α signalling plays a key role in macrophage function. Of note, the molecular mechanisms of how exactly p110α regulates the functions in macrophages have not yet been established.

    3. Reviewer #2 (Public review):

      Summary:

      Cell intrinsic signaling pathways controlling the function of macrophages in inflammatory processes, including in response to infection, injury or in the resolution of inflammation are incompletely understood. In this study, Rosell et al. investigate the contribution of RAS-p110α signaling to macrophage activity. p110α is a ubiquitously expressed catalytic subunit of PI3K with previously described roles in multiple biological processes including in epithelial cell growth and survival, and carcinogenesis. While previous studies have already suggested a role for RAS-p110α signaling in macrophage function, the cell intrinsic impact of disrupting the interaction between RAS and p110α in this central myeloid cell subset is not known.

      Strengths:

      Exploiting a sound previously described genetically engineered mouse model that allows tamoxifen-inducible disruption of the RAS-p110α pathway and using different readouts of macrophage activity in vitro and in vivo, the authors provide data consistent with their conclusion that alteration in RAS-p110α signaling impairs various but selective aspects of macrophage function in a cell-intrinsic manner.

      Weaknesses:

      My main concern is that for various readouts, the difference between wild-type and mutant macrophages in vitro or between wild-type and Pik3caRBD mice in vivo is modest, even if statistically significant. To further substantiate the extent of macrophage function alteration upon disruption of RAS-p110α signaling and its impact on the initiation and resolution of inflammatory responses, the manuscript would benefit from a more extensive assessment of macrophage activity and inflammatory responses in vivo.

      In the in vivo model, all cells have disrupted RAS-p100α signaling, not only macrophages. Given that other myeloid cells besides macrophages contribute to the orchestration of inflammatory responses, it remains unclear whether the phenotype described in vivo results from impaired RAS-p100α signaling within macrophages or from defects in other haematopoietic cells such as neutrophils, dendritic cells, etc.

      Inclusion of information on the absolute number of macrophages, and total immune cells (e.g. for the spleen analysis) would help determine if the reduced frequency of macrophages represents an actual difference in their total number or rather reflects a relative decrease due to an increase in the number of other/s immune cell/s.

      Comments on revisions:

      I thank the authors for addressing my comments.<br /> - I believe that additional in vivo experiments, or the inclusion of controls for the specificity of the inhibitor, which the authors argue are beyond the scope of the current study, are essential to address the weaknesses and limitations stated in my current evaluation.<br /> - While the neutrophil depletion suggests neutrophils are not required for the phenotype, there are multiple other myeloid cells, in addition to macrophages, that could be contributing or accounting for the in vivo phenotype observed in the mutant strain (not macrophage specific).<br /> - Inclusion of absolute cell numbers (in addition to the %) is essential. I do not understand why the authors are not including these data. Have they not counted the cells?<br /> - Lastly, inclusion of representatives staining and gating strategies for all immune profiling measurements carried out by flow cytometry is important. This point has not been addressed, not even in writing.

    1. eLife Assessment

      This manuscript reports important findings that the methyltransferase METTL3 is involved in the repair of abasic sites and uracil in DNA, mediating resistance to floxuridine-driven cytotoxicity. The presented evidence is conclusive for the involvement of m6A in DNA involving single cell imaging and mass spectrometry data. The authors present convincing evidence that the m6A signal does not result from bacterial contamination or RNA.

    2. Reviewer #1 (Public review):

      Summary:

      The authors sought to identify unknown factors involved in the repair of uracil in DNA through a CRISPR knockout screen.

      Strengths:

      The screen identified both known and unknown proteins involved in DNA repair resulting from uracil or modified uracil base incorporation into DNA. The conclusion is that the protein activity of METTL3, which converts A nucleotides to 6mA nucleotides, plays a role in the DNA damage/repair response. The importance of METTL3 in DNA repair, and its colocalization with a known DNA repair enzyme, UNG2, is well characterized.

    3. Reviewer #2 (Public review):

      Summary:

      In this work, the authors performed a CRISPR knockout screen in the presence of floxuridine, a chemotherapeutic agent that incorporates uracil and fluoro-uracil into DNA, and identified unexpected factors, such as the RNA m6A methyltransferase METTL3, as required to overcome floxuridine-driven cytotoxicity in mammalian cells. Interestingly, the observed N6-methyladenosine was embedded in DNA, which has been reported as DNA 6mA in mammalian genomes and is currently confirmed with mass spectrometry in this model. Therefore, this work consolidated the functional role of mammalian genomic DNA 6mA, and supported with solid evidence to uncover the METTL3-6mA-UNG2 axis in response to DNA base damage.

      Strengths:

      In this work, the authors took an unbiased, genome-wide CRISPR approach to identify novel factors involved in uracil repair with potential clinical interest.

      The authors designed elegant experiments to confirm the METTL3 works through genomic DNA, adding the methylation into DNA (6mA) but not the RNA (m6A), in this base damage repair context. The authors employ different enzymes, such as RNase A, RNase H, DNase, and liquid chromatography coupled to tandem mass spectrometry to validate that METTL3 deposits 6mA in DNA in response to agents that increase genomic uracil.

      They also have the Mettl3-KO and the METTL3 inhibition results to support their conclusion.

      Weaknesses:

      The authors used the METTL3 inhibitor and Mettl3-KO to validate the METTL3-6mA-UNG2 functional roles. While not an outright weakness, rescue experiments of the KO line with wild type and the METTL3 catalytic mutant would have further strengthened the evidence.

    4. Author response:

      The following is the authors’ response to the original reviews.

      eLife Assessment  

      This manuscript reports important findings that the methyltransferase METTL3 is involved in the repair of abasic sites and uracil in DNA, mediating resistance to floxuridine-driven cytotoxicity. Convincing evidence shows the involvement of m6A in DNA based on single cell imaging and mass spec data. The authors present evidence that the m6A signal does not result from bacterial contamination or RNA, but the text does not make this overly clear.

      We thank the editors for recognizing the importance of our work and the relevance of METTL3 and 6mA in DNA repair. We agree the evidence presented can be regarded as convincing, in that it includes validation with orthogonal approaches and excludes the source of 6mA being RNA or bacterial contamination.

      To clarify, the identification of 6mA in DNA, upon DNA damage, is based first on immunofluorescence observations using an anti-m6A antibody. In this setting, removal of RNA with RNase treatment fails to reduce the 6mA signal, excluding the possibility that the source of signal is RNA. In contrast, removal of DNA with DNase treatment removes all 6mA signal, strongly suggesting that the species carrying the N6-methyladenosine modification is DNA (Figure 3D, E). Importantly, in Figure 3F, G, we provide orthogonal, quantitative mass spectrometry data that independently confirm this finding. Mass spectrometry-liquid chromatography of DNA analytes, conclusively shows the presence of 6mA in DNA upon treatment with DNA damaging agents and excludes that the source is RNA, based on exact mass. 

      Cells only show the 6mA signal when treated with DNA damaging agents, and the 6mA is absent from untreated cells (Figure 3D, E, H, I). This provides strong evidence that the 6mA signal is not a result of bacterial contamination in our cell lines. Furthermore, our cell lines are routinely tested for mycoplasma contamination. It could be possible that stock solutions of DNA damaging agents may be contaminated, but this would need to be true for all individual drugs and stocks tested, which is highly unlikely. Moreover, the data showing 6mA signal is not significantly different from untreated cells when a DNA damaging agent is combined with a METTL3 inhibitor (Figure 3H, I) provides strong evidence against bacterial contamination in our stocks.  

      In summary, we provide conclusive evidence, based on orthogonal methods, that the METTL3-dependent N6-methyladenosine modification is deposited in DNA, not RNA, in response to DNA damage and have now clarified these points in the results and discussion. 

      Public Reviews:  

      Reviewer #1 (Public review):  

      Summary:  

      The authors sought to identify unknown factors involved in the repair of uracil in DNA through a CRISPR knockout screen.  

      Strengths:  

      The screen identified both known and unknown proteins involved in DNA repair resulting from uracil or modified uracil base incorporation into DNA. The conclusion is that the protein activity of METTL3, which converts A nucleotides to 6mA nucleotides, plays a role in the DNA damage/repair response. The importance of METTL3 in DNA repair, and its colocalization with a known DNA repair enzyme, UNG2, is well characterized.  

      Weaknesses:  

      This reviewer identified no major weaknesses in this study. The manuscript could be improved by tightening the text throughout, and more accurate and consistent word choice around the origin of U and 6mA in DNA. The dUTP nucleotide is misincorporated into DNA, and 6mA is formed by methylation of the A base present in DNA. Using words like 6mA "deposition in DNA" seems to imply it results from incorporation of a methylated dATP nucleotide during DNA synthesis.  

      The increased presence of 6mA during DNA damage could result from methylation at the A base itself (within DNA) or from incorporation of pre-modified 6mA during DNA synthesis. Our data do not directly discriminate between these two mechanisms, and we clarified this point in the discussion.  

      Reviewer #2 (Public review):  

      Summary:  

      In this work, the authors performed a CRISPR knockout screen in the presence of floxuridine, a chemotherapeutic agent that incorporates uracil and fluoro-uracil into DNA, and identified unexpected factors, such as the RNA m6A methyltransferase METTL3, as required to overcome floxuridine-driven cytotoxicity in mammalian cells. Interestingly, the observed N6-methyladenosine was embedded in DNA, which has been reported as DNA 6mA in mammalian genomes and is currently confirmed with mass spectrometry in this model. Therefore, this work consolidated the functional role of mammalian genomic DNA 6mA, and supported with solid evidence to uncover the METTL3-6mA-UNG2 axis in response to DNA base damage.  

      Strengths:  

      In this work, the authors took an unbiased, genome-wide CRISPR approach to identify novel factors involved in uracil repair with potential clinical interest.  

      The authors designed elegant experiments to confirm the METTL3 works through genomic DNA, adding the methylation into DNA (6mA) but not the RNA (m6A), in this base damage repair context. The authors employ different enzymes, such as RNase A, RNase H, DNase, and liquid chromatography coupled to tandem mass spectrometry to validate that METTL3 deposits 6mA in DNA in response to agents that increase genomic uracil.  

      They also have the Mettl3-KO and the METTL3 inhibition results to support their conclusion.  

      Weaknesses:  

      Although this study demonstrates that METTL3-dependent 6mA deposition in DNA is functionally relevant to DNA damage repair in mammalian cells, there are still several concerns and issues that need to be improved to strengthen this research.  

      First, in the whole paper, the authors never claim or mention the mammalian cell lines contamination testing result, which is the fundamental assay that has to be done for the mammalian cell lines DNA 6mA study.  

      Our cell lines are routinely tested for bacterial contamination, specifically mycoplasma, and we state this information in the revised manuscript. 

      Importantly, we do not observe 6mA in untreated cells, strongly suggesting that the 6mA signal observed is dependent on the presence of DNA damage and not caused by contamination in the cell lines (Figure 3D, E, H, I). While it could be possible that stock solutions of DNA damaging agents may be contaminated, this would need to be the case for all individual drugs and stocks tested that induce 6mA, which is very unlikely. Finally, the data showing 6mA signal is not significantly different from untreated cells when a DNA damaging agent is combined with a METTL3 inhibitor (Figure 3 H, I) provides strong evidence against bacterial contamination in our drug stocks.  

      Second, in the whole work, the authors have not supplied any genomic sequencing data to support their conclusions. Although the sequencing of DNA 6mA in mammalian models is challenging, recent breakthroughs in sequencing techniques, such as DR-Seq or NT/NAME-seq, have lowered the bar and improved a lot in the 6mA sequencing assay. Therefore, the authors should consider employing the sequencing methods to further confirm the functional role of 6mA in base repair.  

      While we agree that it could be important to understand the precise genomic location of 6mA in relation to DNA damage, this is outside the scope of the current study. Moreover, this exercise may prove unproductive. If 6mA is enriched in DNA at damage sites or as DNA is replicated, the genomic mapping of 6mA is likely to be stochastic. If stochastic, it would be impossible to obtain the read depth necessary to map 6mA accurately. 

      Third, the authors used the METTL3 inhibitor and Mettl3-KO to validate the METTL36mA-UNG2 functional roles. However, the catalytic mutant and rescue of Mettl3 may be the further experiments to confirm the conclusion.  

      We believe this to be an excellent suggestion from Reviewer #2 but we are unable to perform the proposed experiment at this time. We encourage future studies to explore the rescue experiment.  

      Reviewer #3 (Public review):  

      Summary:  

      The authors are showing evidence that they claim establishes the controversial epigenetic mark, DNA 6mA, as promoting genome stability.  

      Strengths:  

      The identification of a poorly understood protein, METTL3, and its subsequent characterization in DDR is of high quality and interesting.  

      Weaknesses:  

      (1) The very presence of 6mA (DNA) in mammalian DNA is still highly controversial and numerous studies have been conclusively shown to have reported the presence of 6mA due to technical artifacts and bacterial contamination. Thus, to my knowledge there is no clear evidence for 6mA as an epigenetic mark in mammals, and consequently, no evidence of writers and readers of 6mA. None of this is mentioned in the introduction. Much of the introduction can be reduced, but a paragraph clearly stating the controversy and lack of evidence for 6mA in mammals needs to be added, otherwise, the reader is given an entirely distorted view of the field.  

      These concerns must also be clearly in the limitations section and even in the results section which fails to nuance the authors' findings. 

      We agree with the reviewer that the presence and potential function of 6mA in mammalian DNA has been debated. Importantly, the debate regarding the presence and quantity of 6mA in DNA has been previously restricted to undamaged, baseline conditions. In complete agreement with this notion, we do not detect appreciable levels of 6mA in untreated cells. We revised the introduction section to present the debate about 6mA in DNA. We, however, want to highlight that our study provides, for the first time, convincing evidence (based on two orthogonal methods) that 6mA is present in DNA in response to a stimulus, DNA damage. We do not claim or provide any data that suggest 6mA is a baseline epigenetic mark.  

      (2) What is the motivation for using HT-29 cells? Moreover, the materials and methods do not state how the authors controlled for bacterial contamination, which has been the most common cause of erroneous 6mA signals to date. Did the authors routinely check for mycoplasma? 

      HT-29 is a cell line of colorectal origin and chemotherapeutic agents that introduce uracil and uracil derivatives in DNA, as those used in this study, are relevant for the treatment of colorectal cancer. As indicated above, we do not observe 6mA in untreated cells, strongly suggesting that the 6mA signal observed is dependent on DNA damage and not caused by a potential bacterial contamination (Figure 3D, E, H, I). Additionally, our cell lines are routinely tested for bacterial contamination, specifically mycoplasma. 

      (3) The single cell imaging of 6mA in various cells is nice. The results are confirmed by mass spec as an orthogonal approach. Another orthogonal and quantitative approach to assessing 6mA levels would be PacBio. Similarly, it is unclear why the authors have not performed dot-blots of 6mA for genomic DNA from the given cell lines.

      We are confused by this point since an orthogonal approach to detect 6mA, mass spectrometry-liquid chromatography, was employed. This method does not use an antibody and confirms the increase of 6mA in DNA when cells were treated with DNA damaging agents. This data is presented in Figure 3F, G. 

      It is sensible to hypothesize that the localization of 6mA is consistent with DNA replication (like uracil deposition). In this event, the genomic mapping of 6mA is likely to be stochastic. This would make quantification with PacBio sequencing difficult because it would be very challenging to achieve the appropriate read depth to call a modified base. 

      Dot blots rely on an antibody and thus are not truly orthogonal to our immunofluorescence-based measurements. We preferred the mass spectrometry-liquid chromatography approach we took as a true orthogonal approach.  

      (4) The results of Figure 3 need further investigation and validation. If the results are correct the authors are suggesting that the majority of 6mA in their cell lines is present in the DNA, and not the RNA, which is completely contrary to every other study of 6mA in mammalian cells that I am aware of. This could suggest that the antibody is not, in fact, binding to 6mA, but to unmodified adenine, which would explain why the signal disappears after DNAse treatment. Indeed, binding of 6mA to unmethylated DNA is a commonly known problem with most 6mA antibodies and is well described elsewhere.  

      Based on this and the following comment, we are convinced that Reviewer #3 has overlooked two critical elements of our study:

      First, the immunofluorescence work presented in Figure 3, showing 6mA signal in response to DNA damage, uses cells that were pre-extracted to remove excess cytoplasmic RNA. This method is often used in immunofluorescence experiments of this kind. The pre-extraction method removes most of the cytoplasmic content, and the majority of the cytoplasmic m6A RNA signal. Supplementary Figure 3D shows cells that have not been pre-extracted prior to staining. These images show the cytoplasmic m6A signal is abundant if we do not perform the pre-extraction step. 

      If the antibody used to label 6mA significantly reacted with unmodified adenine, we would expect a large signal in untreated or untreated and denatured conditions. In contrast, an increase in 6mA is not observed in either case.

      Second, the orthogonal approach we employed, mass spectrometry coupled with liquid chromatography, measures 6mA DNA analytes specifically by exact mass. This approach does not depend on an antibody and yields results consistent with those from the immunofluorescence experiments. 

      (5) Given the lack of orthologous validation of the observed DNA 6mA and the lack of evidence supporting the presence of 6mA in mammalian DNA and consequently any functional role for 6mA in mammalian biology, the manuscript's conclusions need to be toned down significantly, and the inherent difficulty in assessing 6mA accurately in mammals acknowledged throughout.  

      As discussed in response to prior comments, Figure 3 does provide two independent and orthologous methods that demonstrate 6mA presence in DNA specifically, and not RNA, in response to DNA damage. Complementary and orthogonal datasets are presented using either immunofluorescence microscopy or mass spectrometry-liquid chromatography of extracted DNA. The latter method does not rely on an antibody and can discriminate 6mA DNA versus RNA based on exact mass. We revised the text to clarify that Figure 3F, G is a completely orthogonal approach. 

      Recommendations for the authors:

      Reviewer #2 (Recommendations for the authors):  

      The authors cited most of the related publications; however, the reviewer suggested that three 2015 papers in Cell (Dahua Chen's, Yang Shi's, and Chuan He's) and the 2016 Nature (Andrew Xiao's) article are worth citing here because those are the milestone works reported the genomic DNA 6mA, for the first wave, in eukaryotic and mammalian genomes.  

      Furthermore, in Tao P. Wu and Andrew Z. Xiao's 2016 Nature article, the result has already emphasized the genomic DNA 6mA is enriched in the H2A.X sites; therefore, that work indicated the link between DNA damage and repair and 6mA's functional role. The authors may add some comments or discussion on this point.  

      Last but not least, the authors may also need to discuss the reported evidence of DNA 6mA's function in mitochondria.  

      We thank the reviewer for these suggestions. We revised our introduction and include additional references and discussion points, as suggested by the reviewer. 

      Reviewer #3 (Recommendations for the authors):  

      Minor points:  

      (1) In general, the manuscript is too verbose, and the amount of text can be dramatically reduced/sharpened. The introduction in particular is too long. 

      We revised the manuscript and reduced text when appropriate.

      (2) Each results section can also be condensed to improve clarity significantly. Indeed the results section reads like a 'Result & Discussion' section, which is then followed by a Discussion. Maybe the discussion section can be shortened to a 'conclusion'.

      We revised the results section when appropriate and reworked the discussion.

      Importantly, we revised the text related to Figure 3 as it does appear that Reviewer #3 did not appreciate key results present in this figure, specifically the orthogonal, mass spectrometry approach validating the discovery of 6mA DNA species (Figure 3F, G). We added a schematic as Figure 3F to further clarify this point as well. 

      (3) The accession number for sequencing data in GEO data should be provided.  

      The accession numbers is now provided in the manuscript. GSE282260.

      (4) All figures are unnecessarily small and in some cases, supporting figures from the supplementary data should be moved into the main figure to improve clarity. 

      The figures are of high image quality and can be enlarged easily. If there are specific figures that the reviewer believes will improve clarity, we would be happy to move them.

    1. eLife Assessment

      This paper addresses the important question of quantifying epistasis patterns, which affect the predictability of evolution, by reanalyzing a recently published combinatorial deep mutational scan experiment. The findings are that epistasis is fluid, i.e. strongly background dependent, but that fitness effects of mutations are predictable based on the wild-type phenotype. However, these potentially interesting claims are inadequately supported by the analysis, because measurement noise is not accounted for, arbitrary cutoffs are used, and global nonlinearities are not sufficiently considered. If the results continue to hold after these major improvements in the analysis, they should be of interest to all biologists working in the field of fitness landscapes.

    2. Reviewer #1 (Public review):

      This paper describes a number of patterns of epistasis in a large fitness landscape dataset recently published by Papkou et al. The paper is motivated by an important goal in the field of evolutionary biology to understand the statistical structure of epistasis in protein fitness landscapes, and it capitalizes on the unique opportunities presented by this new dataset to address this problem.

      The paper reports some interesting previously unobserved patterns that may have implications for our understanding of fitness landscapes and protein evolution. In particular, Figure 5 is very intriguing. However, I have two major concerns detailed below. First, I found the paper rather descriptive (it makes little attempt to gain deeper insights into the origins of the observed patterns) and unfocused (it reports what appears to be a disjointed collection of various statistics without a clear narrative. Second, I have concerns with the statistical rigor of the work.

      (1) I think Figures 5 and 7 are the main, most interesting, and novel results of the paper. However, I don't think that the statement "Only a small fraction of mutations exhibit global epistasis" accurately describes what we see in Figure 5. To me, the most striking feature of this figure is that the effects of most mutations at all sites appear to be a mixture of three patterns. The most interesting pattern noted by the authors is of course the "strong" global epistasis, i.e., when the effect of a mutation is highly negatively correlated with the fitness of the background genotype. The second pattern is a "weak" global epistasis, where the correlation with background fitness is much weaker or non-existent. The third pattern is the vertically spread-out cluster at low-fitness backgrounds, i.e., a mutation has a wide range of mostly positive effects that are clearly not correlated with fitness. What is very interesting to me is that all background genotypes fall into these three groups with respect to almost every mutation, but the proportions of the three groups are different for different mutations. In contrast to the authors' statement, it seems to me that almost all mutations display strong global epistasis in at least a subset of backgrounds. A clear example is C>A mutation at site 3.

      1a. I think the authors ought to try to dissect these patterns and investigate them separately rather than lumping them all together and declaring that global epistasis is rare. For example, I would like to know whether those backgrounds in which mutations exhibit strong global epistasis are the same for all mutations or whether they are mutation- or perhaps position-specific. Both answers could be potentially very interesting, either pointing to some specific site-site interactions or, alternatively, suggesting that the statistical patterns are conserved despite variation in the underlying interactions.

      1b. Another rather remarkable feature of this plot is that the slopes of the strong global epistasis patterns seem to be very similar across mutations. Is this the case? Is there anything special about this slope? For example, does this slope simply reflect the fact that a given mutation becomes essentially lethal (i.e., produces the same minimal fitness) in a certain set of background genotypes?

      1c. Finally, how consistent are these patterns with some null expectations? Specifically, would one expect the same distribution of global epistasis slopes on an uncorrelated landscape? Are the pivot points unusually clustered relative to an expectation on an uncorrelated landscape?

      1d. The shapes of the DFE shown in Figure 7 are also quite interesting, particularly the bimodal nature of the DFE in high-fitness (HF) backgrounds. I think this bimodality must be a reflection of the clustering of mutation-background combinations mentioned above. I think the authors ought to draw this connection explicitly. Do all HF backgrounds have a bimodal DFE? What mutations occupy the "moving" peak?

      1e. In several figures, the authors compare the patterns for HF and low-fitness (LF) genotypes. In some cases, there are some stark differences between these two groups, most notably in the shape of the DFE (Figure 7B, C). But there is no discussion about what could underlie these differences. Why are the statistics of epistasis different for HF and LF genotypes? Can the authors at least speculate about possible reasons? Why do HF and LF genotypes have qualitatively different DFEs? I actually don't quite understand why the transition between bimodal DFE in Figure 7B and unimodal DFE in Figure 7C is so abrupt. Is there something biologically special about the threshold that separates LF and HF genotypes? My understanding was that this was just a statistical cutoff. Perhaps the authors can plot the DFEs for all backgrounds on the same plot and just draw a line that separates HF and LF backgrounds so that the reader can better see whether the DFE shape changes gradually or abruptly.

      1f. The analysis of the synonymous mutations is also interesting. However I think a few additional analyses are necessary to clarify what is happening here. I would like to know the extent to which synonymous mutations are more often neutral compared to non-synonymous ones. Then, synonymous pairs interact in the same way as non-synonymous pair (i.e., plot Figure 1 for synonymous pairs)? Do synonymous or non-synonymous mutations that are neutral exhibit less epistasis than non-neutral ones? Finally, do non-synonymous mutations alter epistasis among other mutations more often than synonymous mutations do? What about synonymous-neutral versus synonymous-non-neutral. Basically, I'd like to understand the extent to which a mutation that is neutral in a given background is more or less likely to alter epistasis between other mutations than a non-neutral mutation in the same background.

      (2) I have two related methodological concerns. First, in several analyses, the authors employ thresholds that appear to be arbitrary. And second, I did not see any account of measurement errors. For example, the authors chose the 0.05 threshold to distinguish between epistasis and no epistasis, but why this particular threshold was chosen is not justified. Another example: is whether the product s12 × (s1 + s2) is greater or smaller than zero for any given mutation is uncertain due to measurement errors. Presumably, how to classify each pair of mutations should depend on the precision with which the fitness of mutants is measured. These thresholds could well be different across mutants. We know, for example, that low-fitness mutants typically have noisier fitness estimates than high-fitness mutants. I think the authors should use a statistically rigorous procedure to categorize mutations and their epistatic interactions. I think it is very important to address this issue. I got very concerned about it when I saw on LL 383-388 that synonymous stop codon mutations appear to modulate epistasis among other mutations. This seems very strange to me and makes me quite worried that this is a result of noise in LF genotypes.

    3. Reviewer #2 (Public review):

      Significance:

      This paper reanalyzes an experimental fitness landscape generated by Papkou et al., who assayed the fitness of all possible combinations of 4 nucleotide states at 9 sites in the E. coli DHFR gene, which confers antibiotic resistance. The 9 nucleotide sites make up 3 amino acid sites in the protein, of which one was shown to be the primary determinant of fitness by Papkou et al. This paper sought to assess whether pairwise epistatic interactions differ among genetic backgrounds at other sites and whether there are major patterns in any such differences. They use a "double mutant cycle" approach to quantify pairwise epistasis, where the epistatic interaction between two mutations is the difference between the measured fitness of the double-mutant and its predicted fitness in the absence of epistasis (which equals the sum of individual effects of each mutation observed in the single mutants relative to the reference genotype). The paper claims that epistasis is "fluid," because pairwise epistatic effects often differs depending on the genetic state at the other site. It also claims that this fluidity is "binary," because pairwise effects depend strongly on the state at nucleotide positions 5 and 6 but weakly on those at other sites. Finally, they compare the distribution of fitness effects (DFE) of single mutations for starting genotypes with similar fitness and find that despite the apparent "fluidity" of interactions this distribution is well-predicted by the fitness of the starting genotype.

      The paper addresses an important question for genetics and evolution: how complex and unpredictable are the effects and interactions among mutations in a protein? Epistasis can make the phenotype hard to predict from the genotype and also affect the evolutionary navigability of a genotype landscape. Whether pairwise epistatic interactions depend on genetic background - that is, whether there are important high-order interactions -- is important because interactions of order greater than pairwise would make phenotypes especially idiosyncratic and difficult to predict from the genotype (or by extrapolating from experimentally measured phenotypes of genotypes randomly sampled from the huge space of possible genotypes). Another interesting question is the sparsity of such high-order interactions: if they exist but mostly depend on a small number of identifiable sequence sites in the background, then this would drastically reduce the complexity and idiosyncrasy relative to a landscape on which "fluidity" involves interactions among groups of all sites in the protein. A number of papers in the recent literature have addressed the topics of high-order epistasis and sparsity and have come to conflicting conclusions. This paper contributes to that body of literature with a case study of one published experimental dataset of high quality. The findings are therefore potentially significant if convincingly supported.

      Validity:

      In my judgment, the major conclusions of this paper are not well supported by the data. There are three major problems with the analysis.

      (1) Lack of statistical tests. The authors conclude that pairwise interactions differ among backgrounds, but no statistical analysis is provided to establish that the observed differences are statistically significant, rather than being attributable to error and noise in the assay measurements. It has been established previously that the methods the authors use to estimate high-order interactions can result in inflated inferences of epistasis because of the propagation of measurement noise (see PMID 31527666 and 39261454). Error propagation can be extreme because first-order mutation effects are calculated as the difference between the measured phenotype of a single-mutant variant and the reference genotype; pairwise effects are then calculated as the difference between the measured phenotype of a double mutant and the sum of the differences described above for the single mutants. This paper claims fluidity when this latter difference itself differs when assessed in two different backgrounds. At each step of these calculations, measurement noise propagates. Because no statistical analysis is provided to evaluate whether these observed differences are greater than expected because of propagated error, the paper has not convincingly established or quantified "fluidity" in epistatic effects.

      (2) Arbitrary cutoffs. Many of the analyses involve assigning pairwise interactions into discrete categories, based on the magnitude and direction of the difference between the predicted and observed phenotypes for a pairwise mutant. For example, the authors categorize as a positive pairwise interaction if the apparent deviation of phenotype from prediction is >0.05, negative if the deviation is <-0.05, and no interaction if the deviation is between these cutoffs. Fluidity is diagnosed when the category for a pairwise interaction differs among backgrounds. These cutoffs are essentially arbitrary, and the effects are assigned to categories without assessing statistical significance. For example, an interaction of 0.06 in one background and 0.04 in another would be classified as fluid, but it is very plausible that such a difference would arise due to error alone. The frequency of epistatic interactions in each category as claimed in the paper, as well as the extent of fluidity across backgrounds, could therefore be systematically overestimated or underestimated, affecting the major conclusions of the study.

      (3) Global nonlinearities. The analyses do not consider the fact that apparent fluidity could be attributable to the fact that fitness measurements are bounded by a minimum (the fitness of cells carrying proteins in which DHFR is essentially nonfunctional) and a maximum (the fitness of cells in which some biological factor other than DHFR function is limiting for fitness). The data are clearly bounded; the original Papkou et al. paper states that 93% of genotypes are at the low-fitness limit at which deleterious effects no longer influence fitness. Because of this bounding, mutations that are strongly deleterious to DHFR function will therefore have an apparently smaller effect when introduced in combination with other deleterious mutations, leading to apparent epistatic interactions; moreover, these apparent interactions will have different magnitudes if they are introduced into backgrounds that themselves differ in DHFR function/fitness, leading to apparent "fluidity" of these interactions. This is a well-established issue in the literature (see PMIDs 30037990, 28100592, 39261454). It is therefore important to adjust for these global nonlinearities before assessing interactions, but the authors have not done this.

      This global nonlinearity could explain much of the fluidity claimed in this paper. It could explain the observation that epistasis does not seem to depend as much on genetic background for low-fitness backgrounds, and the latter is constant (Figure 2B and 2C): these patterns would arise simply because the effects of deleterious mutations are all epistatically masked in backgrounds that are already near the fitness minimum. It would also explain the observations in Figure 7. For background genotypes with relatively high fitness, there are two distinct peaks of fitness effects, which likely correspond to neutral mutations and deleterious mutations that bring fitness to the lower bound of measurement; as the fitness of the background declines, the deleterious mutations have a smaller effect, so the two peaks draw closer to each other, and in the lowest-fitness backgrounds, they collapse into a single unimodal distribution in which all mutations are approximately neutral (with the distribution reflecting only noise).<br /> Global nonlinearity could also explain the apparent "binary" nature of epistasis. Sites 4 and 5 change the second amino acid, and the Papkou paper shows that only 3 amino acid states (C, D, and E) are compatible with function; all others abolish function and yield lower-bound fitness, while mutations at other sites have much weaker effects. The apparent binary nature of epistasis in Figure 5 corresponds to these effects given the nonlinearity of the fitness assay. Most mutations are close to neutral irrespective of the fitness of the background into which they are introduced: these are the "non-epistatic" mutations in the binary scheme. For the mutations at sites 4 and 5 that abolish one of the beneficial mutations, however, these have a strong background-dependence: they are very deleterious when introduced into a high-fitness background but their impact shrinks as they are introduced into backgrounds with progressively lower fitness. The apparent "binary" nature of global epistasis is likely to be a simple artifact of bounding and the bimodal distribution of functional effects: neutral mutations are insensitive to background, while the magnitude of the fitness effect of deleterious mutations declines with background fitness because they are masked by the lower bound. The authors' statement is that "global epistasis often does not hold." This is not established. A more plausible conclusion is that global epistasis imposed by the phenotype limits affects all mutations, but it does so in a nonlinear fashion.

      In conclusion, most of the major claims in the paper could be artifactual. Much of the claimed pairwise epistasis could be caused by measurement noise, the use of arbitrary cutoffs, and the lack of adjustment for global nonlinearity. Much of the fluidity or higher-order epistasis could be attributable to the same issues. And the apparently binary nature of global epistasis is also the expected result of this nonlinearity.

    4. Reviewer #3 (Public review):

      Summary:

      The authors have studied a previously published large dataset on the fitness landscape of a 9 base-pair region of the folA gene. The objective of the paper is to understand various aspects of epistasis in this system, which the authors have achieved through detailed and computationally expensive exploration of the landscape. The authors describe epistasis in this system as "fluid", meaning that it depends sensitively on the genetic background, thereby reducing the predictability of evolution at the genetic level. However, the study also finds two robust patterns. The first is the existence of a "pivot point" for a majority of mutations, which is a fixed growth rate at which the effect of mutations switches from beneficial to deleterious (consistent with a previous study on the topic). The second is the observation that the distribution of fitness effects (DFE) of mutations is predicted quite well by the fitness of the genotype, especially for high-fitness genotypes. While the work does not offer a synthesis of the multitude of reported results, the information provided here raises interesting questions for future studies in this field.

      Strengths:

      A major strength of the study is its detailed and multifaceted approach, which has helped the authors tease out a number of interesting epistatic properties. The study makes a timely contribution by focusing on topical issues like the prevalence of global epistasis, the existence of pivot points, and the dependence of DFE on the background genotype and its fitness. The methodology is presented in a largely transparent manner, which makes it easy to interpret and evaluate the results.

      The authors have classified pairwise epistasis into six types and found that the type of epistasis changes depending on background mutations. Switches happen more frequently for mutations at functionally important sites. Interestingly, the authors find that even synonymous mutations in stop codons can alter the epistatic interaction between mutations in other codons. Consistent with these observations of "fluidity", the study reports limited instances of global epistasis (which predicts a simple linear relationship between the size of a mutational effect and the fitness of the genetic background in which it occurs). Overall, the work presents some evidence for the genetic context-dependent nature of epistasis in this system.

      Weaknesses:

      Despite the wealth of information provided by the study, there are some shortcomings of the paper which must be mentioned.

      (1) In the Significance Statement, the authors say that the "fluid" nature of epistasis is a previously unknown property. This is not accurate. What the authors describe as "fluidity" is essentially the prevalence of certain forms of higher-order epistasis (i.e., epistasis beyond pairwise mutational interactions). The existence of higher-order epistasis is a well-known feature of many landscapes. For example, in an early work, (Szendro et. al., J. Stat. Mech., 2013), the presence of a significant degree of higher-order epistasis was reported for a number of empirical fitness landscapes. Likewise, (Weinreich et. al., Curr. Opin. Genet. Dev., 2013) analysed several fitness landscapes and found that higher-order epistatic terms were on average larger than the pairwise term in nearly all cases. They further showed that ignoring higher-order epistasis leads to a significant overestimate of accessible evolutionary paths. The literature on higher-order epistasis has grown substantially since these early works. Any future versions of the present preprint will benefit from a more thorough contextual discussion of the literature on higher-order epistasis.

      (2) In the paper, the term 'sign epistasis' is used in a way that is different from its well-established meaning. (Pairwise) sign epistasis, in its standard usage, is said to occur when the effect of a mutation switches from beneficial to deleterious (or vice versa) when a mutation occurs at a different locus. The authors require a stronger condition, namely that the sum of the individual effects of two mutations should have the opposite sign from their joint effect. This is a sufficient condition for sign epistasis, but not a necessary one. The property studied by the authors is important in its own right, but it is not equivalent to sign epistasis.

      (3) The authors have looked for global epistasis in all 108 (9x12) mutations, out of which only 16 showed a correlation of R^2 > 0.4. 14 out of these 16 mutations were in the functionally important nucleotide positions. Based on this, the authors conclude that global epistasis is rare in this landscape, and further, that mutations in this landscape can be classified into one of two binary states - those that exhibit global epistasis (a small minority) and those that do not (the majority). I suspect, however, that a biologically significant binary classification based on these data may be premature. Unsurprisingly, mutational effects are stronger at the functional sites as seen in Figure 5 and Figure 2, which means that even if global epistasis is present for all mutations, a statistical signal will be more easily detected for the functionally important sites. Indeed, the authors show that the means of DFEs decrease linearly with background fitness, which hints at the possibility that a weak global epistatic effect may be present (though hard to detect) in the individual mutations. Given the high importance of the phenomenon of global epistasis, it pays to be cautious in interpreting these results.

      (4) The study reports that synonymous mutations frequently change the nature of epistasis between mutations in other codons. However, it is unclear whether this should be surprising, because, as the authors have already noted, synonymous mutations can have an impact on cellular functions. The reader may wonder if the synonymous mutations that cause changes in epistatic interactions in a certain background also tend to be non-neutral in that background. Unfortunately, the fitness effect of synonymous mutations has not been reported in the paper.

      (5) The authors find that DFEs of high-fitness genotypes tend to depend only on fitness and not on genetic composition. This is an intriguing observation, but unfortunately, the authors do not provide any possible explanation or connect it to theoretical literature. I am reminded of work by (Agarwala and Fisher, Theor. Popul. Biol., 2019) as well as (Reddy and Desai, eLife, 2023) where conditions under which the DFE depends only on the fitness have been derived. Any discussion of possible connections to these works could be a useful addition.

    5. Author response:

      Thank you for sharing a detailed review of our manuscript titled, Variations and predictability of epistasis on an intragenic fitness landscape. We have now carefully gone through the reviewers’ and the editor’s comments and have the following preliminary responses.

      (1) Measurement noise in the folA fitness landscape. All three reviewers and the editors raise the important matter of incorporating measurement noise in the fitness landscape. The paper by Papkou and coworkers makes the fitness measurements of the landscape in six independent repeats. They show that the fitness data is highly correlated in each repeat, and use the weighted mean of the repeats to report their results. They do not study how measurement noise influences their findings. The results by Papkou and coworkers were our starting point, and hence, we built on the landscape properties reported in their study. As a result, we also analyse our results working with the same mean of the six independent measurements.

      The main result of the work by Papkou and coworkers is that largest subgraph in the landscape has 514 fitness peaks. 

      We revisit this result by quantifying how measurement noise changes this number. By doing this, we note the subgraph contains only 127 peaks which are statistically significant. We define a sequence as a peak when its corresponding fitness is greater than all its one-distance neighbours with a p-value < 0.05. This shows that, as pointed out in the reviews, incorporating noise in the landscape results significantly changes how we view the landscape – a facet not included in Papkou et al and the current version of our manuscript. 

      Not incorporating measurement noise means that the entire landscape has 4055 peaks. When measurement noise is included in the analysis, this number reduces to 137, out of which 136 are high fitness backgrounds (functional). 

      In the revised version of our manuscript, we will incorporate measurement noise in our analysis. Through this, we will also address the concern regarding the use of an arbitrary cut-off to study “fluid” epistasis. However, we note that arbitrary cut-offs to define DFEs have been recently used (Sane et al., PNAS, 2023).

      We also note that previous work with large scale landscapes (Wu et al, eLife, 2016) also reported a fitness landscape with a single experiment, with no repeats. 

      (2) Global nonlinearities and higher-order leading to fluid epistasis. Attempts at building models for higher-order epistasis from empirical data have largely been confined to landscapes of a limited data size. For example, Sailer & Harms, Genetics, 2017 propose models for higher-order epistasis from seven empirical data sets, each with less than a 100 data points. Another recent attempt (Park et al, Nat Comm, 2024) proposes rule for protein structure-function with 20 fitness landscapes. In this study, only one landscape which used fitness as a phenotype had ~160000 data points (of which only 42% were included for analysis). All other data sets which used fitness as a phenotype contained less than 10000 data points. While these statistical proposals of how higher-order epistasis operates exist, none of them are reliant of large scale, exhaustive network, like the one proposed by Papkou and coworkers.  

      In the edited manuscript, we will replace our arbitrary cut-off with results of statistical tests carried out based on measurement noise. 

      Global non-linearities shape evolutionary responses. We would like to emphasize that the goal of this work to study and understand how these global non-linearities result in patterns on a large fitness landscape by presenting the sum total of these fundamental factors in shaping statistical patterns. 

      While we understand that we may not have sufficiently explained the effects of global non-linearities on our results, we do not agree with the reviewer’s conclusion that our results are artifacts of these non-linearities. We will expand on the role of these nonlinearities on the patterns that we observe (like, fitness being bounded, as pointed out by reviewer 2, or differential impact of a mutation in functional vs. non-functional variants).

      We also speculate that changing our arbitrary cut-off (selection coefficient of 0.05) to measurement noise will not alter our results qualitatively. 

      The question we address in our work is, therefore, how does the nature of epistasis change with genetic background over a large, exhaustive landscape. The nature of epistasis between two mutations is analysed in all 4<sup>7</sup> backgrounds. The causative agents for the change in epistasis will be context-dependent, depending on the precise nature of the two mutations and the background. For instance, a certain background might simply introduce a Stop codon in the sequence. Notwithstanding these precise, local mechanistic explanations, we seek to answer how epistasis changes statistically in a sequence. Investigating statistical patterns which explain switch in nature of epistasis in deep, exhaustive landscapes is a long-term goal of this research.

      (3) Last, in our revised manuscript, we will address the reviewers’ other minor comments on the various aspects of the manuscript.

    1. eLife Assessment

      This study makes a valuable contribution to understanding Bayesian inference in dynamic environments by demonstrating how humans integrate prior beliefs with sensory evidence, revealing an overestimation of environmental volatility while accurately tracking noise. The evidence is solid, supported by robust model fitting and principled factorial model set analyses, though limitations in sample size and inconclusive findings on memory capacity tradeoffs reduce the overall impact. Future work should expand validation across datasets, enhance model comparisons, and explore the generalizability of reduced Bayesian frameworks to strengthen the conclusions and broader relevance of the study.

    2. Reviewer #1 (Public review):

      Summary

      Behavioural adjustments to different sources of uncertainty remain a hot topic in many fields including reinforcement learning. The authors present valuable findings suggesting that human participants integrate prior beliefs with sensory evidence to improve their predictions in dynamically changing environments involving perceptual decision-making, pinpointing to hallmarks of Bayesian inference. Fitting of a reduced Bayesian model to participant choice behaviour reveals that decision-makers overestimate environmental volatility, but were reasonably accurate in terms of tracking environmental noise.

      Strengths

      Using a perceptual decision-making task in which participants were presented with sequences of noisy observation in environments with constant volatility and variable noise, the authors demonstrate solid evidence in favour of reduced Bayesian models that can account for participant choice behaviour when its generative parameters are fitted freely. The work nicely complements recent work demonstrating the fitting of a full Bayesian model to human reinforcement learning. The authors' approach to the fitting of the model in a principled/factorial manner that is exhaustive performs the model comparison and highlights the need for further work in evaluating the model's performance in environments outside of its generative parameters. Overall the work further highlights the utility of using perceptual decision-making for Bayesian inference questions.

      Weaknesses

      Although data sharing and reanalysis of data are extremely welcome, particularly considering their utility for open science, the small sample size (N= 29) of the original dataset somewhat restricts the authors' ability to show more conclusive findings when it comes to deciphering the optimal memory capacity of the fitted models. It is likely that the relatively small sample size also contributes to certain key hypotheses not being confirmed intuitively, for example, the expected negative relationship between hazard rates and log (noise). The notion that the participants rely on priors to a greater extent in low noise environments relative to high noise may also indicate that they might misattribute noise as volatility, as higher noise in the environment usually obscures the information content of outcomes, and in the case of pure random/noisy sequences, it should increase reliance to priors as new sensory evidence becomes unreliable.

    3. Reviewer #2 (Public review):

      Summary:

      Meijer et al reanalyze behavioral data from a task in which people made predictions about the next in a sequence of localized sounds with the goal of understanding the computations through which people combine sensory experiences into a prior used for perception. The authors combine basic analyses of experimental data with model simulations and development and fitting of a factorial model set that includes a prominent model of change-point detection that has previously been shown to approximate Bayesian inference at a reduced computational cost and provide a good match to human prediction data (reduced Bayesian model). The authors present a number of findings, including a demonstration of key qualitative markers for Bayesian change-point detection, a tendency in humans to over-rely on recent observations, a lack of an inverse relationship between fit values of hazard rate and fit values of noise, support for a number of assumptions in the reduced Bayesian model, and a lack of evidence for reliance on memory systems beyond the extremely minimal requirements of that model.

      Strengths:

      The paper asks an important question and takes a number of useful steps toward answering it. In particular, the factorial model set constructed to examine a number of explicit assumptions in the models typically fit to change-point predictive inference task data was a very useful innovation, and in some cases showed clearly that assumptions in the model are necessary or at least better than the proposed alternatives. In particular, the paper develops a notion of memory capacity that allows for a continuum of models differing in their tradeoffs between computational cost and predictive precision. Another strength of the paper is that it relies on data that avoids sequential biases that can contaminate reported beliefs in more standard predictive inference tasks.

      Weaknesses:

      The primary weakness of the paper is that most of the definitive findings reported within it have already been reported elsewhere. That humans increase the influence of surprising outcomes indicative of change points, or to say this another way, decrease their reliance on prior information in such cases, has been fairly well established, as has the discovery that humans tend to overuse recent outcomes when making predictions. The most novel aspect of the paper, the exploration of reductions of the Bayesian ideal observer that rely on differing memory capacities, yielded results that are somewhat difficult to interpret, particularly because it is not clear that the task analyzed is diagnostic of the memory capacity term in the model, or if so, what the qualitative hallmarks of a high/low memory capacity model reduction might be.

    1. eLife Assessment

      This manuscript presents a useful mean-field model for a network of Hodgkin-Huxley neurons retaining the equations for ion exchange between the intracellular and extracellular space. The mean-field model derived in this work relies on approximations and heuristic arguments that, on the one hand, allow a closed-form derivation of the mean-field equations, but also raise questions about their justifications and the degree to which the results agree with experiments as well as direct numerical simulations. Therefore, the evidence for the utility of this approach is at present incomplete.

    2. Reviewer #1 (Public review):

      Summary:

      In this manuscript, the authors derive a mean-field model for a network of Hodgkin-Huxley neurons retaining the equations for ion exchange between the intracellular and extracellular space.

      The mean-field model derived in this work relies on approximations and heuristic arguments that, on the one hand, allow a closed-form derivation of the mean-field equations, and on the other hand restrict its validity to a limited regime of activity corresponding to quasi-synchronous neuronal populations. Therefore, rather than an exact mean-field representation, the model provides a description of a mesoscopic population of connected neurons driven by ion exchange dynamics.

      Strengths:

      The idea of deriving a mean-field model that relates the slow-timescale biophysical mechanism of ion exchange and transportation in the brain to the fast-timescale electrical activities of large neuronal ensembles.

      Weaknesses:

      The idea underlying this work is not completely implemented in practice.

      The derived mean field model does not show a one-to-one correspondence with the neural network simulations, except in strongly synchronous regimes. The agreement with the in vitro experiment is hardly evident, both for the mean-field model and for the network model. The assumptions made to derive the closed-form equations of the mean-field model have not been justified by any biological reason, they just allow for the mathematical derivation. The final form of the mean-field equations does not clarify whether or not microscopic variables are used together with macroscopic variables in an inconsistent mixture.

    3. Reviewer #2 (Public review):

      Summary:

      The authors aim to develop a neural mass model characterized by a few collective variables mimicking the dynamics of a network of Hodgkin - Huxley neurons encompassing ion-exchange mechanisms. They describe in detail the derivation of the mean-field model, then they compare experimental results obtained for the hippocampus of a mouse with the neural network simulations and the mean-field results. Furthermore, they report a bifurcation analysis of the developed model and simulation of a small network containing various coupled neural masses, somehow moving towards the simulation of an entire connectome.

      Strengths:

      The author attempts to develop a mean-field model for a globally coupled network of heterogeneous Hodgkin-Huxley neurons with an explicit ion exchange mechanism between the cell interior and exterior.

      Weaknesses:

      (1) It seems that the reduction methodology that is employed is not the most suitable one for the single-neuron model they are considering.<br /> (2) The authors' derivation of the neural mass model is based on several assumptions, and not all well justified.<br /> (3) The formulation of the mean-field derivation is unnecessarily complicated. It could be heavily simplified by following previously published approaches to derive biologically realistic neural masses.<br /> (4) The model seems to work only for highly synchronized situations and not for the standard asynchronous evolution usually observed in neural circuits.

      General Statements:

      The authors honestly declared the many limitations of their approach. It is assumed that the results of the mean-field are somehow inconsistent with the neural network simulations as expected.

      The authors suggest employing this model for the simulations on the whole connectome to follow seizure propagation, however, I believe that the Epileptor remains superior in this respect to this model. That indeed includes biophysical parameters but their correspondence with the ones employed in the network dynamics remains elusive, due to the many assumptions required to derive this mean-field model. Furthermore, it is more complicated than the Epileptor, I do not think that the present model will be largely employed by the community.

    1. eLife Assessment

      This important study uses diffusion magnetic resonance imaging to non-invasively map the white matter fibres connecting the zona incerta and cortex in humans. The authors present convincing evidence to indicate that these connections are organized along a rostro-caudal axis. The findings will be of interest to researchers interested in neuroanatomy and cortico-subcortical connectivity.

    2. Reviewer #1 (Public review):

      Summary:

      This is a study that used 7T diffusion MRI in subjects from a Human Connectome Project dataset to characterize the zona incerta, an area of gray matter whose involvement has been demonstrated in a broad range of behavioral and physiologic functions. The authors employ tractography to model white matter tracts that involve connections with the ZI and use clustering techniques to segment the ZI into distinct subregions based on similar patterns of connectivity. The authors report a rostral-caudal organization of the ZI's streamlines where rostrally-projecting tracts are rostrally-positioned in the ZI and caudally-projecting tracts are caudally-positioned in the ZI.

      Strengths:

      The paper presents robust findings that demonstrate subregions of the human ZI that appear to be structurally distinct using a combination of spectral clustering and diffusion map embedding methods. The results of this work can contribute to our understanding of the anatomy and structural connectivity of the ZI, allowing us to further explore its role as a neuromodulatory target for various neurological disorders.

      Weaknesses:

      There should be further discussion of the clustering methods employed and why they are appropriate for the pertinent data. Additionally, the limitations of analyzing solely the cortical connections of the zona incerta should be addressed, as anatomical studies of the ZI have shown significant involvement of the ZI in tracts projecting to deep brain regions.

    3. Reviewer #2 (Public review):

      Summary:

      Haast et al. investigated the organization of the zona incerta (ZI) in the human brain based on its structural connectivity to the neocortex. They found that the ZI is organized according to a primary rostro-caudal gradient, where the rostral ZI is more strongly connected to the prefrontal cortex and the caudal ZI to the sensorimotor cortex. They also found that the central region of the ZI is differently connected to the neocortex compared with the rostral and caudal regions, and could be important as a deep brain stimulation target for the treatment of essential tremors.

      Strengths:

      I think the overall quality of this work is great, and the results are presented in a very clear and organized manner. I particularly appreciate the effort that the authors put into validating the results using 7T and 3T data, as well as test-retest data.

      Weaknesses:

      That being said, I was left with a couple of concerns after reading the paper.

      (1) Although the authors discussed animal evidence for a dorsal-ventral organization of the ZI, I thought that the evidence they presented for it in this paper was not so convincing. In Figure S5, the second gradient (G2) shows a clear dorsoventral pattern, but this pattern seems to primarily separate the ZI and H fields rather than show an internal topology of the ZI. This is more likely the case given that there are two bands (superior and inferior) of high G2 values surrounding a single band (middle) of low G2 values. The evidence for the rostrocaudal gradient, on the other hand, is quite convincing.

      (2) HCP data is still too advanced for clinical translation. Although 3T is becoming more and more prevalent for presurgical planning, the HCP 3T dataset is acquired with a voxel size of 1.25mm, which is a far higher resolution than the typical clinical scan. It would be very useful for clinical readers to see what individual subject replicability looks like if the data were acquired at the more typical voxel size of 2mm. This could be achieved by replicating the analysis on a downsampled version of the HCP data that more closely resembles clinical data. This is understandably a large undertaking, so it could be left to future validation work.

  2. Jan 2025
    1. eLife Assessment

      This important work proposes a neural network model of interactions between the prefrontal cortex and basal ganglia to implement adaptive resource allocation in working memory, where the gating strategies for storage are adjusted by reinforcement learning. Numerical simulations provide convincing evidence for the superiority of the model in improving effective capacity, optimizing resource management, and reducing error rates, as well as for its human-like performance. This work will be of broad interest to computational and cognitive neuroscientists, and may also interest machine-learning researchers who seek to develop brain-inspired machine-learning algorithms for memory.

    2. Reviewer #1 (Public review):

      Summary:

      In this research, Soni and Frank investigate the network mechanisms underlying capacity limitations in working memory from a new perspective, with a focus on Visual Working Memory (VWM). The authors have advanced beyond the classical neural network model, which incorporates the prefrontal cortex and basal ganglia (PBWM), by introducing an adaptive chunking variant. This model is trained using a biologically-plausible, dopaminergic reinforcement learning framework. The adaptive chunking mechanism is particularly well-suited to the VWM tasks involving continuous stimuli and elegantly integrates the 'slot' and 'resource' theories of working memory constraints. The chunk-augmented PBWM operates as a slot-like system with resource-like limitations.

      Through numerical simulations under various conditions, Soni and Frank demonstrate the performance of the chunk-augmented PBWM model surpass the no-chunk control model. The improvements are evident in enhanced effective capacity, optimized resource management, and reduced error rates. The retention of these benefits, even with increased capacity allocation, suggests that working memory limitations are due to a combination of factors, including the efficient credit assignment that are learned flexibly through reinforcement learning. In essence, this work addresses fundamental questions related to a computational working memory limitation using a biologically-inspired neural network, thus has implications for conditions such as Parkinson's disease, ADHD and schizophrenia.

      Strengths:

      The integration of mechanistic flexibility, reconciling two theories for WM capacity into a single unified model, results in a neural network that is both more adaptive and human-like. Building on the PBWM framework ensures the robustness of the findings. The addition of the chunking mechanism tailors the original model for continuous visual stimuli. Chunk-stripe mechanisms contribute to the 'resource' aspect, while input-stripes contribute to the 'slot' aspect. This combined network architecture enables flexible and diverse computational functions, enhancing performance beyond that of the classical model.

      Moreover, unlike previous studies that design networks for specific task demands, the proposed network model can dynamically adapt to varying task demands by optimizing the chunking gating policy through RL.

      The implementation of a dopaminergic reinforcement learning protocol, as opposed to a hard-wired design, leads to the emergence of strategic gating mechanisms that enhance the network's computational flexibility and adaptability. These gating strategies are vital for VWM tasks and are developed in a manner consistent with ecological and evolutionary learning held by human. Further examination of how reward prediction error signals, both positive and negative, collaborate to refine gating strategies reveals the crucial role of reward feedback in fine-tuning the working memory computations and the model's behavior, aligning with the current neuroscientific understanding that reward matters.

      Assessing the impact of a healthy balance of dopaminergic RPE signals on information manipulation holds implications for patients with altered striatal dopaminergic signaling.

      Comments on revisions:

      In the revised version, the authors have thoroughly addressed all the questions raised in my previous review. They have clarified the model architecture, provided detailed explanations of the training process, and elaborated on the convergence of the optimization.

      Additionally, Reviewer 2 made a very constructive suggestion: Can related cognitive functions or phenomena emerge from the model? The newly added analysis and results highlighting the recency effect directly address this question and significantly strengthen the paper.

    3. Reviewer #2 (Public review):

      Summary:

      This paper utilizes a neural network model to investigate how the brain employs an adaptive chunking strategy to effectively enhance working memory capacity, which is a classical and significant question in cognitive neuroscience. By integrating perspectives from both the 'slot model' and 'limited resource models,' the authors adopted a neural network model encompassing the prefrontal cortex and basal ganglia, introduced an adaptive chunking strategy, and proposed a novel hybrid model. The study demonstrates that the brain can adaptively bind various visual stimuli into a single chunk based on the similarity of color features (a continuous variable) among items in visual working memory, thereby improving working memory efficiency. Additionally, it suggests that the limited capacity of working memory arises from the computational characteristics of the neural system, rather than anatomical constraints.

      Strengths:

      The neural network model utilized in this paper effectively integrates perspectives from both slot models and resource models (i.e., resource-like constraints within a slot-like system). This methodological innovation provides a better explanation for the limited capacity of working memory. By simulating the neural networks of the prefrontal cortex and basal ganglia, the model demonstrates how to optimize working memory storage and retrieval strategies through reinforcement learning (i.e., the efficient management of access to and from working memory). This biological simulation offers a novel perspective on human working memory and provides new explanations for the working memory difficulties observed in patients with Parkinson's disease and other disorders. Furthermore, the effectiveness of the model has been validated through computational simulation experiments, yielding reliable and robust predictions.

      Comments on revisions:

      The authors have already answered all my questions.

    4. Author response:

      The following is the authors’ response to the original reviews.

      eLife Assessment

      This important work proposes a neural network model of interactions between the prefrontal cortex and basal ganglia to implement adaptive resource allocation in working memory, where the gating strategies for storage are adjusted by reinforcement learning. Numerical simulations provide convincing evidence for the superiority of the model in improving effective capacity, optimizing resource management, and reducing error rates, as well as solid evidence for its human-like performance. The paper could be strengthened further by a more thorough comparison of model predictions with human behavior and by improved clarity in presentation. This work will be of broad interest to computational and cognitive neuroscientists, and may also interest machine-learning researchers who seek to develop brain-inspired machine-learning algorithms for memory.

      We thank the reviewers for their thorough and constructive comments, which have helped us clarify, augment and solidify our work. Regarding the suggestion to include a “more thorough comparison with with human behavior”, we believe this comment reflects one of the reviewer’s suggestion to compare with sequential order effects. We now include a new section with simulations showing that the network exhibits clear recency effects in accordance with the literature, and where such recency effects are known to be related to WM interference and not due to passive decay. Overall our work makes substantial contact with human behavioral patterns that have been documented in the human literature (and which as far as we know have not been jointly captured by any one model), such as the shape of the error distributions, including probability of recall and variable precision;  attraction to recently presented items,  sensitivity to reinforcement history, set-size dependent chunking, recency effects,  dopamine manipulation effects, as well of a range of human data linking capacity limitations to frontostriatal function. It also provides a theoretical proposal for the well established phenomenon of capacity limitations in humans, suggesting that they arise due to difficulty in WM management.

      Below we address each reviewer individually, responding to each comment and providing the relevant location in the paper that the changes and additions were made. Reviewer responses are included in blue/bold for clarity.  

      Public Reviews:

      Reviewer 1:

      Thank you for your comments. We appreciate your statements of the strengths of this paper and your suggestions to improve this paper.

      First, the method section appears somewhat challenging to follow. To enhance clarity, it might be beneficial to include a figure illustrating the overall model architecture. This visual aid could provide readers with a clearer understanding of the overall network model.

      Additionally, the structure depicted in Figure 2 could be potentially confusing. Notably, the absence of an arrow pointing from the thalamus to the PFC and the apparent presence of two separate pathways, one from sensory input to the PFC and another from sensory input to the BG and then to the thalamus, may lead to confusion. While I recognize that Figure 2 aims to explain network gating, there is room for improvement in presenting the content accurately.

      As suggested, we added a figure (new figure 2) illustrating the overall model architecture before expanding it to show the chunking circuitry. This figure also shows the projections from thalamus to PFC (we preserve the previous figure 2, now figure 3, as an example sequence of network gating decisions, in more abstract form to help facilitate a functional understanding of the sequence of events without too much clutter). We also made several other general clarifications to the methods sections to make it more transparent and easier to follow, as per your suggestions.   

      Still, for the method part, it would enhance clarity to explicitly differentiate between predesigned (fixed) components and trainable components. Specifically, does the supplementary material state that synaptic connection weights in striatal units (Go&NoGo) are trained using XCAL, while other components, such as those in the PFC and lateral inhibition, are not trained (I found some sentences in 'Limitations and Future Directions')?

      We have now explicitly specified learned and fixed components. We have further explained the role of XCAL and how striatal Go/NoGo weights are trained. We have also added clarification on how gating policies are learned via eligibility traces and synaptic tags.

      I'm not sure about the training process shown in Figure 8. It appears that the training may not have been completed, given that the blue line representing the chunk stripe is still ascending at the endpoint. The weights depicted in panel d) seem to correspond with those shown in panels b) and c), no? Then, how is the optimization process determined to be finished? Alternatively, could it be stated that these weight differences approach a certain value asymptotically? It would be better to clarify the convergence criteria of the optimization process.

      The training process has been clarified and we specify (in the last paragraph of the Base PBWM Model) how we determine when training is complete. We also can confirm that the network behavior has stabilized in learning even if the Go/NoGo weights continue to grow over time for the chunked layer (due to imperfect performance and reinforcement of the chunk gating strategy).

      Reviewer 2:

      Thank you for your comments. We appreciate your notes on the strengths of the paper and your suggestions to help improve the paper.

      The model employs a spiking neural network, which is relatively complex. Additionally, while this paper validates the effectiveness of chunking strategies used by the brain to enhance working memory efficiency through computational simulations, further comparison with related phenomena observed in cognitive neuroscience experiments on limited working memory capacity, such as the recency effect, is necessary to verify its generalizability.

      Thank you for proposing we add in more connections with human WM. Based on your specific recommendation, we have included the section “Network recapitulates human sequential effects in working memory.” where we discuss recency effects in human working memory and how our model recapitulates this effect. We have also made the connections to human data and human work more explicit throughout the manuscript (Figure 4c). As noted in response to the assessment, we believe our model does make contact with a wide variety of cognitive neuroscience data in human WM, such as the shape of the error distributions,  including probability of recall and variable precision;  attraction to recently presented items,  sensitivity to

      reinforcement history, set-size dependent chunking, recency effects, and dopamine manipulation effects, as well of a range of human data linking capacity limitations to frontostriatal function. It also provides a theoretical proposal for the well established phenomenon of capacity limitations in humans, suggesting that they arise due to difficulty in WM management.

      Recommendations For The Authors:

      Reviewer 1:

      I appreciate the authors' clear discussion of the limitations of this work in the section "Limitations and Future Directions". The development of a comprehensive model framework to overcome these constraints should require a separate paper, though, I am curious if the authors have attempted any experiments, such as using two identically designed chunking layers, that could partially support the assumptions presented in the paper.

      Expanding the number of chunking layers is a great future direction. We felt that it was most effective for this paper to begin with a minimal set up with proof of concept. We hypothesize that, given our results, a reinforcement learning algorithm would be able to learn to select the best level of abstraction (degree of chunking) in more continuous form, but would require more experience across a range of tasks to do so.

      I'm not sure whether it's appropriate that "Frontostriatal Chunking Gating..." precedes "Dopamine Balance is...", maybe it would be better to reverse the order thus avoiding the need to mention the role of dopamine before delving into the details. Additionally, including a summary at the end of the Introduction, outlining how the paper is organized, could provide readers with a clear roadmap of the forthcoming content.

      We appreciate this suggestion. After careful thought, we wanted to preserve the order because we felt it was important to make the direct connection between set size and stripe usage following the discussion on performance based on increasing stripes.  

      The authors could improve the overall polish of the paper. The equations in the Method section are somewhat confusing: Eq. (2) appears incorrect, as it lacks a weight w_i and n should presumably be in the denominator. For Eq. (3), the comma should be replaced with ']'... It would be advisable to cross-reference these equations with the original O'Reilly and Frank paper for consistency.

      Thank you for pointing out the errors in the method equations- those equations were indeed rendering incorrectly. We have fixed this problem.  

      Additionally, there are frequent instances of missing figure and reference citations (many '?'s), and it would be beneficial to maintain consistent citation formatting throughout the paper: sometimes citations are presented as "key/query coding (Traylor, Merullo, Frank, and Pavlick, 2024; see also Swan and Wyble, 2014)", while other times they are written as "function (O'Reilly & Frank, 2006)"...

      Lastly, there is an empty '3.1' section in the supplementary material that should be addressed.

      The citation issues were fixed. The supplementary information was cleaned and the missing section was removed. Thank you for mentioning these errors.  

      Reviewer 2:

      Thank you for the following recommendations and suggestions. We respond to each individual point based on the numbering system used in your review.  

      (1) This paper utilizes the experimental paradigm of visual working memory, in which different visual stimuli are sequentially loaded into the working memory system, and the accuracy of memory for these stimuli is calculated.

      The authors could further plot the memory accuracy curve as the number of items (N) increases, under both chunking and non-chunking strategies. This would allow for the examination of whether memory accuracy suddenly declines at a specific value of N (denoted as Nc), thereby determining the limited capacity of working memory within this experimental framework, which is about 4 different items or chunks. Additionally, it could be investigated whether the value of Nc is larger when the chunking strategy is applied.

      We have included an additional plot (Probability of Recall) as a supplemental figure to Figure 5 to explore the probability of recall as a function of set size for both chunking and no chunking models.  This plot shows that the chunking model increases probability of recall when set size exceeds allocated capacity (but that nevertheless both models show decreases in recall with set size, consistent with the literature).

      (2) The primacy effect or recency effect observed in the experiments and traditional working memory models, including the slot model and the limited resource model, should be examined to see if it also appears in this model.

      The literature on human working memory shows a prevalent recency effect (but not a primacy effect, which is thought to be due to episodic memory, and which is not included in our model). We have added a section showing that our model demonstrates clear recency effects.

      (3) The construction of the model and the single neuron dynamics involved need further refinement and optimization:

      Model Description: The details of the model construction in the paper need to be further elaborated to help other researchers better understand and apply the model in reproducing or extending research. Specifically:

      a) The construction details of different modules in the model (such as Input signal, BG, striatum, superficial PFC, deep PFC) and the projection relationships between different modules. Adding a diagram to illustrate the network construction would be beneficial.

      To aid in the understanding of the model construction and model components, we have included an additional figure (Figure 1: Base Model) that explains the key layers and components of the model.  We have also altered the overall model figures to show more clearly that the inputs project to both PFC and striatum, to highlight that information is temporarily represented in superficial PFC layers even before striatal gating, which is needed for storage after the input decays.

      We have expanded the methods and equations and we also provide a link to the model github for purposes of reproducibility and sharing.  

      A base model figure was added to specify key connections.  

      a) The numbers of excitatory and inhibitory neurons within different modules and the connections between neurons.

      We added clarification on the type of connections between layers (specifying which are fixed and learned). We have also added the size of layers in a new appendix section “Layer Sizes and Inner Mechanics”

      b) The dynamics of neurons in different modules need to be elaborated, including the description of the dynamic equations of variables (such as x) involved in single neuron equations.

      Single neuron dynamics are explained in equations 1-4. Equations 5-6 explain how activation travels between layers. The specific inhibitory dynamics in the chunking layer are elaborated in Figure 4. PBWM Model and Chunking Layer Details. The Appendix section “Neural model  implementational details” states the key equations, neural information and connectivity. Since there is a large corpus of background information underlying these models, we have linked the Emergent github and specifically the Computational Cognitive Neuroscience textbook which has a detailed description of all equations. For the sake of paper length and understability, we chose the most relevant equations that distinguish our model.  

      c) The selection of parameters in the model, especially those that significantly affect the model's performance.

      The appendix section hyperparameter search details some of the key parameters and why those values were chosen.  

      d) The model employs a sequential working memory paradigm, the forms of external stimuli involved in the encoding and recalling phases (including their mathematical expressions, durations, strengths, and other parameters) need to be elaborated further.

      We appreciate this comment. We have expanded the Appendix section “Continuous Stimuli” to include the details of stimuli presentation (including durations etc).  

      (4) The figures in the paper need optimization. For example, the size of the schematic diagram in Figure 2 needs to be enlarged, while the size of text such as "present stimulus 1, 2, recall stimulus 1" needs to be reduced. Additionally, the citation of figures in the main text needs to be standardized. For example, Figure 1b, Figure 1c, etc., are not cited in the main text.

      The task sequence figure (original Figure 2) has been modified and following your suggestions, text sizes have been modified.  

      (5) Section 3.1 in the appendix is missing.

      Supplemental section 3.1 is removed.

    1. eLife Assessment

      This report used a new double knockout mouse model to investigate the role of two neuropeptides, substance P and CGRPa, in pain signaling. There is convincing evidence that double knockout of these two molecules, both of which have historically been associated with pain, does not affect nociception or acute pain behaviors in males and females. This finding is fundamental, as it challenges the hypothesis that these peptides are essential for pain transmission, even when targeted together. This paper will be of interest to those interested in the neurobiology of pain and/or neuropeptide function.

    2. Reviewer #2 (Public review):

      Summary,

      The paper aimed to examine the effect of co-ablating Substance P and CGRPα peptides on pain using Tac1 and Calca double knockout (DKO) mice. The authors observed no significant changes in acute, inflammatory, and neuropathic pain. These results suggest that Substance P and CGRPα peptides do not play a major role in mediating pain in mice. Moreover, they reveal that the lack of behavioral phenotype cannot be explained by the redundancy between the two peptides, which are often co-expressed in the same neuron

      Strengths,

      The paper uses a straightforward approach to address a significant question in the field. The authors confirm the absence of Substance P and CGRPα peptides at the levels of DRG, spinal cord, and midbrain. Subsequently, they employ a comprehensive battery of behavioral tests to examine pain phenotypes, including acute, inflammatory, and neuropathic pain. Additionally, they evaluate neurogenic inflammation by measuring edema and extravasation, revealing no changes in DKO mice. The data are compelling, and the study's conclusions are well-supported by the results. The manuscript is succinct and well-presented.

    3. Reviewer #3 (Public review):

      In this study, the authors aimed to determine the role of a global double knockout (DKO) of substance P and CGRPα in modulating acute and chronic pain transmission. After successfully generating and validating the DKO mouse model, they conducted a series of behavioral pain assessments to evaluate the role of these neuropeptides in acute and chronic pain. Despite the well-established involvement of substance P and CGRPα in chronic pain, their findings revealed that the global loss of both neuropeptides did not affect the transmission of either acute or chronic pain.

      A major strength of the paper is that they validated their double knockout mouse model before using a comprehensive array of both acute and chronic pain tests to reach their conclusions. One minor weakness is that their n numbers for some of the studies conducted are low.

      The conclusions made by the authors are largely supported by their results and the authors successfully achieved their aim of investigating the role of simultaneous inhibition of substance P and CGRPα in pain transmission.

      This study offers valuable insights into our understanding of the pain pathways. Both Substance P and CGRPα neuropeptides and their receptors were considered key players in pain signaling due to their high expression in pain-responsive neurons. However, targeting these peptides in clinical trials has not been successful. By investigating the simultaneous inhibition of substance P and CGRPα through the generation of Tac1 and Calca double knockout (DKO) mice, the authors addressed an important gap in the field. Their comprehensive assessment of pain behaviors across a range of acute and chronic pain models revealed an unexpected outcome: the absence of both neuropeptides did not significantly alter pain responses. This finding is pivotal, as it challenges the hypothesis that these peptides are essential for pain transmission, even when targeted together.

      Comments on revisions:

      All my previous concerns have been addressed.

    4. Author response:

      The following is the authors’ response to the original reviews.

      Reviewer #1 (Public Review):

      MacDonald et al., investigated the consequence of double knockout of substance P and CGRPα on pain behaviors using a newly created mouse model. The investigators used two methods to confirm knockout of these neuropeptides: traditional immunolabeling and a neat in vitro assay where sensory neurons from either wildtype or double knock are co-cultured with substance P "sniffer cells", HEK cells stably expressing NKR1 (a substance P receptor), GCaMP6s and Gα15. It should be noted that functional assays confirming CGRPα knockout were not performed. Subsequently, the authors assayed double knockout mice (DKO) and wildtype (WT) mice in numerous behavioral assays using different pain models, including acute pain and itch stimuli, intraplanar injection of Complete Freund's Adjuvant, prostaglandin E2, capsaicin, AITC, oxaliplatin, as well as the spared nerve injury model. Surprisingly, the authors found that pain behaviors did not differ between DKO and WT mice in any of the behavioral assays or pain paradigms. Importantly, female and male mice were included in all analyses. These data are important and significant, as both substance P and CGRPα have been implicated in pain signaling, though the magnitude of the effect of a single knockout of either gene has been variable and/or small between studies.

      The conclusions of the study are largely supported by the data; however, additional experimental controls and analyses would strengthen the authors claims.

      We thank the reviewer for their insightful comments and have answered them below.

      (1) The authors note that single knockout models of either substance P or CGRPα have produced variable effects on pain behaviors that are study-dependent. Therefore, it would have strengthened the study if the authors included these single knockout strains in a side-by-side analysis (in at least some of the behavioral assays), as has been done in prior studies in the field when using double- or triple-knockout mouse models (for example, see PMID: 33771873). If in the authors hands, single knockouts of either peptide also show no significant differences in pain behaviors, then the finding that double knockouts also do not show significant differences would be less surprising.

      In our study, we found no phenotypic differences between WT and DKO mice, suggesting Substance P and CGRPα are largely dispensable for pain behavior. We agree that if we had we observed significant changes in behavior, it would have been interesting to examine the effects of knocking out each gene individually to determine which peptide is responsible for the phenotype. However, given the double deletion had no effect, we can predict that loss of each alone would have no or minor effects. In line with this, a more recent study that comprehensively phenotyped the Calca KO mouse found no deficits in a range of danger related behaviors (PMID: 34376756). Overall, as we are reporting negative data about the Double KO, we do not believe extensive studies of the single KOs is necessary to support the findings of our paper.

      (2) It is unclear why the authors only show functional validation of substance P knockout using "sniffer" cells, but not CGRPα. Inclusion of this experiment would have added an additional layer of rigor to the study.

      Imaging of CGRPα release is more challenging using the ‘sniffer’ approach because functional CGRP receptors require the expression of two genes: Calcrl (or Calcr) along with Ramp1. We now have succeeded in generating a new stable cell line expressing Calcrl and Ramp1, along with GCaMPs and human Galpha15 and include new data in the revised Figure 1F-H and Figure Supplement 1B. These cells respond robustly to CGRPalpha, but not to SP. In contrast, the existing SP cell line responds to SP but not CGRPalpha. Capsaicin evokes a strong response in these cells in co-culture with DRGs. This response is dramatically reduced in the DKO. This data therefore confirms our mice have a loss of CGRPalpha signaling as indicated by IHC.

      (3) The authors should be a bit more reserved in the claims made in the manuscript. The main claim of the study is that "CGRPα and substance P are not required for pain transmission." However, the authors also note that neuropeptides can have opposing effects that may produce a net effect of no change. In my view, the data presented show that double knockout of substance P and CGRPα do not affect somatic pain behaviors, but do not preclude a role for either of these molecules in pain signaling more generally. Indeed, the authors also note that these neuropeptides could be involved in nociceptor crosstalk with the immune or vascular systems to promote headache. The authors only assayed pain responses to glabrous skin stimulation. How the DKO mice would behave in orofacial pain assays, migraine assays, visceral pain assays, or bone/joint pain assays, for example, was not tested. I do not suggest the authors include these experiments, only that they address the limitations/weaknesses of their study more thoroughly.

      The reviewer makes an important point that we agree with. Our study assesses acute and chronic pain in peptide DKO mice lacking Substance P and CGRPα. Most of our data focuses on the hindpaw as pain in the paw is the gold-standard approach for phenotyping pain targets and numerous well-validated chronic pain models have been developed for this body site.  However, to extend the conclusions to other tissues, we did also look at visceral pain and GI distress using acetic acid and LiCl models (Figure 2J and Figure 2 supplement). We agree with the reviewer that given the utility of CGRP monoclonal antibodies, migraine experiments would be interesting for future studies using these mice, a point we highlight in the discussion. Bone/joint pain is also clearly important from a translational perspective, but outside the scope of the current study.

      (4) A more minor but important point, the authors do not describe the nature of the WT animals used. Are the littermates or a separately maintained colony of WT animals? The WT strain background should be included in the methods section.

      The WT strain are C57/BL6j from Jackson Lab. This has been added to the methods.

      Reviewer #2 (Public Review):

      Summary:

      The paper aimed to examine the effect of co-ablating Substance P and CGRPα peptides on pain using Tac1 and Calca double knockout (DKO) mice. The authors observed no significant changes in acute, inflammatory, and neuropathic pain. These results suggest that Substance P and CGRPα peptides do not play a major role in mediating pain in mice. Moreover, they reveal that the lack of behavioral phenotype cannot be explained by the redundancy between the two peptides, which are often co-expressed in the same neuron

      Strengths:

      The paper uses a straightforward approach to address a significant question in the field. The authors confirm the absence of Substance P and CGRPα peptides at the levels of DRG, spinal cord, and midbrain. Subsequently, they employ a comprehensive battery of behavioral tests to examine pain phenotypes, including acute, inflammatory, and neuropathic pain. Additionally, they evaluate neurogenic inflammation by measuring edema and extravasation, revealing no changes in DKO mice. The data are compelling, and the study's conclusions are well-supported by the results. The manuscript is succinct and well-presented.

      We thank the reviewer for their enthusiasm for the importance of our work.

      Reviewer #3 (Public Review):

      In this study, the authors were assessing the role of double global knockout of substance P and CGPRα on the transmission of acute and chronic pain. The authors first generated the double knockout (DKO) mice and validated their animal model. This is then followed by a series of acute and chronic pain assessments to evaluate if the global DKO of these neuropeptides are important in modulating acute and chronic pain behaviors. Authors found that these DKO mice Substance P and CGRPα are not required for the transmission of acute and chronic pain although both neuropeptides are strongly implicated in chronic pain. This study does provide more insight into the role of these neuropeptides on chronic pain processing, however, more work still needs to be done. (see the comments below).

      We thank the reviewer for their detailed and constructive feedback, and below outline the steps we have taken to answer their concerns.

      (1) In assessing the double KO (result #1), why are different regions of the brains shown for substance P and CGRPα (for example, midbrain for substance P and amygdala for CGRPα)? Since the authors mentioned that these peptides co-expressed in the brain (as in the introduction), shouldn't the same brain regions be shown for both IHC? It would be ideal if the authors could show both regions (midbrain and amygdala) in addition to the DRG and spinal cord for both peptides in their findings.<br /> In addition, since this is double KO, the authors should show more representative IHC-stained brain regions (spanning from the anterior to posterior).

      We could not co-stain both SP and CGRP in the same sections as the DKO mouse has endogenous GFP and RFP fluorescence, limiting us to one channel (far red). Specifically, we use a Calca KO that is a Cre:GRP knock-in/knockout (Chen et al 2018, PMID30344042) and Tac1 KO is a tagRFP knock-in/knockout (Wu et al 2018 PMID29485996). This is why we show different brain sections.

      (2) It is also unclear as to why the authors only assessed the loss of substance P signaling in the double KO mice. Shouldn't the same be done for CGRPα signaling? Either the authors assess this, or the authors have to provide clear explanations as to why only substance P signaling was assessed.

      As noted in our response to Reviewer 1, imaging of CGRP release is more challenging using the ‘sniffer’ approach because functional CGRP receptors require the expression of two genes: Calcrl (or Calcr) along with Ramp1. We have now generated this cell line and performed the experiment (see revised Figure 1 and Figure 1 Supplement).

      (3) Has these animal's naturalistic behavior been assessed after the double KO (food intake, sleep, locomotion for example)? I think this is important as changes to these naturalistic behaviors can affect pain processes or outcomes.

      We agree that assessment of naturalistic behavior including food intake, sleep and locomotion would be interesting to look at in DKO mice. However, our study is focused on acute and chronic pain behavior of these animals, and therefore a comprehensive phenotypic assessment of naturalistic home-cage behavior is outside the scope of our study.

      (4) Figure 2H: The authors acknowledge that there is a trend to decrease with capsaicin-evoked coping-like responses. However, a close look at the graph suggests that the lack of significance could be driven by 1 mouse. Have the authors run an outlier test? Alternatively, the authors should consider adding more n to these experiments to verify their conclusions.

      We were reluctant to add more animals searching for significance. Instead, we investigated the potential phenotype further by looking at cfos staining in the cord and found no differences (Figure 2, supplement 1). This result suggests loss of the two peptides does not grossly disrupt capsaicin evoked pain signal transmission between the nociceptor and post-synaptic dorsal neurons in the spinal cord.

      (5) Similarly, the values for WT in the evoked cFos activity (Figure 2- Suppl Figure 1) are pretty variable. Considering that the n number is low (n = 5), authors should consider adding more n.<br /> Also, since the n number is low in this experiment (eg. 5 vs 4), does this pass the normality test to run a parametric unpaired t-test? Either the authors increase their n numbers or run the appropriate statistical test.

      As described in the statistical tables, the Shapiro-Wilk test indicates these data do pass the normality test. Therefore, we retain the use of the unpaired t test, which demonstrates no significant difference between the groups.

      (6) In most of the results, authors ran a parametric test despite the low n number. Authors have to ensure that they are carrying out the appropriate statistical test for their dataset and n number.

      We now provide a table of the statistical results, which provides detailed information about all statistical tests performed in this study. For experiments where we make a single comparison between the two distributions (WT vs DKO), we have run a Shapiro-Wilk test. Where the data from both groups pass the normality test, we retain the use of the unpaired t test. Where the Shapiro-Wilk test indicates data from either group are unlikely to be normally distributed, we now use a Mann-Whitney U test to compare the groups, as this non-parametric test makes no assumptions about the underlying distribution.

      Many experiments involved two factors (genotype, and e.g. temperature, drug, time-point). These data were analyzed in the original submission using 2-WAY ANOVA or Repeated Measures 2-WAY ANOVA, followed by post-hoc Sidak’s tests to compute p values adjusted for multiple comparisons. Because there is no widely agreed non-parametric alternative to 2-WAY ANOVA for analyzing data with two factors and that enables us to account for multiple comparisons, we used 2-WAY ANOVA as is typically used in the field for these kinds of experiments. We reasoned sticking with the 2-WAY ANOVA was the best course of action based on information provided by the statistical software used for this study - https://www.graphpad.com/support/faq/with-two-way-anova-why-doesnt-prism-offer-a-nonparametric-alternative-test-for-normality-test-for-homogeneity-of-variances-test-for-outliers/

      We note that regardless of the test, our conclusion that there are no major changes in acute or chronic pain behaviors are clear and strongly supported.

      (7) Along the same line of comment with the previous, authors should increase the n number for DKO for staining (Figure 4) as n number is only 3 and there is variability in the cFos quantification in the ipsilateral side.

      We believe this is not necessary as the finding is clear that there is no difference.

      (8) Authors should provide references for statement made in Line 319-321 as authors mentioned that there are accumulating evidence indicating that secretion of these neuropeptides from nociceptor peripheral terminals modulates immune cells and the vasculature in diverse tissues.

      We now provide several references to primary papers and reviews supporting this statement.

      (9) Authors state that the sample size used was similar to those from previous studies, but no references were provided. Also, even though the sample sizes used were similar, I believe that the right statistic test should be used to analyze the data.

      We have now cited several classic studies phenotyping mouse KOs in pain in the methods that used similar sample sizes. As detailed above, we have taken the reviewer’s feedback on board and performed normality testing to ensure the correct statistical test is used for each experiment.

      (10) In the discussion, the authors noted that knocking out of a gene remains the strongest test of whether the molecule is essential for a biological phenomenon. At the same time, it was acknowledged that Substance P infusion into the spinal cord elicits pain, but it is analgesic in the brain. The authors might want to expand more on this discussion, including how we can selectively assess the role of these neuropeptides in areas of interest. For example, knocking out both Substance P and CGRPα in selected areas instead of the global KO since there are reported compensatory effects.

      This is highlighted in the closing paragraph: “Emerging approaches to image and manipulate these molecules (Girven et al., 2022; Kim et al., 2023), as well as advances in quantitating pain behaviors (Bohic et al., 2023; MacDonald and Chesler, 2023), may ultimately reveal the fundamental roles of neuropeptides in generating our experience of pain.” The Kim preprint (now published, and so the citation has been updated in the text) describes a method of inactivating neuropeptide transmission in select brain regions in a cell-type specific manner.

      Recommendations for the authors:

      Reviewer #2 (Recommendations For The Authors):

      I do not have any major comments. My minor comments are as follows:

      (1) What was the control group for all behavioral studies? Was it WT from an independent colony or one of the littermates was used for generating controls?

      We used C57/Bl6 mice from Jax. This is now mentioned in methods.

      (2) In Fig. 2H, it seems that the effect will become significant if several mice are added.

      We are reluctant to add mice searching for significance. Sample sizes were determined before we collected the data blind.

      (3) There is no figure 3, but two figures 4.

      Thank you. This has been corrected.

      (4) Multiple typos in the legend for figure 4 (lines 234-254). Line 242 (& n=8 (3M, 3F)), line 243 (swelling and plasma), line 252 ((n=8 for) & n=6 for DKO (4M, 4F)).

      Thank you. This has been corrected.

      (5) In Figure 4 (lines 273-285), the contralateral side is mentioned in B but no images are shown.

      Thank you. We removed the mention.

      (6) Although ligand knockouts cannot be compared directly with receptor inhibition, the readers could benefit from discussing studies of receptor ablation and/or pharmacological inhibition.

      We do discuss the classic studies of receptor KO, and the clinical data on receptor blockers here –

      “However, selective antagonists of the Substance P receptor NKR1 failed to relieve chronic pain in human clinical trials (Hill, 2000). Although CGRP monoclonal antibodies and receptor blockers have proven effective for subsets of migraine patients, their usefulness for other types of pain in humans is unclear (De Matteis et al., 2020; Jin et al., 2018). In line with this, knockout mice deficient in Substance P, CGRPα or their receptors have been reported to display some pain deficits, but the analgesic effects are neither large nor consistent between studies (Cao et al., 1998; De Felipe et al., 1998; Guo et al., 2012; Salmon et al., 2001, 1999; Zimmer et al., 1998).” 

      Reviewer #3 (Recommendations For The Authors):

      Minor comments:

      (1) Figure 1E: What does chambers mean? Additionally, are the 12 chambers equally from the male and female samples (6 from male and 6 from female)?

      We have changed this to well. Each replicate is an individual well from 8 well chamber slide. In all these experiments, the wells are approximately evenly distributed by mouse, because from each mouse we cultured around 8 wells’ worth of DRGs.

      (2) Figure 1D: What does low and high mean in the Hargreaves test?

      These refer to a low and high active intensity of the radiant heat stimulus. Number is now described in the methods. 40 and 55 in the intensity units used by the instrument.

      (3) Figure 2-Suppl Figure 1: Authors should provide a bigger image of the image so that it is clearer to the readers.

      We think the image is of a reasonable size and comparable to the images used elsewhere in the paper.

      (4) Authors should consider labeling their supplementary figures in running numbers or combining supplementary figures together to avoid confusion. For example, Figure 2-Supplementary Figure 1 and Figure 2- Supplementary Figure 2 can be combined as just Supplementary Figure 2.

      We agree with the reviewer this would be clearer, but we have followed eLife’s convention for labelling and numbering supplements.

      (5) Figure 3 is mislabeled as Figure 4.

      Thank you. We have corrected this.

      (6) Only female mice were used in the CFA experiment, which does not go in line with the rest of the results which consist of both sexes.

      We have repeated the experiment with additional male mice. To be consistent with the von frey data, these were followed for 7 days, and so the figure now shows a 7 day time course.

      (7) Typo in line 243. The word "and" is subscript.

      Thank you. We have corrected this.

      (8) There is a typo in the legend for Figure 4 where E is labeled I, G is labeled as F, and J is labeled as J.

      Thank you. We have corrected this.

      (9) Authors should specify what "several weeks" means (Line 263).

      It means three weeks. We tested to 21 days. We will replace with three.

      (10) Authors should specify what "one day" means (Line 267). For example, how many days after the intraplantar oxaliplatin treatment? Also, authors should justify why that specific time point was selected or have a reference for it.

      This means one day after - 24 hours. Please see PMID: 33693512. Two references are provided in them methods.

      (11) Figure 4 legend: authors should again be specific on what "prolonged" entails (Line 277).

      We have replaced prolonged with 30 minutes brushing. Specifically, 3 x 10 min stim period, with 1 min rest between stim. It is in the methods.

      (12) In the methods section, authors state that both male and female mice were used for all experiments. However, only female mice were used in the CFA experiment (see minor comment #6). Authors should verify and correct this.

      This is correct. We only used female mice for one of the groups. We have since repeated with males, now included in the data.

      (13) Authors should be more specific in the methods section on how long the habituation per day, how many days and what were the mice habituation to (experimenter, room, chamber, etc)?

      As noted in the methods, mice are habituated for at least an hour to the chambers, and thus implicitly to the room. We do not perform explicit habituation to the investigator such as repeated handling.

      (14) Authors need to provide more information on the semi-automated procedure they are referring to in Line 397. Also, authors should also provide the criteria for cFos quantification (eg. Intensity, etc). If this has been published before, they should provide the reference.

      We have added this. We used the ‘Find maxima’ and ‘Analyze particles’ functions in FIJI, followed by a manual curation step.

      (15) How much acetone was applied and how was it applied to the paw? (Line 495)

      We used the same applicator (1ml syringe with a well at the top) to generate a droplet of acetone that was used for all mice. This has been added to methods.

      (16) Authors should specify the amount of capsaicin injected in μl (Line 500).

      20 ul. We have added this.

      (17) Authors should explain or reference why they are analyzing the 15 min interval between 5 and 20 minutes for injection (Line507-508).

      Acetic acid behaviour lasts around 30 mins in our hands. We chose the 15 minute interval because it reduces burdensome hand scoring time by 50% versus doing the whole 30 mins. We reasoned that in the first 5 mins post injection the animal behaviour may be contaminated by stress related to handling, injection and return to chamber. Thus, 5 and 20 minutes provided a sensible time-frame for scoring the behavior when it is at its peak.

      (18) Authors have to provide more information/explanation on how they decide on the conditioned taste aversion protocol. Like why they do 30 mins exposure to a single water-containing bottle followed 90 mins exposure to both bottles. If this has been published before, they should provide the reference.

      We read dozens of different published protocols in the literature, and piloted one that was something of an amalgam of some of them with various adaptations of convenience. Because it worked on our first attempt, we stuck to it. The advantage of the CTA assay is it is incredibly robust to changes in the specificities of the paradigm, evincing the clear survival value of learning to avoid tastes that make you sick.

      (19) Authors again should provide more detail in their methods section.

      a. Specify the time frame that they are assessing here (Line 533).

      This can be seen in the Figure. 0 to 120 mins. We have added it to the methods.

      b. How long were the mice allowed to recover post-SNI before mechanical allodynia was assessed (Line 545)?

      This is apparent in the figures. 2 days to 21 days. We have added it to the methods.

      c. How much of the oxaliplatin was injected into the mice?

      40 ug / 40 ul (see PMID:33693512)

      Editors note: Reviewers agreed that addressing the concerns about power, outliers, and statistics, as well as functional validation of CGRPα would raise the strength of evidence to compelling, and inclusion of comparison to single KO would raise it to exceptional.

      Should you choose to revise your manuscript, please check to ensure full statistical reporting including exact p-values wherever possible alongside the summary statistics (test statistic and df) and 95% confidence intervals. These should be reported for all key questions and not only when the p-value is less than 0.05.

    1. eLife Assessment

      This important study provides convincing data from in vitro models and patient-derived samples to demonstrate how modulation of GSK3 activity can reprogram macrophages, revealing potential therapeutic applications in inflammatory diseases such as severe COVID-19. The study stands out for its clear and systematic presentation, strong experimental approach, and the relevance of its findings to the field of immunology.

    2. Reviewer #1 (Public review):

      The manuscript by Rios et al. investigates the potential of GSK3 inhibition to reprogram human macrophages, exploring its therapeutic implications in conditions like severe COVID-19. The authors present convincing evidence that GSK3 inhibition shifts macrophage phenotypes from pro-inflammatory to anti-inflammatory states, thus highlighting the GSK3-MAFB axis as a potential therapeutic target. Using both GM-CSF- and M-CSF-dependent monocyte-derived macrophages as model systems, the study provides extensive transcriptional, phenotypic, and functional characterizations of these reprogrammed cells. The authors further extend their findings to human alveolar macrophages derived from patient samples, demonstrating the clinical relevance of GSK3 inhibition in macrophage biology.

      The experimental design is sound, leveraging techniques such as RNA-seq, flow cytometry, and bioenergetic profiling to generate a comprehensive dataset. The study's integration of multiple model systems and human samples strengthens its impact and relevance. The findings not only offer insights into macrophage plasticity but also propose novel therapeutic strategies for macrophage reprogramming in inflammatory diseases.

      Strengths:

      (1) Robust Experimental Design: The use of both in vitro and ex vivo models adds depth to the findings, making the conclusions applicable to both experimental and clinical settings.<br /> (2) Thorough Data Analysis: The extensive use of RNA-seq and gene set enrichment analysis (GSEA) provides a clear transcriptional signature of the reprogrammed macrophages.<br /> (3) Relevance to Severe COVID-19: The study's focus on macrophage reprogramming in the context of severe COVID-19 adds clinical significance, especially given the relevance of macrophage-driven inflammation in this disease.

      Weaknesses:

      There are no significant weaknesses in the study.

    3. Reviewer #2 (Public review):

      Summary:

      The study by Rios and colleagues provides the scientific community with a compelling exploration of macrophage plasticity and its potential as a therapeutic target. By focusing on the GSK3-MAFB axis, the authors present a strong case for macrophage reprogramming as a strategy to combat inflammatory and fibrotic diseases, including severe COVID-19. Using a robust and comprehensive methodology, in this study it is conducted a broad transcriptomic and functional analyses and offers valuable mechanistic insights while highlighting its clinical relevance

      Strengths:

      Well performed and analyzed

      Weaknesses:

      Additional analyses, including mechanistic studies, would increase the value of the study.

    4. Author response:

      Regarding a future revised version, we plan to:

      • refer to the "MoMac-VERSE" study according to the original report.

      • modify incorrectly formatted references.

      • modify the text to acknowledge the heterogeneity and variability in the response of primary cells to the GSK3 inhibitor.

      • improve the explanation of the reanalysis of single cell RNAseq data in Figure 7 (ref. 47, GSE120833), and re-adapt the graphs of the scRNA-Seq data using different plot parameters (e.g., reduction = "umap.scvi") to provide a more friendly-user visualization including bona fide macrophage markers for each subpopulation.

      • include statistical analyses in each one of the figure legends

      • perform additional analyses (e.g., dose-response and kinetics of CHIR-99021 effects) and mechanistic studies (e.g., role of proteasome) to further dissect the re-programming ability of the GSK3/MAFB axis.

    1. eLife Assessment

      This study provides valuable insights into the behavioral, computational, and neural mechanisms of regime shift detection, by identifying distinct roles for the frontoparietal network and ventromedial prefrontal cortex in sensitivity to signal diagnosticity and transition probabilities, respectively. The findings are supported by solid evidence, including an innovative task design, robust behavioral modeling, and well-executed model-based fMRI analyses, though claims of neural selectivity would benefit from more rigorous statistical comparisons. Overall, this work advances our understanding of how humans adapt belief updating in dynamic environments and offers a framework for exploring biases in decision-making under uncertainty.

    2. Reviewer #1 (Public review):

      Summary:

      The study examines human biases in a regime-change task, in which participants have to report the probability of a regime change in the face of noisy data. The behavioral results indicate that humans display systematic biases, in particular, overreaction in stable but noisy environments and underreaction in volatile settings with more certain signals. fMRI results suggest that a frontoparietal brain network is selectively involved in representing subjective sensitivity to noise, while the vmPFC selectively represents sensitivity to the rate of change.

      Strengths:

      (1) The study relies on a task that measures regime-change detection primarily based on descriptive information about the noisiness and rate of change. This distinguishes the study from prior work using reversal-learning or change-point tasks in which participants are required to learn these parameters from experiences. The authors discuss these differences comprehensively.

      (2) The study uses a simple Bayes-optimal model combined with model fitting, which seems to describe the data well.

      (3) The authors apply model-based fMRI analyses that provide a close link to behavioral results, offering an elegant way to examine individual biases.

      Weaknesses:

      My major concern is about the correlational analysis in the section "Under- and overreactions are associated with selectivity and sensitivity of neural responses to system parameters", shown in Figures 5c and d (and similarly in Figure 6). The authors argue that a frontoparietal network selectively represents sensitivity to signal diagnosticity, while the vmPFC selectively represents transition probabilities. This claim is based on separate correlational analyses for red and blue across different brain areas. The authors interpret the finding of a significant correlation in one case (blue) and an insignificant correlation (red) as evidence of a difference in correlations (between blue and red) but don't test this directly. This has been referred to as the "interaction fallacy" (Niewenhuis et al., 2011; Makin & Orban de Xivry 2019). Not directly testing the difference in correlations (but only the differences to zero for each case) can lead to wrong conclusions. For example, in Figure 5c, the correlation for red is r = 0.32 (not significantly different from zero) and r = 0.48 (different from zero). However, the difference between the two is 0.1, and it is likely that this difference itself is not significant. From a statistical perspective, this corresponds to an interaction effect that has to be tested directly. It is my understanding that analyses in Figure 6 follow the same approach.

      Relevant literature on this point is:

      Nieuwenhuis, S, Forstmann, B & Wagenmakers, EJ (2011). Erroneous analyses of interactions in neuroscience: a problem of significance. Nat Neurosci 14, 1105-1107. https://doi.org/10.1038/nn.2886

      Makin TR, Orban de Xivry, JJ (2019). Science Forum: Ten common statistical mistakes to watch out for when writing or reviewing a manuscript. eLife 8:e48175. https://doi.org/10.7554/eLife.48175

      There is also a blog post on simulation-based comparisons, which the authors could check out: https://garstats.wordpress.com/2017/03/01/comp2dcorr/

      I recommend that the authors carefully consider what approach works best for their purposes. It is sometimes recommended to directly compare correlations based on Monte-Carlo simulations (cf Makin & Orban). It might also be appropriate to run a regression with the dependent variable brain activity (Y) and predictors brain area (X) and the model-based term of interest (Z). In this case, they could include an interaction term in the model:

      Y = \beta_0 + \beta_1 \cdot X + \beta_2 \cdot Z + \beta_3 \cdot X \cdot Z

      The interaction term reflects if the relationship between the model term Z and brain activity Y is conditional on the brain area of interest X.

      Another potential concern is that some important details about the parameter estimation for the system-neglect model are missing. In the respective section in the methods, the authors mention a nonlinear regression using Matlab's "fitnlm" function, but it remains unclear how the model was parameterized exactly. In particular, what are the properties of this nonlinear function, and what are the assumptions about the subject's motor noise? I could imagine that by using the inbuild function, the assumption was that residuals are Gaussian and homoscedastic, but it is possible that the assumption of homoscedasticity is violated, and residuals are systematically larger around p=0.5 compared to p=0 and p=1.

      Relatedly, in the parameter recovery analyses, the authors assume different levels of motor noise. Are these values representative of empirical values?

      The main study is based on N=30 subjects, as are the two control studies. Since this work is about individual differences (in particular w.r.t. to neural representations of noise and transition probabilities in the frontoparietal network and the vmPFC), I'm wondering how robust the results are. Is it likely that the results would replicate with a larger number of subjects? Can the two control studies be leveraged to address this concern to some extent?

      It seems that the authors have not counterbalanced the colors and that subjects always reported the probability of the blue regime. If so, I'm wondering why this was not counterbalanced.

    3. Reviewer #2 (Public review):

      Summary:

      This paper focuses on understanding the behavioral and neural basis of regime shift detection, a common yet hard problem that people encounter in an uncertain world. Using a regime-shift task, the authors examined cognitive factors influencing belief updates by manipulating signal diagnosticity and environmental volatility. Behaviorally, they have found that people demonstrate both over and under-reaction to changes given different combinations of task parameters, which can be explained by a unified system-neglect account. Neurally, the authors have found that the vmPFC-striatum network represents current belief as well as belief revision unique to the regime detection task. Meanwhile, the frontoparietal network represents cognitive factors influencing regime detection i.e., the strength of the evidence in support of the regime shift and the intertemporal belief probability. The authors further link behavioral signatures of system neglect with neural signals and have found dissociable patterns, with the frontoparietal network representing sensitivity to signal diagnosticity when the observation is consistent with regime shift and vmPFC representing environmental volatility, respectively. Together, these results shed light on the neural basis of regime shift detection especially the neural correlates of bias in belief update that can be observed behaviorally.

      Strengths:

      (1) The regime-shift detection task offers a solid ground to examine regime-shift detection without the potential confounding impact of learning and reward. Relatedly, the system-neglect modeling framework provides a unified account for both over or under-reacting to environmental changes, allowing researchers to extract a single parameter reflecting people's sensitivity to changes in decision variables and making it desirable for neuroimaging analysis to locate corresponding neural signals.

      (2) The analysis for locating brain regions related to belief revision is solid. Within the current task, the authors look for brain regions whose activation covary with both current belief and belief change. Furthermore, the authors have ruled out the possibility of representing mere current belief or motor signal by comparing the current study results with two other studies. This set of analyses is very convincing.

      (3) The section on using neuroimaging findings (i.e., the frontoparietal network is sensitive to evidence that signals regime shift) to reveal nuances in behavioral data (i.e., belief revision is more sensitive to evidence consistent with change) is very intriguing. I like how the authors structure the flow of the results, offering this as an extra piece of behavioral findings instead of ad-hoc implanting that into the computational modeling.

      Weaknesses:

      (1) The authors have presented two sets of neuroimaging results, and it is unclear to me how to reason between these two sets of results, especially for the frontoparietal network. On one hand, the frontoparietal network represents belief revision but not variables influencing belief revision (i.e., signal diagnosticity and environmental volatility). On the other hand, when it comes to understanding individual differences in regime detection, the frontoparietal network is associated with sensitivity to change and consistent evidence strength. I understand that belief revision correlates with sensitivity to signals, but it can probably benefit from formally discussing and connecting these two sets of results in discussion. Relatedly, the whole section on behavioral vs. neural slope results was not sufficiently discussed and connected to the existing literature in the discussion section. For example, the authors could provide more context to reason through the finding that striatum (but not vmPFC) is not sensitive to volatility.

      (2) More details are needed for behavioral modeling under the system-neglect framework, particularly results on model comparison. I understand that this model has been validated in previous publications, but it is unclear to me whether it provides a superior model fit in the current dataset compared to other models (e.g., a model without \alpha or \beta). Relatedly, I wonder whether the final result section can be incorporated into modeling as well - i.e., the authors could test a variant of the model with two \betas depending on whether the observation is consistent with a regime shift and conduct model comparison.

    4. Author response:

      eLife Assessment

      This study provides valuable insights into the behavioral, computational, and neural mechanisms of regime shift detection, by identifying distinct roles for the frontoparietal network and ventromedial prefrontal cortex in sensitivity to signal diagnosticity and transition probabilities, respectively. The findings are supported by solid evidence, including an innovative task design, robust behavioral modeling, and well-executed model-based fMRI analyses, though claims of neural selectivity would benefit from more rigorous statistical comparisons. Overall, this work advances our understanding of how humans adapt belief updating in dynamic environments and offers a framework for exploring biases in decision-making under uncertainty.

      Thank you for reviewing our manuscript. We appreciate the editors’ assessment and the reviewers’ constructive comments. Below we address the reviewers’ comments. In particular, we addressed Reviewer 1’s comments on (1) neural selectivity by performing statistical comparisons and (2) parameter estimation by providing more details on how the system-neglect model was parameterized. We addressed Reviewer 2’s comments on (1) our neuroimaging results regarding frontoparietal network and (2) model comparisons.  

      Public Reviews:

      Reviewer #1 (Public review):

      Summary:

      The study examines human biases in a regime-change task, in which participants have to report the probability of a regime change in the face of noisy data. The behavioral results indicate that humans display systematic biases, in particular, overreaction in stable but noisy environments and underreaction in volatile settings with more certain signals. fMRI results suggest that a frontoparietal brain network is selectively involved in representing subjective sensitivity to noise, while the vmPFC selectively represents sensitivity to the rate of change.

      Strengths:

      (1) The study relies on a task that measures regime-change detection primarily based on descriptive information about the noisiness and rate of change. This distinguishes the study from prior work using reversal-learning or change-point tasks in which participants are required to learn these parameters from experiences. The authors discuss these differences comprehensively.

      Thank you for recognizing our contribution to the regime-change detection literature and our effort in discussing our findings in relation to the experience-based paradigms.

      (2) The study uses a simple Bayes-optimal model combined with model fitting, which seems to describe the data well.

      Thank you for recognizing the contribution of our Bayesian framework and system-neglect model.

      (3) The authors apply model-based fMRI analyses that provide a close link to behavioral results, offering an elegant way to examine individual biases.

      Thank you for recognizing our execution of model-based fMRI analyses and effort in using those analyses to link with behavioral biases.

      Weaknesses:

      My major concern is about the correlational analysis in the section "Under- and overreactions are associated with selectivity and sensitivity of neural responses to system parameters", shown in Figures 5c and d (and similarly in Figure 6). The authors argue that a frontoparietal network selectively represents sensitivity to signal diagnosticity, while the vmPFC selectively represents transition probabilities. This claim is based on separate correlational analyses for red and blue across different brain areas. The authors interpret the finding of a significant correlation in one case (blue) and an insignificant correlation (red) as evidence of a difference in correlations (between blue and red) but don't test this directly. This has been referred to as the "interaction fallacy" (Niewenhuis et al., 2011; Makin & Orban de Xivry 2019). Not directly testing the difference in correlations (but only the differences to zero for each case) can lead to wrong conclusions. For example, in Figure 5c, the correlation for red is r = 0.32 (not significantly different from zero) and r = 0.48 (different from zero). However, the difference between the two is 0.1, and it is likely that this difference itself is not significant. From a statistical perspective, this corresponds to an interaction effect that has to be tested directly. It is my understanding that analyses in Figure 6 follow the same approach.

      Relevant literature on this point is:

      Nieuwenhuis, S, Forstmann, B & Wagenmakers, EJ (2011). Erroneous analyses of interactions in neuroscience: a problem of significance. Nat Neurosci 14, 1105-1107. https://doi.org/10.1038/nn.2886

      Makin TR, Orban de Xivry, JJ (2019). Science Forum: Ten common statistical mistakes to watch out for when writing or reviewing a manuscript. eLife 8:e48175. https://doi.org/10.7554/eLife.48175

      There is also a blog post on simulation-based comparisons, which the authors could check out: https://garstats.wordpress.com/2017/03/01/comp2dcorr/

      I recommend that the authors carefully consider what approach works best for their purposes. It is sometimes recommended to directly compare correlations based on Monte-Carlo simulations (cf Makin & Orban). It might also be appropriate to run a regression with the dependent variable brain activity (Y) and predictors brain area (X) and the model-based term of interest (Z). In this case, they could include an interaction term in the model:

      Y = \beta_0 + \beta_1 \cdot X + \beta_2 \cdot Z + \beta_3 \cdot X \cdot Z

      The interaction term reflects if the relationship between the model term Z and brain activity Y is conditional on the brain area of interest X.

      Thank you for this great suggestion. We tested the difference in correlation both parametrically and nonparametrically. Their results were identical. In our parametric test, we used the Fisher z transformation to transform the difference in correlation coefficients to the z statistic (Fisher, 1921). That is, for two correlation coefficients, r<sub>blue</sub> (the correlation between behavioral slope, and neural slope estimated at change-consistent signals; sample size n<sub>blue</sub>) and  r<sub>red</sub>, (the correlation between behavioral slope, and neural slope estimated at change-consistent signals; sample size n<sub>red</sub>), the z statistic of the difference in correlation is given by

      We found that among the five ROIs in the frontoparietal network, two of them, namely the left IFG and left IPS, the difference in correlation was significant (one-tailed z test; left IFG: z=1.8355, p=0.0332; left IPS: z=2.3782, p=0.0087). For the remaining three ROIs, the difference in correlation was not significant (dmPFC: z=0.7594, p=0.2238 ; right IFG: z=0.9068, p=0.1822; right IPS: z=1.3764, p=0.0843). We chose one-tailed test because we already know the correlation under the blue signals was significantly greater than 0. Hence the alternative hypothesis is that r<sub>blue</sub>r<sub>red</sub>>0.

      In our nonparametric test, we performed nonparametric bootstrapping to test for the difference in correlation. That is, we resampled with replacement the dataset (subject-wise) and used the resampled dataset to compute the difference in correlation. We then repeated the above for 100,000 times so as to obtain the distribution of the correlation difference. We then tested for significance and estimated p-value based on this distribution. Consistent with our parametric tests, here we also found that the difference in correlation was significant in left IFG and left IPS (left IFG: r<sub>blue</sub>r<sub>red</sub>=0.46, p=0.0496; left IPS: r<sub>blue</sub>r<sub>red</sub>=0.5306, p=0.0041), but was not significant in dmPFC, right IFG, and right IPS (dmPFC: r<sub>blue</sub>r<sub>red</sub>=0.1634, p=0.1919; right IFG: r<sub>blue</sub>r<sub>red</sub>=0.2123, p=0.1681; right IPS: r<sub>blue</sub>r<sub>red</sub>=0.3434, p=0.0631).

      We will update these results in the revised manuscript. In summary, we found that the left IFG and left IPS in the frontoparietal network differentially responded to signals consistent with change (blue signals) compared with signals inconsistent with change (red signals). First, the neural sensitivity to signal diagnosticity measured when signals consistent with change appeared (blue signals) significantly correlated with individual subjects’ behavioral sensitivity to signal diagnosticity (r<sub>blue</sub>). By contrast, neural sensitivity to signal diagnosticity measured when signals inconsistent with change appeared did not significantly correlate with behavioral sensitivity (r<sub>red</sub>). Second, the difference in correlation, r<sub>blue</sub>r<sub>red</sub>, was statistically significant between correlation obtained at signals consistent with change and correlation obtained at signals inconsistent with change.

      Another potential concern is that some important details about the parameter estimation for the system-neglect model are missing. In the respective section in the methods, the authors mention a nonlinear regression using Matlab's "fitnlm" function, but it remains unclear how the model was parameterized exactly. In particular, what are the properties of this nonlinear function, and what are the assumptions about the subject's motor noise? I could imagine that by using the inbuild function, the assumption was that residuals are Gaussian and homoscedastic, but it is possible that the assumption of homoscedasticity is violated, and residuals are systematically larger around p=0.5 compared to p=0 and p=1. Relatedly, in the parameter recovery analyses, the authors assume different levels of motor noise. Are these values representative of empirical values?

      We thank the reviewer for this excellent point. The reviewer touched on model parameterization, assumption of noise, and parameter recovery analysis, which we answered below.

      On our model was parameterized

      We parameterized the model according to the system-neglect model in Eq. (2) and estimated the alpha parameter separately for each level of transition probability and the beta parameter separately for each level of signal diagnosticity. As a result, we had a total of 6 parameters (3 alpha and 3 beta parameters) in the model. The system-neglect model is then called by fitnlm so that these parameters can be estimated. The term ‘nonlinear’ regression in fitnlm refers to the fact that you can specify any model (in our case the system-neglect model) and estimate its parameters when calling this function. In our use of fitnlm, we assume that the noise is Gaussian and homoscedastic (the default option).

      On the assumptions about subject’s motor noise

      We wish to emphasize that we did not call the noise ‘motor’ because it can be estimation noise as well. Regardless, in the context of fitnlm, we assume that the noise is Gaussian and homoscedastic.

      On the possibility that homoscedasticity is violated

      In the revision, we plan to examine this possibility (residuals larger around p=0.5 compared with p=0 and p=1).

      On whether the noise levels in parameter recovery analysis are representative of empirical values

      To address the reviewer’s question, we conducted a new analysis using maximum likelihood estimation to estimate the noise level of each individual subject. We proceeded in the following steps. First, for each subject separately, we used the parameter estimates of the system-neglect model to compute the period-wise probability estimates of regime shift. As a reminder, we referred to a ‘period’ as the time when a new signal appeared during a trial (for a given transition probability and signal diagnosticity). Each trial consisted of 10 successive periods. Second, we computed the period-wise likelihood, the probability of observing the subject’s actual probability estimate given the probability estimate predicted by the system-neglect model and the noise level. Here we define noise as the standard deviation of a Gaussian distribution centered at the model-predicted probability estimate. We then summed over all periods the negative logarithm of likelihood and used MATLAB’s minimization algorithm (the ‘fmincon’ function) to obtain the noise estimate that minimized the sum of negative log likelihood (thus the noise estimate that maximized the sum of log likelihood). Across subjects, we found that the mean noise estimate was 0.1480 and ranged from 0.0816 to 0.3239. The noise estimate of each subject can be seen in the figure below.

      Author response image 1.

      Compared with our original parameter recovery analysis where the maximum noise level was set at 0.1, our data indicated that some subjects’ noise was larger than this value. Therefore, we expanded our parameter recovery analysis to include noise levels beyond 0.1 to up to 0.3. We found good parameter recovery across different levels of noise, with the Pearson correlation coefficient between the input parameter values used to simulate data and the estimated parameter values greater 0.95 (Supplementary Fig. S3). The results will be updated in Supplementary Fig. S3.

      Author response image 2.

      Parameter recovery. We simulated probability estimates according to the system-neglect model. We used each subject’s parameter estimates as our choice of parameter values used in the simulation. Using simulated data, we estimated the parameters (𝛼 and 𝛽) in the system-neglect model. To examine parameter recovery, we plotted the parameter values we used to simulate the data against the parameter estimates we obtained based on simulated data and computed their Pearson correlation. Further, we added different levels of Gaussian white noise with standard deviation 𝜎 = 0.01, 0.05, 0.1,0.2, 0.3 to the simulated data to examine parameter recovery and show the results respectively in Fig. A, B, C, D, and E. For each noise level, we show the parameter estimates in the left two graphs. In the right two graphs, we plot the parameter estimates based on simulated data against the parameter values used to simulate the data. A. Noise 𝜎 = 0.01. B. Noise 𝜎 = 0.05. C. Noise 𝜎 = 0.1. D. Noise 𝜎 = 0.2. E. Noise 𝜎 = 0.3.

      We will update the parameter recovery section (p. 44) and Supplementary Figure S3 to incorporate these new results:

      “We implemented 5 levels of noise with σ={0.01,0.05,0.1,0.2,0.3} and examined the impact of noise on parameter recovery for each level of noise. These noise levels covered the range of empirical noise levels we estimated from the subjects. To estimate each subject’s noise level, we carried out maximum likelihood estimation in the following steps. First, for each subject separately, we used the parameter estimates of the system-neglect model to compute the period-wise probability estimates of regime shift. As a reminder, we referred to a ‘period’ as the time when a new signal appeared during a trial (for a given transition probability and signal diagnosticity). Each trial consisted of 10 successive periods. Second, we computed the period-wise likelihood, the probability of observing the subject’s actual probability estimate given the probability estimate predicted by the system-neglect model and the noise level. Here we define noise as the standard deviation of a Gaussian distribution centered at the model-predicted probability estimate. We then summed over all periods the negative natural logarithm of likelihood and used MATLAB’s minimization algorithm (the ‘fmincon’ function) to obtain the noise estimate that minimized the sum of negative log likelihood (thus the noise estimate that maximized the sum of log likelihood). Across subjects, we found that the mean noise estimate was 0.1480 and ranged from 0.0816 to 0.3239 (Supplementary Figure S3).”

      The main study is based on N=30 subjects, as are the two control studies. Since this work is about individual differences (in particular w.r.t. to neural representations of noise and transition probabilities in the frontoparietal network and the vmPFC), I'm wondering how robust the results are. Is it likely that the results would replicate with a larger number of subjects? Can the two control studies be leveraged to address this concern to some extent?

      It would be challenging to use the control studies to address the robustness concern. The control studies were designed to address the motor confounds. They were less suitable, however, for addressing the individual difference issue raised by the reviewer. We discussed why this is the case below.

      The two control studies did not allow us to examine individual differences – in particular with respect to neural selectivity of noise and transition probability – and therefore we think it is less likely to leverage the control studies. Having said that, it is possible to look at neural selectivity of noise (signal diagnosticity) in the first control experiment where subjects estimated the probability of blue regime in a task where there was no regime change (transition probability was 0). However, the fact that there were no regime shifts in the first control experiment changed the nature of the task. Instead of always starting at the Red regime in the main experiment, in the first control experiment we randomly picked the regime to draw the signals from. It also changed the meaning and the dynamics of the signals (red and blue) that would appear. In the main experiment the blue signal is a signal consistent with change, but in the control experiment this is no longer the case. In the main experiment, the frequency of blue signals is contingent upon both noise and transition probability where blue signals are less frequent than red signals because of the small transition probabilities. But in the first control experiment, the frequency of blue signals is not less frequent because the regime was blue in half of the trials. Due to these differences, we do not see how analyzing the control experiments could help in establishing robustness because we do not have a good prediction as to whether and how the neural selectivity would be impacted by these differences.

      We can address the issue of robustness through looking at the effect size. In particular, with respect to individual differences in neural sensitivity of transition probability and signal diagnosticity, since the significant correlation coefficients between neural and behavioral sensitivity were between 0.4 and 0.58 for signal diagnosticity in frontoparietal network (Fig. 5C), and -0.38 and -0.37 for transition probability in vmPFC (Fig. 5D), the effect size of these correlation coefficients was considered medium to large (Cohen, 1992). Cohen, J. (1992). A power primer. Psychological Bulletin, 112(1), 155-159.

      It seems that the authors have not counterbalanced the colors and that subjects always reported the probability of the blue regime. If so, I'm wondering why this was not counterbalanced.

      We are aware of the reviewer’s concern. The first reason we did not do these (color counterbalancing and report blue/red regime balancing) was to not confuse the subjects in an already complicated task. Balancing these two variables also comes at the cost of sample size, which was the second reason we did not do it. Although we can elect to do these balancing at the between-subject level to not impact the task complexity, we could have introduced another confound that is the individual differences in how people respond to these variables. This is the third reason we were hesitant to do these counterbalancing.

      Reviewer #2 (Public review):

      Summary:

      This paper focuses on understanding the behavioral and neural basis of regime shift detection, a common yet hard problem that people encounter in an uncertain world. Using a regime-shift task, the authors examined cognitive factors influencing belief updates by manipulating signal diagnosticity and environmental volatility. Behaviorally, they have found that people demonstrate both over and under-reaction to changes given different combinations of task parameters, which can be explained by a unified system-neglect account. Neurally, the authors have found that the vmPFC-striatum network represents current belief as well as belief revision unique to the regime detection task. Meanwhile, the frontoparietal network represents cognitive factors influencing regime detection i.e., the strength of the evidence in support of the regime shift and the intertemporal belief probability. The authors further link behavioral signatures of system neglect with neural signals and have found dissociable patterns, with the frontoparietal network representing sensitivity to signal diagnosticity when the observation is consistent with regime shift and vmPFC representing environmental volatility, respectively. Together, these results shed light on the neural basis of regime shift detection especially the neural correlates of bias in belief update that can be observed behaviorally.

      Strengths:

      (1) The regime-shift detection task offers a solid ground to examine regime-shift detection without the potential confounding impact of learning and reward. Relatedly, the system-neglect modeling framework provides a unified account for both over or under-reacting to environmental changes, allowing researchers to extract a single parameter reflecting people's sensitivity to changes in decision variables and making it desirable for neuroimaging analysis to locate corresponding neural signals.

      Thank you for recognizing our task design and our system-neglect computational framework in understanding change detection.

      (2) The analysis for locating brain regions related to belief revision is solid. Within the current task, the authors look for brain regions whose activation covary with both current belief and belief change. Furthermore, the authors have ruled out the possibility of representing mere current belief or motor signal by comparing the current study results with two other studies. This set of analyses is very convincing.

      Thank you for recognizing our control studies in ruling out potential motor confounds in our neural findings on belief revision.

      (3) The section on using neuroimaging findings (i.e., the frontoparietal network is sensitive to evidence that signals regime shift) to reveal nuances in behavioral data (i.e., belief revision is more sensitive to evidence consistent with change) is very intriguing. I like how the authors structure the flow of the results, offering this as an extra piece of behavioral findings instead of ad-hoc implanting that into the computational modeling.

      Thank you for appreciating how we showed that neural insights can lead to new behavioral findings.

      Weaknesses:

      (1) The authors have presented two sets of neuroimaging results, and it is unclear to me how to reason between these two sets of results, especially for the frontoparietal network. On one hand, the frontoparietal network represents belief revision but not variables influencing belief revision (i.e., signal diagnosticity and environmental volatility). On the other hand, when it comes to understanding individual differences in regime detection, the frontoparietal network is associated with sensitivity to change and consistent evidence strength. I understand that belief revision correlates with sensitivity to signals, but it can probably benefit from formally discussing and connecting these two sets of results in discussion. Relatedly, the whole section on behavioral vs. neural slope results was not sufficiently discussed and connected to the existing literature in the discussion section. For example, the authors could provide more context to reason through the finding that striatum (but not vmPFC) is not sensitive to volatility.<br />

      We thank the reviewer for the valuable suggestions.

      With regard to the first comment, we wish to clarify that we did not find frontoparietal network to represent belief revision. It was the vmPFC and ventral striatum that we found to represent belief revision ( in Fig. 3). For the frontoparietal network, we identified its involvement in our task through finding that its activity correlated with strength of change evidence (Fig. 4) and individual subjects’ sensitivity to signal diagnosticity (Fig. 5). Conceptually, these two findings reflect how individuals interpret the signals (signals consistent or inconsistent with change) in light of signal diagnosticity. This is because (1) strength of change evidence is defined as signals (+1 for signal consistent with change, and -1 for signal inconsistent with change) multiplied by signal diagnosticity and (2) sensitivity to signal diagnosticity reflects how individuals subjectively evaluate signal diagnosticity. At the theoretical level, these two findings can be interpreted through our computational framework in that both the strength of change evidence and sensitivity to signal diagnosticity contribute to estimating the likelihood of change (Eqs. 1 and 2). We added a paragraph in Discussion to talk about this.

      We will add on p. 35:

      “For the frontoparietal network, we identified its involvement in our task through finding that its activity correlated with strength of change evidence (Fig. 4) and individual subjects’ sensitivity to signal diagnosticity (Fig. 5). Conceptually, these two findings reflect how individuals interpret the signals (signals consistent or inconsistent with change) in light of signal diagnosticity. This is because (1) strength of change evidence is defined as signals (+1 for signal consistent with change, and -1 for signal inconsistent with change) multiplied by signal diagnosticity and (2) sensitivity to signal diagnosticity reflects how individuals subjectively evaluate signal diagnosticity. At the theoretical level, these two findings can be interpreted through our computational framework in that both the strength of change evidence and sensitivity to signal diagnosticity contribute to estimating the likelihood of change (Eqs. 1 and 2).”

      With regard to the second comment, we added discussion on the behavioral and neural slope comparison. We pointed out previous papers conducting similar analysis (Vilares et al., 2012; Ting et al., 2015; Yang & Wu, 2020), their findings and how they relate to our results. Vilares et al. found that sensitivity to prior information (uncertainty in prior distribution) in the orbitofrontal cortex (OFC) and putamen correlated with behavioral measure of sensitivity to prior. In the current study, transition probability acts as prior in the system-neglect framework (Eq. 2) and we found that ventromedial prefrontal cortex represents subjects’ sensitivity to transition probability. Together, these results suggest that OFC and vmPFC are involved in the subjective evaluation of prior information in both static (Vilares et al., 2012) and dynamic environments (current study). In addition, we added to the literature by showing that distinct from vmPFC in representing sensitivity to transition probability or prior, the frontoparietal network represents how sensitive individual decision makers are to the diagnosticity of signals in revealing the true state (regime) of the environment.

      We will add on p. 36:

      “In the current study, our psychometric-neurometric analysis focused on comparing behavioral sensitivity with neural sensitivity to the system parameters (transition probability and signal diagnosticity). We measured sensitivity by estimating the slope of behavioral data (behavioral slope) and neural data (neural slope) in response to the system parameters. Previous studies had adopted a similar approach (Vilares et al., 2012; Ting et al., 2015; Yang & Wu, 2020). For example, Vilares et al. (2012) found that sensitivity to prior information (uncertainty in prior distribution) in the orbitofrontal cortex (OFC) and putamen correlated with behavioral measure of sensitivity to the prior. In the current study, transition probability acts as prior in the system-neglect framework (Eq. 2) and we found that ventromedial prefrontal cortex represents subjects’ sensitivity to transition probability. Together, these results suggest that OFC and vmPFC are involved in the subjective evaluation of prior information in both static (Vilares et al., 2011) and dynamic environments (current study). In addition, we added to the literature by showing that distinct from vmPFC in representing sensitivity to transition probability or prior, the frontoparietal network represents how sensitive individual decision makers are to the diagnosticity of signals in revealing the true state (regime) of the environment.” 

      (2) More details are needed for behavioral modeling under the system-neglect framework, particularly results on model comparison. I understand that this model has been validated in previous publications, but it is unclear to me whether it provides a superior model fit in the current dataset compared to other models (e.g., a model without \alpha or \beta). Relatedly, I wonder whether the final result section can be incorporated into modeling as well - i.e., the authors could test a variant of the model with two \betas depending on whether the observation is consistent with a regime shift and conduct model comparison.

      Thank you for the great suggestion.

      To address the reviewer’s question on model comparison, we tested 4 variants of the system-neglect model and incorporated them into the final result section. The original system-neglect model and its four models are:

      – Original system-neglect model: 6 total parameters, 3 beta parameters (one for each level of signal diagnosticity) and 3 alpha parameters (one for each level of transition probability).  

      – M1: System-neglect model with signal-dependent beta parameters (alpha parameters, and beta parameters separately estimated at change-consistent and change-inconsistent signals): 9 total parameters, 3 beta parameters for change-consistent signals, 3 beta parameters for change-inconsistent signals, and 3 alpha parameters.

      – M2: System-neglect model with signal-dependent alpha parameters (alpha parameters separately estimated at change-consistent and change-inconsistent signals, and beta parameters): 9 total parameters, 3 alpha parameters for change-consistent signals, 3 alpha parameters for change-inconsistent signals, and 3 beta parameters.

      – M3: System-neglect model without alpha parameters (only the beta parameters): 3 total parameters, all are beta parameters (one for each level of signal diagnosticity).

      – M4: System-neglect model without beta parameters (only the alpha parameters): 3 total parameters, all are alpha parameters (one for each level of transition probability).

      We compared these four models with the original system-neglect model. In the figure below, we plot  where  is the Akaike Information Criterion (AIC) of one of the new models minus the AIC of the original model. ∆AIC<0  indicates that the new model is better than the original model. By contrast, ∆AIC>0 suggests that the new model is worse than the original model.

      Author response image 3.

      When we separately estimated the beta parameter (M1) for change-consistent signals and change-inconsistent signals, we found that its AIC is significantly smaller than the original model (p<0.01). The same was found for the model where we separately estimated the alpha parameters for change-consistent and change-inconsistent signals (M2). When we took out either the alpha (M3) or the beta parameters (M4), we found that these models were worse than the original model (p<0.01). In summary, we found that models where we separately estimated the alpha/beta parameters for change-consistent and change-inconsistent signals were better than the original model. This is consistent with the insight the neural data provided.

      To show these results, we will add a new figure (Figure 7) in the revised manuscript.

    1. eLife Assessment

      This valuable study implicates a specific Wolbachia gene in driving the male-killing phenotype in a moth: This is a contribution to a growing body of literature from the authors in which they authors have nicely teased apart the loci responsible for male killing across diverse insects. The conclusions are supported by solid evidence.

    2. Reviewer #1 (Public review):

      Summary:

      Insects and their relatives are commonly infected with microbes that are transmitted from mothers to their offspring. A number of these microbes have independently evolved the ability to kill the sons of infected females very early in their development; this male killing strategy has evolved because males are transmission dead-ends for the microbe. A major question in the field has been to identify the genes that cause male killing and to understand how they work. This has been especially challenging because most male-killing microbes cannot be genetically manipulated. This study focuses on a male-killing bacterium called Wolbachia. Different Wolbachia strains kill male embryos in beetles, flies, moths, and other arthropods. This is remarkable because how sex is determined differs widely in these hosts. Two Wolbachia genes have been previously implicated in male-killing by Wolbachia: oscar (in moth male-killing) and wmk (in fly male-killing). The genomes of some male-killing Wolbachia contain both of these genes, so it is a challenge to disentangle the two.

      This paper provides strong evidence that oscar is responsible for male-killing in moths. Here, the authors study a strain of Wolbachia that kills males in a pest of tea, Homona magnanima. Overexpressing oscar, but not wmk, kills male moth embryos. This is because oscar interferes with masculinizer, the master gene that controls sex determination in moths and butterflies. Interfering with the masculinizer gene in this way leads the (male) embryo down a path of female development, which causes problems in regulating the expression of genes that are found on the sex chromosomes.

      Strengths:

      The authors use a broad number of approaches to implicate oscar, and to dissect its mechanism of male lethality. These approaches include: a) overexpressing oscar (and wmk) by injecting RNA into moth eggs, b) determining the sex of embryos by staining female sex chromosomes, c) determining the consequences of oscar expression by assaying sex-specific splice variants of doublesex, a key sex determination gene, and by quantifying gene expression and dosage of sex chromosomes, using RNASeq, and d) expressing oscar along with masculinizer from various moth and butterfly species, in a silkmoth cell line. This extends recently published studies implicating oscar in male-killing by Wolbachia in Ostrinia corn borer moths, although the Homona and Ostrinia oscar proteins are quite divergent. Combined with other studies, there is now broad support for oscar as the male-killing gene in moths and butterflies (i.e. order Lepidoptera). So an outstanding question is to understand the role of wmk. Is it the master male-killing gene in insects other than Lepidoptera and if so, how does it operate?

      Weaknesses:

      I found the transfection assays of oscar and masculinizer in the silkworm cell line (Figure 4) to be difficult to follow. There are also places in the text where more explanation would be helpful for non-experts.

    3. Reviewer #2 (Public review):

      Summary:

      Wolbachia are maternally transmitted bacteria that can manipulate host reproduction in various ways. Some Wolbachia induce male killing (MK), where the sons of infected mothers are killed during development. Several MK-associated genes have been identified in Homona magnanima, including Hm-oscar and wmk-1-4, but the mechanistic links between these Wolbachia genes and MK in the native host are still unclear.

      In this manuscript, Arai et al. show that Hm-oscar is the gene responsible for Wolbachia-induced MK in Homona magnanima. They provide evidence that Hm-Oscar functions through interactions with the sex determination system. They also found that Hm-Oscar disrupts sex determination in male embryos by inducing female-type dsx splicing and impairing dosage compensation. Additionally, Hm-Oscar suppresses the function of Masc. The manuscript is well-written and presents intriguing findings. The results support their conclusions regarding the diversity and commonality of MK mechanisms, contributing to our understanding of the mechanisms and evolutionary aspects of Wolbachia-induced MK.

      Comments on revisions:

      The authors have already addressed the reviewer's concerns.

    4. Reviewer #3 (Public review):

      Summary:

      Overall, this is a clearly written manuscript with nice hypothesis testing in a non-model organism that addresses the mechanism of Wolbachia-mediated male killing. The authors aim to determine how five previously identified male-killing genes (encoded in the prophage region of the wHm Wolbachia strain) impact the native host, Homona magnanima moths. This work builds on the authors' previous studies in which<br /> (1) they tested the impact of these same wHm genes via heterologous expression in Drosophila melanogaster<br /> (2) also examined the activity of other male-killing genes (e.g., from the wFur Wolbachia strain in its native host: Ostrinia furnacalis moths).

      Advances here include identifying which wHm gene most strongly recapitulates the male-killing phenotype in the native host (rather than in Drosophila), and the finding that the Hm-Oscar protein has the potential for male-killing in a diverse set of lepidopterans, as inferred by the cell-culture assays.

      Strengths:

      Strengths of the manuscript include the reverse genetics approaches to dissect the impact of specific male-killing loci, and use of a "masculinization" assay in Lepidopteran cell lines to determine the impact of interactions between specific masc and oscar homologs.

      Weaknesses:

      It is clear from Figure 1 that the combinations of wmk homologs do not cause male killing on their own here. While I largely agree with the author's conclusions that oscar is the primary MK factor in this system, I don't think we can yet rule out that wmk(s) may work synergistically or interactively with oscar in vivo. This might be worth a small note in the discussion. (eg at line 294 'indicating that wmk likely targets factors other than masc." - this could be downstream of the impacts of oscar; perhaps dependent on oscar-mediated impacts on masc first).

      Regarding the perceived male-bias in Figure 2a: I think readers might be interpreting "unhatched" as "total before hatching". You could eliminate ambiguity by perhaps splitting the bars into male and female, and then within a bar, coloring by hatched versus unhatched. But this is a minor point, and I think the updated text helps clarify this.

      The new Figure 4b looks to be largely redundant with the oscar information in Figure 1a.

      Updated statistical comparisons for the RNA-seq analysis are helpful. However these analyses are based on single libraries (albeit each a pool of many individuals), so this is still a weaker aspect of the manuscript.

      The new information on masc similarity is useful (Fig 4d) - if the authors could please include a heatmap legend for the colors, that would be helpful. Also, please avoid green and red in the same figure when key for interpretation.

      Figure 1A "helix-turn-helix" is misspelled. ("tern").

    5. Author response:

      The following is the authors’ response to the original reviews.

      Public Reviews:

      Reviewer #1 (Public review):

      Summary:

      Insects and their relatives are commonly infected with microbes that are transmitted from mothers to their offspring. A number of these microbes have independently evolved the ability to kill the sons of infected females very early in their development; this male killing strategy has evolved because males are transmission dead-ends for the microbe. A major question in the field has been to identify the genes that cause male killing and to understand how they work. This has been especially challenging because most male-killing microbes cannot be genetically manipulated. This study focuses on a male-killing bacterium called Wolbachia. Different Wolbachia strains kill male embryos in beetles, flies, moths, and other arthropods. This is remarkable because how sex is determined differs widely in these hosts. Two Wolbachia genes have been previously implicated in male-killing by Wolbachia: oscar (in moth male-killing) and wmk (in fly male-killing). The genomes of some male-killing Wolbachia contain both of these genes, so it is a challenge to disentangle the two.

      This paper provides strong evidence that oscar is responsible for male-killing in moths. Here, the authors study a strain of Wolbachia that kills males in a pest of tea, Homona magnanima. Overexpressing oscar, but not wmk, kills male moth embryos. This is because oscar interferes with masculinizer, the master gene that controls sex determination in moths and butterflies. Interfering with the masculinizer gene in this way leads the (male) embryo down a path of female development, which causes problems in regulating the expression of genes that are found on the sex chromosomes.

      We would like to thank you for evaluating our manuscript.

      Strengths:

      The authors use a broad number of approaches to implicate oscar, and to dissect its mechanism of male lethality. These approaches include:

      (1) Overexpressing oscar (and wmk) by injecting RNA into moth eggs.

      (2) Determining the sex of embryos by staining female sex chromosomes.

      (3) Determining the consequences of oscar expression by assaying sex-specific splice variants of doublesex, a key sex determination gene, and by quantifying gene expression and dosage of sex chromosomes, using RNASeq.

      (4) Expressing oscar along with masculinizer from various moth and butterfly species, in a silkmoth cell line.

      This extends recently published studies implicating oscar in male-killing by Wolbachia in Ostrinia corn borer moths, although the Homona and Ostrinia oscar proteins are quite divergent. Combined with other studies, there is now broad support for oscar as the male-killing gene in moths and butterflies (i.e. order Lepidoptera). So an outstanding question is to understand the role of wmk. Is it the master male-killing gene in insects other than Lepidoptera and if so, how does it operate?

      Thank you for your comments. Wolbachia strains often carry wmk genes, but as observed in this study, the homologs in Homona showed no apparent MK ability. These showed strong male lethality in D. melanogaster, but it is still unclear whether the genes are the master male-killing gene in Diptera. It is also possible that the genes show toxicities in other lepidopteran insects as well as in other insect taxa. Further functional validation assays in different insects are warranted to clarify whether wmk shows toxicity in different insect taxa. We have also discussed the functions of wmk in the Discussion section (lines 301-306).

      Weaknesses:

      I found the transfection assays of oscar and masculinizer in the silkworm cell line (Figure 4) to be difficult to follow. There are also places in the text where more explanation would be helpful for non-experts (see recommendations).

      Thank you for your suggestion. We have thoroughly revised the manuscript to address all the questions, comments and suggestions you raised in “recommendations”. In particular, we have revised the section on the transfection assays of Oscar and Masc in Bm-N4 cells (result section “Hm-oscar suppresses the masculinizing functions of lepidopteran masc genes” starts on line 214 and Fig. 4; materials and methods section ”Transfection assays and quantification of BmIMP<sup>M</sup>”, starts on line 483). We have also provided more detailed explanations for non-experts in some contexts (in response to your recommendation). We believe that the resulting revisions have significantly improved the quality and comprehensiveness of our manuscript.

      Reviewer #2 (Public review):

      Summary:

      Wolbachia are maternally transmitted bacteria that can manipulate host reproduction in various ways. Some Wolbachia induce male killing (MK), where the sons of infected mothers are killed during development. Several MK-associated genes have been identified in Homona magnanima, including Hm-oscar and wmk-1-4, but the mechanistic links between these Wolbachia genes and MK in the native host are still unclear.

      In this manuscript, Arai et al. show that Hm-oscar is the gene responsible for Wolbachia-induced MK in Homona magnanima. They provide evidence that Hm-Oscar functions through interactions with the sex determination system. They also found that Hm-Oscar disrupts sex determination in male embryos by inducing female-type dsx splicing and impairing dosage compensation. Additionally, Hm-Oscar suppresses the function of Masc. The manuscript is well-written and presents intriguing findings. The results support their conclusions regarding the diversity and commonality of MK mechanisms, contributing to our understanding of the mechanisms and evolutionary aspects of Wolbachia-induced MK.

      We would like to thank you for evaluating our manuscript.

      Strengths/weaknesses:

      (1) The authors found that transient overexpression of Hm-oscar, but not wmk-1-4, in Wolbachia-free H. magnanima embryos induces female-biased sex ratios. These results are striking and mirror the phenotype of the wHm-t infected line (WT12). However, Table 1 lists the "male ratio," while the text presents the "female ratio" with standard deviation. For consistency, the calculation term should be uniform, and the "ratio" should be listed for each replicate.

      We have revised the first results section (Hm-oscar induces female-biased sex ratios, starting from line 147) accordingly to maintain the consistency in the calculation term. In the revised manuscript, the 'male ratio' is now consistently used, in alignment with Fig. 1. In addition, we have included all sex ratio information (number of males and females) in the supplementary data file for transparency and clarity.

      (2) The error bars in Figure 3 are quite large, and the figure lacks statistical significance labels. The authors should perform statistical analysis to demonstrate that Hm-oscar-overexpressed male embryos have higher levels of Z-linked gene expression.

      The large error bar on each chromosome (Fig.3a-d) likely reflect the overall variation in expression levels across different transcripts. Accordingly, we have included statistical data for Figure 3 based on the Steel-Dwass test for expression levels. However, displaying statistical significance directly on the whisker plots would make the figure too cluttered due to the numerous combinations. Instead, we have provided all the statistical data in the supplementary data file. To further support the claim that Z-linked genes are more highly expressed in wHm-t-infected/Hb-Oscar-injected embryos, we have included the expression data for a Z-linked gene tpi, along with its statistical data in the revised manuscript (Fig. 3e, lines 210-212).

      (3) The authors demonstrated that Hm-Oscar suppresses the masculinizing functions of lepidopteran Masc in BmN-4 cells derived from the female ovaries of Bombyx mori. They should clarify why this cell line was chosen and its biological relevance. Additionally, they should explain the rationale for evaluating the expression levels of the male-specific BmIMP variant and whether it is equivalent to dsx.

      Thank you for your suggestion. We selected BmN-4 cell line because previous studies have established it as a reliable model for investigating the functions of lepidopteran masc genes and the interactions between masc and Oscar genes (Katsuma et al., 2019; 2022). In addition, BmIMP<sup>M</sup> is a male-specific regulator of the male-type dsx, making it an ideal target for assessing the 'maleness' induced by transfection of the masc gene in female-derived BmN-4 cells (Suzuki et al., 2010; Katsuma et al., 2015). We have included more detailed background information in the revised manuscript and have thoroughly revised this section (Hm-oscar suppresses the masculinizing functions of lepidopteran masc genes, starting at line 214) and Figure 4 for better clarity.

      (4) Although the authors show that Hm-oscar is involved in Wolbachia-induced MK in Homona magnanima and interacts with the sex determination system in lepidopteran insects, the precise molecular mechanism of Hm-oscar-induced MK remains unclear. Further studies are needed to elucidate how Hm-oscar regulates Homona magnanima genes to induce MK, though this may be beyond the scope of the current manuscript.

      Based on our findings and previous studies in Homona, Ostrinia and Bombyx (Arai et al., 2023a; Katsuma et al., 2023; Kiuchi et al., 2014), we hypothesize that the molecular mechanisms underlying _w_Hm-induced MK are likely linked to impaired dosage compensation caused by the inhibition of Masc function by the Hm-Oscar protein. While the precise mechanisms remain unclear, unbalanced Z-linked gene expression due to the impaired dosage compensation (i.e., 2-fold higher Z-linked gene expression compared to normal males) is known to be lethal for lepidopteran males (Kiuchi et al., 2014; Fukui et al., 2015; Visser et al., 2021). We have outlined this hypothesis in the Discussion section (lines 245-254).

      Reviewer #3 (Public review):

      Summary:

      Overall, this is a clearly written manuscript with nice hypothesis testing in a non-model organism that addresses the mechanism of Wolbachia-mediated male killing. The authors aim to determine how five previously identified male-killing genes (encoded in the prophage region of the wHm Wolbachia strain) impact the native host, Homona magnanima moths. This work builds on the authors' previous studies in which:

      (1) They tested the impact of these same wHm genes via heterologous expression in Drosophila melanogaster.

      (2) They examined the activity of other male-killing genes (e.g., from the wFur Wolbachia strain in its native host: Ostrinia furnacalis moths).

      Advances here include identifying which wHm gene most strongly recapitulates the male-killing phenotype in the native host (rather than in Drosophila), and the finding that the Hm-Oscar protein has the potential for male-killing in a diverse set of lepidopterans, as inferred by the cell-culture assays.

      Strengths:

      Strengths of the manuscript include the reverse genetics approaches to dissect the impact of specific male-killing loci, and the use of a "masculinization" assay in Lepidopteran cell lines to determine the impact of interactions between specific masc and oscar homologs.

      We would like to thank you for evaluating our manuscript.

      Weaknesses:

      My major comments are related to the lack of statistics for several experiments (and the data normalization process), and opportunities to make the manuscript more broadly accessible.

      Thank you for your suggestions. We have thoroughly revised the manuscript to provide clearer explanations for non-experts. In addition, we have included more detailed statistical data for Figure 3 and Figure 4 based on the Steel-Dwass tests. For Figure 3a-d, displaying statistical significance directly on the whisker plots would make the figure too cluttered due to the numerous combinations. Therefore, we have provided all the statistical data in the supplementary data file. To further support the claim that Z-linked genes are more highly expressed in w_Hm-t-infected/Hm-Oscar-injected embryos, we have included the expression data for a Z-linked gene _tpi, along with its statistical data in the revised manuscript (Fig.3e, lines 210-212). Regarding Figure 4, we have revised the Figure based on the reviewer’s suggestions, and provided more detailed information on how the expression data were analyzed (Transfection assays and quantification of BmIMP<sup>M</sup>, lines 495-520). We have also included more detailed background information on the assay system (Hm-oscar suppresses the masculinizing functions of lepidopteran masc genes, lines 215-237). Although we did not observe statistical significance based on the Steel-Dwass test, likely due to limited replicates, the observed changes in the IMP gene expression remain clearly evident.

      The manuscript I think would be much improved by providing more details regarding some of the genes and cross-lineage comparisons. I know some of this is reported in previous publications, but some summary and/or additional analysis would make this current manuscript much more approachable for a broader audience, and help guide readers to specific important findings. For example, a graphic and/or more detail on how the wmk/oscar homologs (within and across Wolbachia strains) differ (e.g., domains, percent divergence, etc) would be helpful for contextualizing some of the results. I recognize the authors discuss this in parts (e.g., lines 223-227), but it does require some bouncing between sections to follow. Similarly, the experiments presented in Figure 4 indicate that Hm-oscar has broad spectrum activity: how similar are the masc proteins from these various lepidopterans? Are they highly conserved? Rapidly evolving? Do the patterns of masc protein evolution provide any hints at how Oscar might be interacting with masc?

      Thank you for your valuable suggestion. To address this, we have included a visualization of the structural differences between the Oscar and wmk homologs in Figure 1a of the revised manuscript. In addition, we have included more detailed information for these genes and revised the introduction (lines 110-114; 124-137) and discussion (lines 255-266) to provide a clearer and more comprehensive overview. We have also described the similarity of the Masc proteins and Oscar proteins that we used, which is now reflected in the revised Figure 4b and 4d. More detailed information on these proteins is available in the supplementary data. Notably, Masc proteins exhibit high sequence variability with conserved domains (Figure 4d). Previous study identified the N-terminal region of Masc as crucial for the Oscar function (Katsuma et al., 2022). The wide spectrum of the actions of Hm-Oscar likely stems from these conserved structures of Masc, but the effects might have undergone evolutionary tuning through interactions with the native host as discussed in lines 293-294.

      It is clear from Figure 1 that the combinations of wmk homologs do not cause male killing on their own. Did the authors test if any of the wmk homologs impact the MK phenotype of oscar? It looks like a previous study tested this in wFur (noted in lines 250-252), but given that the authors also highlight the differences between the wFur-oscar and Hm-oscar proteins, this may be worth testing in this system. Related to this, what is the explanation for why there would be 4 copies of wmk in Hm?

      Thank you for your valuable suggestion. Unfortunately, we have not yet tested the effects of co-expression of wmk and Oscar. Due to a technical issue, the mixing of multiple constructs results in a reduced amount of mRNA (i.e. mixing wmk-3 and Hm-Oscar at the same concentration results in a 2-fold lower concentration in mRNA for both genes compared to mono-injected groups). In addition, we have previously tested injecting mRNA at the twofold higher concentration (i.e. 2 ug/ul mRNA), which resulted in very low hatchability regardless of the genes. Katsuma et al (2022) tested the effect of wmk on the sex determination system, but did not test the effect of co-injection/transfection of wmk and Oscar. Considering the results of this and previous studies (Katsuma et al., 2022; Arai et al., 2023), it is likely that the targets of the wmk and oscar genes are different (as discussed in lines 267-289). Co-injection of wmk and oscar may not produce additive effects. Nevertheless, we would like to test the results in future studies using the Drosophila system as well.

      As you point out, it is an interesting point that the moth-derived MK Wolbachia w_Hm-t encodes four _wmk genes, although they have no apparent effect on host survival. The exact functional relevance of these wmk homologs remains unclear. However, they may play a role in Wolbachia biology as transcriptional regulators, given that they encode HTH domains. Wolbachia generally encode several wmk homologs in their genome, regardless of whether they induce MK. This suggests that the functions of the wmk genes may be 'suppressed' in certain Wolbachia-host systems. The wmk and Hm-oscar genes are located within a prophage region, and some wmk genes are tandemly arrayed (as described in Arai et al., 2023). These wmk homologs may have increased in number by horizontal phage transfer, and the region containing wmk and adjacent sequences may act as a genomic island for virulence. So far, the function of wmk homologs has only been tested in D. melanogaster and H. magnanima, and further studies in other Wolbachia-host systems are highly warranted to test whether wmk exerts MK effects in other insect models. These points have been briefly discussed in the revised manuscript (lines 301-306; 318-320).

      Why are some of the broods male-biased (2/3) rather than ~50:50? (Lines 170-175, Figure 2a). For example, there is a strong male bias in un-hatched oscar-injected and naturally infected embryos, whereas the control uninfected embryos have normal 50:50 sex ratios. It is difficult to interpret the rate of male-killing given that the sex ratios of different sets of zygotes are quite variable.

      The observed male-biased sex ratios in unhatched embryos are due to the occurrence of MK during embryogenesis. In the unhatched groups, the skew towards males reflects that fact that the male embryos were targeted and killed by Wolbachia/Oscar, leading to a surplus of unhatched male embryos. Conversely, hatched individuals show a higher proportion of females because many of the males were eliminated during embryogenesis. Thus, the unhatched embryos are more male-biased, while the hatched individuals are more female-biased in the Hm-oscar/_w_Hm-t treated groups. We have revised the relevant section (Males are killed mainly at the embryonic stage, lines 179-186) and provided more detailed information to clarify this explanation.

      Figure 2b - it appears there are both male and female bands in the HmOsc male lane. I think this makes sense (likely a partial phenotype due to the nature of the overexpression approach), but this is worth highlighting, especially in the context of trying to understand how much of the MK phenotype might be recapitulated through these methods. Related, there is no negative control for this PCR.

      Thank you for your suggestion. As you noted, a faint dsx-M band is visible in the Hm-oscar treated group in Figure 2b. This is consistent with previous findings by Arai et al. (2023), which reported that male embryos with low-density w_Hm-t showed double bands of _dsx-M and dsx-F, similar to what we observed in this study. This information has been included in the revised manuscript in lines 196-198, as follows:

      “Notably, male embryos expressing Hm-oscar also exhibited weak male-type dsx splicing in addition to the female-type splicing, resembling the previously observed pattern in male embryos infected with low-titer _w_Hm-t (Arai et al., 2023a).”

      Also, we appreciate your comment regarding the missing of negative control. The figure has now been revised as we realised that the negative control lane had been lost during the preparation of the figure. We also included the relevant molecular marker information in both the figure legends and Figure 2b.

      It appears the RNA-seq analysis (Figure 3) is based on a single biological replicate for each condition. And, there are no statistical comparisons that support the conclusions of a shift in dosage compensation. Finally, it is unclear what exactly is new data here: the authors note "The expression data of the wHm-t-infected and non-infected groups were also calculated based on the transcriptome data included in Arai et al. (2023a)" - So, are the data in Figure 3c and 3d a re-print of previous data? The level of dosage compensation inferred by visually comparing the control conditions in 3b and 3d does not appear consistent. With only one biological replicate library per condition, what looks like a re-print of previous data, and no statistical comparisons, this is a weakly supported conclusion.

      Thank you for your suggestion. In this study, we generated the RNA-seq data for the Hm-oscar/GFP-injected groups, but did not sequence the w_Hm-t-infected/NSR lines. Instead, the previously generated RNA-seq data of _w_Hm-t-infected/NSR (Arai et al., 2023) were re-analyzed (rather than simply reprinted) to evaluate whether the expression patterns of _Hm-oscar-injected and w_Hm-t-infected groups are similar. We have revised the Results section (_Hm-oscar impairs dosage compensation in male embryos, lines 200-212), the Materials and methods section (Quantification of Z chromosome-linked genes, lines 454-456), and the figure legends to provide more precise information about this analysis.

      Although we did not perform replicates for the RNA-seq comparisons, it is important to note that each RNA-seq sample contains 50-60 male/female individuals. We believe the results are still robust and clearly indicative of the trends we observe. This was further supported by the quantification of Hmtpi gene expression, which we have visualized in Figure 3e (and lines 210-212). As you noted, the expression patterns in Figure 3b (GFP injected) and Figure 3d (NSR) are not completely identical. This discrepancy may be due to the differences between injection treatments and natural infections. Nevertheless, both treatments are consistent in showing that gene expressions on the Z chromosome (Chr01 and Chr15) are not upregulated.

      We have also added more detailed statistical data for Figure 3 based on the Steel-Dwass tests. For Figure 3a-d, however, showing the statistical significance directly on the whisker plots would create excessive clutter due to the numerous combinations of chromosomes. Instead, we have provided the full statistical data in the supplementary data file. Furthermore, to support/strengthen our conclusion that Z-linked genes are highly expressed in w_Hm-t-infected/_Hm-Oscar-injected embryos, we have included expression data for the Z-linked gene tpi, along with statistical data, in the revised manuscript (Fig. 3e, lines 210-212).

      In Figure 4: There are no statistics to support the conclusions presented here. Additionally, the data have gone through a normalization process, but it is difficult to follow exactly how this was done. The control conditions appear to always be normalized to 100 ("The expression levels of BmImpM in the Masc and Hm-Oscar/Oscar co-transfected cells were normalized by setting each Masc-transfected cell as 100"). I see two problems with this approach:

      (1) This has eliminated all of the natural variation in BmImpM expression, which is likely not always identical across cells/replicates.

      (2) How then was the percentage of BmImpM calculated for each of the experimental conditions? Was each replicate sample arbitrarily paired with a control sample? This can lead to very different outcomes depending on which samples are paired with each other. The most appropriate way to calculate the change between experimental and control would be to take the difference between every single sample (6 total, 3 control, 3 experimental) and the mean of the control group. The mean of the control can then be set at 100 as the authors like, but this also maintains the variability in the dataset and then eliminates the issue of arbitrary pairings. This approach would also then facilitate statistical comparisons which is currently missing.

      Thank you for your suggestion. As you pointed out in (1), the previous analysis did indeed eliminate the natural variation in BmIMP-M expression. In the revised manuscript and Figure 4, we have reanalyzed the data following your suggestion and have described the variation across replicates.

      For (2), the data shown in the previous manuscript were normalized to 100 for each Masc-treated group. In doing so, each replicate sample was arbitrarily paired with a control sample from the same cell lot to account for variations that might occur due to differences in cell lots. However, following your recommendation, we have revised the figure to set the average of the Hm-masc treated group to 100, rather than using arbitrary pairings. More detailed normalization procedures have been provided in the section 'Transfection assays and quantification of BmIMP' (lines 483-520). Additionally, we have provided more detailed background information on the assay system in lines 218-223. Although we did not observe statistical significance based on the Steel-Dwass test, likely due to the limited number of replicates, the differences in IMP gene expression between the Masc-treated and Masc&Hm-oscar-treated groups remain evident.

      Recommendations for the authors:

      Reviewer #1 (Recommendations for the authors):

      Line 38: change to: 'Wolbachia are maternally transmitted'.

      Revised accordingly (line 38).

      Line 69: remove 'seemingly'.

      Revised accordingly (line 69).

      Paragraph starting line 123: I don't think this is so clear to a reader who is not familiar with the work and system. It would be helpful to more clearly explain that candidate male-killing genes from Wolbachia that infect Homona were inserted into Drosophila melanogaster, and that their expression was then induced, with interesting patterns (and that it can be a bit difficult to interpret the transgenic expression of genes from a moth male-killer that are inserted into a fly). Also, the sentence about the combined action of cifA and cifB in Drosophila cytoplasmic incompatibility is also confusing to a non-expert. I would suggest removing it.

      Thank you for your suggestion. We have revised the paragraph (lines 124-139) to provide clearer background information, making it easier for non-experts to follow. We have also removed the sentence regarding the combined effect of cifA and cifB to improve the flow and overall clarity.

      Line 170: what is the explanation for the male-biased sex ratio instead of 50-50?

      The male-biased sex ratio occurs because MK happens during embryogenesis. Unhatched embryos include males that were killed by Wolbachia/Oscar, resulting in a higher proportion of unhatched male embryos. Conversely, the hatched individuals display a female bias, as most of the males were eliminated during embryogenesis. Thus, the unhatched embryos are more male-biased, while the hatched individuals are more female-biased in the Hm-oscar/_w_Hm-t treated groups. We have revised the section “Males are killed mainly at the embryonic stage” (lines 170-186) to include more detailed information explaining this phenomenon.

      Line 190: please explain what are the Z chromosomes in Bombyx and Homona and Lepidoptera in general (chromosomes 1 and 15?), as this is not so clear for a non-expert.

      Thank you for your suggestion. I have revised the section (lines 200-212) to include more precise background information about the chromosome constitutions in lines 202-204 as follows:

      “Unlike other lepidopteran species, Tortricidae, including H. magnanima, generally possess a large Z chromosome that is homologous to B. mori chromosomes 1 (Z) and 15 (autosome).”

      Line 222: please explain oscar diversity and classification in more detail, as this is not so clear for a non-expert.

      Thank you for your suggestion. We have revised the sentences to provide clearer background information on the diversity of oscar genes (lines 255-264).

      Figure 4: I found this difficult to follow. Why are there 2 rows (HmOscar and Oscar)? Does oscar here refer to oscar from Ostrinia? I am also a bit confused about the baseline control of Masc in these cell lines. If I understand Lepidoptera sex determination, then these cell lines are expressing high levels of female-specific piRNAs that suppress Masc. How specific are these piRNAs (i.e. do Bombyx piRNAs suppress Mascs from other Lepidoptera)? How much extra Masc will override endogenous piRNA? Information is lost by setting Masc expression to 100% in each separate comparison.

      Yes, the Oscar indicates the w_Fur-encoded _oscar (Oscar from Ostrinia) that was tested to compare function with the Homona-derived Hm-oscar gene. In addition, following the reviewer's suggestions, we have revised the figure and included more detailed information on how we adjusted the expressions in the M&M section.

      A previous study (Shoji et al., 2017, RNA 23:86–97) demonstrated that the Fem piRNA (29 bp) in Bombyx mori requires a 17 bp complementary sequence from its 5' region for its function. However, in species other than B. mori, no significant homology (i.e., over 17 bp matches) was found between the B. mori Fem piRNA and the masc genes analyzed in this study. Therefore, it is likely that the Fem piRNA expressed in BmN-4 cells is unable to suppress the masculinizing function driven by masc genes in other lepidopteran species. In addition, we did not quantify the levels of piRNA in this system, but the expression levels of masc are probably too high to be suppressed.

      Figure 4 legend: spelling of Spodoptera.

      Revised accordingly.

      Reviewer #2 (Recommendations for the authors):

      In Figure 2, what is the dsx splicing type for the hatched male in the Hm-oscar-injected group and the wHm-t infected line? Dsx-F or dsx-M?

      Thank you for your suggestion. Unfortunately, we have not tested splicing in the hatched male neonates (1st instar larvae), partly due to difficulties in obtaining sufficient material for RNA extraction. Based on the previous publication in the Ostrinia system, where Oscar-bearing w_Sca induces MK, the hatched males (ZZ) exhibit female type _dsx as observed in the male embryos (Herran et al., 2022). The hatched Homona males may show double bands for dsx-M and dsx-F as observed in this study.

      The size of the markers (in kilobase pairs) should be indicated in Figure 2.

      We have accordingly included the marker information in the revised Figure 2b and the figure legends.

      In Figure 3, could the authors identify which genes exhibit higher expression levels in the Hm-oscar-injected group and the wHm-t infected line? Could they provide hints for the possible mechanism of male-killing?

      In the RNA-seq data shown in Figure 3a-d, we observed that both the Hm-oscar-injected and w_Hm-infected groups generally exhibited upregulated expression of Z-linked genes. Rather than the upregulation or downregulation of a specific gene, we consider that global upregulation of Z-linked genes, caused by improper dosage compensation, is lethal for males. The Z chromosome contains various genes involved in key biological processes such as endocrine function and detoxification, and disruption of these processes may contribute to male lethality. Additionally, in this revised manuscript, we have provided more detailed information on the expression level of the Z-linked gene _tpi. We have also discussed the potential mechanisms of MK in the Discussion section (lines 245-254).

      The format of the references should be consistent. Gene and species names should be italicized.

      We have accordingly formatted.

      Reviewer #3 (Recommendations for the authors):

      The authors use the term "upstream" (e.g., Oscar suppressed the function of masculinizer, the upstream male sex determinant...), which was sometimes confusing. In many cases, it reads as though the masculinizer was upstream of oscar, but what I think the authors are trying to convey is that masculinizer is a primary sex-determining factor.

      Thank you for your suggestion. We have accordingly revised the term.

      Line 101: which insect is wFur from?

      It is from Ostrinia furnacalis - line 104 has been revised.

      Figure 1: it would be helpful to indicate the statistical results on the figure.

      Accordingly, we have added statistical data (binominal test) for Figure 1. The data for the Steel-Dwass test have been included in the supplementary data.

      Figure 2b: please label the ladder on the gel.

      Thank you for your suggestion. We have accordingly labeled the DNA ladder on the gel.

    1. eLife Assessment

      This study provides compelling data regarding the molecular characterization of a rare tumor type with few treatment options. This fundamental work significantly advances our mechanistic understanding of solitary fibrous tumours, a critical first step towards targeted precision medicine approaches. The results of this study will be of broad interest to cancer biologists and experimental oncologists.

    2. Joint Public Review:

      Solitary Fibrous Tumors (SFTs) are a rare malignancy defined by NAB2-STAT6 fusions. Because the molecular understanding of the disease is largely lacking, there are currently no targeted treatment approaches. Using primary tumor and adjacent normal tissue samples and cells inducibly expressing NAB2-STAT6, Hill et al. perform a detailed characterization of the transcriptomic and epigenomic NAB2-STAT6 SFT signatures. They identify enrichment or EGR1/NAB2 (but not STAT6) sites bound by the fusion protein and increased expression of EGR1 targets. Their studies indicate that NAB2-STAT6 fusion may direct the nuclear translocation of NAB2 and EGR1 proteins and potentially NAB1. Transcriptionally, NAB2-STAT6 SFTs most closely resemble neuroendocrine tumors.

      This pioneering study provides critical insight into the molecular pathogenesis of SFTs, pivotal for the future development of mechanistically informed treatment approaches. The study is rigorously executed and well-written. This new knowledge is an important addition to the field.

    3. Author response:

      The following is the authors’ response to the original reviews.

      Response to the Joint Public Review:

      We are indebted to eLife’s reviewing process for helping us improve our manuscript and for highlighting that our study provides new molecular insights into SFT pathogenesis.  

      Response to Reviewers:

      (1) The authors state that "NAB2-STAT6 localization is exclusively driven by EGR1 binding" yet WT1 motives are also consistently enriched. Can you please touch upon the potential involvement of WT1 (or lack thereof, and why)?

      Our data suggest that EGR1 is the primary driver of NAB2-STAT6 localization. In fact, EGR1 is the most significantly enriched motif (Fig. 4) at NAB2-STAT6 binding sites and we detect an interaction between the fusion protein and EGR1 (Fig. 5). Conversely, we did not identify an interaction between NAB2-STAT6 and WT1. However, WT1 also belongs to the C2H2 zinc finger subclass and recognizes a motif bearing striking similarities to the EGR1/2 consensus. EGR1 has been previously described to bind WT1 motifs and to function as an activator of WT1 targets (as opposed to WT1 repressive abilities). See https://www.jbc.org/article/S0021-9258(20)74720-4/fulltext and https://www.sciencedirect.com/science/article/pii/S0378111901005935.

      (2) In the description of Figure 5C the authors observe nuclear staining of both NAB2 and STAT6 following NAB2-STAT6 fusion induction. They interpret this as the fusion stimulates nuclear translocation of endogenous NAB2. This statement can only be rigorously made if the authors can unequivocally demonstrate that their antibody exclusively detects endogenous NAB2 and not the NAB2 portion of the fusion. As presented, a more likely interpretation is that the NAB2 staining detects NAB2-STAT6 fusion protein. Since there is some cytoplasmic NAB2 signal still present, the findings in Figure 5c do not support nor disprove nuclear translocation of endogenous NAB2. It may be prudent to remove this section. Figure 5B is currently the best direct evidence of nuclear translocation.

      We agree with the reviewer that Fig. 5C does not rigorously show that NAB2-STAT6 fusion proteins drag endogenous NAB2 into the nucleus. The immunostaining reveals that wt NAB2 localization is overwhelmingly cytoplasmic at steady-state conditions (and prior to expression of the fusion protein). Instead, Figure 5B shows that endogenous NAB2 translocates to the nucleus upon NAB2-STAT6 expression. Additionally, figure 5A (along with Suppl. Fig. 5 E-F) demonstrates that endogenous NAB2 co-precipitates with NAB2-STAT6 fusions in nuclear extracts of U2OS and HEK293T cells. We have rephrased the paragraph accordingly.

      (3) Figure 5D: for the interpretation of the presented data to hold up, namely, NAB1 nuclear translocation upon NAB2-STAT6 expression, it is important to demonstrate that NAB1 antibodies do not cross-react with NAB2 given the similarity between NAB1 and NAB2. Without such control, another likely interpretation of the results in Figure 5D is that NAB1 antibody detects the NAB2 portion of the overexpressed fusion protein. This needs to be acknowledged in the text.

      We had similar concerns, therefore we confirmed that the NAB1 antibody does not cross react with NAB2 by immunoblot (see figure below). We overexpressed FLAG-NAB2, HA-NAB1 and GFP constructs in HEK293T cells, we performed immunoprecipitation with either HA or FLAG from whole cell extracts followed by western blot using anti-NAB2 and anti-NAB1 polyclonal antibodies. We did not observe cross-reactivity of these antibodies. We acknowledged antibody validation in the revised text.

      Author response image 1.

      (4) Also, to support the notion that NAB2-STAT6 fusion promotes nuclear translocation of the entire complex, an imaging approach detecting EGR1 similar to Figure 5C-D would be helpful. EGR1 staining also avoids the potential pitfall of NAB1/2 antibodies detecting NAB2-STAT6 overexpressed fusion instead of endogenous proteins.

      We agree with the reviewer that this would be a helpful approach. Unfortunately, none of the commercially available EGR1 antibodies that we tested were suitable for immunocytochemistry, as they either failed to show a proper signal or were marred by high nonspecific background signal.

      (5) The authors found increased mRNA expression of certain cytokines and secreted neuropeptides in SFTs. While this may be consistent with a secretory phenotype, additional evidence such as detection of elevated levels of these proteins in tumor lysates or in culture media is necessary to formally make this claim. Please rephrase.

      We have rephrased our claims as suggested. The revised text is now as follows: “​​We also identified a distinct secretory gene signature associated with SFTs. In fact, IGF2 is the most upregulated gene, via activation of an intronic enhancer by EGR1. IGF2 was pinpointed as the cause of hypoglycemia occurring in a very small subset of SFTs (Doege–Potter syndrome)(52). Our data suggest that IGF2 (and IGF1) upregulation is a common feature of all SFTs. In addition to insulin-like growth factors, STFs may secrete a host of peptides with diverse functions in neuronal processes, chemotaxis, and growth stimulation. The previously unrecognized neuronal features and the putative secretory phenotype of STFs set them apart from mesenchymal malignancies and relate them to neuroendocrine malignancies such as pheochromocytoma, oligodendroglioma and neuroblastoma.”

      (6) GSEA with 500 randomly selected genes from target datasets needs a more detailed description to clarify the method.

      To improve clarity, we added the following description: “Gene set enrichment analysis (GSEA) was done with 500 randomly selected genes from the given set of genes across the C2 collection of the human molecular signatures database or custom signatures using the GSEA function in clusterProfiler package in R (v4.6.2).

      (7) In the IP-MS description, please double check the NaCl concentration in the second extraction step - 0.5mM seems low. Also, in the IP part, a buffer recipe appears to have been incorrectly pasted.

      We thank the reviewer for identifying this typo. Indeed, we used 0.5M NaCl instead of 0.5mM. We have corrected the co-IP buffer recipe accordingly.

    1. eLife Assessment

      This valuable contribution combines high-resolution histology with magnetic resonance imaging in a novel way to study the organisation of the human amygdala. The main findings convincingly show the axes of microstructural organisation within the amygdala and how they map onto the functional organisation. Overall, the approach taken in this paper showcases the utility of combining multiple modalities at different spatial scales to help understand brain organisation.

    2. Reviewer #1 (Public review):

      The paper by Auer et. makes several contributions:

      (1) The study developed a novel approach to map the microstructural organization of the human amygdala by applying radiomics and dimensionality reduction techniques to high-resolution histological data from the BigBrain dataset.

      (2) The method identified two main axes of microstructural variation in the amygdala, which could be translated to in vivo 7 Tesla MRI data in individual subjects.

      (3) Functional connectivity analysis using resting-state fMRI suggests that microstructurally defined amygdala subregions had distinct patterns of functional connectivity to cortical networks, particularly the limbic, frontoparietal, and default mode networks.

      (4) Meta-analytic decoding was used to suggest that the superior amygdala subregion's connectivity is associated with autobiographical memory, while the inferior subregion was linked to emotional face processing.

      (5) Overall, the data-driven, multimodal approach provides an account of amygdala microstructure and possibly function that can be applied at the individual subject level, potentially advancing research on amygdala organization.

    3. Reviewer #2 (Public review):

      Summary:

      This study bridges a micro- to macroscale understanding of the organization of the amygdala. First, using a data-driven approach, the authors identify structural clusters in the human amygdala from high-resolution post-mortem histological data. Next, multimodal imaging data to identify structural subunits of the amygdala and the functional networks in which they are involved. This approach is exciting because it permits the identification of both structural amygdalar subunits, and their functional implications, in individual subjects. There are, however, some differences in the macro and microscale levels of organization that should be addressed.

      Strengths:

      The use of data-driven parcellation on a structure that is important for human emotion and cognition, and the combination of this with high-resolution individual imaging-based parcellation, is a powerful and exciting approach, addressing both the need for a template-level understanding of organization as well as a parcellation that is valid for individuals. The functional decoding of rsfMRI permits valuable insight into the functional role of structural subunits. Overall, the combination of micro to macro, structure, and function, and general organization to individual relevance is an impressive holistic approach to brain mapping.

    4. Author response:

      The following is the authors’ response to the original reviews.

      Reviewer #1:

      The paper by Auer et. makes several contributions: (1) The study developed a novel approach to map the microstructural organization of the human amygdala by applying radiomics and dimensionality reduction techniques to high-resolution histological data from the BigBrain dataset. (2) The method identified two main axes of microstructural variation in the amygdala, which could be translated to in vivo 7 Tesla MRI data in individual subjects. (3) Functional connectivity analysis using resting-state fMRI suggests that microstructurally defined amygdala subregions had distinct patterns of functional connectivity to cortical networks, particularly the limbic, frontoparietal, and default mode networks. (4) Meta-analytic decoding was used to suggest that the superior amygdala subregion's connectivity is associated with autobiographical memory, while the inferior subregion was linked to emotional face processing. (5) Overall, the data-driven, multimodal approach provides an account of amygdala microstructure and possibly function that can be applied at the individual subject level, potentially advancing research on amygdala organization.

      We thank the Reviewer for the positive comments and insightful evaluation of the work.

      (1.1) Although these are meritorious contributions there are some concerns that I will summarize below. The paper makes little-to-no contact with the monkey literature regarding the anatomy of amygdala subregions, their functionality, and their patterns of anatomical connectivity. This is surprising because such literature on non-human primates is a very important starting point for understanding the human amygdala. I recommend taking a careful look at the work by Helen Barbas, among others. There are too many papers to cite but a notable example is: Ghashghaei, H. T., Hilgetag, C. C., & Barbas, H. (2007). Sequence of information processing for emotions based on the anatomic dialogue between prefrontal cortex and amygdala. Neuroimage, 34(3), 905-923. The work of Amaral is also highly relevant.

      As suggested, we included the important work of Amaral et al. as well as Ghashghaei et al. highlighting its contribution to mapping the intricate anatomy and function of the amygdala in non-human primates. We comment on this in the Introduction of the manuscript. Please see P.3.

      “Early research on the amygdala in non-human primates has been instrumental in understanding its intricate structure, function and patterns of anatomical connectivity (Amaral and Price 1984; Ghashghaei et al. 2007). This foundational study highlights the amygdala’s different subdivisions, most notably the basomedial nucleus (BM), basolateral nucleus (BL), and central nucleus (Ce) (Amaral et al. 1992). Furthermore, this work describes a dense network between these subdivisions and the prefrontal cortex, most strongly found in the posterior orbitofrontal and anterior cingulate areas.”

      (1.2) Furthermore, the authors subscribe to a model with LB, CM, and SF sectors. How does the SF sector relate to monkey anatomy?

      The overall organization of these subregions is largely conserved between humans and monkeys, reflecting their evolutionary relationship. While the basic subregional organization is conserved, there are still some important structural and functional differences between human and monkey amygdalae. For example, the SF subregion, often described in humans includes parts of the cortical nuclei (VCo), anterior amygdaloid area (AAA), amygdalohippocampal transition area (AHi), amygdalopiriform transition area (APir) as well as the lateral olfactory tract (LOT). This remark was added in the Discussion, on P.12:

      “However, this region has been previously described as consisting of three main subdivisions: LB, CM, and SF, each composed of smaller subnuclei with distinct connectivity patterns and functions (Amunts et al. 2005; Ball et al. 2007; Bzdok et al. 2013; de Olmos and Heimer 1999). These subregions are largely conserved between humans and monkeys, reflecting their evolutionary relationship. However, there are still some considerable differences such as in the SF subregion, where its description in monkeys additionally contains the lateral olfactory tract (LOT) (De Olmos 1990).”

      (1.3) The authors use meta-analytical decoding via NeuroSynth. If the authors like those results of course they should keep them but the quality of coordinate reporting in the literature is insufficient to conclude much in the context of amygdala subregion function in my opinion. I believe the results reported are at most "somewhat suggestive".

      We agree with the Reviewer that use of data from NeuroSynth poses unique challenges, particularly relating to investigations of a small structure such as the amygdala. However, to clarify, these analyses decode the cortex-wide functional connectivity patterns of amygdala subregions and not activations within subregions defined by our microanatomical analyses. Additionally, comments from Reviewer 2 suggested expanding the NeuroSynth decoding to the contralateral hemisphere. As such, we decided to keep this analysis in the main manuscript but rephrase the interpretation of these findings in the Discussion to emphasize their exploratory nature on P.13:

      “Functional decoding of subregional functional connectivity patterns indicated possible dissociations in cognitive (e.g., memory) and affective (e.g., emotional face processing) functions of the amygdala, echoing previous accounts of this region’s involvement in associative processing of emotional stimuli. Notably, these findings link the functional connectivity profile of a subregion partially co-localizing with LB to emotional face processing. The LB subregion has been previously linked to associative processing related to the integration of sensory information (Bzdok et al. 2013; Ghods-Sharifi, St Onge, and Floresco 2009; Pessoa 2010; Winstanley et al. 2004; Boyer 2008), which is consistent with the association with visual emotional information processing identified in the present work.”

      (1.4) Another significant concern has to do with the results in Figure 3. The red and yellow clusters identified are quite distinct but the differences in functional connectivity are very modest. Figure 3C reveals very similar functional connectivity with the networks investigated. This is very surprising, and the authors should include a careful comparison with related findings in the literature. Overall, there is limited comparison between the observed results and those obtained via other methods. On a more pessimistic note, the results of Figure 3 seem to question the validity of the general approach.

      We agree with the Reviewer that we can indeed observe considerable overlap between functional connectivity profiles of amygdala subregions. The amygdala is a relatively small structure, leading to likely interconnectivity between its subregions (Bzdok et al. 2013) in addition to considering BOLD signal autocorrelation within this region. In addition, functional signals in the amygdala are affected by relatively lower signal-to-noise ratio (SNR), a limitation extending to temporobasal and mesiotemporal regions. Despite these challenges, our technique remained sensitive to detect subtle differences in connectivity patterns even in this small group of subjects in this restricted subcortical territory.

      In the revised manuscript, we further highlight these caveats in the Discussion (P.13):

      “Although these findings are promising, we also observe considerable overlap between functional connectivity networks of both our defined subregions. Indeed, the amygdala is a relatively small structure, leading to likely interconnectivity between its subregions and locally high signal autocorrelation. Functional connectivity and microstructure in the amygdala are certainly related, however previous work suggests they do not perfectly overlap (Bzdok et al. 2013). In addition, this region is affected by relatively low signal-to-noise ratio (SNR), as is observed in broader temporobasal and mesiotemporal territories.”

      (1.5) Some statements in the Discussion feel unwarranted. For example, "significant dissociation in functional connectivity to prefrontal structures that support self-referential, reward-related, and socio-affective processes." This feels way beyond what can be stated based on the analyses performed.

      We agree that this interpretation may reach beyond the analyses performed and reported findings. We have adjusted this portion of the text accordingly in our Discussion on functional connectivity findings (P.13):

      “Qualitatively, we found that the subregion defined by the highest 25% of U1 values mainly overlapped with what is commonly defined as the superficial and centromedial subregions, whereas the lowest 25% U1 values subregion overlapped mostly with the laterobasal division. Interestingly, CM and SF characterized subregions showed significantly stronger functional connectivity to prefrontal structures. This finding aligns with previous work demonstrating unique affiliations between the CM subregion and anterior cingulate and frontal cortices (Kapp, Supple, and Whalen 1994; Barbour et al. 2010), as well as between the SF subregion and the orbitofrontal cortex (Goossens et al. 2009; Caparelli et al. 2017; Pessoa 2010; Klein-Flügge et al. 2022).”

      Additionally, we have also edited our Discussion to ensure that our interpretations are grounded in the analyses conducted, while framing the findings as potential avenues for future work. Please see P.13.

      “Functional decoding of functional connectivity results indicated possible dissociations in cognitive (e.g., memory) and affective (e.g., emotional face processing) functions of the amygdala, echoing previous accounts of this region’s functional specialization and subregional segregation of associative processing of emotional stimuli.”

      Recommendations for the authors:

      (1.6) Figure 1 has panels A-I but only A-D are discussed in the caption. The orientation of the slices is not indicated which makes it very hard to follow for most readers.

      The subpanels are now referred to in the revised Results. We also added a notation on the orientation of the slices and described them accordingly in our Figure 1 description. (P.5-6):

      “(A) The amygdala was segmented from the 100-micron resolution BigBrain dataset using an existing subcortical parcellation (Xiao et al. 2019). Slice orientation is consistent across all panels in this figure.”

      (1.7) Some figure references in the text seem to be incorrect; please check that the text refers to the correct figure number and panel.

      We thank the Reviewer for pointing this out. We thoroughly revised the correspondence between figure panel labels and their referencing in the text.

      Reviewer #2:

      This study bridges a micro- to macroscale understanding of the organization of the amygdala. First, using a data-driven approach, the authors identify structural clusters in the human amygdala from high-resolution post-mortem histological data. Next, multimodal imaging data to identify structural subunits of the amygdala and the functional networks in which they are involved. This approach is exciting because it permits the identification of both structural amygdalar subunits, and their functional implications, in individual subjects. There are, however, some differences in the macro and microscale levels of organization that should be addressed.

      Strengths:

      The use of data-driven parcellation on a structure that is important for human emotion and cognition, and the combination of this with high-resolution individual imaging-based parcellation, is a powerful and exciting approach, addressing both the need for a template-level understanding of organization as well as a parcellation that is valid for individuals. The functional decoding of rsfMRI permits valuable insight into the functional role of structural subunits. Overall, the combination of micro to macro, structure, and function, and general organization to individual relevance is an impressive holistic approach to brain mapping.

      We thank the Reviewer for their constructive and helpful feedback on our work.

      Weaknesses:

      (2.1) UMAP 1, as calculated from the histological data, appears to correlate well across individuals, and decently with the MRI data, although the medial-lateral coordinate axis is an outlier. UMAP 2, on the other hand, does not appear to correlate well with imaging data or across individuals. This does pose a problem with the claim that this paper bridges micro- and macroscale parcellations. One might certainly expect, however, that different levels of organization might parcellate differently, but the authors should address this in the discussion and offer ways forward.

      Data driven methods hold several advantages for the quantitative extraction of signal from the underlying data in an observer-independent manner. However, these techniques are also sensitive to potential idiosyncrasies in the data. In the present work, our main analyses rely on the processing of a histological dataset (BigBrain) providing a unique opportunity for high-resolution analysis of amygdala histology and in vivo translation of findings leveraging ultra-high field MRI (n=10). However, both datasets are limited by their small sample size (n=1 for BigBrain and n=10 for MICA-PNI). As a result, we speculate that signal variations captured by U2 may be sensitive to artifacts or subject-specific sources of variance. Moving forward, this hypothesis could be assessed in future work via the analysis of larger histological and neuroimaging datasets to better track recurring features picked up by U2 or the association of these unique topographies with behavioural markers.

      As suggested, we included a section in our Discussion highlighting this shortcoming and the importance for larger datasets moving forward. Please see P.11-12.

      “However, it is important to note that both datasets analyzed in this work are limited by their small sample size (n=1 for BigBrain and n=10 for MICA-PNI). We speculate that the signal variations captured by U2 may be sensitive to artifacts or subject-specific sources of variance, potentially explaining why it was not consistent between subjects and modalities. Moving forward, this hypothesis could be assessed in future work via the analysis of larger histological and neuroimaging datasets to better track recurring features picked up by U2 or the association of these unique topographies with behavioural markers.”

      (2.1) It would be interesting to see functional decoding for the right amygdala. This could be included in the supplementary material. A discussion of differences in the results in the two hemispheres could be illuminating.

      In accordance with the Reviewer’s suggestion, we added Supplementary figure S2 exploring the decoding of connectivity profiles of the right amygdala stratified by its cytoarchitectural embedding with UMAP.

      Upon analysis, dissociation in functional connectivity patterns over the right amygdala were less evident, leading to overall similar functional decoding across the two clusters. We refer to this Supplementary Figure in our Discussion on P.13.

      “For the right amygdala, dissociation in functional connectivity patterns were more subtle, leading to overall similar functional decoding across the two clusters. (Figure S2)”

      (2.3) The authors acknowledge that this mapping matches some but not all subunits that have been previously described in the amygdala. It would be helpful to neuroanatomists if the authors could discuss these differences in more detail in the discussion, to identify how this mapping differs and what the implications of this are.

      In our work, we focus on mapping the three well characterized amygdala subregions, specifically the superficial (SF), centromedial (CM) and laterobasal (LB) subdivisions. Qualitative histological accounts have indeed delineated multiple subunits within these subregions which we now describe in the revised manuscript. Due to the lower resolution of in vivo MRI data used in this work relative to post mortem histology, we focused our analyses on larger subregions that could be more reliably mapped to native quantitative T1 spaces of each participant. We now overview this issue in the Discussion. Please see P.12.

      “Although qualitative histological accounts have indeed delineated multiple subunits within these general regions, the present work focuses on three subdivisions (Amunts et al. 2005) to account for resolution disparities when translating our findings to in vivo MRI data. The LB subdivision includes the basomedial nucleus (Bm), basolateral nucleus (BL), lateral nucleus (LA) and paralaminar nucleus (PL). Moving medially, the CM subdivision includes the central (Ce) and medial nuclei (Me), while the SF subdivision includes the anterior amygdaloid area (AAA), amygdalohippocampal transition area (AHi), amygdalopiriform transition area (APir), and ventral cortical nucleus (VCo) (Heimer et al. 1999). However, disagreement on the precise attribution of nuclei to broader subdivisions motivated our investigations of probabilistic subunits of the amygdala (Kedo et al. 2018). The development of new tools to segment amygdala subnuclei in vivo offers opens opportunities for future work to further validate our framework at the precision of these nuclei within subjects (Saygin et al. 2017).”

      (2.4) The acronym UMAP is not explained. A brief explanation and description would be useful to the reader.

      We moved the expanded acronym from the Methods to the first instance of the term UMAP in our paper, found in the Introduction. As suggested, we also added a sentence describing the technique. Please see P.6.

      “We then applied Uniform Manifold Approximation and Projection (UMAP), a non-linear dimensionality reduction technique that preserves the local and global structure of high-dimensional data by projecting it into a lower-dimensional space (Becht et al. 2018), to the resulting 20-feature matrix to derive a 2-dimensional embedding of amygdala cytoarchitecture (Figure 1D).”

    1. eLife Assessment

      This important study provides insights into the role of the cerebellum in fear conditioning, addressing a key gap in the literature. The evidence presented is solid overall, although the theoretical framing and clarity of the results can be improved and some concerns remain about the reliability of results based on small numbers of trials. This work will be of interest to both the extinction learning and cerebellar research communities.

    2. Reviewer #1 (Public review):

      Nio and colleagues address an important question about how the cerebellum and ventral tegmental area (VTA) contribute to the extinction learning of conditioned fear associations. This work tackles a critical gap in the existing literature and provides new insights into this question in humans through the use of high-field neuroimaging with robust methodology. The presented results are novel and will broadly interest both the extinction learning and cerebellar research communities. As such, this is a very timely and impactful manuscript. However, there are several points that could be addressed during the review process to strengthen the claims and enhance their value for readers and the broader scientific community.

      Points to Address:

      (1) Reward Interpretation and Skin Conductance Responses (SCR):<br /> A central premise of the manuscript is that 'unexpected omissions of expected aversive events' are rewarding, which plays a critical role in extinction learning. The authors also suggest that the cerebellum is involved in reward processing. However, it is unclear how this conclusion can be directly drawn from their task, which does not explicitly model 'reward.' Instead, the interpretation relies on SCR, which seems more indicative of association or prediction rather than reward per se. Is SCR a valid metric of reward experienced during the extinction of feared associations? Or could these findings reflect processes tied more closely to predictive learning? Please, discuss.

      (2) Reinforcement Agent and SCR Modeling:<br /> The modeling approach with the deep reinforcement agent treats SCR as a personalized expectation of shock for a given trial. However, this interpretation seems misaligned with participants' actual experience - they are aware of the shock but exhibit evolving responses to it over time. Why is this operationalization useful or valid? It would benefit the manuscript to provide a clearer justification for this approach.

      (3) Clarity and Visualization of Results:<br /> The results section is challenging to follow, and the visualization and quantification of findings could be significantly improved. Terms like 'trending' appear frequently - what does this mean, and is it worth reporting? Adding clear statistical quantifications alongside additional visualizations (e.g., bar or violin plots of group means within specific subregions within the cerebellum, or grouped mean activity in VTA and DCN) would enhance clarity and allow readers to better assess the distribution and systematicity of effects. Furthermore, the figures are overly complex and difficult to read due to the heavy use of abbreviations. Consider splitting figures by either phase of the experiment or regions, and move some details to the supplemental material for improved readability.

      (4) Theoretical Context for Paradigm Phases:<br /> The manuscript benefits from the comprehensive experimental paradigm, which includes multiple phases (acquisition, extinction, recall, reacquisition, re-extinction). This design has great potential for providing a more holistic view of conditioned fear learning and extinction. However, the manuscript lacks clarity on what insights can be drawn from these distinct phases. What theoretical framework underpins the different stages, and how should the results be interpreted in this context? At present, the findings seem like a display of similar patterns across phases without sufficient interpretation. Providing a stronger theoretical rationale and reorganizing the results by experimental phase could significantly improve readability and impact.

      (5) Cerebellum-VTA Connectivity Analysis:<br /> The authors argue that the cerebellum modulates VTA activity, yet they perform the PPI analysis in the reverse direction. Why does this make sense? In their DCM analysis, they found a bidirectional relationship (both cerebellum - VTA and VTA-cerebellum), yet the discussion focused on connectivity from the cerebellum to VTA. A more careful interpretation of the connectivity findings would be useful - especially the strong claims in the discussion on the cerebellum providing the reward signal to the VTA should be tempered.

    3. Reviewer #2 (Public review):

      Summary:

      Building upon the group's previous work, this study used a 3-day threat acquisition, extinction, recall, reextinction, and reacquisition paradigm with 7T imaging to probe the mechanism by which the cerebellum contributes to fear extinction learning. The authors hypothesise this may be via its connection to the VTA, a known modulator of fear extinction due to its role in reward processing. Using complementary analysis methods, the authors demonstrate that activity with the cerebellum, DNC, and VTA is modulated by predictions about the occurrence of the US, which shows regional specificity. They show trend-level evidence that there is increased functional connectivity between the cerebellum and VTA during all phases of the paradigm with unexpected omissions. They also present a DCM which indicates that the cerebellum could positively modulate VTA activity during extinction learning. This study adds to a growing literature supporting the role of the historically overlooked cerebellum in the control of emotions and suggests that an interaction between the cerebellum and VTA should be considered in the existing model of the fear extinction network.

      Strengths:

      The authors address their research question using a number of complementary methods, including parametric modulation by model-derived expectation parameters, PPI, and DCM, in a logical and easily understood way. I feel the authors provide a balanced interpretation of their findings, presenting numerous interpretations and offering insight with regard to reward vs attention or unsigned prediction errors and the directionality of the interaction they identify. The manuscript is a timely addition to growing literature highlighting the role of the cerebellum in fear conditioning, and emotion generation and regulation more generally.

      Weaknesses:

      Subjective and skin conductance responses do not completely support the success of the learning paradigm. For example, CS+/CS- differentiation in both domains persisted after extinction training. I do not feel that this negates the findings of this manuscript, though it raises questions about the parametric modulators used, and the interpretation of the neural mechanisms proposed if they do not strongly relate to updated subjective appraisals (the goal of extinction therapy). My interpretation of the manuscript suggests there are some key results based upon contrasts that have as few as three events; I am a little unsure about the power and reliability of these effects, though I await author clarification on this matter. There are a number of unaddressed deviations from the pre-registered protocol that I have asked the authors to elaborate upon.

    4. Author response:

      Reviewer 1:

      (1) Reward Interpretation and Skin Conductance Responses (SCR):

      The reviewer raises a valid point, as the model from which we derive prediction errors describes predictive learning—specifically, the occurrence of shock—without incorporating additional reward learning effects. SCRs are used to fit the model’s hyperparameters but do not directly measure reward; rather, they serve as a marker of arousal.

      In our paradigm, SCRs are measured during CS presentation and primarily reflect predictive learning, as they are closely linked to contingency awareness. The association between estimated prediction errors during unexpected US omissions and reward remains reliant on existing literature.

      In the revised manuscript, we will further elaborate on these points to clarify the distinction between predictive learning and direct reward processing, while contextualizing our findings within the broader literature on reward signaling and fear extinction.

      (2) Reinforcement Agent and SCR Modeling:

      Notably, we do not use SCR as a personalized expectation measure due to its limited reliability at the individual level; instead, the model's hyperparameters are fitted on the entire SCR dataset, yielding per-trial prediction and prediction error estimates for each CS sequence rather than for individual participants.

      (3) Clarity and Visualization of Results:

      We recognize that the presentation of our results can be improved and will take steps to enhance figure clarity, also ensuring that trend-level results are clearly distinguished.

      (4) Theoretical Context for Paradigm Phases:

      Regarding the differences across experimental phases, we recognize the theoretical significance of these distinctions. However, our primary focus is on identifying commonalities in unexpected US omission responses across phases rather than emphasizing phase-specific differences. Nevertheless, we will provide a brief clarification on phase differences to enhance the manuscript’s interpretability.

      (5) Cerebellum-VTA Connectivity Analysis:

      Furthermore, we acknowledge that our conclusion regarding the modulation of the dopaminergic system by the cerebellum should be framed more cautiously. We will temper our claims to better reflect the bidirectional and potentially indirect nature of cerebellum-VTA interactions. Additionally, we plan to include PPI results using a cerebellar seed showing the VTA, potentially in the supplementary material.

      Reviewer 2:

      (1) Success of extinction learning based on Self-reports and SCRs?

      The reviewer points to a problem, which is inherent to extinction learning: The initial fear association is not erased, but merely inhibited, and is prone to return. Although the recall phase follows the extinction phase, we did not expect a complete inhibition of the conditioned response; instead, spontaneous recovery is expected. In fact, the spontaneous recovery observed in the recall phase provided us with an additional opportunity to investigate unexpected US omissions, which was our primary focus.

      (2) Concerns on reliability of event-based contrasts using three events:

      Regarding concerns about the reliability of analyses based on three events, we believe that the consistency of our parametric modulation analysis— which incorporates all events— combined with the three-event analysis results, provides further support for the observed patterns. We are currently discussing ways of additional analysis for further verification of the reliability of using three events.

      (3) Deviations from preregistration:

      Finally, we will carefully review all deviations from our preregistration to ensure transparency. Any methodological or analytical changes will be explicitly addressed in the revised manuscript.

    1. eLife Assessment

      This research addresses an important and timely topic in cancer treatment, as the authors present a novel computational tool, 'retriever,' which has the potential to revolutionize personalized cancer treatment strategies by predicting effective drug combinations for triple-negative breast cancer. The strength of the evidence presented is solid, as evidenced by the systematic testing of 152 drug response profiles and 11,476 drug combinations.

    2. Reviewer #1 (Public review):

      Summary:

      Identifying drugs that target specific disease phenotypes remains a persistent challenge. Many current methods are only applicable to well-characterized small molecules, such as those with known structures. In contrast, methods based on transcriptional responses offer broader applicability because they do not require prior information about small molecules. Additionally, they can be rapidly applied to new small molecules. One of the most promising strategies involves the use of "drug response signatures"-specific sets of genes whose differential expression can serve as markers for the response to a small molecule. By comparing drug response signatures with expression profiles characteristic of a disease, it is possible to identify drugs that modulate the disease profile, indicating a potential therapeutic connection.

      This study aims to prioritize potential drug candidates and to forecast novel drug combinations that may be effective in treating triple-negative breast cancer (TNBC). Large consortia, such as the LINCS-L1000 project, offer transcriptional signatures across various time points after exposing numerous cell lines to hundreds of compounds at different concentrations. While this data is highly valuable, its direct applicability to pathophysiological contexts is constrained by the challenges in extracting consistent drug response profiles from these extensive datasets. The authors use their method to create drug response profiles for three different TNBC cell lines from LINCS.

      To create a more precise, cancer-specific disease profile, the authors highlight the use of single-cell RNA sequencing (scRNA-seq) data. They focus on TNBC epithelial cells collected from 26 diseased individuals compared to epithelial cells collected from 10 healthy volunteers. The authors are further leveraging drug response data to develop inhibitor combinations.

      Strengths:

      The authors of this study contribute to an ongoing effort to develop automated, robust approaches that leverage gene expression similarities across various cell lines and different treatment regimens, aiming to predict drug response signatures more accurately. The authors are trying to address the gap that remains in computational methods for inferring drug responses at the cell subpopulation level.

      Weaknesses:

      One weakness is that the authors do not compare their method to previous studies. The authors develop a drug response profile by summarizing the time points, concentrations, and cell lines. The computational challenge of creating a single gene list that represents the transcriptional response to a drug across different cell lines and treatment protocols has been previously addressed. The Prototype Ranked List (PRL) procedure, developed by Iorio and co-authors (PNAS, 2010, doi:10.1073/pnas.1000138107), uses a hierarchical majority-voting scheme to rank genes. This method generates a list of genes that are consistently overexpressed or downregulated across individual conditions, which then hold top positions in the PRL. The PRL methodology was used by Aissa and co-authors (Nature Comm 2021, doi:10.1038/s41467-021-21884-z) to analyze drug effects on selective cell populations using scRNA-seq datasets. They combined PRL with Gene Set Enrichment Analysis (GSEA), a method that compares a ranked list of genes like PRL against a specific set of genes of interest. GSEA calculates a Normalized Enrichment Score (NES), which indicates how well the genes of interest are represented among the top genes in the PRL. Compared to the method described in the current manuscript, the PRL method allows for the identification of both upregulated and downregulated transcriptional signatures relevant to the drug's effects. It also gives equal weight to each cell line's contribution to the drug's overall response signature.

      The authors performed experimental validation of the top two identified drugs; however, the effect was modest. In addition, the effect on TNBC cell lines was cell-line specific as the identified drugs were effective against BT20, whose transcriptional signatures from LINCS were used for drug identification, but not against the other two cell lines analyzed. An incorrect choice of genes for the signature may result in capturing similarities tied to experimental conditions (e.g., the same cell line) rather than the drug's actual effects. This reflects the challenges faced by drug response signature methods in both selecting the appropriate subset of genes that make up the signature and in managing the multiple expression profiles generated by treating different cell lines with the same drug.

    3. Reviewer #2 (Public review):

      Summary:

      In their study, Osorio and colleagues present 'retriever,' an innovative computational tool designed to extract disease-specific transcriptional drug response profiles from the LINCS-L1000 project. This tool has been effectively applied to TNBC, leveraging single-cell RNA sequencing data to predict drug combinations that may effectively target the disease. The public review highlights the significant integration of extensive pharmacological data with high-resolution transcriptomic information, which enhances the potential for personalized therapeutic applications.

      Strengths:

      A key finding of the study is the prediction and validation of the drug combination QL-XII-47 and GSK-690693 for the treatment of TNBC. The methodology employed is robust, with a clear pathway from data analysis to experimental confirmation.

      Weaknesses:

      However, several issues need to be addressed. The predictive accuracy of 'retriever' is contingent upon the quality and comprehensiveness of the LINCS-L1000 and single-cell datasets utilized, which is an important caveat as these datasets may not fully capture the heterogeneity of patient responses to treatment. While the in vitro validation of the drug combinations is promising, further in vivo studies and clinical trials are necessary to establish their efficacy and safety. The applicability of these findings to other cancer types also warrants additional investigation. Expanding the application of 'retriever' to a broader range of cancer types and integrating it with clinical data will be crucial for realizing its potential in personalized medicine. Furthermore, as the study primarily focuses on kinase inhibitors, it remains to be seen how well these findings translate to other drug classes.

    4. Author response:

      Reviewer 1:

      Summary:

      Identifying drugs that target specific disease phenotypes remains a persistent challenge. Many current methods are only applicable to well-characterized small molecules, such as those with known structures. In contrast, methods based on transcriptional responses offer broader applicability because they do not require prior information about small molecules. Additionally, they can be rapidly applied to new small molecules. One of the most promising strategies involves the use of “drug response signatures”-specific sets of genes whose differential expression can serve as markers for the response to a small molecule. By comparing drug response signatures with expression profiles characteristic of a disease, it is possible to identify drugs that modulate the disease profile, indicating a potential therapeutic connection.

      This study aims to prioritize potential drug candidates and to forecast novel drug combinations that may be effective in treating triple-negative breast cancer (TNBC). Large consortia, such as the LINCS-L1000 project, offer transcriptional signatures across various time points after exposing numerous cell lines to hundreds of compounds at different concentrations. While this data is highly valuable, its direct applicability to pathophysiological contexts is constrained by the challenges in extracting consistent drug response profiles from these extensive datasets. The authors use their method to create drug response profiles for three different TNBC cell lines from LINCS.

      To create a more precise, cancer-specific disease profile, the authors highlight the use of single-cell RNA sequencing (scRNA-seq) data. They focus on TNBC epithelial cells collected from 26 diseased individuals compared to epithelial cells collected from 10 healthy volunteers. The authors are further leveraging drug response data to develop inhibitor combinations.

      Strengths:

      The authors of this study contribute to an ongoing effort to develop automated, robust approaches that leverage gene expression similarities across various cell lines and different treatment regimens, aiming to predict drug response signatures more accurately. The authors are trying to address the gap that remains in computational methods for inferring drug responses at the cell subpopulation level.

      Weaknesses:

      One weakness is that the authors do not compare their method to previous studies. The authors develop a drug response profile by summarizing the time points, concentrations, and cell lines. The computational challenge of creating a single gene list that represents the transcriptional response to a drug across different cell lines and treatment protocols has been previously addressed. The Prototype Ranked List (PRL) procedure, developed by Iorio and co-authors (PNAS, 2010, doi:10.1073/pnas.1000138107), uses a hierarchical majority-voting scheme to rank genes. This method generates a list of genes that are consistently overexpressed or downregulated across individual conditions, which then hold top positions in the PRL. The PRL methodology was used by Aissa and co-authors (Nature Comm 2021, doi:10.1038/s41467-021-21884-z) to analyze drug effects on selective cell populations using scRNA-seq datasets. They combined PRL with Gene Set Enrichment Analysis (GSEA), a method that compares a ranked list of genes like PRL against a specific set of genes of interest. GSEA calculates a Normalized Enrichment Score (NES), which indicates how well the genes of interest are represented among the top genes in the PRL. Compared to the method described in the current manuscript, the PRL method allows for the identification of both upregulated and downregulated transcriptional signatures relevant to the drug’s effects. It also gives equal weight to each cell line’s contribution to the drug’s overall response signature.

      The authors performed experimental validation of the top two identified drugs; however, the effect was modest. In addition, the effect on TNBC cell lines was cell-line specific as the identified drugs were effective against BT20, whose transcriptional signatures from LINCS were used for drug identification, but not against the other two cell lines analyzed. An incorrect choice of genes for the signature may result in capturing similarities tied to experimental conditions (e.g., the same cell line) rather than the drug’s actual effects. This reflects the challenges faced by drug response signature methods in both selecting the appropriate subset of genes that make up the signature and managing the multiple expression profiles generated by treating different cell lines with the same drug.

      We appreciate the reviewer’s thoughtful feedback and their suggestion to refer to the Prototype Ranked List (PRL) manuscript. Unfortunately, since this methodology for the PRL isn’t implemented in an open-source package, direct comparison with our approach is challenging. Nonetheless, we investigated whether using ranks would yield similar results for the most likely active drug pairs identified by retriever. To do this, we calculated and compared the rankings of the average effect sizes provided by retriever. Although the Spearman (ρ \= 0.98) correlation coefficient was high, we observed that key genes are disadvantaged when using ranks compared to effect sizes. This difference is particularly evident in the gene set enrichment analysis, where using average ranks identified only one pathway as statistically significantly enriched. The code to replicate these analyses is available at https://github.com/dosorio/L1000-TNBC/blob/main/Code/.

      Author response image 1.

      Given the similarity in purpose between retriever and the PRL approach, we have added the following statement to the introduction: “Previously, this goal was approached using a majority-voting scheme to rank genes across various cell types, concentrations, and time points. This approach generates a prototype ranked list (PRL) that represents the consistent ranks of genes across several cell lines in response to a specific drug.”

      Regarding the experimental validation, we believe there is a misunderstanding about the evidence we provided. We would like to claridy that we used three different TNBC cell lines: CAL120, BT20, and DU4475. It’s important to note that CAL120 and DU4475 were not included in the signature generation process. Despite this, we observed effects that exceeded the additive effects expectations, particularly in the CAL120 cell line (Figure 5, Panel F).

      Reviewer 2:

      Summary:

      In their study, Osorio and colleagues present ‘retriever,’ an innovative computational tool designed to extract disease-specific transcriptional drug response profiles from the LINCS-L1000 project. This tool has been effectively applied to TNBC, leveraging single-cell RNA sequencing data to predict drug combinations that may effectively target the disease. The public review highlights the significant integration of extensive pharmacological data with high-resolution transcriptomic information, which enhances the potential for personalized therapeutic applications.

      Strengths:

      A key finding of the study is the prediction and validation of the drug combination QL-XII-47 and GSK-690693 for the treatment of TNBC. The methodology employed is robust, with a clear pathway from data analysis to experimental confirmation.

      Weaknesses:

      However, several issues need to be addressed. The predictive accuracy of ’retriever’ is contingent upon the quality and comprehensiveness of the LINCS-L1000 and single-cell datasets utilized, which is an important caveat as these datasets may not fully capture the heterogeneity of patient responses to treatment. While the in vitro validation of the drug combinations is promising, further in vivo studies and clinical trials are necessary to establish their efficacy and safety. The applicability of these findings to other cancer types also warrants additional investigation. Expanding the application of ’retriever’ to a broader range of cancer types and integrating it with clinical data will be crucial for realizing its potential in personalized medicine. Furthermore, as the study primarily focuses on kinase inhibitors, it remains to be seen how well these findings translate to other drug classes.

      We thank the reviewer for their thoughtful and constructive feedback. We appreciate your insights and agree that several important considerations need to be addressed.

      We recognize that the predictive accuracy of retriever depends on the LINCS-L1000 and single-cell datasets. These resources may not fully represent the complete range of transcriptional responses to disease and treatment across different patients. As you mentioned, this is an important limitation. However, we believe that by extrapolating the evaluation of the most likely active compound to each individual patient, we can help address this issue. This approach will provide valuable insights into which patients in the study are most likely to respond positively to treatment.

      On the in-vitro validation of drug combinations, we agree that while promising, these results are not sufficient on their own to establish clinical efficacy. Additional in-vivo studies will be essential in assessing the therapeutic potential and safety of these combinations, and clinical trials will be an important next step to validate the translational impact of our findings.

      Lastly, we appreciate the reviewer’s comment about the focus of our study on kinase inhibitors. This result was unexpected, as we tested the full set of compounds from the LINCS-L1000 project. We agree that exploring other top candidates, including different drug classes, will be important for assessing how broadly retriever approach can be applied.

    1. eLife Assessment

      This important study provides new evidence on the role of norepinephrine (NE) release in the hippocampus in response to environmental transitions (event boundaries), providing a potential link between NE signaling and the segmentation of episodic memories. The work is solid, employing innovative techniques such as fiber photometry with the GRAB-NE sensor for NE measurement, the analysis of public electrophysiology hippocampal datasets, and well-controlled experiments. While further analysis could strengthen some claims, this work offers insights into memory, neuromodulation, and hippocampal function.

    2. Reviewer #1 (Public review):

      Summary:

      This study investigates the role of norepinephrine (NE) signaling in the hippocampus during event transitions, positing that NE release serves as a mechanism for marking event boundaries to facilitate episodic memory segmentation. The authors use a genetically encoded fluorescent indicator (GRABNE) to measure NE release with high temporal precision, correlating these signals with changes in hippocampal firing dynamics. By integrating photometry data, behavioral analyses, and analysis of neuronal activity from publicly available datasets, the work addresses fundamental questions about the relationship between neuromodulatory signals and memory encoding.

      Strengths:

      The authors present a compelling framework linking NE signaling to event boundaries, offering insight into how episodic memory segmentation may occur in the brain. The writing is clear and the data are well-described. It is easy to follow. The pharmacological validation of the GRABNE sensor enhances confidence in their NE measurements, an important methodological strength given the potential limitations of fluorescence-based neuromodulatory indicators. Moreover, the authors carefully disentangle NE signals from confounding behavioral variables, providing evidence that NE release is time-locked to event boundaries rather than movement or arousal-related behaviors. This level of analytical rigor strengthens their central claims. Additionally, the observation of NE signal dynamics that decay over hundreds of seconds is interesting, as it aligns with timescales relevant to hippocampal plasticity reported in prior literature.

      Weaknesses:

      While the authors establish correlations between NE signaling and hippocampal activity changes, causation is not demonstrated. Future studies using perturbative approaches (e.g., optogenetic or chemogenetic manipulation of NE release) would be necessary to establish a direct causal link. Furthermore, the persistence of NE signals over long timescales (hundreds of seconds) raises questions about its role in encoding rapid event boundaries, as it is unclear how this prolonged signaling might affect memory encoding for closely spaced events. The lack of a discussion about how NE dynamics would operate in such scenarios weakens the proposed framework. Finally, while the authors acknowledge the limitations of the GRABNE sensor, a more detailed exploration of how sensor sensitivity might influence their results would enhance the interpretation of their findings.

    3. Reviewer #2 (Public review):

      Summary:

      The authors use a genetically encoded fluorescent sensor, GRABNE, to measure NE dynamics in the dorsal hippocampus of mice in response to multiple behavioral manipulations. A non-linear model and regression were used to quantitatively assess the contribution of multiple behavioral covariates to changes in NE signaling, with the result that NE signal dynamics were best predicted by time from event transitions, with the signal exponentially decaying over a period of seconds to minutes after transitions. Event transitions were implemented as a transfer from a home cage to a novel arena, a transfer to a familiar linear track, and the introduction of novel objects. Additional experiments showed that spatial context transitions dominate NE signaling over novel object presentations, and experience accelerates the decay of the NE signal after spatial context transitions. Correspondingly, the hippocampal CA1 spatial code takes minutes to stabilize after context transition in both novel and familiar spaces.

      Strengths:

      A strength of the study is the use of the NE sensor with sub-second resolution, non-linear modeling, and regression to identify the prominent variable of interest as time from event transition, and multiple behavioral controls. The use of multiple behavioral designs to investigate the effect of familiarity, experience, and interaction of spatial context transitions and novel object introduction is a strength. Relating the dynamics of NE signal decay to the rate of CA1 spatial code changes is also a strength.

      Weaknesses:

      A minor weakness is that the concept of an event boundary needs to be more broadly discussed. The manuscript uses event transitions such as spatial context changes and novel object introduction to implement an event boundary. However, especially in episodic memory studies in humans, event structure and boundaries have also been shown to occur through the automatic segmentation of experiences into discrete events (Baldassano et al., Neuron, 2017; Radvansky and Zacks, Curr. Opi. Behav. Sci, 2017). The rodent experiments in the current manuscript explicitly introduce event boundaries through changes in context or objects, which can potentially be conflated with novelty. A discussion of these differences, and whether NE can also have a role in event boundary transitions based on automatic segmentation of experiences, will add to the impact of the manuscript.

    4. Reviewer #3 (Public review):

      Summary

      The manuscript investigates the role of norepinephrine (NE) release in the rodent hippocampus during event boundaries, such as transitions between spatial contexts and the introduction of novel objects. It also explores how NE release is altered by experience and how novelty drives the amplitude and decay times of extracellular NE. By utilizing the GRABNE sensor for sub-second resolution measurement of NE, the authors demonstrate that NE release is driven primarily by the time elapsed since an event boundary and is independent of behaviors like movement or reward. The study further explores how hippocampal neural representations are altered over time, showing that these representations stabilize shortly after event transitions, potentially linking NE release to episodic memory encoding.

      Strengths

      Overall, the work provides novel insights into the interplay between NE signaling and hippocampal activity and presents an intriguing hypothesis on how NE release may help push hippocampal activity into unique attractor states to encode novel experiences. The experiments are well-controlled, and the analysis is well-presented, with a detailed and engaging discussion that points towards several new and exciting research directions. The use of several behavioral paradigms to demonstrate the strongest predictor of NE release is a strength, as well as the regression analysis to disambiguate the contribution of other correlated variables. The suggestion that NE does not select ensembles for subsequent replay is also an interesting result.

      Weaknesses

      The authors have not convincingly established a link between hippocampal neural activity and NE release, showing qualitative rather than quantitative correlations. Therefore, at this stage, the role of NE on hippocampal function remains speculative.

      Another general concern is that the smoothing/ kinetics of the sensor impacts the regression analyses. Most of the other variables, such as speed, acceleration, and even reward time points are highly dynamic and it is possible that the limitations of the sensor decorrelate the signal from (potentially) causal variables, therefore resulting in the time since the event start having the most explanatory power for most of the analyses.

      More broadly, the figure legends should be expanded to better describe error bounds, mean vs median, sample sizes, and averaging choices for plots.

      There are also some concerns regarding the nearest neighbor analysis and the reported differences in the rate of reactivations after familiar and novel environments, as outlined below.

      (1) Lines 657-658. How far away in time can the top three nearest neighbor time points be? Must they lie in different trials, or can they also be within the same trial? Is there a systematic difference in the average time lags for the nearest neighbors over the course of the session?

      The authors should only allow nearest neighbors to be in a different lap because systematic changes in behavior (running fast initially) might force earlier time bins in a certain location to match with a different trial, while the later time bins can be from within the same trial if the mice are moving slower and stay in the same spatial bin location longer. The authors should also provide information on how the averaging is performed because there are several axes of variability - spatial bin locations, sessions, different environments, and animals.

      (2) Figure 8: These results are very interesting. However, I am confused by the differences between Figure 8B and D because the significant reactivations in A and C are very similar. The 1-minute and 10-minute windows seem somewhat arbitrary and prone to noise and variability. Perhaps the authors should fit a slope for the curves on A and C and compare whether the slope/ intercept are significantly different between the novel and familiar environments.

    1. eLife Assessment

      This study examined the important question of how neurons code temporal information across the hippocampus, dorsal striatum, and orbitofrontal cortex. Using a behavioral task in the rat that requires discrimination between short and long time intervals, the authors conclude that time intervals are represented in all three regions and that synchronized activity of time-coding cells across the brain regions is coordinated by theta rhythms. However, several weaknesses are noted, and in its current form, the study provides incomplete evidence for understanding how temporal information is processed and coordinated throughout these brain networks.

    2. Reviewer #1 (Public review):

      Summary:

      It is known that neuronal activity in several brain regions encodes interval time. However, how interval time is encoded across distributed brain regions remains unclear. By simultaneously recording neuronal activity from the hippocampal CA1, dorsal striatum, and orbitofrontal cortex during a temporal bisection task, the authors showed that elapsed time during the interval period is encoded similarly across these regions and that the neuronal activity of time cells across these regions tends to be synchronized within 100 ms. Using Bayesian decoding, they demonstrated that the interval time decoded from the firing activity of time cells in these regions correlated with the rats' decisions and that the times decoded from the neuronal activity of different brain regions were correlated. The sound experiments and analyses support most of the main conclusions of this paper.

      Strengths:

      They used a temporal bisection task in which the effects of time and distance can be dissociated. The test trials successfully revealed the relationship between the interval time estimated by Bayesian decoding and the animal's judgment of long versus short interval times. Simultaneous recording of neuronal activity from the hippocampal CA1, dorsal striatum, and orbitofrontal cortex, which is technically challenging, allowed comparison of interval time encoding across brain regions and the degree of synchrony between neurons from different brain regions.

      Weaknesses:

      Some analyses were not explained in detail, making it difficult to assess whether their results support the authors' conclusions.

    3. Reviewer #2 (Public review):

      Summary:

      In this work, the authors examined how neural activity related to temporal information is distributed and coordinated throughout the hippocampus, dorsal striatum, and orbitofrontal cortex. Rats were forced to run for fixed time intervals on a treadmill and make a decision based on whether the interval was long (10s) or short (5s). Under these conditions time cells were observed across all examined brain regions. The primary finding of the authors is that synchronized activity between time cells across brain regions is entrained into the theta cycle. This observation is used to support the central claim that the sharing of temporal information is mediated by the theta oscillation.

      Strengths:

      By simultaneously recording several brain regions in an interval discrimination task, the authors provide a valuable dataset for understanding how temporal information is processed and distributed throughout relevant networks.

      Weaknesses:

      Several methodological concerns should be addressed and a more focused analysis should be performed to strengthen the central claims of this work.

      Major Concerns

      (1) The restriction to only use time cells to understand temporal information processing. Other mechanisms of encoding time, like population clocks and ramping, have been characterized in the striatum and frontal cortex, and these dynamics might contain more temporal information than the subset of cells that meet the statistical criteria for being a time cell. Furthermore, time cells in the OFC, and DS in particular, appear to be heavily biased towards the beginning of treadmill running. This raises the question of whether temporal information can be encoded by neurons other than time cells in these two regions.

      (2) The results of the Bayesian decoding analysis should be expanded on. In particular, the performance of each decoder above the chance level is not quantified. Comparing the performance of decoders trained on all cells to the performance of decoders trained on time cells alone would partially address the question of whether or not time cells are the only cells that can encode temporal information in the DS and OFC.

      (3) The decoding results for the test trials appear different from the results in the authors' previous publication (Shimbo et. al., 2021). There, differences in decoded time between the selected-long and selected-short trials emerged after 5s, the duration of the short trials. This was to be expected given the following two reasons. First, from the task design, it is unclear that the animal can distinguish trial types (long, short, or test) until after the first 5 seconds of treadmill running, making it logical for differences in decoded time to emerge only after this point. Second, time cell activity was identical in the first 5s of the long and short trials as shown in Figure 2A. Here, however, the differences in decoded time during the selected-long and selected-short test trials emerge within the first 2s of treadmill running. Could the authors explain this discrepancy?

      Furthermore, in Figure 6B, at 3 seconds of running time, the decoded time for selected-long and selected-short trials shows a difference of nearly 2 seconds, with no further increase as running time progresses. In contrast, at 2 seconds of running time, there is no significant difference in decoded time for DS and OFC, while CA1 shows a slight increase in the decoded time for selected-long trials. This pattern suggests a sudden jump in the encoded time for selected-long trials between 2 and 3 seconds. However, without explicitly showing the raw data, it is difficult to interpret this result and other results from the decoding analysis.

      Minor Concerns

      (1) It is not clear how the Bayes decoder was trained. Does the training data come entirely from the long trials?

      (2) For Figure 5D, even if only one of two neurons in a pair has its spike rate modulated by theta, wouldn't the expectation be that synchronous spike events between these two neurons would be modulated by theta as well? This analysis might benefit from shuffling methods to determine if the mean resultant length of synchronous spike events is larger than the chance level.

      (3) In Figure 5A, the authors suggest that 'the synchronization of time cells was modulated by theta oscillation.' However, it is unclear whether the population exhibits a preferred theta phase or the phase preference only occurs at the individual cell level. If there is no preference on the population level, how would the authors interpret this result?

    4. Reviewer #3 (Public review):

      Summary:

      This study examines neural activity recorded simultaneously in the hippocampus, dorsal striatum, and orbitofrontal cortex as rats performed an interval timing task. The analyses primarily focus on the activity of "time cells" which are neurons that fire at specific moments during the intervals. In this experiment, the intervals consist of periods when animals are running on a treadmill before selecting the arm associated with the interval duration. The results show that the theta oscillations induced by this running behavior were observed across the three regions and that this strong oscillation modulated the activity of neurons across regions. While these findings are correlative in nature, they provide an important characterization of activity patterns across regions during complex behavior. However, more research is needed to determine whether these activity patterns specifically contribute to temporal coding.

      Strengths:

      (1) Overall, the paper is very well written. Although I have specific concerns about the review of the relevant literature and the interpretation of the results (see below), I do want to commend the authors for their efforts toward presenting this complex work in an accessible manner.

      (2) The study is well designed and the quality of the electrophysiological data collected from multiple brain regions in such a challenging behavioral experiment is impressive. This work is a technical tour de force.

      (3) The analyses are very thorough, statistically rigorous, and clearly explained and visualized. The authors provide a thoughtful mixture of example data (at the level of individual cells or animals) and aggregated data (at the group or session level) to properly explain and quantify the activity patterns of interest.

    1. eLife Assessment

      This is an important study providing convincing evidence that increased blood pressure variability impairs myogenic tone and diminishes baroreceptor reflex. The study also provides evidence that blood pressure variability blunts functional hyperemia and contributes to cognitive decline. The authors use appropriate and validated methodology in line with the current state-of-the-art.

    2. Reviewer #1 (Public review):

      This study examined the effect of blood pressure variability on brain microvascular function and cognitive performance. By implementing a model of blood pressure variability using an intermittent infusion of AngII for 25 days, the authors examined different cardiovascular variables, cerebral blood flow, and cognitive function during midlife (12-15-month-old mice). Key findings from this study demonstrate that blood pressure variability impairs baroreceptor reflex and impairs myogenic tone in brain arterioles, particularly at higher blood pressure. They also provide evidence that blood pressure variability blunts functional hyperemia and impairs cognitive function and activity. Simultaneous monitoring of cardiovascular parameters, in vivo imaging recordings, and the combination of physiological and behavioral studies reflect rigor in addressing the hypothesis. The experiments are well-designed, and the data generated are clear. I list below a number of suggestions to enhance this important work:

      (1) Figure 1B: It is surprising that the BP circadian rhythm is not distinguishable in either group. Figure 2, however, shows differences in circadian rhythm at different timepoints during infusion. Could the authors explain the lack of circadian effect in the 24-h traces?

      (2) While saline infusion does not result in elevation of BP when compared to Ang II, there is an evident "and huge" BP variability in the saline group, at least 40mmHg within 1 hour. This is a significant physiological effect to take into consideration, and therefore it warrants discussion.

      (3) The decrease in DBP in the BPV group is very interesting. It is known that chronic Ang II increases cardiac hypertrophy, are there any changes to heart morphology, mass, and/or function during BPV? Can the the decrease in DBP in BPV be attributed to preload dysfunction? This observation should be discussed.

      (4) Examining the baroreceptor reflex during the early and late phases of BPV is quite compelling. Figures 3D and 3E clearly delineate the differences between the two phases. For clarity, I would recommend plotting the data as is shown in panels D and E, rather than showing the mathematical ratio. Alternatively, plotting the correlation of ∆HR to ∆SBP and analyzing the slopes might be more digestible to the reader. The impairment in baroreceptor reflex in the BPV during high BP is clear, is there any indication whether this response might be due to loss of sympathetic or gain of parasympathetic response based on the model used?

      (5) Figure 3B shows a drop in HR when the pump is ON irrespective of treatment (i.e., independent of BP changes). What is the underlying mechanism?

      (6) The correlation of ∆diameter vs MAP during low and high BP is compelling, and the shift in the cerebral autoregulation curve is also a good observation. I would strongly recommend that the authors include a schematic showing the working hypothesis that depicts the shift of the curve during BPV.

      (7) Functional hyperemia impairment in the BPV group is clear and well-described. Pairing this response with the kinetics of the recovery phase is an interesting observation. I suggest elaborating on why BPV group exerts lower responses and how this links to the rapid decline during recovery.

      (8) The experimental design for the cognitive/behavioral assessment is clear and it is a reasonable experiment based on previous results. However, the discussion associated with these results falls short. I recommend that the authors describe the rationale to assess recognition memory, short-term spatial memory, and mice activity, and explain why these outcomes are relevant in the BPV context. Are there other studies that support these findings? The authors discussed that no changes in alternation might be due to the age of the mice, which could already exhibit cognitive deficits. In this line of thought, what is the primary contributor to behavioral impairment? I think that this sentence weakens the conclusion on BPV impairing cognitive function and might even imply that age per se might be the factor that modulates the various physiological outcomes observed here. I recommend clarifying this section in the discussion.

      (9) Why were only male mice used?

      (10) In the results for Figure 3: "Ang II evoked significant increases in SBP in both control and BPV groups;...". Also, in the figure legend: "B. Five-minute average HR when the pump is OFF or ON (infusing Ang II) for control and BPV groups...." The authors should clarify this as the methods do not state a control group that receives Ang II.

    3. Reviewer #2 (Public review):

      Summary:

      Blood pressure variability has been identified as an important risk factor for dementia. However, there are no established animal models to study the molecular mechanisms of increased blood pressure variability. In this manuscript, the authors present a novel mouse model of elevated BPV produced by pulsatile infusions of high-dose angiotensin II (3.1ug/hour) in middle-aged male mice. Using elegant methodology, including direct blood pressure measurement by telemetry, programmable infusion pumps, in vivo two-photon microscopy, and neurobehavioral tests, the authors show that this BPV model resulted in a blunted bradycardic response and cognitive deficits, enhanced myogenic response in parenchymal arterioles, and a loss of the pressure-evoked increase in functional hyperemia to whisker stimulation.

      Strengths:

      As the presentation of the first model of increased blood pressure variability, this manuscript establishes a method for assessing molecular mechanisms. The state-of-the-art methodology and robust data analysis provide convincing evidence that increased blood pressure variability impacts brain health.

      Weaknesses:

      One major drawback is that there is no comparison with another pressor agent (such as phenylephrine); therefore, it is not possible to conclude whether the observed effects are a result of increased blood pressure variability or caused by direct actions of Ang II. Ang II is known to have direct actions on cerebrovascular reactivity, neuronal function, and learning and memory. Given that Ang II is increased in only 15% of human hypertensive patients (and an even lower percentage of non-hypertensive), the clinical relevance is diminished. Nonetheless, this is an important study establishing the first mouse model of increased BPV.

    1. eLife Assessment

      The authors show that: 1) following brief peripheral optogenetic stimulation of forepaw proprioceptors in mice, sensory-evoked responses in primary motor cortex (M1) are delayed relative to primary somatosensory cortex (S1); 2) the responses in both cortical areas follow a triphasic pattern of activation-suppression-activation; 3) directly activating cortical parvalbumin-positive (PV) inhibitory interneurons mimicked both the suppression and rebound components of the sensory-evoked response; and 4) partially suppressing activity in S1 reduces the sensory-evoked response in M1. The conclusions are convincing and build on prior work on cortical circuits related to the mouse forelimb from this group (Yamawaki et al., 2021, eLife, doi:10.7554/eLife.66836). More rigorously determining whether the peripheral stimulation approach used evokes movements would strengthen the conclusions. It is also possible that these effects would differ for peripheral mechanoreceptor stimulation. Overall, this in vivo work assessing sensory responses in forepaw-related cortical circuits represents a valuable comparison to previously published work.

    2. Reviewer #1 (Public review):

      Summary:

      Building on previous in vitro synaptic circuit work (Yamawaki et al., eLife 10, 2021), Piña Novo et al. utilize an in vivo optogenetic-electrophysiological approach to characterize sensory-evoked spiking activity in the mouse's forelimb primary somatosensory (S1) and motor (M1) areas. Using a combination of a novel "phototactile" somatosensory stimuli to the mouse's hand and simultaneous high-density linear array recordings in both S1 and M1, the authors report in awake mice that evoked cortical responses follow a triphasic peak-suppression-rebound pattern response. They also find that M1 responses are delayed and attenuated relative to S1. Further analysis revealed a 20-fold difference in subcortical versus corticocortical propagation speeds. They also report that PV interneurons in S1 are strongly recruited by hand stimulation. Furthermore, they report that selective activation of PV cells can produce a suppression and rebound response similar to "phototactile" stimuli. Lastly, the authors demonstrate that silencing S1 through local PV cell activation reduces M1 response to hand stimulation, suggesting S1 may directly drive M1 responses.

      Strengths:

      The study was technically well done, with convincing results. The data presented are appropriately analyzed. The author's findings build on a growing body of both in vitro and in vivo work examining the synaptic circuits underlying the interactions between S1 and M1. The paper is well-written and illustrated. Overall, the study will be useful to those interested in forelimb S1-M1 interactions.

      Weaknesses:

      Although the results are clear and convincing, one weakness is that many results are consistent with previous studies in other sensorimotor systems, and thus not all that surprising. For example, the findings that sensory stimulation results in delayed and attenuated responses in M1 relative to S1 and that PV inhibitory cells in S1 are strongly recruited by sensory stimulation are not novel (e.g., Bruno et al., J Neurosci 22, 10966-10975, 2002; Swadlow, Philos Trans R Soc Lond B Biol Sci 357, 1717-1727, 2002; Gabernet et al., Neuron 48, 315-327, 2005; Cruikshank et al., Nat Neurosci 10, 462-468, 2007; Ferezou et al., Neuron 56, 907-923, 2007; Sreenivasan et al., Neuron 92, 1368-1382, 2016; Yu et al., Neuron 104, 412-427 e414, 2019). Furthermore, the observation that sensory processing in M1 depends upon activity in S1 is also not novel (e.g., Ferezou et al., Neuron 56, 907-923, 2007; Sreenivasan et al., Neuron 92, 1368-1382, 2016). The authors do a good job highlighting how their results are consistent with these previous studies.

      Perhaps a more significant weakness, in my opinion, was the missing analyses given the rich dataset collected. For example, why lump all responsive units and not break them down based on their depth? Given superficial and deep layers respond at different latencies and have different response magnitudes and durations to sensory stimuli (e.g., L2/3 is much more sparse) (e.g., Constantinople et al., Science 340, 1591-1594, 2013; Manita et al., Neuron 86, 1304-1316, 2015; Petersen, Nat Rev Neurosci 20, 533-546, 2019; Yu et al., Neuron 104, 412-427 e414, 2019), their conclusions could be biased toward more active layers (e.g., L4 and L5). These additional analyses could reveal interesting similarities or important differences, increasing the manuscript's impact. Given the authors use high-density linear arrays, they should have this data.

      Similarly, why not isolate and compare PV versus non-PV units in M1? They did the photostimulation experiments and presumably have the data. Recent in vitro work suggests PV neurons in the upper layers (L2/3) of M1 are strongly recruited by S1 (e.g., Okoro et al., J Neurosci 42, 8095-8112, 2022; Martinetti et al., Cerebral cortex 32, 1932-1949, 2022). Does the author's data support these in vitro observations?

      It would have also been interesting to suppress M1 while stimulating the hand to determine if any part of the S1 triphasic response depends on M1 feedback. I appreciate the control experiment showing that optical hand stimulation did not evoke forelimb movement. However, this appears to be an N=1. How consistent was this result across animals, and how was this monitored in those animals? Can the authors say anything about digit movement? A light intensity of 5 mW was used to stimulate the hand, but it is unclear how or why the authors chose this intensity. Did S1 and M1 responses (e.g., amplitude and latency) change with lower or higher intensities? Was the triphasic response dependent on the intensity of the "phototactile" stimuli?

    3. Reviewer #2 (Public review):

      Summary:

      Communication between sensory and motor cortices is likely to be important for many aspects of behavior, and in this study, the authors carefully analyse neuronal spiking activity in S1 and M1 evoked by peripheral paw stimulation finding clear evidence for sensory responses in both cortical regions

      Strengths:

      The experiments and data analyses appear to have been carefully carried out and clearly represented.

      Weaknesses:

      (1) Some studies have found evidence for excitatory projection neurons expressing PV and in particular some excitatory pyramidal cells can be labelled in PV-Cre mice. The authors might want to check if this is the case in their study, and if so, whether that might impact any conclusions.

      (2) I think the analysis shown in Figure S1 apparently reporting the absence of movements evoked by the forepaw stimulation could be strengthened. It is unclear what is shown in the various panels. I would imagine that an average of many stimulus repetitions would be needed to indicate whether there is an evoked movement or not. This could also be state-dependent and perhaps more likely to happen early in a recording session. Videography could also be helpful.

      (3) Some similar aspects of the evoked responses, including triphasic dynamics, have been reported in whisker S1 and M1, and the authors might want to cite Sreenivasan et al., 2016.

    4. Reviewer #3 (Public review):

      Summary:

      This is a solid study of stimulus-evoked neural activity dynamics in the feedforward pathway from mouse hand/forelimb mechanoreceptor afferents to S1 and M1 cortex. The conclusions are generally well supported, and match expectations from previous studies of hand/forelimb circuits by this same group (Yamawaki et al., 2021), from the well-studied whisker tactile pathway to whisker S1 and M1, and from the corresponding pathway in primates. The study uses the novel approach of optogenetic stimulation of PV afferents in the periphery, which provides an impulse-like volley of peripheral spikes, which is useful for studying feedforward circuit dynamics. These are primarily proprioceptors, so results could differ for specific mechanoreceptor populations, but this is a reasonable tool to probe basic circuit activation. Mice are awake but not engaged in a somatosensory task, which is sufficient for the study goals.

      The main results are:<br /> (1) brief peripheral activation drives brief sensory-evoked responses at ~ 15 ms latency in S1 and ~25 ms latency in M1, which is consistent with classical fast propagation on the subcortical pathway to S1, followed by slow propagation on the polysynaptic, non-myelinated pathway from S1 to M1;<br /> (2) each peripheral impulse evokes a triphasic activation-suppression-rebound response in both S1 and M1;<br /> (3) PV interneurons carry the major component of spike modulation for each of these phases;<br /> (4) activation of PV neurons in each area (M1 or S1) drives suppression and rebound both in the local area and in the other downstream area;<br /> (5) peripheral-evoked neural activity in M1 is at least partially dependent on transmission through S1.

      All conclusions are well-supported and reasonably interpreted. There are no major new findings that were not expected from standard models of somatosensory pathways or from prior work in the whisker system.

      Strengths:

      This is a well-conducted and analyzed study in which the findings are clearly presented. This will provide important baseline knowledge from which studies of more complex sensorimotor processing can build.

      Weaknesses:

      A few minor issues should be addressed to improve clarity of presentation and interpretation:

      (1) It is critical for interpretation that the stimulus does not evoke a motor response, which could induce reafference-based activity that could drive, or mask, some of the triphasic response. Figure S1 shows that no motor response is evoked for one example session, but this would be stronger if results were analyzed over several mice.

      (2) The recordings combine single and multi-units, which is fine for measures of response modulation, but not for absolute evoked firing rate, which is only interpretable for single units. For example, evoked firing rate in S1 could be higher than M1, if spike sorting were more difficult in S1, resulting in a higher fraction of multi-units relative to M1. Because of this, if reporting of absolute firing rates is an essential component of the paper, Figs 3D and 4E should be recalculated just for single units.

      (3) In Figure 5B, the average light-evoked firing rate of PV neurons seems to come up before time 0, unlike the single-trial rasters above it. Presumably, this reflects binning for firing rate calculation. This should be corrected to avoid confusion.

      (4) In Figure 6A bottom, please clarify what legends "W. suppression" and "W. rebound" mean.

    1. eLife Assessment

      This important study examines heterochromatin domain dynamics using a model system that allows reversible transition from an embryonic stem cell to a 2-cell-like state. The authors present a solid resource to the research community that will further the understanding of changes in the chromatin-bound proteome during the 2C-to-ESC transition. However, conclusions related to the functional roles of the interaction between the SWI/SNF complex component SMARCAD1 and the DNA Topoisomerase II Binding protein (TOPBP1) remain incomplete.

    2. Reviewer #1 (Public review):

      In this study, the authors investigate the molecular mechanisms driving the establishment of constitutive heterochromatin during embryonic development. The experiments have been meticulously conducted and effectively address the proposed hypotheses.

      The methodology stands out for its robustness, utilizing:<br /> i) an efficient system for converting ESCs to 2C-like cells via Dux overexpression;<br /> ii) a global approach through IPOTD, which unveils the chromatome at distinct developmental stages; and<br /> iii) STORM technology, enabling high-resolution visualization of DNA decompaction. These tools collectively provide clear and comprehensive insights that support the study's conclusions.

      The work makes a significant contribution to the field, offering valuable insights into chromatin-bound proteins at critical stages of embryonic development. These findings may also inform our understanding of processes beyond heterochromatin maintenance.

      The revised manuscript shows improvement, particularly through enhanced discussion and the addition of new references addressing the cooperation of SMARCAD1 and TOPBP1. All my previous concerns have been thoroughly addressed by the authors. However, I believe that, as this reviewer suggested, the inclusion of a model that summarizes the main findings of the study and discusses the potential mechanisms involved, would enhance the clarity and understanding of the message the manuscript aims to convey.

    3. Reviewer #2 (Public review):

      As noted in the original review, the study by Sebastian-Perez addresses an important research question using a tractable model system to examine the earliest drivers of heterochromatin formation during embryogenesis. Moreover, the proteomic analyses provide a valuable resource to the research community to understand changes in the chromatin-bound proteome during the 2C-to-ESC transition. From there, they carry out more detailed analyses of TOPBP1, which shows substantive changes in chromatin association in 2C-like cells, and a potential interacting protein SMARCAD1, which shows only modest changes in chromatin association. While I appreciate that the authors have revised the manuscript to some extent to address the minor points raised, the major over-arching issue of how TOPBP1 and SMARCAD1 function in the 2C-like state is still a concern.

    4. Reviewer #3 (Public review):

      The manuscript entitled "SMARCAD1 and TOPBP1 contribute to heterochromatin maintenance at the transition from the 2C-like to the pluripotent state" by Sebastian-Perez et al. adopted the iPOTD method to compare the chromatin-bound proteome in ESCs and 2CLCs induced by Dux overexpression. The authors identified 397 chromatin-bound proteins enriched specifically in non-2CLCs, among which they further investigated TOPBP1 due to its potential role in chromocenter reorganization. SMARCD1, a known interacting protein of TOPBP1, was also investigated in parallel. The authors report increased size and decreased number of H3K9me3-heterochromatin foci in Dux-induced 2CLCs. Remarkably, depletion of either TOPBP1 or SMARCD1 resulted in similar phenotypes. However, the absence of these proteins did not affect the entry into or exit from the 2C-like state. The authors further showed that both TOPBP1 and SMARCD1 are essential for early embryonic development.

      This manuscript provides valuable insights into the features of 2CLCs regarding H3K9me3-heterochromatin reorganization. However, the findings are largely descriptive. Mechanistic studies are required in future studies, such as: 1) how SMARCD1 associates with H3K9me3 and contributes to heterochromatin maintenance, 2) how TOPBP1 regulates the expression of SMARCD1 and facilitates its localization in heterochromatin foci, 3) whether the remodelling of chromocenter directly influence the transitions between ESCs and 2CLCs.

    5. Author response:

      The following is the authors’ response to the original reviews.

      Reviewer #1 (Public Review):

      In the present work the authors explore the molecular driving events involved in the establishment of constitutive heterochromatin during embryo development. The experiments have been carried out in a very accurate manner and clearly fulfill the proposed hypotheses.

      Regarding the methodology, the use of: i) an efficient system for conversion of ESCs to 2C-like cells by Dux overexpression; ii) a global approach through IPOTD that reveals the chromatome at each stage of development and iii) the STORM technology that allows visualization of DNA decompaction at high resolution, helps to provide clear and comprehensive answers to the conclusion raised.

      The contribution of the present work to the field is very important as it provides valuable information on chromatin-bound proteins at key stages of embryonic development that may help to understand other relevant processes beyond heterochromatin maintenance.

      The study could be improved through a more mechanistic approach that focuses on how SMARCAD1 and TOPBP1 cooperate and how they functionally connect with H3K9me3, HP1b and heterochromatin regulation during embryonic development. For example, addressing why topoisomerase activity is required or whether it connects (or not) to SWI/SNF function and the latter to heterochromatin establishment, are questions that would help to understand more deeply how SMARCAD1 and TOPBP1 operate in embryonic development.

      We would like to thank the reviewer for the positive evaluation of our work and the methodology we employed. We greatly appreciated the reviewer’s recognition of our study to “provide valuable information on chromatin-bound proteins at key stages of embryonic development that may help to understand other relevant processes beyond heterochromatin maintenance”. While we acknowledge the value of including mechanistic studies, such an addition would require a substantial amount of experimental work that exceeds our current resources.

      Reviewer #1 (Recommendations For The Authors):

      In my opinion, the authors could improve the study by deciphering -to a certain extent- the possible mechanism by which SMARCAD1 and TOPBP1 are cooperating in their system to establish H3K9me3 and consequently heterochromatin; and whether it is different (or not) from that already reported in yeast (ref 27). In fact, is it only SMARCAD1 that participates in this process or the whole SWI/SNF complex? Could the lack of SMARCAD1 compromise the proper assembly of the SWI/SNF complex? In this regard, a model describing the main findings of the study and the discussion of the possible mechanisms involved -based on the current bibliography- would be appreciated. This, although speculative, would illustrate the range of possibilities that could be operating in the maintenance of heterochromatin during embryonic development. In conclusion, it would be great if the authors could link -mechanistically- the dots connecting SMARCD1, TOPBP1, H3K9me3/HP1/heterochromatin.

      As suggested by the reviewer and to enrich the discussion, we have included some additional sentences and references in the revised discussion section.

      As a minor point, In Figure 3A, left panel it appears that the protein precipitating with H3K9me3 reacts with TOPBP1 but its molecular weight does not exactly match to the TOPBP1 band found in the input. The authors should clarify this point and it is also recommended that IPs and inputs are run in the same gel. Please replace Figure 3A right panel.

      Following the reviewer’s suggestion and to improve the reading flow, we have restructured the order of the figures and removed the original Figure 3A. The revised Figure 3A-C panel illustrates the SMARCAD1 association with H3K9me3 in ESCs and 2C- cells, while capturing the reduced SMARCAD1-H3K9me3 association in 2C<sup>+</sup> cells.

      Reviewer #2 (Public Review):

      The manuscript by Sebastian-Perez describes determinants of heterochromatin domain formation (chromocenters) at the 2-cell stage of mouse embryonic development. They implement an inducible system for transition from ESC to 2C-like cells (referred to as 2C<sup>+</sup>) together with proteomic approaches to identify temporal changes in associated proteins. The conversion of ESCs to 2C<sup>+</sup> is accompanied by dissolution of chromocenter domains marked by HP1b and H3K9me3, which reform upon transition back to the 2C-like state. The innovation in this study is the incorporation of proteomic analysis to identify chromatin-associated proteins, which revealed SMARCAD1 and TOPBP1 as key regulators of chromocenter formation.

      In the model system used, doxycycline induction of DUX leads to activation of EGFP reporter regulated by the MERVL-LTR in 2C<sup>+</sup> cells that can be sorted for further analysis. A doxycycline-inducible luciferase cell line is used as a control and does not activate the MERVL-LTR GFP reporter. The authors do see groups of proteins anticipated for each developmental stage that suggest the overall strategy is effective.

      The major strengths of the paper involve the proteomic screen and initial validation. From there, however, the focus on TOPBP1 and SMARCAD1 is not well justified. In addition, how data is presented in the results section does not follow a logical flow. Overall, my suggestion is that these structural issues need to be resolved before engaging in comprehensive review of the submission. This may be best achieved by separating the proteomic/morphological analyses from the characterization of TOPBP1 and SMARCAD1.

      We appreciate the reviewer’s positive evaluation of our inducible system to trigger the transition from ESCs to 2C-like cells, and the strength of the chromatin proteomics we conducted. In response to the reviewer’s suggestion, we have reorganized the order of the figures, particularly Figure 1 and Figure 2, and revised the text to improve readability and flow.

      Reviewer #2 (Recommendations For The Authors):

      There are some very interesting components to the study but, as noted, the narrative requires changes and the rationale for focusing on TOPBP1 and SMARCAD1 is not strong at present. Specific comments are noted below

      (1) Inclusion of authentic 2C cells for comparative chromocenter analysis (or at least a more fulsome discussion of how the system has been benchmarked in previous studies).

      We have included more detail in the revised methods section, in the “Cell lines and culture conditions” paragraph. We have added: “The Dux overexpression system was benchmarked according to previously reported features. Dux overexpression resulted in the loss of DAPI-dense chromocenters and the loss of the pluripotency transcription factor OCT4 (fig. S1E) (6, 7), upregulation of specific genes of the 2-cell transcriptional program such as endogenous Dux, MERVL, and major satellites (MajSat) (fig. S1F) (6, 7, 11, 26, 58), and accumulation in the G2/M cell cycle phase (fig. S1G), with a reduced S phase consistent in several clonal lines (fig. S1H) (15).”

      (2) In Figure 1A, the text indicates a loss of chromocenters, but it may be better described as decompaction because the DAPI/H3K9me3 staining shows diffuse/expanded structures (this is in fact how it is described in relation to Figure 2).

      We have changed the text accordingly, now describing it as “decompaction”.

      (3) Table S1 has 6 separate tabs but these are not specified in the text. It would be useful to separate the 397 proteins unique to Luc and 2C- cells since they form much of the basis for the remaining analysis. This approach also assumes it is the absence of a protein in the 2C<sup>+</sup> that accounts for the lack of chromocenters (noting there are 510 proteins unique to the 2C<sup>+</sup> state that are not discussed).

      We have referenced the supplementary table as Table S1 in the text for simplicity. It includes: Table S1A - List of Protein Groups identified by mass spectrometry in -EdU, Luc, 2C- and 2C<sup>+</sup> cells; Table S1B - Input data for SAINT analysis; Table S1C - SAINT results of the comparison 2C- vs Luc and 2C<sup>+</sup> vs Luc; Table S1D - SAINT results of the comparison Luc vs 2C- and 2C<sup>+</sup> vs 2C-; Table S1E - SAINT results of the comparison Luc vs 2C<sup>+</sup> and 2C- vs 2C<sup>+</sup>; and Table S1F - Total number of PSM per protein in the different cells and conditions tested.

      (4) Since there is no change in H3K9me3 levels, loss of SUV420H2 from 2C<sup>+</sup> chromatin (figure 1G) coupled with potential changes in H4K20me3 could contribute the morphological differences. SUV420H2 is known to regulate chromocenter clustering in a way the requires H4K20me3 but this is not addressed or cited (PUBMED: 23599346).

      As suggested by the reviewer, we have added additional sentences and references in the revised manuscript.

      (5) In Figure 1C, there does appear to be overlap between the 2C<sup>+</sup> and 2C- populations (while the Luc population is distinct) even though they are morphologically distinct when imaged in Figure 2A. The 2C- cells are thought to be an intermediate, low Dux expressing population.

      Chromatome profiling through genome capture provides a snapshot of the chromatin-bound proteome in the analyzed samples (shown in revised Fig. 2B). As indicated by the reviewer and previously reported in the literature, 2C- cells are an intermediate population before reaching 2C<sup>+</sup> cells. For this study, we have focused on H3K9me3 morphological changes. Even though 2C- and 2C<sup>+</sup> cells are distinct with respect to H3K9me3 morphology (shown in revised Fig. 1B), analysis of the chromatome data from hundreds of chromatin-bound proteins revealed some overlap between these two populations. However, replicates from the same population tend to cluster together, for example, 2C<sup>+</sup> rep1 and 2C<sup>+</sup> rep3, and 2C- rep1 and 2C- rep2. Collectively, these data suggest that a defined subset of coordinated changes in the chromatome likely triggers the transition from 2C- to 2C<sup>+</sup> cells. Further experimental investigation of the chromatome dataset during the 2C-like transition would be interesting, however, we believe it is beyond the scope of this study.

      (6) Data with SUV39H1 and 2 is difficult to accommodate; what about other H3K9 methyltransferases or proteins such as TRIM28 (KAP1) and SETDB1 (this comes up in the discussion but is not assessed in the results section).

      We agree that investigating the role of TRIM28 (KAP1) and SETDB1 in this experimental setting could be of interest, however, we believe that these experiments go beyond the scope of the presented study.

      (7) Rationale for choosing TOPBP1 needs to be improved. How do TOPBP1 levels relate to TOPI/TOP2A/TOP2B levels across the 3 cell populations? By what criteria does topoisomerase inhibitor treatment increase 2C<sup>+</sup> like cells? Moreover, to what extent will inhibiting topoisomerases lead to global heterochromatin and cell cycle changes regardless of cell type.

      Following the reviewer’s suggestion, we have included some additional references throughout the text to strengthen our rationale for selecting TOPBP1, given its well-established critical role in DNA replication and repair. Additionally, we have revised the results and discussion sections to include new sentences that propose a potential mechanism by which topoisomerase inhibitors may indirectly recruit TOPBP1 to facilitate DNA repair, ultimately leading to an increase in 2C<sup>+</sup> cells.

      (8) Likewise, the decision to look at SMARCAD1 based solely on its interaction with TOPBP1 seems somewhat arbitrary and it did not seem to come up as of interest in the iPOTD analysis. Moreover, they were not able to validate the interaction with their own analyses.

      We have revised the text to clarify the connection further.

      (9) The flow of results is confusing. The first section concludes with a focus on TOPBP1 and SMARCAD1, then progresses to morphological characterization of heterochromatin regions in the next two sections before returning to TOPBP1 and SMARCAD1. It seems like it would make more sense to describe the model system and morphological characterization at the beginning of the results section and then transition to the proteomic analysis and characterization of TOPBP1 and SMARCAD1 (with the expectation that the rationale be improved).

      As suggested by the reviewer, we have reordered the figures, particularly Figure 1 and Figure 2, and rephased the text to improve the overall reading flow.

      (10) There has been considerable work done on characterizing chromatin structure, epigenetic changes, and morphology during early embryonic development. It is therefore difficult to see what validating some of these changes in the inducible model is adding much in the way of new knowledge. It may, but this is not articulated in the current text.

      As detailed before, we have rephrased the text to improve the overall reading flow, which we hope has improved the understanding of the impact of our results.

      (11) It is difficult to disentangle broader effects of both TOPBP1 and SMARCAD1 from those described here; they may induce phenotypes, but these may not be unique to this model system.

      We agree with the reviewer, but to address this point would require additional experiments which would go beyond the scope of the presented study.

      (12) One of the issues with this assay is global chromatin recovery; it is not focused on heterochromatin compartments. The statement "We identified a total of 2396 proteins, suggesting an efficient pull-down of chromatin-associated factors (fig. S2D and Table S1)" does not demonstrate efficiency. Additional functional annotation would be required to establish this claim, including what fraction are known chromatin-associated proteins (with a focus on the heterochromatin compartment).

      We have changed the text accordingly. The resulting statement reads as: “We identified a total of 2396 proteins, suggesting an effective pull-down of putative chromatin-associated factors (fig. S2D and Table S1)”.

      Reviewer #3 (Public Review):

      The manuscript entitled "SMARCAD1 and TOPBP1 contribute to heterochromatin maintenance at the transition from the 2C-like to the pluripotent state" by Sebastian-Perez et al. adopted the iPOTD method to compare the chromatin-bound proteome in ESCs and 2C-like cells generated by Dux overexpression. The authors identified 397 chromatin-bound proteins enriched only in ESC and 2C- cells, among which they further investigated TOPBP1 due to its potential role in controlling chromocenter reorganization. SMARCD1, a known interacting protein of TOPBP1, was also investigated in parallel. The authors observed increased size and decreased number of H3K9me3-heterochromatin foci in Dux-induced 2C<sup>+</sup> cells. Interestingly, depletion of TOPBP1 or SMARCD1 also led to increased size and decreased number of H3K9me3 foci. However, depletion of these proteins did not affect entry into or exit from the 2C-like state. Nevertheless, the authors showed that both TOPBP1 and SMARCD1 are required for early embryonic development.

      Although this manuscript provides new insights into the features of 2C-like cells regarding H3K9me3-heterochromatin reorganization, it remains largely descriptive at this stage. It does not provide new insights into the following important aspects: 1) how SMARCD1 associates with H3K9me3 and contributes to heterochromatin maintenance, 2) how TOPBP1 regulates the expression of SMARCD1 and facilitates its localization in heterochromatin foci, 3) whether the remodelling of chromocenter is causally related to the mutual transitions between ESCs and 2C-like cells. Furthermore, some results are over-interpreted. Additional experiments and analyses are needed to increase the strength of mechanistic insights and to support all claims in the manuscript.

      We would like to thank the reviewer for their positive and thorough evaluation of our manuscript. We have revised the text and hope that the overall flow is now clearer. Moreover, while we acknowledge the value of including mechanistic studies, such an addition would require a substantial amount of experimental work that exceeds our current resources. 

      Reviewer #3 (Recommendations For The Authors):

      Major points:

      (1) Fig.2: the DNA decompaction of the chromatin fibers shown in 2C<sup>+</sup> cells may be more related to a relaxed 3D chromatin conformation (Zhu, NAR 2021; Olbrich, Nat Commun 2021) than chromatin accessibility. The authors should discuss this point.

      As suggested by the reviewer, we have included some additional sentences and references in the revised manuscript to address this concern.

      (2) Chemical inhibition of topoisomerases resulted in an increase in the percentage of 2C<sup>+</sup> cells. Does depletion of TOPBP1 also resulted in increased percentage of 2C<sup>+</sup> cells? Please include this result in Fig. 3E. Additionally, it should be noted that DDR and p53 have been reported to activate Dux (Stashpaz, eLife 2020; Grow, Nat Genet 2021), and thus, may contribute to the increased percentage of 2C<sup>+</sup> cells observed upon topoisomerase inhibition. This point should be discussed in the manuscript.

      To address this concern, we have included some additional sentences and references in the revised manuscript.

      (3) Fig 3A: the TOPBP1 band in the IP sample is questionable, and therefore the conclusion that TOPBP1 is associated with H3K9me3 is difficult to draw from Fig 3A. Additionally, the authors mentioned that association of TOPBP1 and SMARCAD1 is undetected in ESCs, likely due to the suboptimal efficiency of available antibodies. As these are key conclusions in this study, the authors are suggested to try other commercially available TOPBP1 antibodies (e.g., Abcam #ab-105109, used by ElInati, PNAS 2017) or knock-in tags to perform the co-IP experiment.

      Following the reviewer’s suggestion and to improve the reading flow, we have restructured the order of figures and removed the original Figure 3A. The revised Figure 3A-C panel illustrates the SMARCAD1 association with H3K9me3 in ESCs and 2C- cells, while capturing the reduced SMARCAD1-H3K9me3 association in 2C<sup>+</sup> cells.

      (4) Fig. 3C-D, Fig. S3D: the authors claimed reduction of both SMARCAD1 expression and its co-localization with H3K9me3 foci in 2C<sup>+</sup> cells, but did not perform mechanistic studies. It is important to know if TOPBP1 expression also decreases in 2C<sup>+</sup> cells. Additionally, it is unclear if the reduced co-localization of SMARCAD1 with H3K9me3 foci results from its altered nuclear localization or simply from reduced expression level? In either case, please provide some mechanistic insights.

      While we acknowledge the value of including mechanistic studies, such an addition would require a substantial amount of experimental work that exceeds our current resources. 

      (5) Fig. 3K, Fig. S4D-E: does SMARCAD1 expression decrease upon TOPBP1 depletion? Statistical analysis of SMARCAD1 intensity in Fig. S4E is needed, and a Western blot analysis is strongly suggested. Additionally, it is unclear if the reduced co-localization of SMARCAD1 with H3K9me3 foci results from its altered nuclear localization or simply from reduced expression level? In Fig. 3K, TOPBP1-depleted cells appear to show decreased size and increased number of H3K9me3 foci, which is inconsistent with Fig. S4B-C. The authors should clarify this discrepancy. Furthermore, statistics should be performed to determine whether Smarcad1/Topbp1 knockdown could further increase the size and decrease the number of H3K9me3 foci in 2C<sup>+</sup> cells. This would provide additional evidence for the involvement of these proteins in heterochromatin maintenance.

      We did not observe Smarcad1 downregulation after Topbp1 knockdown (shown in fig. S4A). In Figs. S4B and S4C, we observed that the number of H3K9me3 foci decreased, and their area became larger after knocking down either Smarcad1 or Topbp1, compared to scramble controls. These results align with the reviewer’s comment. Additionally, it should be noted that these findings were derived from the quantification of tens of cells and hundreds of foci, as indicated in the figure legend. This resulted in statistical significance after applying the test indicated in the figure legend.

      (6) Fig. 3J is suggested to be moved to Fig. 4. Additionally, performing immunostaining of SMARCAD1, TOPBP1, and H3K9me3 during pre-implantation development would provide valuable information on their protein-level dynamics, interactions, and functions in early embryos. This would further strengthen the conclusions drawn in the manuscript.

      We agree that performing these additional experiments would provide additional valuable information, however this would require a substantial amount of experimental work that exceeds our current resources.

      (7) Fig. 4 and Fig. S5: the authors observed reduced H3K9me3 signal in the Smarcad1 MO embryos at the 8-cell stage, but claim that they failed to examine Topbp1 MO embryos at the 8-cell stage due to their developmental arrest at the 4-cell stage. However, based on Fig. 4A, not all Topbp1 MO embryos were arrested at the 4-cell stage, and it is still possible to examine the H3K9me3 signal in 8-cell Topbp1 MO embryos, which is critical for demonstrating its function in early embryos. Also, how to interpret the increased HP1b signal in Topbp1 MO embryos?

      For Topbp1 silencing, we observed an even more severe phenotype compared to Smarcad1 MO. All the Topbp1 MO-injected embryos (100 %) arrested at the 4-cell stage and did not develop further (shown in Fig. 4A and 4B). Therefore, the severity of the Topbp1 morpholino phenotype posed a technical challenge in evaluating the H3K9me3 signal in 8-cell Topbp1 MO embryos, as none of the injected embryos developed beyond the 4-cell stage.

      We believe the increased HP1b signal in Topbp1 MO embryos could indicate potential alterations in chromatin organization and heterochromatin stability. Specifically, we observed remodeling of heterochromatin in both 2-cell and 4-cell Topbp1 MO arrested embryos compared to controls, as evidenced by the spreading and increased HP1b signal (shown in fig. S5F-S5I). Further investigations could enhance our understanding of the underlying defects in Topbp1 knockdown embryos, extending beyond heterochromatin-related errors.

      Minor points:

      (1) Page 4, the third row from the bottom: please revise the sentence.

      We have reviewed the text and it now reads correctly in the revised manuscript.

      (2) Fig. 1C: The authors claimed "Luc replicates clustered separately from 2C<sup>+</sup> and 2C- conditions", however, Luc rep3 is apparently clustered with 2C conditions.

      (3) The GFP signal in Fig. S1E is confusing.

      (4) Please include ESC in Fig. 2D-E. Also label the colors in Fig. 2E.

      As indicated in the figure legend of the revised Fig. 1F: “Cells with a GFP intensity score > 0.2 are colored in green. Black dots indicate 2C- cells and green dots indicate 2C<sup>+</sup> cells.”

      (5) Fig. 2G: Transposition of the heatmap (show genes in rows) is suggested to improve readability.

      (6) Page 7, the third row from the bottom: incorrect citation of Fig. 1K.

      Thank you for spotting this incorrect citation. We have corrected it in the revised manuscript.

      (7) Page 8, row 15, Fig. S3D should be cited to support the decreased expression of SMARCAD1 in 2C<sup>+</sup> cells.

      We have cited the corresponding supplementary figure S3D in the mentioned sentence.

      (8) Fig. 2H: what is the difference between "2C-" and "ESC-like"?

      We named 2C- to those cells not expressing the GFP reporter in the transition from ESCs to 2C<sup>+</sup> cells. We named ESC-like cells to those cells that do not express the GFP reporter during exit, meaning from sorted and purified 2C<sup>+</sup> to a GFP negative state.

      (9) Fig. S4A-C: compared with shTopbp1#2, shTopbp1#1 appears to be slightly more effective in knockdown, but less dramatic changes in the size/number of H3K9me3 foci.

      (10) Fig. 4: please show the effectiveness of Topbp1 MO by Immunostaining of TOPBP1.

      (11) Fig. 4C: please label the developmental stage as in Fig. 4E and 4G.

      We have added a “8-cell” label in the Figure 4C, as suggested by the reviewer.

    1. eLife Assessment

      This important study shows that Type 3 secretion translocons in Edwardsiella tarda and other bacteria activate the NAIP-NLRC4 inflammasome. The data from cellular and biochemical experiments showing that EseB is required for activation of the NLRC4 inflammasome are convincing. This paper is broadly relevant to those investigating host-pathogen interactions in diverse organisms.

    2. Reviewer #1 (Public review):

      Summary:

      In this study, Zaho and colleagues investigate inflammasome activation by E. tarda infections. They show that E. tarda induces the activation of the NLRC4 inflammasome as well as the non-canonical pathway in human THP1 macrophages. Further dissecting NLRC4 activation, the find that T3SS translocon components eseB, eseC and eseD are necessary for NLRC4 activation, and that delivery of purified eseB is sufficient to trigger NAIP-dependnet NLRC4 activation. Sequence analysis reveals that eseB shares homology within the C-terminus with T3SS needle and rod proteins, leading the authors to test if this region is necessary for inflammasome activation. They show that the eseB CT is required and that it mediates interaction with NAIP. Finally, they that homologs of eseB in other bacteria also share the same sequence and that they can activate NLRC4 in a HEK293T cell overexpression system.

      Strengths:

      This is a very nice study that convincingly shows that eseB and its homologs can be recognized by the human NAIP/NLRC4 inflammasome. The experiments are well-designed, controlled and described, and the papers is convincing as a whole.

      Weaknesses:

      The authors need to discuss their study in the context of previous papers that have shown an important role for E. tarda flagellin in inflammasome activation and test whether flagellin and/or E. tarda T3SSs needle or rod can activate NLRC4.

      The authors show that eseB and its homologs can activate NLRC4, but there are also other translocon proteins that are very different such as YopB or PopB. and share little homology with eseB. It would be nice to include a section comparing the different type 3 secretion systems. are there 2 different families of T3SSs, those that feature translocon components that are recognized by NAIP-NLRC4 and those that cannot be recognized?

      Comments on revisions:

      The authors have addressed my concern with additional experiments, which strengthen the authors' conclusions.

    3. Reviewer #2 (Public review):

      Summary:

      This work by Zhao et al. demonstrates the role of the Edwardsiella tarda type 3 secretion system translocon in activating human macrophage inflammation and pyroptosis. The authors show the requirement of both the bacterial translocon proteins and particular host inflammasome components for E. tarda-induced pyroptosis. In addition, the authors show that the C-terminal region of the translocon protein, EseB, is both necessary and sufficient to induce pyroptosis when present in the cytoplasm. The most terminal region of EseB was determined to be highly conserved among other T3SS-encoding pathogenic bacteria and a subset of these exhibited functionally similar effects on inflammasome activation. Overall, the data support the conclusions and interpretations and provide valuable insights into interactions between bacterial T3SS components and the host immune system., thereby expanding our understanding of E. tarda pathogenesis.

      Strengths:

      The authors use established and reliable molecular biology and bacterial genetics strategies to characterize the roles of the bacterial T3SS translocon and host inflammasome pathways to E. tarda-induced pyroptosis in human macrophages. These observations are naturally expanded upon by demonstrating the specific regions of EseB that are required for inflammasome activation and the conservation of this sequence and function among other pathogenic bacteria.

    4. Author response:

      The following is the authors’ response to the original reviews.

      Reviewer #1 (Public Review):

      Summary:

      In this study, Zhao and colleagues investigate inflammasome activation by E. tarda infections. They show that E. tarda induces the activation of the NLRC4 inflammasome as well as the non-canonical pathway in human THP1 macrophages. Further dissecting NLRC4 activation, they find that T3SS translocon components eseB, eseC and eseD are necessary for NLRC4 activation and that delivery of purified eseB is sufficient to trigger NAIP-dependent NLRC4 activation. Sequence analysis reveals that eseB shares homology within the C-terminus with T3SS needle and rod proteins, leading the authors to test if this region is necessary for inflammasome activation. They show that the eseB CT is required and that it mediates interaction with NAIP. Finally, they that homologs of eseB in other bacteria also share the same sequence and that they can activate NLRC4 in a HEK293T cell overexpression system.

      Strengths:

      This is a very nice study that convincingly shows that eseB and its homologs can be recognized by the human NAIP/NLRC4 inflammasome. The experiments are well designed, controlled and described, and the papers is convincing as a whole.

      Weaknesses:

      The authors need to discuss their study in the context of previous papers that have shown an important role for E. tarda flagellin in inflammasome activation and test whether flagellin and/or E. tarda T3SSs needle or rod can activate NLRC4.

      The authors show that eseB and its homologs can activate NLRC4, but there are also other translocon proteins that are very different such as YopB or PopB. and share little homology with eseB. It would be nice to include a section comparing the different type 3 secretion systems. are there 2 different families of T3SSs, those that feature translocon components that are recognized by NAIP-NLRC4 and those that cannot be recognized?

      (1) The authors need to discuss their study in the context of previous papers that have shown an important role for E. tarda flagellin in inflammasome activation and test whether flagellin and/or E. tarda T3SSs needle or rod can activate NLRC4.

      According to the reviewer’s suggestion, we added the relevant discussion (lines 326-334) and carried out additional experiments to examine whether E. tarda flagellin, needle, and rod could activate NLRC4. The relevant results are shown in Figure S3, Figure S5, and lines 226-230 and 269-274.

      (2) The authors show that eseB and its homologs can activate NLRC4, but there are also other translocon proteins that are very different such as YopB or PopB. and share little homology with eseB. It would be nice to include a section comparing the different type 3 secretion systems. are there 2 different families of T3SSs, those that feature translocon components that are recognized by NAIP-NLRC4 and those that cannot be recognized?

      According to the reviewer’s suggestion, additional experiments were performed to examine the NLRC4-activating potentials of 14 translocator proteins that share low sequence identities with EseB. The relevant results and discussion are shown in Figure S8 and lines 289-301; 364-372, and 377-379.

      Reviewer #2 (Public Review):

      Summary:

      This work by Zhao et al. demonstrates the role of the Edwardsiella tarda type 3 secretion system translocon in activating human macrophage inflammation and pyroptosis. The authors show the requirement of both the bacterial translocon proteins and particular host inflammasome components for E. tarda-induced pyroptosis. In addition, the authors show that the C-terminal region of the translocon protein, EseB, is both necessary and sufficient to induce pyroptosis when present in the cytoplasm. The most terminal region of EseB was determined to be highly conserved among other T3SS-encoding pathogenic bacteria and a subset of these exhibited functionally similar effects on inflammasome activation. Overall, the data support the conclusions and interpretations and provide interesting insights into interactions between bacterial T3SS components and the host immune system.

      Strengths:

      The authors use established and reliable molecular biology and bacterial genetics strategies to characterize the roles of the bacterial T3SS translocon and host inflammasome pathways to E. tarda-induced pyroptosis in human macrophages. These observations are naturally expanded upon by demonstrating the specific regions of EseB that are required for inflammasome activation and the conservation of this sequence among other pathogenic bacteria.

      Weaknesses:

      The functional assessment of EseB homologues is limited to inflammasome activation at the protein level but does not include the effects on cell viability as shown for E. tarda EseB. Confirmation that EseB homologues have similar effects on cell death would strengthen this portion of the manuscript.

      According to the reviewer’s suggestion, the effects of representative EseB homologs on cell death were examined in the revised manuscripts (Figure 5D, Figure S7 and line 289).

      Recommendations for the authors:

      Reviewer #1 (Recommendations For The Authors):

      I only have a few suggestions on how to improve the study:

      Activation of caspase-4 requires entry into the host cytosol. Can this be observed with E. tarda and is it T3SS dependent? The fact that deleting the translocon components abrogates all GSDMD activation (see Fig. 2D) suggests that also Casp4 activation requires an active T3SS. It would be useful for the reader to include some more information on the cellular biology of E. tarda.

      In our study, we found that E. tarda could enter THP-1 cells (Figure S1), and host cell entry was not affected by deletion of eseB-D (Δ_eseB-D_) in the T3SS system (Figure 2B, C). Additional experiments showed that Δ_eseB-D_ abolished the ability of E. tarda to activate Casp4 (Figure S2), implying that Casp4 activation required an active T3SS. Relevant changes in the revised manuscript: lines 223 and 224, 341-342.

      The data presented by the authors suggest that escB is sensed by NLRC4 when overexpressed, they do however not prove that during an infection escB is the main factor that drives NLRC4 activation, since deficiency in escB also abrogated translocation of other potential activators of NLRC4, e.g. flagellin and T3SS needle and rod subunits. I would thus find it essential to properly test if E. tarda flagellin can activate NLRC4 by comparing a WT and flagellin deficient strain, and/or by transfecting or expressing E.t. flagellin in these cells, as well as testing whether E.t. rod and needle subunits act as NLRC4 activators. This is important as previous studies suggested that flagellin is the main activator of cytotoxicity during E. tarda infection.

      Previous studies have shown that flagellin is required for E. tarda-induced macrophage death in fish [1] but not in mice [2]. In the revised manuscript, we performed additional experiments to examine whether E. tarda flagellin, needle, and rod could activate NLRC4. The relevant results are shown in Figure S3, Figure S5, and lines 226-230 and 269-274, and 326-334.

      References

      (1) Xie HX, Lu JF, Rolhion N, Holden DW, Nie P, Zhou Y, et al. Edwardsiella tarda-induced cytotoxicity depends on its type III secretion system and flagellin. Infect Immun. 2014;82(8):3436-45. doi: 10.1128/IAI.01065-13.

      (2) Chen H, Yang D, Han F, Tan J, Zhang L, Xiao J, et al. The bacterial T6SS effector EvpP prevents NLRP3 inflammasome activation by inhibiting the Ca<sup>2+</sup>-dependent MAPK-JNK pathway. Cell Host Microbe. 2017;21(1):47-58. doi: 10.1016/j.chom.2016.12.004.

      Figure 5/S4, please list the names of the eseB homologs. It is cumbersome to have to access GenBank with the accession number to be able to understand what proteins the authors define as homologs of eseB.

      The names were added to the revised Table S2, Figure 5 and Figure S6 (the original Figure S4).

      The authors mention that other translocon proteins, such as YopB/D and PopB/D, were suggested to cause inflammasome activation. How do these compare to eseB and its homologs? Do they share the CT motif?

      Additional experiments were performed to compare the inflammasome activation abilities of EseB and other translocator proteins including YopD and PopD. The relevant results and discussion are shown in Figure S8 and lines 289-301, 364-372, and 377-379.

      It would be nice to show that there are potentially two groups of translocon proteins, one group sharing homology to needle subunits within the CT region and another that is different. A quick look at the sequence of these proteins suggests that they are quite different and much larger than eseB.

      In our study, additional experiments with more translocator proteins indicated that the possession of EseB T6R-like terminal residues does not necessarily guarantee the protein to activate the NLRC4 inflammasome. Relevant results and discussion are shown in lines 289-301, 364-372, and 377-379.

    1. eLife Assessment

      This paper reports important findings on giant organelle complexes containing endosomes and lysosomes (termed endosomal-lysosomal organelles form assembly structures [ELYSAs]) present in mouse oocytes and 1- to 2-cell embryos. The data showing the localization and dynamics of ELYSAs during oocyte/embryo maturation are convincing. This work will be of interest to general cell biologists and developmental biologists.

    2. Reviewer #1 (Public review):

      Satouh et al. report giant organelle complexes in oocytes and early embryos. Although these structures have often been observed in oocytes and early embryos, their exact nature has not been characterized. The authors named these structures "endosomal-lysosomal organelles form assembly structures (ELYSAs)". ELYSAs contain organelles such as endosomes, lysosomes, and probably autophagic structures. ELYSAs are initially formed in the perinuclear region and then seem to migrate to the periphery in an actin-dependent manner. When ELYSAs are disassembled after the 2-cell stage, the V-ATPase V1 subunit is recruited to make lysosomes more acidic and active. The ELYSAs are most likely the same as the "endolysosomal vesicular assemblies (ELVAs)", reported by Elvan Böke's group earlier this year (Zaffagnini et al. doi.org/10.1016/j.cell.2024.01.031). However, it is clear that Satouh et al. identified and characterized these structures independently. These two studies could be complementary. Although the nature of the present study is generally descriptive, this paper provides valuable information about these giant structures. Since the ELYSA described in this paper and ELVA proposed by Elvan Böke appear to be the same structure, it would be helpful to the field if the two groups discuss unifying the nomenclature in the future.

      Comments on latest version:

      In this revised manuscript, the authors have provided additional data supporting their conclusions and also revised the text to more accurately reflect the experimental results.

    3. Reviewer #2 (Public review):

      Satouh et al report the presence of spherical structures composed of endosomes, lysosomes and autophagosomes within immature mouse oocytes. These endolysosomal compartments have been named as Endosomal-LYSosomal organellar Assembly (ELYSA). ELYSAs increase in size as the oocytes undergo maturation. ELYSAs are distributed throughout the oocyte cytoplasm of GV stage immature oocytes but these structures become mostly cortical in the mature oocytes. Interestingly, they tend to avoid the region which contain metaphase II spindle and chromosomes. They show that the endolysosomal compartments in oocytes are less acidic and therefore non-degradative but their pH decreases and become degradative as the ELYSAs begin to disassemble in the embryos post fertilization. This manuscript shows that lysosomal switching does not happen during oocyte development, and the formation of ELYSAs prevent lysosomes from being activated. Structures similar to these ELYSAs have been previously described in mouse oocytes (Zaffagnini et al, 2024) and these vesicular assemblies are important for sequestering protein aggregates in the oocytes but facilitate proteolysis after fertilization. The current manuscript, however, provides further details of endolysosomal disassembly post fertilization. Specifically, the V1-subunit of V-ATPase targeting to the ELYSAs increases the acidity of lysosomal compartments in the embryos. This is a well-conducted study and their model is supported by experimental evidence and data analyses.

      Comments on revisions:

      This revised version of the manuscript has addressed most of my concerns.

    4. Author response:

      The following is the authors’ response to the original reviews.

      Public Reviews:

      Reviewer #1 (Public Review):

      In this manuscript, Satouh et al. report giant organelle complexes in oocytes and early embryos. Although these structures have often been observed in oocytes and early embryos, their exact nature has not been characterized. The authors named these structures "endosomal-lysosomal organelles form assembly structures (ELYSAs)". ELYSAs contain organelles such as endosomes, lysosomes, and probably autophagic structures. ELYSAs are initially formed in the perinuclear region and then migrate to the periphery in an actin-dependent manner. When ELYSAs are disassembled after the 2-cell stage, the V-ATPase V1 subunit is recruited to make lysosomes more acidic and active. The ELYSAs are most likely the same as the "endolysosomal vesicular assemblies (ELVAs)", reported by Elvan Böke's group earlier this year (Zaffagnini et al. doi.org/10.1016/j.cell.2024.01.031). However, it is clear that Satouh et al. identified and characterized these structures independently. These two studies could be complementary. Although the nature of the present study is generally descriptive, this paper provides valuable information about these giant structures. The data are mostly convincing, and only some minor modifications are needed for clarification and further explanation to fully understand the results.

      Reviewer #2 (Public Review):

      Satouh et al report the presence of spherical structures composed of endosomes, lysosomes, and autophagosomes within immature mouse oocytes. These endolysosomal compartments have been named as Endosomal-LYSosomal organellar Assembly (ELYSA). ELYSAs increase in size as the oocytes undergo maturation. ELYSAs are distributed throughout the oocyte cytoplasm of GV stage immature oocytes but these structures become mostly cortical in the mature oocytes. Interestingly, they tend to avoid the region which contains metaphase II spindle and chromosomes. They show that the endolysosomal compartments in oocytes are less acidic and therefore non-degradative but their pH decreases and becomes degradative as the ELYSAs begin to disassemble in the embryos post-fertilization. This manuscript shows that lysosomal switching does not happen during oocyte development, and the formation of ELYSAs prevents lysosomes from being activated. Structures similar to these ELYSAs have been previously described in mouse oocytes (Zaffagnini et al, 2024) and these vesicular assemblies are important for sequestering protein aggregates in the oocytes but facilitate proteolysis after fertilization. The current manuscript, however, provides further details of endolysosomal disassembly post-fertilization. Specifically, the V1-subunit of V-ATPase targeting the ELYSAs increases the acidity of lysosomal compartments in the embryos. This is a well-conducted study and their model is supported by experimental evidence and data analyses.

      Reviewer #3 (Public Review):

      Fertilization converts a cell defined as an egg to a cell defined as an embryo. An essential component of this switch in cell fate is the degradation (autophagy) of cellular elements that serve a function in the development of the egg but could impede the development of the embryo. Here, the authors have focused on the behavior during the egg-to-embryo transition of endosomes and lysosomes, which are cytoplasmic structures that mediate autophagy. By carefully mapping and tracking the intracellular location of well-established marker proteins, the authors show that in oocytes endosomes and lysosomes aggregate into giant structures that they term Endosomal LYSosomal organellar Assembl[ies] (ELYSA). Both the size distribution of the ELYSAs and their position within the cell change during oocyte meiotic maturation and after fertilization. Notably, during maturation, there is a net actin-dependent movement towards the periphery of the oocyte. By the late 2-cell stage, the ELYSAs are beginning to disintegrate. At this stage, the endo-lysosomes become acidified, likely reflecting the activation of their function to degrade cellular components.

      This is a carefully performed and quantified study. The fluorescent images obtained using well-known markers, using both antibodies and tagged proteins, support the interpretations, and the quantification method is sophisticated and clearly explained. Notably, this type of quantification of confocal z-stack images is rarely performed and so represents a real strength of the study. It provides sound support for the conclusions regarding changes in the size and position of the ELYSAs. Another strength is the use of multiple markers, including those that indicate the activity state of the endo-lysosomes. Altogether, the manuscript provides convincing evidence for the existence of ELYSAs and also for regulated changes in their location and properties during oocyte maturation and the first few embryonic cell cycles following fertilization.

      At present, precisely how the changes in the location and properties of the ELYSAs affect the function of the endo-lysosomal system is not known. While the authors' proposal that they are stored in an inactive state is plausible, it remains speculative. Nonetheless, this study lays the foundation for future work to address this question.

      Minor point: l. 299. If I am not mistaken, there is a typo. It should read that the inhibitors of actin polymerization prevent redistribution from the cytoplasm to the cortex during maturation.

      Minor point: A few statements in the Introduction would benefit from clarification. These are noted in the comments to the authors.

      We sincerely appreciate the editorial board of eLife and the reviewers for their helpful and constructive comments on our manuscript. We are pleased that the reviewers acknowledged that we identified and characterized this assembly structure independently. In the revised manuscript, we have carefully considered the reviewers’ comments and conducted additional analysis to address each of them.

      Regarding the typographical errors, we revised the description to fit with our findings and the reviewers’ comments. We also found that the primer sequence was correct, and we carefully checked the accuracy of the entire manuscript.

      We hope that the revised version will now be deemed suitable for publication in eLife.

      Recommendations for the authors:

      Reviewer #1 (Recommendations For The Authors):

      Q. 1) The authors state in the Abstract that ELYSAs contain autophagosome-like membranes in the outer layer. However, this seems to be just speculation based on the LC3 staining results and is not directly shown. Are there autophagosome-like double membrane structures in ELYSAs?

      We appreciate this comment. We also agree with this concern; however, it was difficult to assert that they are autophagosomes based on the observation of the electron micrographs. For this reason, we rephrased it to be "Most ELYSAs are also positive for an autophagy regulator, LC3.” (lines 33). In addition, we revised the notation to LC3-positive structures in the Result and Discussion section (line 165-169, 286).

      Q. 2) The data in Figure 2A, showing a decrease in the number of LAMP1 structures, seems to contradict the data in Figure 1B, showing an apparent increase in LAMP1 structures. Please explain this discrepancy. If the authors did not count structures just below the plasma membrane, please explain the rationale for this.

      We really appreciate the valuable comment. Regarding the number of LAMP1-positive structures, it is not suitable for comparison with Figure 1B, etc., as pointed out by the reviewer, since the distribution of the LAMP1 signal differs from plane to plane. To avoid any potential confusion, we added new images of the Z-projection of the immunostained images that can better reflect the number of positive structures in the whole oocyte/embryo in Figure 2.

      In addition, as the reviewer pointed out, there is a technical difficulty in measuring the LAMP1-positive signal on the plasma membrane or just below it. We explained how and why we had to delete plasma membrane signals in our response #21.

      Q. 3) The actin dependence is not observed in Figure 5C. What is the difference between Figure 5C and 5E? Please explain further.

      We apologize for the lack of clarity; Figures 5C and 5E show the average number of LAMP1-positive structures (5C) and the percentage of the sum of granule volumes in LAMP1 positive structure (5E), respectively, after classifying the LAMP1 positive granules by their diameters.

      We removed Figure 5E for the sake of conciseness since we already mentioned a similar fact in Figure 5C. To clarify the corresponding explanations, we moved figures that were not classified by diameter to Supplementary Figure 8 to improve readability. Moreover, we have rewritten the main text on lines 200–211.

      Q. 4) While the actin inhibitors reduce the number of peripheral LAMP1 structures (Figure 5F), they do not affect their number in the central region (Figure 5G). How can the authors conclude that actin inhibitors inhibit the migration of LAMP1 structures?

      We appreciate the comment. As pointed out, the number of large LAMP1-positive structures in the medial region did not change. Therefore, we have avoided the description that ELYSAs migrate from the middle region to the cell periphery and have unified the description of whether large structures in the periphery occur. Please refer to the subsection title (line 188), the following descriptions (lines 189–199), the related description in the Results (lines 200–211), and the title and the legend of Figure 5.

      Q. 5) The authors show that the V1A subunit associates with the surface of LAMP1 structures as punctate structures (Figure 6B). What are these V1A-positive structures? Is V1A recruited to some specific domains of ELYSAs, or are V1A-positive active lysosomes recruited to ELYSAs? Please provide an interpretation of these data. The phrase "The V1-subunit of V-ATPase is targeted to these structures" (line 262) is not appropriate because it is indistinguishable whether only the V1 subunits are recruited or active lysosomes containing the V1 subunit are recruited.

      Thank you for the valuable comment. Indeed, our analysis, including the analysis of Fig. 8 described on line 262, did not clarify whether free V1A-mCherry molecules accessed the ELYSA periphery or whether lysosomes with V1A-mCherry molecules newly merged into the ELYSA. Therefore, we added this interpretation to lines 232–234 of the Results and revised the Discussion as "The number of membrane structures positive for V1A-mCherry increase upon ELYSA disassembly, indicating further acidification of the endosomal/lysosomal compartment" (lines 292–294).

      Q. 6) Why did the authors use LysoSensor as a marker for ELYSA instead of LAMP1 in Figure 8 and 9? Some reasons should be given.

      There is a clear technical reason for this: when LAMP1-EGFP was expressed in a zygote, it was largely migrated to the plasma membrane before and after the 2-cell stage, making it difficult to capture the change of ELYSAs. To circumvent this difficulty, we used Lysosensor to visualize ELYSAs instead of LAMP1-EGFP. This explanation was added to lines 258–260.

      Q. 7) In Figure 9A, it is not clear whether the activity of LysoSensor-positive structures is lower at this stage compared to other stages. It may be shown in Figure S7, but the data are not clearly visible. A direct comparison would be ideal.

      A new analysis similar to that shown in Fig. 9 for early 2-cells and 4-cells was performed and added to Figure S7. To support direct comparison, the ranges of axes were set to be similar.

      As a result, the quantified MagicRed signal on the isolated LysoSensor-positive punctate structure in MII oocyte was nearly the same as that in early 2-cells and 4-cells. In early 2-cells, LysoSensor gave a signal at the cellular boundary, where MagicRed staining was not observed, confirming that MagicRed activity is higher in the interior than in the cell periphery in post-fertilization embryos. We have included an additional description in the main text (lines 280–282).

      Q. 8) In the phrase "pregnant mare serum gonadotropin or an anti-inhibin antibody" (line 382), is "or" correct?

      When inducing superovulatory stimulation, an anti-inhibin antibody (distributed as CARD HyperOva) can be used as a substitute for PMSG (after additional stimulation with hCG), which results in the production of eggs of similar quality to those of PMSG. This was used in most experiments. To amend the lack of clarity, a reference (Takeo and Nakagata Plos One, 2015) was added to the description of HyperOva (line 417).

      Q. 9) In almost all graphs, please indicate what the X-axis is indicating (not just "number") so that readers can understand what number is being represented without reading the legends.

      We revised the axis titles in all figures.

      Q. 10) Since grayscale images provide better contrast than color images, it is recommended that single-color images be shown in grayscale.

      We replaced all single-color images with grayscale images.

      Reviewer #2 (Recommendations For The Authors):

      Specific comments:

      Q. 11) Figure 1 and S1- Both Rab5 and Rab7 co-localize with LAMP1. However, there seems to be a lot of LAMP1-free Rab5 dots as compared to the Lamp1-free Rab7. As a result, LAMP1 and Rab7 are co-localized more frequently than LAMP1 and Rab5 (video1). Could it be that early endosomes (Rab5+) are yet to be incorporated into ELYSAs? If so, a brief discussion of this phenomenon would be nice.

      Thank you very much for the comment. We agree with the reviewer’s interpretation. In accordance with this suggestion, we clearly stated in the main text: “Although small punctate structures that are RAB5-positive but LAMP1-negative also spread over the cytosol, most giant structures were positive for RAB5 and LAMP1 (Video 1)” (lines 91–93). In the Discussion section, a brief statement was included: “Considering the large number of RAB5-positive and LAMP1-negative punctate structures in MII oocytes, these layers may also reflect the assembly mechanism of the ELYSA” (lines 318–320).

      Q. 12) Video 3 (and Figure 6) clearly shows the dynamics of LAMP1-labelled vesicles during maturation, which is impressive. In contrast to the live cell imaging after LAMP1 mRNA injection, Figure 1 used anti-LAMP1 Ab to detect endogenous levels of LAMP1. It appears that mRNA microinjection causes LAMP1 overexpression causing more (but smaller) vesicles to form. It should be easy to quantify and compare the vesicles in Figure 1 and 6

      We appreciate the comment. As mentioned, injections of EGFP-LAMP1 mRNA are useful for the visualization of LAMP1 dynamics during the maturation phase from GV to MII by live cell imaging, which is not feasible with immunostaining. However, the fluorescence emitted by EGFP-LAMP1 is only a few tenths of that of antibody staining, and because of the technical difficulty of microinjection into GV oocytes, the signal-to-noise ratio sufficient for imaging was merely one in ten oocytes. In addition, live cell imaging of oocytes in Figure 6 had to be carried out with very low excitation light exposure to reduce the toxicity. It was also performed with a low magnification lens and a longer step size in the z-axis. For these reasons, in examining the point raised, we performed an additional 3D object analysis, in the same way as in Figure 2, on the data of IVM oocytes injected with EGFP-LAMP1 mRNA using the same lens as in Figure 1 and with a longer exposure time than in live imaging. The results were compared with the MII data of Figures 1 and 2.

      As a result, as shown in the new Figure S8, more objects with a diameter of 0.2–0.4 µm were found than in the immunostaining data, which fits the reviewer’s point. In addition, the counts were lower for the 0.6–1.0 µm diameter, but there was no significant difference in the number of larger LAMP1 positive structures corresponding to the ELYSA size. We consider that this was appropriate for the original purpose of characterizing the ELYSA formation process. A description of these points has been added to lines 221–225.

      Q. 13) In Figure 4A and B- Seems like not all LAMP1-positive structures were LC3-positive. Is there any size or location within the oocyte that determines LC3 positivity?

      We appreciate the valuable comment. To answer this comment, we proceeded with a new 3D object-based co-localization analysis on Lamp1 and LC3, determined the number, volume, and distribution within the oocyte, and incorporated the results as Supplementary Figure 6. To examine the positivity, we further analyzed the percentage of double-positive structures of all the LAMP1-positive structures. The results showed that their average diameter significantly shifted from 2.36 µm (GV) to 3.78 µm (MII). Moreover, it was clearly indicated that LAMP1-positive structures smaller than 2 µm in diameter are rarely positive for LC3. In terms of location, measuring the distance of the double positive structures from the oocyte center (the cellular geometric center) indicated that they tend to be observed at the periphery of both stages of oocytes (more than 80% in > 30 µm in the MII oocyte). Of note, no clear tendency of double positivity was observed. A description of these points has been added to lines 174–186.

      Q. 14) In discussion, line 256- Small ELYSAs are formed in GV oocytes. Since you haven't checked the smaller-sized, growing oocytes, I suggest rephrasing this sentence as 'are present' rather than 'are formed'.

      We agree with the reviewer’s suggestion and changed it to "present" (line 287).

      Q. 15) Line 188- ELISA should instead be ELYSA

      Thank you for pointing this out. We have found a few more typographical errors, and all of them have been corrected (lines 213 and 321).

      Reviewer #3 (Recommendations For The Authors):

      Q. 16) Line 42: What do you mean by 'zygotic gene expression following the degradation of the cellular components of each maternal and paternal gamete'? ZGA requires this degradation? Please provide supporting references from the literature.

      We apologize for the confusing wording. We meant to say that both ZGA and degradation of parental components are required. To avoid misunderstanding, we have revised “zygotic gene expression as well as the degradation of the cellular components of each maternal and paternal gamete” and inserted a new reference (line 44).

      Q. 17) 50: MII means metaphase II, not meiosis II.

      We corrected the clerical mistake (line 50).

      Q. 18) 51: Define LC3.

      We added the definition of LC3 (line 51-52).

      Q. 19) 60: 'lysosomal activity in oocytes is upregulated by sperm-derived factors as the oocytes grow and mature'. As written, the sentence implies that oocytes grow and mature after fertilization. This may be true for maturation, but I would be surprised to learn that there is growth of the oocyte after fertilization.

      We appreciate this valuable comment.

      The C. elegans lives mainly as a hermaphrodite, which contains a couple of U-shaped gonad arms including the ovary, spermatheca and uterus in the body. Oocytes grow in the ovary and maturate upon receiving major sperm proteins secreted from sperms and ovulated to the spermatheca for fertilization. In 2017, Kenyon’s group reported that major sperm proteins act as sperm-secreted hormones to upregulates the lysosomal activity in oocytes during oocyte growth and maturation. We have revised our manuscript to avoid misunderstanding, to ' lysosomal activity in oocytes is upregulated by major sperm proteins secreted from sperms as the oocytes grow and mature '. (L. 61-66).

      Q. 20) 94 and Figure 1B: While it is clear that many LAMP1 foci at the late 2-cell stage do not also contain RAB5, it seems that the majority of RAB5 loci also stain for LAMP1. This may be a minor point in the context of the paper but could be clarified.

      We could not easily agree with the suggestion because of the possibility that the images might give different impressions on each plane. Therefore, as a way to verify this point, we attempted to quantify the co-localization by reconstructing the 3D puncta information based on the two types of antibody staining data. Unfortunately, as shown in Fig. 1AB, Rab5 had a high cytoplasmic background, and although we were able to extract peaks, we could not reliably recalibrate the three-dimensional punctate structure (please refer to the new Supplementary Fig. 6). Therefore, co-localization on each other's punctate structure (LAMP1/RAB5 vs. RAB5/LAMP1) could not be verified. The validation using specific planes also showed large differences between planes, with overlapping punctate structures counted separately in adjacent planes, making reliable quantification difficult. This is an issue that will be addressed in the future.

      On the other hand, the newly added Z-projection figure (Fig. 1AB) shows that RAB5-positive and LAMP1-negative punctate structures tend to accumulate along the LAMP1-positive punctate structures larger than 1 µm at the late 2-cell stage in all observed embryos; we added this statement on lines 99–101.

      Q. 21) 100-102 and Figure 2A: Does the decrease in the total number of LAMP1 foci refer just to cytoplasmic or also to membrane foci? If the former, what was the reason for not including the membrane in the analysis?

      We appreciate the critical question. The LAMP1 signal on the plasma membrane interfered with the measurement of the signals just below the plasma membrane. The biological cause of this increased signal on the plasma membrane, as shown in Fig. 2E, seemed to be caused by the migration of the LAMP1 signals post-fertilization, which was also reported in a previous paper by Zaffagnini et al. (2024), published in Cell.

      In our analysis, oocytes are giant cells, and confocal imaging has a technical limitation in obtaining the same fluorescent intensity along the z-axis. However, 3D-object analysis requires thresholding based on absolute values. As a result of this situation, the presence of the plasma membrane signal caused punctate structures located close to the membrane to be captured and recognized as a single, very large LAMP1-positive structure, resulting in the loss of the punctate structure that should be measured.

      To avoid this issue, we have used several programs to correct the fluorescence difference along the z-axis; nonetheless, these attempts were unsuccessful. Therefore, as described in the Materials and Methods section, we applied only background subtraction at each z-position and then manually removed the plasma membrane signal (which was thin and continuous at the edges). Furthermore, when the plasma membrane and punctate structure signals overlapped, we paid attention not to remove the signals but to separate them. Thus, we believe that the decrease in the number and volume of LAMP1-positive structures after fertilization is still a phenomenon associated with the shift of LAMP1 to the plasma membrane.

      Q. 22) Figure 2B, F, G: As the x-axis does not represent a continuous variable, adjacent data points should not be connected by a line. The histogram representations in A, C, and E are much easier to understand. I suggest presenting all data in this format.

      We revised the line graphs to bar graphs. Besides, to make the significance among populations clearer, the significances are now expressed using alphabetical indicators.

      Q. 23) Figure 2B, C: It seems that the values for the different stages are expressed relative to the value at MII. Why not use the GV value at the base-line? This would follow the developmental trajectory of the oocyte/embryo more directly and would not (I believe) change the conclusions.

      We appreciated the comment. We meant to express that ELYSA develops most in the MII phase and that it decreases after fertilization, so considering the reviewer’s suggestion, we expressed GV-MII changes based on GV and changes after fertilization based on the MII phase (Fig. 2C, D).

    1. eLife Assessment

      This study convincingly shows that aquaporin-mediated cell migration plays a key role in blood vessel formation during zebrafish development. In particular, the paper implicates hydrostatic pressure and water flow as mechanisms controlling endothelial cell migration during angiogenic sprouting. This fundamental study is highly novel and significantly advances our understanding of cell migration during morphogenesis. As such, this work will be of great interest to developmental and cell biologists working on organogenesis, angiogenesis, and cell migration.

    2. Reviewer #1 (Public review):

      Summary:

      The paper details a study of endothelial cell vessel formation during zebrafish development. The results focus on the role of aquaporins, which mediate the flow of water across the cell membrane, leading to cell movement. The authors show that actin and water flow together drive endothelial cell migration and vessel formation. If any of these two elements are perturbed, there are observed defects in vessels. Overall, the paper significantly improves our understanding of cell migration during morphogenesis in organisms.

      Strengths:

      The data are extensive and are of high quality. There is a good amount of quantification with convincing statistical significance. The overall conclusion is justified given the evidence.

      Weaknesses:

      There are two weaknesses, which if addressed, would improve the paper.

      (1) The paper focuses on aquaporins, which while mediates water flow, cannot drive directional water flow. If the osmotic engine model is correct, then ion channels such as NHE1 are the driving force for water flow. Indeed this water is shown in previous studies. Moreover, NHE1 can drive water intake because the export of H+ leads to increased HCO3 due to reaction between CO2+H2O, which increases the cytoplasmic osmolarity (see Li, Zhou and Sun, Frontiers in Cell Dev. Bio. 2021). If NHE cannot be easily perturbed in zebrafish, it might be of interest to perturb Cl channels such as SWELL1, which was recently shown to work together with NHE (see Zhang, et al, Nat. Comm. 2022).

      After revision, this concern has been addressed.

      (2) In some places the discussion seems a little confusing where the text goes from hydrostatic pressure to osmotic gradient. It might improve the paper if some background is given. For example, mention water flow follows osmotic gradients, which will build up hydrostatic pressure. The osmotic gradients across the membrane are generated by active ion exchangers. This point is often confused in literature and somewhere in the intro, this could be made clearer.

      After revision, this concern has been addressed.

    3. Reviewer #3 (Public review):

      Summary:

      Kondrychyn and colleagues describe the contribution of two Aquaporins Aqp1a.1 and Aqp8a.1 towards angiogenic sprouting in the zebrafish embryo. By whole-mount in situ hybridization, RNAscope and scRNA-seq, they show that both genes are expressed in endothelial cells in partly overlapping spatiotemporal patterns. Pharmacological inhibition experiments indicate a requirement for VEGR2 signaling (but not Notch) in transcriptional activation.

      To assess the role of both genes during vascular development the authors generate genetic mutations. While homozygous single mutants appear normal, aqp1a.1;aqp8a.1 double mutants exhibit defects in EC sprouting and ISV formation.

      At the cellular level, the aquaporin mutants display a reduction of filopodia in number and length. Furthermore, a reduction in cell volume is observed indicating a defect in water uptake.

      The authors conclude, that polarized water uptake mediated by aquaporins is required for the initiation of endothelial sprouting and (tip) cell migration during ISV formation. They further propose that water influx increases hydrostatic pressure within the cells which may facilitate actin polymerization and formation membrane protrusions.

      In the revised version of the manuscript the authors have added data which show that inhibition of swell-induced chloride channels mimics aqp mutant phenotypes, giving credence to the model that water influx via aquaporins is driven by an osmotic gradient.

      Strengths:

      The authors provide a detailed analysis of Aqp1a.1 and Aqp8a.1 during blood vessel formation in vivo, using zebrafish intersomitic vessels as a model. State-of-the-art imaging demonstrates an essential role aquaporins in different aspects of endothelial cell activation and migration during angiogenesis.

      Weaknesses:

      With respect to the connection between Aqp1/8 and actin polymerization/filopodia formation, the evidence appears preliminary and the authors' interpretation is guided by evidence from other experimental systems.

      After revision, the authors have addressed all other concerns

    4. Author response:

      The following is the authors’ response to the original reviews.

      Reviewer #1 (Public Review):

      Summary:

      This paper details a study of endothelial cell vessel formation during zebrafish development. The results focus on the role of aquaporins, which mediate the flow of water across the cell membrane, leading to cell movement. The authors show that actin and water flow together drive endothelial cell migration and vessel formation. If any of these two elements are perturbed, there are observed defects in vessels. Overall, the paper significantly improves our understanding of cell migration during morphogenesis in organisms.

      Strengths:

      The data are extensive and are of high quality. There is a good amount of quantification with convincing statistical significance. The overall conclusion is justified given the evidence.

      Weaknesses:

      There are two weaknesses, which if addressed, would improve the paper.

      (1) The paper focuses on aquaporins, which while mediates water flow, cannot drive directional water flow. If the osmotic engine model is correct, then ion channels such as NHE1 are the driving force for water flow. Indeed this water is shown in previous studies. Moreover, NHE1 can drive water intake because the export of H+ leads to increased HCO3 due to the reaction between CO2+H2O, which increases the cytoplasmic osmolarity (see Li, Zhou and Sun, Frontiers in Cell Dev. Bio. 2021). If NHE cannot be easily perturbed in zebrafish, it might be of interest to perturb Cl channels such as SWELL1, which was recently shown to work together with NHE (see Zhang, et al, Nat. Comm. 2022).

      (2) In some places the discussion seems a little confusing where the text goes from hydrostatic pressure to osmotic gradient. It might improve the paper if some background is given. For example, mention water flow follows osmotic gradients, which will build up hydrostatic pressure. The osmotic gradients across the membrane are generated by active ion exchangers. This point is often confused in literature and somewhere in the intro, this could be made clearer.

      Reviewer #1 (Recommendations For The Authors):

      (1) The paper focuses on aquaporins, which while mediating water flow, cannot drive directional water flow. If the osmotic engine model is correct, then ion channels such as NHE1 are the driving force for water flow. Indeed this water is shown in previous studies. Moreover, NHE1 can drive water intake because the export of H+ leads to increased HCO3 due to the reaction between CO2+H2O, which increases the cytoplasmic osmolarity (see Li, Zhou and Sun, Frontiers in Cell Dev. Bio. 2021). If NHE cannot be easily perturbed in zebrafish, it might be of interest to perturb Cl channels such as SWELL1, which was recently shown to work together with NHE (see Zhang, et al, Nat. Comm. 2022).

      We thank Reviewer #1 for this very important comment and the suggestion to examine the function of ion channels in establishing an osmotic gradient to drive directional flow. We have taken on board the reviewer’s suggestion and examined the expression of NHE1 and SWELL1 in endothelial cells using published scRNAseq of 24 hpf ECs (Gurung et al, 2022, Sci. Rep.). We found that slc9a1a, slc9a6a, slc9a7, slc9a8, lrrc8aa and lrrc8ab are expressed in different endothelial subtypes. To examine the function of NHE1 and SWELL1 in endothelial cell migration, we used the pharmacological compounds, 5-(N-ethyl-Nisopropyl)amiloride (EIPA) and DCPIB, respectively. While we were unable to observe an ISV phenotype after EIPA treatment at 5, 10 and 50µM, we were able to observe impaired ISV formation after DCPIB treatment that was very similar to that observed in Aquaporin mutants. We were very encouraged by these results and proceeded to perform more detailed experiments whose results have yielded a new figure (Figure 6) and are described and discussed in lines 266 to 289 and 396 to 407, respectively, in the revised manuscript.

      (2) In some places the discussion seems a little confusing where the text goes from hydrostatic pressure to osmotic gradient. It might improve the paper if some background is given. For example, mention water flow follows osmotic gradients, which will build up hydrostatic pressure. The osmotic gradients across the membrane are generated by active ion exchangers. This point is often confused in literature and somewhere in the intro, this could be made clearer.

      Thank you for pointing out the deficiency in explaining how osmotic gradients drive water flow to build up hydrostatic pressure. We have clarified this in lines 50, 53 - 54 and 385.

      The two recommendations listed above would improve the paper. They are however not mandatory. The paper would be acceptable with some clarifying rewrites. I am not an expert on zebrafish genetics, so it might be difficult to perturb ion channels in this model organism. Have the authors tried to perturb ion channels in these cells?

      We hope that our attempts at addressing Reviewer’s 1 comments are satisfactory and sufficient to clarify the concerns outlined.

      Reviewer #2 (Public Review):

      Summary:

      Directional migration is an integral aspect of sprouting angiogenesis and requires a cell to change its shape and sense a chemotactic or growth factor stimulus. Kondrychyn I. et al. provide data that indicate a requirement for zebrafish aquaporins 1 and 8, in cellular water inflow and sprouting angiogenesis. Zebrafish mutants lacking aqp1a.1 and aqp8a.1 have significantly lower tip cell volume and migration velocity, which delays vascular development. Inhibition of actin formation and filopodia dynamics further aggravates this phenotype. The link between water inflow, hydrostatic pressure, and actin dynamics driving endothelial cell sprouting and migration during angiogenesis is highly novel.

      Strengths:

      The zebrafish genetics, microscopy imaging, and measurements performed are of very high quality. The study data and interpretations are very well-presented in this manuscript.

      Weaknesses:

      Some of the mechanobiology findings and interpretations could be strengthened by more advanced measurements and experimental manipulations. Also, a better comparison and integration of the authors' findings, with other previously published findings in mice and zebrafish would strengthen the paper.

      We thank Reviewer #2 for the critique that the paper can be strengthened by more advanced measurements and experimental manipulations. One of the technical challenges that we face is how to visualize and measure water flow directly in the zebrafish. We have therefore taken indirect approaches to assess water abundance in endothelial cells in vivo. One approach was to measure the diffusion of GEM nanoparticles in tip cell cytoplasm in wildtype and Aquaporin mutants, but results were inconclusive. The second was to measure the volume of tip cells, which should reflect water in/outflow. As the second approach produced clear and robust differences between wildtype ECs, ECs lacking Aqp1a.1 and Aqp8a.1 and ECs overexpressing Aqp1a.1 (revised Fig. 5), we decided to present these data in this manuscript.

      We have also taken Reviewer 2 advice to better incorporate previously published data in our discussion (see below and lines 374 to 383 of the revised manuscript).

      Reviewer #2 (Recommendations For The Authors):

      I have a few comments that the authors may address to further improve their manuscript analysis, quality, and impact.

      Major comments:

      (1) Citation and discussion of published literature

      The authors have failed to cite and discuss recently published results on the role of aqp1a.1 and aqp8a.1 in ISV formation and caliber in zebrafish (Chen C et al. Cardiovascular Research 2024). That study showed a similar impairment of ISV formation when aqp1a.1 is absent but demonstrated a stronger phenotype on ISV morphology in the absence of aqp8a.1 than the current manuscript by Kondrychyn I et al. Furthermore, Chen C et al show an overall decrease in ISV diameter in single aquaporin mutants suggesting that the cell volume of all ECs in an ISV is affected equally. Given this published data, are ISV diameters affected in single and double mutants in the current study by Kondrochyn I et al? An overall effect on ISVs would suggest that aquaporin-mediated cell volume changes are not an inherent feature of endothelial tip cells. The authors need to analyse/compare and discuss all differences and similarities of their findings to what has been published recently.

      We apologise for having failed and discussed the recently published paper by Chen et al. This has been corrected and discussed in lines 374 to 383.

      In the paper by Chen et al, the authors describe a role of Aqp1a.1 and Aqp8a.1 in regulating ISV diameter (ISV diameter was analysed at 48 hpf) but they did not examine the earlier stages of sprouting angiogenesis between 20 to 30 hpf, which is the focus of our study. We therefore cannot directly compare the ISV phenotypes with theirs. Nevertheless, we recognise that there are differences in ISV phenotypes from 2 dpf. For example, they did not observe incompletely formed or missing ISVs at 2 and 3 dpf, which we clearly observe in our study. This could be explained by differences in the mutations generated. In Chen et al., the sgRNA used targeted the end of exon 2 that resulted in the generation of a 169 amino acid truncated aqp1a.1 protein. However, in our approach, our sgRNA targeted exon 1 of the gene that resulted in a truncated aqp1a.1 protein that is 76 amino acid long. As for the aqp8a.1 zebrafish mutant that we generated, our sgRNA targeted exon 1 of the gene that resulted in a truncated protein that is 73 amino acids long. In Chen et al., the authors did not generate an aqp8a.1 mutant but instead used a crispant approach, which leads to genetic mosaicism and high experimental variability.

      Following the reviewer’s suggestion, we have now measured the diameters of arterial ISVs (aISVs) and venous ISVs (vISVs) in aqp1a.1<sup>-/-</sup>, aqp8a.1<sup>-/-</sup> and aqp1a.1<sup>-/-</sup>;aqp8a.1<sup>-/-</sup> zebrafish. In our lab, we always make a distinction between aISVs and vISVs are their diameters are significantly different from each other. The results are in Fig S11A. While we corroborate a decrease in diameter in both aISVs and vISVs in single aqp1a.1<sup>-/-</sup> and double aqp1a.1<sup>-/-</sup>;aqp8a.1<sup>-/-</sup>.zebrafish, we observed a slight increase in diameter in both aISVs and vISVs in aqp8a.1<sup>-/-</sup> zebrafish at 2 dpf. We also measured the diameter of aISV and vISV in Tg(fli1ep:aqp1a.1-mEmerald) and Tg(fli1ep:aqp8a.1-mEmerald) zebrafish at 2 dpf (Fig S11B) and unlike in Chen et al., we could not detect a difference in the diameter between control and aqp1a.1- or aqp8a.1-overexpressing endothelial cells.

      We also would also like to point out that, because ISVs are incompletely formed or are missing in aqp1a.1<sup>-/-</sup>;aqp8a.1<sup>-/-</sup> zebrafish (Fig. 3G – L), blood flow is most likely altered in the zebrafish trunk of these mutants, and this can have a secondary effect on blood vessel calibre or diameter. In fact, we often observed wider ISVs adjacent to unperfused ISVs (Fig. 3J) as more blood flow enters the lumenized ISV. Therefore, to determine the cell autonomous function of Aquaporin in mediating cell volume changes in vessel diameter regulation, one would need to perform cell transplantation experiments where we would measure the volume of single aqp1a.1<sup>-/-</sup>;aqp8a.1<sup>-/-</sup> endothelial cells in wildtype embryos with normal blood flow. As this is beyond the scope of the present study, we have not done this experiment during the revision process.

      (2) Expression of aqp1a.1 and aqp8a.1

      The quantification shown in Figure 1G shows a relative abundance of expression between tip and stalk cells. However, it seems aqp8a.1 is almost never detected in most tip cells. The authors could show in addition, the % of Tip and stalk cells with detectable expression of the 2 aquaporins. It seems aqp8a1 is really weakly or not expressed in the initial stages. Ofcourse the protein may have a different dynamic from the RNA.

      We would like to clarify that aqp8a.1 mRNA is not detected in tip cells of newly formed ISVs at 20hpf. At 22 hpf, it is expressed in both tip cells (22 out of 23 tip cells analysed) and stalk cells of ISVs at 22hpf. This is clarified in lines 107 - 109. We also include below a graph showing that although aqp8a.1 mRNA is expressed in tip cells, its expression is higher in stalk cells.

      Author response image 1.

      Could the authors show endogenously expressed or tagged protein by antibody staining? The analysis of the Tg(fli1ep:aqp8a.1-mEmerald)rk31 zebrafish line is a good complement, but unfortunately, it does not reveal the localization of the endogenously expressed protein. Do the authors have any data supporting that the endogenously expressed aqp8a.1 protein is present in sprouting tip cells?

      We tested several antibodies against AQP1 (Alpha Diagnostic International, AQP11-A; ThermoFisher Scientific, MA1-20214; Alomone Labs, AQP-001) and AQP8 (Sigma Aldrich, SAB 1403559; Alpha Diagnostic International, AQP81-A; Almone Labs, AQP-008) but unfortunately none worked. As such, we do not have data demonstrating endogenous expression and localisation of Aqp1a.1 and Aqp8a.1 proteins in endothelial cells.

      Could the authors perform F0 CRISPR/Cas9 mediated knockin of a small tag (i.e. HA epitope) in zebrafish and read the endogenous protein localization with anti-HA Ab?

      CRISPR/Cas9 mediated in-frame knock-in of a tag into a genomic locus is a technical challenge that our lab has not established. We therefore cannot do this experiment within the revision period.

      Given the double mutant phenotypic data shown, is aqp8a.1 expression upregulated and perhaps more important in aqp1a.1 mutants?

      In our analysis of aqp1a.1 homozygous zebrafish, there is a slight down_regulation in _aqp8a.1 expression (Fig. S5C). Because the loss of Aqp1a.1 leads to a stronger impairment in ISV formation than the loss of Aqp8a.1 (see Fig. S6F, G, I and J), we believe that Aqp1a.1 has a stronger function than Aqp8a.1 in EC migration during sprouting angiogenesis.

      Regarding the regulation of expression by the Vegfr inhibitor Ki8751, does this inhibitor affect Vegfr/ERK signalling in zebrafish and the sprouting of ISVs significantly?

      ki8751 has been demonstrated to inhibit ERK signalling in tip cells in the zebrafish by Costa et al., 2016 in Nature Cell Biology. In our experiments, treatment with 5 µM ki8751 for 6 hours from 20 hpf also inhibited sprouting of ISVs.

      The data presented suggest that tip cells overexpressing aqp1a.1-mEmerald (Figure 2C) need more than 6 times longer to migrate the same distance as tip cells expressing aqp8a.1mEmerald (Figure 2D). How does this compare with cells expressing only Emerald? A similar time difference can be seen in Movie S1 and Movie S2. Is it just a coincidence? Could aqp8a.1, when expressed at similar levels than aqp1a, be more functional and induce faster cell migration? These experiments were interpreted only for the localization of the proteins, but not for the potential role of the overexpressed proteins on function. Chen C et al. Cardiovascular Research 2024 also has some Aqp overexpression data.

      The still images prepared for Fig. 2 C and D were selected to illustrate the localization of Aqp1a.1-mEmerald and Aqp8a.1-mEmerald at the leading edge of migrating tip cells. We did not notice that the tip cell overexpressing Aqp1a.1-mEmerald (Figure 2C) needed more than 6 times longer to migrate the same distance as the tip cell expressing aqp8a.1-mEmerald (Figure 2D), which the reviewer astutely detected. To ascertain whether there is a difference in migration speed between Aqp1a.1-mEmerald and Aqp8a.1-mEmerald overexpressing endothelial cells, we measured tip cell migration velocity of three ISVs from Tg(fli1ep:aqp1a.1-mEmerald) and Tg(fli1ep:aqp8a.1-mEmerald) zebrafish during the period of ISV formation (24 to 29 hpf) using the Manual Tracking plugin in Fiji. As shown in the graph, there is no significant difference in the migration speed of ECs overexpressing Aqp1a.1-mEmerald and Aqp8a.1-mEmerald, suggesting that Aqp8a.1-overexpressing cells migrate at a similar rate as Aqp1a.1-overexpressing cells. As we have not generated a Tg(fli1ep:mEmerald) zebrafish line, we are unable to determine whether endothelial cells migrate faster in Tg(fli1ep:aqp1a.1mEmerald) and Tg(fli1ep:aqp8a.1-mEmerald) zebrafish compared to endothelial cell expressing only mEmerald. As for the observation that tip cells overexpressing aqp1a.1mEmerald (Figure 2C) need more than 6 times longer to migrate the same distance as tip cells expressing aqp8a.1-mEmerald, we can only surmise that it is coincidental that the images selected “showed” faster migration of one ISV from Tg(fli1ep:aqp8a.1-mEmerald) zebrafish. We do not know whether the Aqp1a.1 and Aqp8a.1 are overexpressed to the same levels in Tg(fli1ep:aqp1a.1mEmerald) and Tg(fli1ep:aqp8a.1-mEmerald) zebrafish.

      We would also like to point out that when we analysed the lengths of ISVs at 28 hpf in aqp1a.1<sup>-/-</sup> and aqp8a.1<sup>-/-</sup> zebrafish, ISVs were shorter in aqp1a.1<sup>-/-</sup> zebrafish compared to aqp8a.1<sup>-/-</sup> zebrafish (Fig. S6 F to J). These results indicate that the loss of Aqp1a.1 function causes slower migration than the loss of aqp8a.1 function, and suggest that Aqp1a.1 induces faster endothelial cell migration that Aqp8a.1.

      Author response image 2.

      The data on Aqps expression after the Notch inhibitor DBZ seems unnecessary, and is at the moment not properly discussed. It is also against what is set in the field. aqp8a.1 levels seem to increase only 24h after DBZ, not at 6h, and still authors conclude that Notch activation inhibits aqp8a.1 expression (Line 138-139). In the field, Notch is considered to be more active in stalk cells, where aqp8a.1 expression seems higher (not lower). Maybe the analysis of tip vs stalk cell markers in the scRNAseq data, and their correlation with Hes1/Hey1/Hey2 and aqp1 vs aqp8 mRNA levels will be more clear than just showing qRT-PCR data after DBZ.

      As our scRNAseq data did not include ECs from earlier during development when ISVs are developing, we have analysed of scRNAseq data of 24 hpf endothelial cells published by Gurung et al, 2022 in Scientific Reports during the revision of this manuscript. However, we are unable to detect separate clusters of tip and stalk cells. As such, we are unable to correlate hes1/hey1/hey2 expression (which would be higher in stalk cells) with that of aqp1a.1/aqp8a.1. Also, we have decided to remove the DBZ-treatment results from our manuscript as we agree with the two reviewers that they are unnecessary.

      The paper would also benefit from some more analysis and interpretation of available scRNAseq data in development/injury/disease/angiogenesis models (zebrafish, mice or humans) for the aquaporin genes characterized here. To potentially raise a broader interest at the start of the paper.

      We thank the reviewer for suggesting examining aquaporin genes in other angiogenesis/disease/regeneration models to expand the scope of aquaporin function. We will do this in future studies.

      (3) Role of aqp1a.1 and aqp8a.1 on cytoplasmic volume changes and related phenotypes

      In Figure 5 the authors show that Aqp1/Aqp8 mutant endothelial tip cells have a lower cytoplasmic volume than tip cells from wildtype fish. If aquaporin-mediated water inflow occurs locally at the leading edge of endothelial tip cells (Figure 2, line 314-318), why doesn't cytoplasmic volume expand specifically only at that location (as shown in immune cells by Boer et al. 2023)? Can the observed reduction in cytoplasmic volume simply be a side-effect of impaired filopodia formation (Figure 4F-I)?

      We believe that water influx not only expands filopodia but also the leading front of tip cells (see bracket region in Fig. 4D), where Aqp1a.1-mEmerald/Aqp8a.1-mEmerald accumulate (Fig. 2), to generate an elongated protrusion and forward expansion of the tip cell. The decrease in cytoplasmic volume observed in the aqp1a.1;aqp8a.1 double mutant zebrafish is a result of decreased formation of these elongated protrusions at the leading front of migration tip cells as shown in Fig. 4E (compare to Fig. 4D), not from just a decrease in filopodia number. In fact, in the method used to quantify cell volume, mEmerald/EGFP localization is limited to the cytoplasm and does not label filopodia well (compare mEmerald/EGFP in green with membrane tagged-mCherry in Fig. 5A - C). The volume measured therefore reflects cytoplasmic volume of the tip cell, not filopodia volume.

      Do the authors have data on cytoplasmic volume changes of endothelial tip cells in latrunculin B treated fish? The images in Figures 6 A,B suggest that there is a difference in cell volume upon lat b treatment only.

      No, unfortunately we have not performed single cell labelling and measurement of tip cells in Latrunculin B-treated embryos. We can speculate that as there is a decrease in actindriven membrane protrusions in this experiment, one would also expect a decrease in cell volume as the reviewer has observed.

      (4) Combined loss of aquaporins and actin-based force generation.

      Lines 331-332 " we show that hydrostatic pressure is the driving force for EC migration in the absence of actin-based force generation"....better leave it more open and stick to the data. The authors show that aquaporin-mediated water inflow partially compensates for the loss of actin-based force generation in cell migration. Not that it is the key driving/rescuing force in the absence of actin-based force.

      We have changed it to “we show that hydrostatic pressure can generate force for EC migration in the absence of actin-based force generation” in line 348.

      (5) Aquaporins and their role in EC proliferation

      In the study by Phnk LK et al. 2013, the authors have shown that proliferation is not affected when actin polymerization or filopodia formation is inhibited. However, in the current manuscript by Kondrychyn I. et al. this has not been analysed carefully. In Movie S4 the authors indicate by arrows tip cells that fail to invade the zebrafish trunk demonstrating a severe defect of sprouting initiation in these mutants. Yet, when only looking at ISVs that reach the dorsal side in Movie S4, it appears that they are comprised of fewer EC nuclei/ISV than the ISVs in Movie S3. At the beginning of DLAV formation, most ISVs in control Movie S3 consist of 3-4 EC nuclei, while in double mutants Movie S4 it appears to be only 2-3 EC nuclei. At the end of the Movie S4, one ISV on the left side even appears to consist of only a single EC when touching the dorsal roof. The authors provide convincing data on how the absence of aquaporin channels affects sprouting initiation and migration speed, resulting in severe delay in ISV formation. However, the authors should also analyse EC proliferation, as it may also be affected in these mutants, and may also contribute to the observed phenotype. We know that effects on cell migration may indirectly change the number of cells and proliferation at the ISVs, but this has not been carefully analysed in this paper.

      We thank the reviewer for highlighting the lack of information on EC number and division in the aquaporin mutants. We have now quantified EC number in ISVs that are fully formed (i.e. connecting the DA or PCV to the DLAV) at 2 and 3 dpf and the results are displayed in Figure S10A and B. At 2 dpf, there is a slight but significant reduction in EC number in both aISVs and vISVs in aqp1a.1<sup>-/-</sup> zebrafish and an even greater reduction in the double aqp1a. aqp1a.1<sup>/-</sup>;aqp8a.1<sup>-/-</sup> zebrafish. No significant change in EC number was observed in aqp8a.1<sup>-/-</sup> zebrafish. EC number was also significantly decreased at 3 dpf for aqp1a.1<sup>-/-</sup>, aqp8a.1<sup>-/-</sup> and aqp1a.1<sup>-/-</sup>;aqp8a.1<sup>-/-</sup> zebrafish. The decreased in EC number per ISV may therefore contribute to the observed phenotype.

      We have also quantified the number of cell divisions during sprouting angiogenesis (from 21 to 30 hpf) to assess whether the lack of Aquaporin function affects EC proliferation. This analysis shows that there is no significant difference in the number of mitotic events between aqp1a.1<sup>+/-</sup>; aqp8a.1<sup>+/-</sup> and aqp1a.1<sup>-/-</sup>;aqp8a.1<sup>-/-</sup> zebrafish (Figure S10 C), suggesting that the reduction in EC number is not caused by a decrease in EC proliferation.

      These new data are reported on lines 198 to 205 of the manuscript.

      Minor comments:

      - Figure 3K data seems not to be necessary and even partially misleading after seeing Figure 3E. Fig. 3E represents the true strength of the phenotype in the different mutants.

      Figure 3K has been removed from Figure 3.

      - Typo Figure 3L (VII should be VI).

      Thank you for spotting this typo. VII has been changed to VI.

      - Line 242: The word "required" is too strong because there is vessel formation without Aqps in endothelial cells.

      This has been changed to “ …Aqp1a.1 and Aqp8a.1 regulate sprouting angiogenesis…” (lines 238 - 239).

      - From Figure S2, the doublets cluster should be removed.

      We have performed a new analysis of 24 hpf, 34hpf and 3 dpf endothelial cells scRNAseq data (the previous analysis did not consist of 24 hpf endothelial cells). The doublets cluster is not included in the UMAP analysis.

      - Better indicate the fluorescence markers/alleles/transgenes used for imaging in Figures 6A-D.

      The transgenic lines used for this experiment are now indicated in the figure (this figure is now Figure 7).

      Reviewer #3 (Public Review):

      Summary:

      Kondrychyn and colleagues describe the contribution of two Aquaporins Aqp1a.1 and Aqp8a.1 towards angiogenic sprouting in the zebrafish embryo. By whole-mount in situ hybridization, RNAscope, and scRNA-seq, they show that both genes are expressed in endothelial cells in partly overlapping spatiotemporal patterns. Pharmacological inhibition experiments indicate a requirement for VEGR2 signaling (but not Notch) in transcriptional activation.

      To assess the role of both genes during vascular development the authors generate genetic mutations. While homozygous single mutants appear normal, aqp1a.1;aqp8a.1 double mutants exhibit defects in EC sprouting and ISV formation.

      At the cellular level, the aquaporin mutants display a reduction of filopodia in number and length. Furthermore, a reduction in cell volume is observed indicating a defect in water uptake.

      The authors conclude, that polarized water uptake mediated by aquaporins is required for the initiation of endothelial sprouting and (tip) cell migration during ISV formation. They further propose that water influx increases hydrostatic pressure within the cells which may facilitate actin polymerization and formation membrane protrusions.

      Strengths:

      The authors provide a detailed analysis of Aqp1a.1 and Aqp8a.1 during blood vessel formation in vivo, using zebrafish intersomitic vessels as a model. State-of-the-art imaging demonstrates an essential role in aquaporins in different aspects of endothelial cell activation and migration during angiogenesis.

      Weaknesses:

      With respect to the connection between Aqp1/8 and actin polymerization/filopodia formation, the evidence appears preliminary and the authors' interpretation is guided by evidence from other experimental systems.

      Reviewer #3 (Recommendations For The Authors):

      Figure 1 H, J:

      The differential response of aqp1/-8 to ki8751 vs DBZ after 6h treatment is quite obvious. Why do the authors show the effect after 24h? The effect is more likely than not indirect.

      We agree with the reviewer and we have now removed 24 hour Ki8751 treatment and all DBZ treatments from Figure 1.

      Figure 2:

      According to the authors' model anterior localization of Aqp1 protein is critical. The authors perform transient injections to mosaically express Aqp fusion proteins using an endothelial (fli1) promoter. For the interpretation, it would be helpful to also show the mCherry-CAAX channel in separate panels. From the images, it is not possible to discern how many cells we are looking at. In particular the movie in panel D may show two cells at the tip of the sprout. A marker labelling cell-cell junctions would help. Furthermore, the authors are using a strong exogenous promoter, thus potentially overexpressing the fusion protein, which may lead to mislocalization. For Aqp1a.1 an antibody has been published to work in zebrafish (e.g. Kwong et al., Plos1, 2013).

      We would like to clarify that we generated transgenic lines - Tg(fli1ep:aqp1a.1-mEmerald) and Tg(fli1ep:aqp8a.1-mEmerald) - to visualize the localization of Aqp1a.1 and Aqp8a.1 in endothelial cells, and the images displayed in Fig. 2 are from the transgenic lines (not transient, mosaic expression).

      To aid visualization and interpretation, we have now added mCherry-CAAX only channel to accompany the Aqp1a.1/Aqp8a.1-mEmerald channel in Fig. 2A and B. To discern how many cells there are in the ISVs at this stage, we have crossed Tg(fli1ep:aqp1a.1-mEmerald) and Tg(fli1ep:aqp8a.1-mEmerald) zebrafish to TgKI(tjp1a-tdTomato)<sup>pd1224</sup> (Levic et al., 2021) to visualize ZO1 at cell-cell junction. However, because tjp1-tdTomato is expressed in all cell types including the skin that lies just above the ISV and the signal in ECs in ISVs is very weak at 22 to 25 hpf, it was very difficult to obtain good quality images that can properly delineate cell boundaries to determine the number of cells in the ISVs at this early stage. Instead, we have annotated endothelial cell boundaries based on more intense mCherryCAAX fluorescence at cell-cell borders, and from the mosaic expression of mCherryCAAX that is intrinsic to the  Tg(kdrl:ras-mCherry)<sup>s916</sup> zebrafish line.

      In Fig. 2D, there are two endothelial cells in the ISV during the period shown but there is only 1 cell occupying the tip cell position i.e. there is one tip cell in this ISV. Unlike the mouse retina where it has been demonstrated that two endothelial cells can occupy the tip cell position side-by-side (Pelton et al., 2014), this is usually not observed in zebrafish ISVs. This is demonstrated in Movie S3, where it is clear that one nucleus (belonging to the tip cell) occupies the tip of the growing ISV. The accumulation of intracellular membranes is often observed in tip cells that may serve as a reservoir of membranes for the generation of membrane protrusions at the leading edge of tip cells.

      We agree that by generating transgenic Tg(fli1ep:aqp1a.1-mEmerald) and Tg(fli1ep:aqp8a.1mEmerald) zebrafish, Aqp1a.1 and Aqp8a.1 are overexpressed that may affect their localization. The eel anti-Aqp1a.1 antibody used in (Kwong et la., 2013) was a gift from Dr. Gordon Cramb, Univ. of St Andrews, Scotland and it was first published in 2001. This antibody is not available commercially. Instead, we have tried to several other antibodies against AQP1 (Alpha Diagnostic International , AQP11-A; ThermoFisher Scientific, MA120214; Alomone Labs, AQP-001) and AQP8 (Sigma Aldrich, SAB 1403559; Alpha Diagnostic International, AQP81-A; Almone Labs, AQP-008) but unfortunately none worked. As such, we cannot compare localization of Aqp1a.1-mEmerald and Aqp8a.1-mEmerald with the endogenous proteins.

      Figure 3:

      E: the quantification is difficult to read. Wouldn't it be better to set the y-axis in % of the DV axis? (see also Figure S6).

      We would like to show the absolute length of the ISVs, and to illustrate that the ISV length decreases from anterior to posterior of the zebrafish trunk. We have increased the size of Fig. 3E to enable easier reading of the bars.

      K: This quantification appears arbitrary.

      We have removed this panel from Figure 3.

      G-J: The magenta channel is difficult to see. Is the lifeact-mCherry mosaic? In panel J there appears to be a nucleus between the sprout and the DLAV. It would be helpful to crop the contralateral side of the image.

      No, the Tg(fli1:Lifeact-mCherry) line is not mosaic. The “missing” vessels are not because of mosaicism in transgene but because of truncated ISVs that is a phenotype of loss Aquaporin function. We have changed the magenta channel to grey and hope that by doing so, the reviewer will be able to see the shape of the blood vessels more clearly. We would like to leave the contralateral side in the images, as it shows that the defective vessel is only on one side of body. Furthermore, when we tried to remove it (reducing the number of Z-stacks) neighbour ISV looks incomplete because the embryos were not mounted flat. To clarify what the nucleus between the sprout and the DLAV is, we have indicated that it is that of the contralateral ISV.

      L: I do not quite understand the significance of the different classes of phenotypes. Do the authors propose different morphogenetic events or contexts of how these differences come about?

      Here, we report the different types of ISV phenotypes that we observe in 3 dpf aqp1a.1<sup>-/-</sup>; aqp8a.1<sup>-/-</sup> zebrafish (Fig. 3 and Fig. S7). As demonstrated in Fig. 4, most of the phenotypes can be explained by the delayed emergence of tip cells from the dorsal aorta and slower tip cell migration. However, in some instances, we also observed retraction of tip cells (Movie S4) and failure of tip cells to emerge from the dorsal aorta or endothelial cell death (see attached figure on page 14), which can give rise to the Class II phenotype. In the dominant class I phenotype (in contrast to class II), secondary sprouting from the posterior cardinal vein is unaffected, and the secondary sprout migrates dorsally passing the level of horizontal myoseptum but cannot complete the formation of vISV (it stops beneath the spinal cord). The Class III phenotype appears to result from a failure of the secondary sprout to fuse with the regressed primary ISV. In the Class IV phenotype, the ventral EC does not maintain a connection to the dorsal aorta. We did not examine how Class III and IV phenotypes arise in detail in this current study.

      Author response image 3.

      Figure 4:

      This figure nicely demonstrates the defects in cell behavior in aqp mutants.

      In panel F it would be helpful to show the single channels as well as the merge.

      We have now added single channels for PLCd1PH and Lifeact signal in panels F and G.

      In Figure 1 the authors argue that the reduction of Aqp1/8 by VEGFR2 inhibition may account for part of that phenotype. In turn, the aqp phenotype seems to resemble incomplete VEGFR2 inhibition. The authors should check whether expression Aqp1Emerald can partially rescue ki8751 inhibition.

      To address the reviewer’s comment, we have treated Tg(fli1ep:Aqp1-Emerald) embryos with ki8751 from 20 hpf for 6 hours but we were unable to observe a rescue in sprouting. It could be because VEGFR2 inhibition also affects other downstream signalling pathways that also control cell migration as well as proliferation.

      Based on previous studies (Loitto et al.; Papadopoulus et al.) the authors propose that also in ISVs aquaporin-mediated water influx may promote actin polymerization and thereby filopodia formation. However, while the effect on filopodia number and length is well demonstrated, the underlying cause is less clear. For example, filopodia formation could be affected by reduced cell polarization. This can be tested by using a transgenic golgi marker (Kwon et al., 2016).

      We have examined tip cell polarity of wildtype, aqp1a.1<sup>-/-</sup> and  aqp8a. 1<sup>-/-</sup> embryos at 24-26 hpf by analysing Golgi position relative to the nucleus. We were unable to analyze polarity in  aqp1a.1<sup>rk28/rk28</sup>; aqp8a.1<sup>rk29/rk29</sup> embryos as they exist in an mCherry-containing transgenic zebrafish line (the Golgi marker is also tagged to mCherry). The results show that tip cell polarity is similar, if not more polarised, in aqp1a.1<sup>-/-</sup> and  aqp8a. 1<sup>-/-</sup> embryos when compared to wildtype embryos (Fig. S10D). This new data is discussed in lines 234 to 237.

      Figure 5:

      Panel D should be part of Figure 4.

      Panel 5D is now in panel J of Figure 4 and described in lines 231 and 235.

    1. eLife Assessment

      This important study presents compelling observational data supporting a role for transcription and polysome accumulation in the separation of newly replicated bacterial chromosomes. The study is generally thorough and rigorous in nature, although there are several instances where revisions would help clarify for the reader that the evidence is primarily circumstantial in nature and that a direct causal relationship between polysome accumulation has yet to be tested. With regard to the latter, the model's predictions could possibly be tested by examining the impact of translation inhibitors on nucleoid organisation. The authors could also compare the radial dimensions of the nucleoid with cell width to confirm that the nucleoid is radially confined across all conditions, a critical assumption of the model.

    2. Reviewer #1 (Public review):

      Summary:

      This paper is an elegant, mostly observational work, detailing observations that polysome accumulation appears to drive nucleoid splitting and segregation. Overall I think this is an insightful work with solid observations.

      Strengths:

      The strengths of this paper are the careful and rigorous observational work that leads to their hypothesis. They find the accumulation of polysomes correlates with nucleoid splitting, and that the nucleoid segregation occurring right after splitting correlates with polysome segregation. These correlations are also backed up by other observations:

      (1) Faster polysome accumulation and DNA segregation at faster growth rates.<br /> (2) Polysome distribution negatively correlating with DNA positioning near asymmetric nucleoids.<br /> (3) Polysomes form in regions inaccessible to similarly sized particles.

      These above points are observational, I have no comments on these observations leading to their hypothesis.

      Weaknesses:

      It is hard to state weaknesses in any of the observational findings, and furthermore, their two tests of causality, while not being completely definitive, are likely the best one could do to examine this interesting phenomenon.

      Points to consider / address:

      Notably, demonstrating causality here is very difficult (given the coupling between transcription, growth, and many other processes) but an important part of the paper. They do two experiments toward demonstrating causality that help bolster - but not prove - their hypothesis. These experiments have minor caveats, my first two points.

      (1) First, "Blocking transcription (with rifampicin) should instantly reduce the rate of polysome production to zero, causing an immediate arrest of nucleoid segregation". Here they show that adding rifampicin does indeed lead to polysome loss and an immediate halting of segregation - data that does fit their model. This is not definitive proof of causation, as rifampicin also (a) stops cell growth, and (b) stops the translation of secreted proteins. Neither of these two possibilities is ruled out fully.

      1a) As rifampicin also stops all translation, it also stops translational insertion of membrane proteins, which in many old models has been put forward as a possible driver of nucleoid segregation, and perhaps independent of growth. This should at last be mentioned in the discussion, or if there are past experiments that rule this out it would be great to note them.

      1b) They address at great length in the discussion the possibility that growth may play a role in nucleoid segregation. However, this is testable - by stopping surface growth with antibiotics. Cells should still accumulate polysomes for some time, it would be easy to see if nucleoids are still segregated, and to what extent, thereby possibly decoupling growth and polysome production. If successful, this or similar experiments would further validate their model.

      (2) In the second experiment, they express excess TagBFP2 to delocalize polysomes from midcell. Here they again see the anticorrelation of the nucleoid and the polysomes, and in some cells, it appears similar to normal (polysomes separating the nucleoid) whereas in others the nucleoid has not separated. The one concern about this data - and the differences between the "separated" and "non-separated" nuclei - is that the over-expression of TagBFP2 has a huge impact on growth, which may also have an indirect effect on DNA replication and termination in some of these cells. Could the authors demonstrate these cells contain 2 fully replicated DNA molecules that are able to segregate?

      (3) What is not clearly stated and is needed in this paper is to explain how polysomes do (or could) "exert force" in this system to segregate the nucleoid: what a "compaction force" is by definition, and what mechanisms causes this to arise (what causes the "force") as the "compaction force" arises from new polysomes being added into the gaps between them caused by thermal motions.

      They state, "polysomes exert an effective force", and they note their model requires "steric effects (repulsion) between DNA and polysomes" for the polysomes to segregate, which makes sense. But this makes it unclear to the reader what is giving the force. As written, it is unclear if (a) these repulsions alone are making the force, or (b) is it the accumulation of new polysomes in the center by adding more "repulsive" material, the force causes the nucleoids to move. If polysomes are concentrated more between nucleoids, and the polysome concentration does not increase, the DNA will not be driven apart (as in the first case) However, in the second case (which seems to be their model), the addition of new material (new polysomes) into a sterically crowded space is not exerting force - it is filling in the gaps between the molecules in that region, space that needs to arise somehow (like via Brownian motion). In other words, if the polysome region is crowded with polysomes, space must be made between these polysomes for new polysomes to be inserted, and this space must be made by thermal (or ATP-driven) fluctuations of the molecules. Thus, if polysome accumulation drives the DNA segregation, it is not "exerting force", but rather the addition of new polysomes is iteratively rectifying gaps being made by Brownian motion.

      The authors use polysome accumulation and phase separation to describe what is driving nucleoid segregation. Both terms are accurate, but it might help the less physically inclined reader to have one term, or have what each of these means explicitly defined at the start. I say this most especially in terms of "phase separation", as the currently huge momentum toward liquid-liquid interactions in biology causes the phrase "phase separation" to often evoke a number of wider (and less defined) phenomena and ideas that may not apply here. Thus, a simple clear definition at the start might help some readers.

      (4) Line 478. "Altogether, these results support the notion that ectopic polysome accumulation drives nucleoid dynamics". Is this right? Should it not read "results support the notion that ectopic polysome accumulation inhibits/redirects nucleoid dynamics"?

      (5) It would be helpful to clarify what happens as the RplA-GFP signal decreases at midcell in Figure 1- is the signal then increasing in the less "dense" parts of the cell? That is, (a) are the polysomes at midcell redistributing throughout the cell? (b) is the total concentration of polysomes in the entire cell increasing over time?

      (6) Line 154. "Cell constriction contributed to the apparent depletion of ribosomal signal from the mid-cell region at the end of the cell division cycle (Figure 1B-C and Movie S1)" - It would be helpful if when cell constriction began and ended was indicated in Figures 1B and C.

      (7) In Figure 7 they demonstrate that radial confinement is needed for longitudinal nucleoid segregation. It should be noted (and cited) that past experiments of Bacillus l-forms in microfluidic channels showed a clear requirement role for rod shape (and a given width) in the positing and the spacing of the nucleoids.<br /> Wu et al, Nature Communications, 2020 . "Geometric principles underlying the proliferation of a model cell system" https://dx.doi.org/10.1038/s41467-020-17988-7

      (8) "The correlated variability in polysome and nucleoid patterning across cells suggests that the size of the polysome-depleted spaces helps determine where the chromosomal DNA is most concentrated along the cell length. This patterning is likely reinforced through the displacement of the polysomes away from the DNA dense region"

      It should be noted this likely functions not just in one direction (polysomes dictating DNA location), but also in the reverse - as the footprint of compacted DNA should also exclude (and thus affect) the location of polysomes

      (9) Line 159. Rifampicin is a transcription inhibitor that causes polysome depletion over time. This indicates that all ribosomal enrichments consist of polysomes and therefore will be referred to as polysome accumulations hereafter". Here and throughout this paper they use the term polysome, but cells also have monosomes (and 2 somes, etc). Rifampicin stops the assembly of all of these, and thus the loss of localization could occur from both. Thus, is it accurate to state that all transcription events occur in polysomes? Or are they grouping all of the n-somes into one group?

    3. Reviewer #2 (Public review):

      Summary:

      The authors perform a remarkably comprehensive, rigorous, and extensive investigation into the spatiotemporal dynamics between ribosomal accumulation, nucleoid segregation, and cell division. Using detailed experimental characterization and rigorous physical models, they offer a compelling argument that nucleoid segregation rates are determined at least in part by the accumulation of ribosomes in the center of the cell, exerting a steric force to drive nucleoid segregation prior to cell division. This evolutionarily ingenious mechanism means cells can rely on ribosomal biogenesis as the sole determinant for the growth rate and cell division rate, avoiding the need for two separate 'sensors,' which would require careful coupling.

      Strengths:

      In terms of strengths; the paper is very well written, the data are of extremely high quality, and the work is of fundamental importance to the field of cell growth and division. This is an important and innovative discovery enabled through a combination of rigorous experimental work and innovative conceptual, statistical, and physical modeling.

      Weaknesses:

      In terms of weaknesses, I have three specific thoughts.

      Firstly, my biggest question (and this may or may not be a bona fide weakness) is how unambiguously the authors can be sure their ribosomal labeling is reporting on polysomes, specifically. My reading of the work is that the loss of spatial density upon rifampicin treatment is used to infer that spatial density corresponds to polysomes, yet this feels like a relatively indirect way to get at this question, given rifampicin targets RNA polymerase and not translation. It would be good if a more direct way to confirm polysome dependence were possible.

      Second, the authors invoke a phase separation model to explain the data, yet it is unclear whether there is any particular evidence supporting such a model, whether they can exclude simpler models of entanglement/local diffusion (and/or perhaps this is what is meant by phase separation?) and it's not clear if claiming phase separation offers any additional insight/predictive power/utility. I am OK with this being proposed as a hypothesis/idea/working model, and I agree the model is consistent with the data, BUT I also feel other models are consistent with the data. I also very much do not think that this specific aspect of the paper has any bearing on the paper's impact and importance.

      Finally, the writing and the figures are of extremely high quality, but the sheer volume of data here is potentially overwhelming. I wonder if there is any way for the authors to consider stripping down the text/figures to streamline things a bit? I also think it would be useful to include visually consistent schematics of the question/hypothesis/idea each of the figures is addressing to help keep readers on the same page as to what is going on in each figure. Again, there was no figure or section I felt was particularly unclear, but the sheer volume of text/data made reading this quite the mental endurance sport! I am completely guilty of this myself, so I don't think I have any super strong suggestions for how to fix this, but just something to consider.

    4. Reviewer #3 (Public review):

      Summary:

      Papagiannakis et al. present a detailed study exploring the relationship between DNA/polysome phase separation and nucleoid segregation in Escherichia coli. Using a combination of experiments and modelling, the authors aim to link physical principles with biological processes to better understand nucleoid organisation and segregation during cell growth.

      Strengths:

      The authors have conducted a large number of experiments under different growth conditions and physiological perturbations (using antibiotics) to analyse the biophysical factors underlying the spatial organisation of nucleoids within growing E. coli cells. A simple model of ribosome-nucleoid segregation has been developed to explain the observations.

      Weaknesses:

      While the study addresses an important topic, several aspects of the modelling, assumptions, and claims warrant further consideration.

      Major Concerns:

      Oversimplification of Modelling Assumptions:

      The model simplifies nucleoid organisation by focusing on the axial (long-axis) dimension of the cell while neglecting the radial dimension (cell width). While this approach simplifies the model, it fails to explain key experimental observations, such as:

      (1) Inconsistencies with Experimental Evidence:

      The simplified model presented in this study predicts that translation-inhibiting drugs like chloramphenicol would maintain separated nucleoids due to increased polysome fractions. However, experimental evidence shows the opposite-separated nucleoids condense into a single lobe post-treatment (Bakshi et al 2014), indicating limitations in the model's assumptions/predictions. For the nucleoids to coalesce into a single lobe, polysomes must cross the nucleoid zones via the radial shells around the nucleoid lobes.

      (2) The peripheral localisation of nucleoids observed after A22 treatment in this study and others (e.g., Japaridze et al., 2020; Wu et al., 2019), which conflicts with the model's assumptions and predictions. The assumption of radial confinement would predict nucleoids to fill up the volume or ribosomes to go near the cell wall, not the nucleoid, as seen in the data.

      (3) The radial compaction of the nucleoid upon rifampicin or chloramphenicol treatment, as reported by Bakshi et al. (2014) and Spahn et al. (2023), also contradicts the model's predictions. This is not expected if the nucleoid is already radially confined.

      (4) Radial Distribution of Nucleoid and Ribosomal Shell:

      The study does not account for well-documented features such as the membrane attachment of chromosomes and the ribosomal shell surrounding the nucleoid, observed in super-resolution studies (Bakshi et al., 2012; Sanamrad et al., 2014). These features are critical for understanding nucleoid dynamics, particularly under conditions of transcription-translation coupling or drug-induced detachment. Work by Yongren et al. (2014) has also shown that the radial organisation of the nucleoid is highly sensitive to growth and the multifork nature of DNA replication in bacteria.

      The omission of organisation in the radial dimension and the entropic effects it entails, such as ribosome localisation near the membrane and nucleoid centralisation in expanded cells, undermines the model's explanatory power and predictive ability. Some observations have been previously explained by the membrane attachment of nucleoids (a hypothesis proposed by Rabinovitch et al., 2003, and supported by experiments from Bakshi et al., 2014, and recent super-resolution measurements by Spahn et al.).

      Ignoring the radial dimension and membrane attachment of nucleoid (which might coordinate cell growth with nucleoid expansion and segregation) presents a simplistic but potentially misleading picture of the underlying factors.

      This reviewer suggests that the authors consider an alternative mechanism, supported by strong experimental evidence, as a potential explanation for the observed phenomena:<br /> Nucleoids may transiently attach to the cell membrane, possibly through transertion, allowing for coordinated increases in nucleoid volume and length alongside cell growth and DNA replication. Polysomes likely occupy cellular spaces devoid of the nucleoid, contributing to nucleoid compaction due to mutual exclusion effects. After the nucleoids separate following ter separation, axial expansion of the cell membrane could lead to their spatial separation.

      Incorporating this perspective into the discussion or future iterations of the model may provide a more comprehensive framework that aligns with the experimental observations in this study and previous work.

      Simplification of Ribosome States:<br /> Combining monomeric and translating ribosomes into a single 'polysome' category may overlook spatial variations in these states, particularly during ribosome accumulation at the mid-cell. Without validating uniform mRNA distribution or conducting experimental controls such as FRAP or single-molecule measurements to estimate the proportions of ribosome states based on diffusion, this assumption remains speculative.

    5. Author response:

      Public Reviews:

      Reviewer #1 (Public review):

      Summary:

      This paper is an elegant, mostly observational work, detailing observations that polysome accumulation appears to drive nucleoid splitting and segregation. Overall I think this is an insightful work with solid observations.

      Thank you for your appreciation and positive comments. In our view, an appealing aspect of this proposed biophysical mechanism for nucleoid segregation is its self-organizing nature and its ability to intrinsically couple nucleoid segregation to biomass growth, regardless of nutrient conditions.

      Strengths:

      The strengths of this paper are the careful and rigorous observational work that leads to their hypothesis. They find the accumulation of polysomes correlates with nucleoid splitting, and that the nucleoid segregation occurring right after splitting correlates with polysome segregation. These correlations are also backed up by other observations:

      (1) Faster polysome accumulation and DNA segregation at faster growth rates.

      (2) Polysome distribution negatively correlating with DNA positioning near asymmetric nucleoids.

      (3) Polysomes form in regions inaccessible to similarly sized particles.

      These above points are observational, I have no comments on these observations leading to their hypothesis.

      Thank you!

      Weaknesses:

      It is hard to state weaknesses in any of the observational findings, and furthermore, their two tests of causality, while not being completely definitive, are likely the best one could do to examine this interesting phenomenon.

      It is indeed difficult to prove causality in a definitive manner when the proposed coupling mechanism between nucleoid segregation and gene expression is self-organizing, i.e., does not involve a dedicated regulatory molecule (e.g., a protein, RNA, metabolite) that we could have depleted through genetic engineering to establish causality. We are grateful to the reviewer for recognizing that our two causality tests are the best that can be done in this context.

      Points to consider / address:

      Notably, demonstrating causality here is very difficult (given the coupling between transcription, growth, and many other processes) but an important part of the paper. They do two experiments toward demonstrating causality that help bolster - but not prove - their hypothesis. These experiments have minor caveats, my first two points.

      (1) First, "Blocking transcription (with rifampicin) should instantly reduce the rate of polysome production to zero, causing an immediate arrest of nucleoid segregation". Here they show that adding rifampicin does indeed lead to polysome loss and an immediate halting of segregation - data that does fit their model. This is not definitive proof of causation, as rifampicin also (a) stops cell growth, and (b) stops the translation of secreted proteins. Neither of these two possibilities is ruled out fully.

      That’s correct; cell growth also stops when gene expression is inhibited, which is consistent with our model in which gene expression within the nucleoid promotes nucleoid segregation and biomass growth (i.e., cell growth), inherently coupling these two processes. This said, we understand the reviewer’s point: the rifampicin experiment doesn’t exclude the possibility that protein secretion and cell growth drive nucleoid segregation. We are assuming that the reviewer is envisioning an alternative model in which sister nucleoids would move apart because they would be attached to the membrane through coupled transcription-translation-protein secretion (transertion) and the membrane would expand between the separating nucleoids, similar to the model proposed by Jacob et al in 1963 (doi:10.1101/SQB.1963.028.01.048). There are several observations arguing against this cell elongation/transertion model.

      (1) For this alternative mechanism to work, membrane growth must be localized at the middle of the splitting nucleoids (i.e., midcell position for slow growth and ¼ and ¾ cell positions for fast growth) to create a directional motion. To our knowledge, there is no evidence of such localized membrane incorporation. Furthermore, even if membrane growth was localized at the right places, the fluidity of the cytoplasmic membrane (PMID: 6996724, 20159151, 24735432, 27705775) would be problematic. To circumvent the membrane fluidity issue, one could potentially evoke an additional connection to the rigid peptidoglycan, but then again, peptidoglycan growth would have to be localized at the middle of the splitting nucleoid. However, peptidoglycan growth is dispersed early in the cell division cycle when the nucleoid splitting happens in fast growing cells and only appears to be zonal after the onset of cell constriction (PMID: 35705811, 36097171, 2656655).

      (2) Even if we ignore the aforementioned caveats, Paul Wiggins’s group ruled out the cell elongation/transertion model by showing that the rate of cell elongation is slower than the rate of chromosome segregation (PMID: 23775792). In the revised manuscript, we wil clarify this point and provide confirmatory data showing that the cell elongation rate is indeed slower than the nucleoid segregation rate, indicating that it cannot be the main driver.

      (3) Furthermore, our correlation analysis comparing the rate of nucleoid segregation to the rate of either cell elongation or polysome accumulation argues that polysome accumulation plays a larger role than cell elongation in nucleoid segregation. These data were already shown in Figure 1H and Figure 1 – figure supplement 3 of the original manuscript but were not highlighted in this context. We will revise the text to clarify this point.

      (4) The asymmetries in nucleoid compaction that we described in our paper are predicted by our model. We do not see how they could be explained by cell growth or protein secretion.

      (5) We also show that polysome accumulation at ectopic sites (outside the nucleoid) results in correlated nucleoid dynamics, consistent with our proposed mechanism. These nucleoid dynamics cannot be explained by cell growth or protein secretion (transertion).

      (1a) As rifampicin also stops all translation, it also stops translational insertion of membrane proteins, which in many old models has been put forward as a possible driver of nucleoid segregation, and perhaps independent of growth. This should at last be mentioned in the discussion, or if there are past experiments that rule this out it would be great to note them.

      It is not clear to us how the attachment of the DNA to the cytoplasmic membrane could alone create a directional force to move the sister nucleoids. We agree that old models have proposed a role for cell elongation (providing the force) and transertion (providing the membrane tether).  Please see our response above for the evidence (from the literature and our work) against it. This was mentioned in the introduction and Results section, but we agree that this was not well explained. We will add experimental data and revise the text to clarify these points.

      (1b) They address at great length in the discussion the possibility that growth may play a role in nucleoid segregation. However, this is testable - by stopping surface growth with antibiotics. Cells should still accumulate polysomes for some time, it would be easy to see if nucleoids are still segregated, and to what extent, thereby possibly decoupling growth and polysome production. If successful, this or similar experiments would further validate their model.

      We reviewed the literature and could not find a drug that stops cell growth without stopping gene expression. Any drug that affects the membrane integrity or potential stops gene expression, which requires ATP.  However, our experiment in which we drive polysome accumulation at ectopic sites decouples polysome accumulation from cell growth. In this experiment, by redirecting most of chromosome gene expression to a single plasmid-encoded gene, we reduce the rate of cell growth but still create a large accumulation of polysomes at an ectopic location. This ectopic polysome accumulation is sufficient to affect nucleoid dynamics in a correlated fashion. In the revised manuscript, we will clarify this point and add model simulations to show that our experimental observations are predicted by our model.

      (2) In the second experiment, they express excess TagBFP2 to delocalize polysomes from midcell. Here they again see the anticorrelation of the nucleoid and the polysomes, and in some cells, it appears similar to normal (polysomes separating the nucleoid) whereas in others the nucleoid has not separated. The one concern about this data - and the differences between the "separated" and "non-separated" nuclei - is that the over-expression of TagBFP2 has a huge impact on growth, which may also have an indirect effect on DNA replication and termination in some of these cells. Could the authors demonstrate these cells contain 2 fully replicated DNA molecules that are able to segregate?

      We will perform the requested experiment.

      (3) What is not clearly stated and is needed in this paper is to explain how polysomes do (or could) "exert force" in this system to segregate the nucleoid: what a "compaction force" is by definition, and what mechanisms causes this to arise (what causes the "force") as the "compaction force" arises from new polysomes being added into the gaps between them caused by thermal motions.

      They state, "polysomes exert an effective force", and they note their model requires "steric effects (repulsion) between DNA and polysomes" for the polysomes to segregate, which makes sense. But this makes it unclear to the reader what is giving the force. As written, it is unclear if (a) these repulsions alone are making the force, or (b) is it the accumulation of new polysomes in the center by adding more "repulsive" material, the force causes the nucleoids to move. If polysomes are concentrated more between nucleoids, and the polysome concentration does not increase, the DNA will not be driven apart (as in the first case) However, in the second case (which seems to be their model), the addition of new material (new polysomes) into a sterically crowded space is not exerting force - it is filling in the gaps between the molecules in that region, space that needs to arise somehow (like via Brownian motion). In other words, if the polysome region is crowded with polysomes, space must be made between these polysomes for new polysomes to be inserted, and this space must be made by thermal (or ATP-driven) fluctuations of the molecules. Thus, if polysome accumulation drives the DNA segregation, it is not "exerting force", but rather the addition of new polysomes is iteratively rectifying gaps being made by Brownian motion.

      We apologize for the understandable confusion. In our picture, the polysomes and DNA (conceptually considered as small plectonemic segments) basically behave as dissolved particles. If these particles were noninteracting, they would simply mix. However, both polysomes and DNA segments are large enough to interact sterically. So as density increases, steric avoidance implies a reduced conformational entropy and thus a higher free energy per particle. We argue (based on Miangolarra et al. PNAS 2021 PMID: 34675077 and Xiang et al. Cell 2021 PMID: 34186018) that the demixing of polysomes and DNA segments occurs because DNA segments pack better with each other than they do with polysomes. This raises the free energy cost associated with DNA-polysome interactions compared to DNA-DNA interactions.  We model this effect by introducing a term in the free energy χ_np, which refer to as a repulsion between DNA and polysomes, though as explained above it arises from entropic effects. At realistic cellular densities of DNA and polysomes this repulsive interaction is strong enough to cause the DNA and polysomes to phase separate.

      This same density-dependent free energy that causes phase separation can also give rise to forces, just in the way that a higher pressure on one side of a wall can give rise to a net force on the wall. Indeed, the “compaction force” we refer to is fundamentally an osmotic pressure difference. At some stages during nucleoid segregation, the region of the cell between nucleoids has a higher polysome concentration, and therefore a higher osmotic pressure, than the regions near the poles. This results in a net poleward force on the sister nucleoids that drives their migration toward the poles. This migration continues until the osmotic pressure equilibrates. Therefore, both phase separation (due to the steric repulsion described above) and nonequilibrium polysome production and degradation (which creates the initial accumulation of polysomes around midcell) are essential ingredients for nucleoid segregation.

      This will be clarified in the revised text, with the support of additional simulation results.

      The authors use polysome accumulation and phase separation to describe what is driving nucleoid segregation. Both terms are accurate, but it might help the less physically inclined reader to have one term, or have what each of these means explicitly defined at the start. I say this most especially in terms of "phase separation", as the currently huge momentum toward liquid-liquid interactions in biology causes the phrase "phase separation" to often evoke a number of wider (and less defined) phenomena and ideas that may not apply here. Thus, a simple clear definition at the start might help some readers.

      Phase separation means that the DNA-polysome steric repulsion is strong enough to drive their demixing, which creates a compact nucleoid. As mentioned in a previous point, this effect is captured in the free energy by the χ_np term, which is an effective repulsion between DNA and polysomes, though as explained above it arises from entropic effects.

      In the revised manuscript, we will illustrate this with our theoretical model by initializing a cell with a diffuse nucleoid and low polysome concentration. For the sake of simplicity, we assume that the cell does not elongate. We observe that the DNA-polysome steric repulsion is sufficient to compact the nucleoid and place it at mid-cell.

      (4) Line 478. "Altogether, these results support the notion that ectopic polysome accumulation drives nucleoid dynamics". Is this right? Should it not read "results support the notion that ectopic polysome accumulation inhibits/redirects nucleoid dynamics"?

      We think that this is correct; the ectopic polysome accumulation drives nucleoid dynamics. In our theoretical model, we can introduce polysome production at fixed sources to mimic the experiments where ectopic polysome production is achieved by high plasmid expression (Fig. 6). The model is able to recapitulate the two main phenotypes observed in experiments. These new simulation results will be added to the revised manuscript.

      (5) It would be helpful to clarify what happens as the RplA-GFP signal decreases at midcell in Figure 1- is the signal then increasing in the less "dense" parts of the cell? That is, (a) are the polysomes at midcell redistributing throughout the cell? (b) is the total concentration of polysomes in the entire cell increasing over time?

      It is a redistribution—the RplA-GFP signal remains constant in concentration from cell birth to division (Figure 1 – Figure Supplement 1E). This will be clarified in the revised text.

      (6) Line 154. "Cell constriction contributed to the apparent depletion of ribosomal signal from the mid-cell region at the end of the cell division cycle (Figure 1B-C and Movie S1)" - It would be helpful if when cell constriction began and ended was indicated in Figures 1B and C.

      Good idea. We will add markers to indicate the start of cell constriction. We will also indicate that cell birth and division correspond to the first and last images/timepoint in Fig. 1B and C, respectively.

      (7) In Figure 7 they demonstrate that radial confinement is needed for longitudinal nucleoid segregation. It should be noted (and cited) that past experiments of Bacillus l-forms in microfluidic channels showed a clear requirement role for rod shape (and a given width) in the positing and the spacing of the nucleoids.

      Wu et al, Nature Communications, 2020 . "Geometric principles underlying the proliferation of a model cell system" https://dx.doi.org/10.1038/s41467-020-17988-7

      Good point. We will add this reference. Thank you.

      (8) "The correlated variability in polysome and nucleoid patterning across cells suggests that the size of the polysome-depleted spaces helps determine where the chromosomal DNA is most concentrated along the cell length. This patterning is likely reinforced through the displacement of the polysomes away from the DNA dense region"

      It should be noted this likely functions not just in one direction (polysomes dictating DNA location), but also in the reverse - as the footprint of compacted DNA should also exclude (and thus affect) the location of polysomes

      We agree that the effects could go both ways at this early stage of the story. We will revise the text accordingly.  

      (9) Line 159. Rifampicin is a transcription inhibitor that causes polysome depletion over time. This indicates that all ribosomal enrichments consist of polysomes and therefore will be referred to as polysome accumulations hereafter". Here and throughout this paper they use the term polysome, but cells also have monosomes (and 2 somes, etc). Rifampicin stops the assembly of all of these, and thus the loss of localization could occur from both. Thus, is it accurate to state that all transcription events occur in polysomes? Or are they grouping all of the n-somes into one group?

      In the discussion, we noted that our term “polysomes” also includes monosomes for simplicity, but we agree that the term should have been defined much earlier. This will be done in the revised manuscript.

      Thank you for the valuable comments and suggestions!

      Reviewer #2 (Public review):

      Summary:

      The authors perform a remarkably comprehensive, rigorous, and extensive investigation into the spatiotemporal dynamics between ribosomal accumulation, nucleoid segregation, and cell division. Using detailed experimental characterization and rigorous physical models, they offer a compelling argument that nucleoid segregation rates are determined at least in part by the accumulation of ribosomes in the center of the cell, exerting a steric force to drive nucleoid segregation prior to cell division. This evolutionarily ingenious mechanism means cells can rely on ribosomal biogenesis as the sole determinant for the growth rate and cell division rate, avoiding the need for two separate 'sensors,' which would require careful coupling.

      Terrific summary! Thank you for your positive assessment.

      Strengths:

      In terms of strengths; the paper is very well written, the data are of extremely high quality, and the work is of fundamental importance to the field of cell growth and division. This is an important and innovative discovery enabled through a combination of rigorous experimental work and innovative conceptual, statistical, and physical modeling.

      Thank you!

      Weaknesses:

      In terms of weaknesses, I have three specific thoughts.

      Firstly, my biggest question (and this may or may not be a bona fide weakness) is how unambiguously the authors can be sure their ribosomal labeling is reporting on polysomes, specifically. My reading of the work is that the loss of spatial density upon rifampicin treatment is used to infer that spatial density corresponds to polysomes, yet this feels like a relatively indirect way to get at this question, given rifampicin targets RNA polymerase and not translation. It would be good if a more direct way to confirm polysome dependence were possible.

      The heterogeneity of ribosome distribution inside E. coli cells has been attributed to polysomes by many labs (PMID: 25056965, 38678067, 22624875, 31150626, 34186018, 10675340).  The attribution is also consistent with single-molecule tracking experiments showing that slow-moving ribosomes (polysomes) are excluded by the nucleoid whereas fast-diffusing ribosomes (free ribosomal subunits) are distributed throughout the cytoplasm (PMID: 25056965, 22624875).

      Furthermore, inhibition of translation initiation with kasugamycin treatment, which decreases the pool of polysomes, results in a homogenization of ribosomes and expansion of the nucleoid (see Author response image 1). This further supports the rifampicin experiments. Given that the attribution of ribosome heterogeneity to polysomes is well accepted in the field, we would prefer to not include these kasugamycin data in the revised manuscript because long-term exposure to this drug leads to nucleoid re-compaction (PMID: 25250841 and PMID: 34186018). This secondary effect may possibly be due to a dysregulated increase in synthesis of naked rRNAs (PMID: 14460744, PMID: 2114400, and PMID: 2448483) or ribosome aggregation, which we are currently investigating.

      Author response image 1.

      Effects of kasugamycin treatment on the intracellular distribution of ribosomes and nucleoids. Representative single cell (CJW7323) growing in M9gluCAAT.  Kasugamycin (3 mg/mL) was added at time = 0 min. Show is the early response (0-30 min) to the drug characterized by the homogenization of the ribosomal RplA-GFP fluorescence and the expansion of the HupA-mCherry-labeled nucleoids. For each segmented cell, the RplA-GFP and HupA-mCherry signals were normalized by the average fluorescence.

      Second, the authors invoke a phase separation model to explain the data, yet it is unclear whether there is any particular evidence supporting such a model, whether they can exclude simpler models of entanglement/local diffusion (and/or perhaps this is what is meant by phase separation?) and it's not clear if claiming phase separation offers any additional insight/predictive power/utility. I am OK with this being proposed as a hypothesis/idea/working model, and I agree the model is consistent with the data, BUT I also feel other models are consistent with the data. I also very much do not think that this specific aspect of the paper has any bearing on the paper's impact and importance.

      We appreciate the reviewer’s comment, but the output of our reaction-diffusion model is a bona fide phase separation (spinodal decomposition). So, we feel that we need to use the term when reporting the modeling results. Inside the cell, the situation is more complicated. As the reviewer points out, there likely are entanglements (not considered in our model) and other important factors (please see our discussion on the model limitations). This said, we will revise our text to clarify our terms and proposed mechanism.

      Finally, the writing and the figures are of extremely high quality, but the sheer volume of data here is potentially overwhelming. I wonder if there is any way for the authors to consider stripping down the text/figures to streamline things a bit? I also think it would be useful to include visually consistent schematics of the question/hypothesis/idea each of the figures is addressing to help keep readers on the same page as to what is going on in each figure. Again, there was no figure or section I felt was particularly unclear, but the sheer volume of text/data made reading this quite the mental endurance sport! I am completely guilty of this myself, so I don't think I have any super strong suggestions for how to fix this, but just something to consider.

      We agree that there is a lot to digest. We will add schematics and a didactic simulation. We will also try to streamline the text.

      Reviewer #3 (Public review):

      Summary:

      Papagiannakis et al. present a detailed study exploring the relationship between DNA/polysome phase separation and nucleoid segregation in Escherichia coli. Using a combination of experiments and modelling, the authors aim to link physical principles with biological processes to better understand nucleoid organisation and segregation during cell growth.

      Strengths:

      The authors have conducted a large number of experiments under different growth conditions and physiological perturbations (using antibiotics) to analyse the biophysical factors underlying the spatial organisation of nucleoids within growing E. coli cells. A simple model of ribosome-nucleoid segregation has been developed to explain the observations.

      Weaknesses:

      While the study addresses an important topic, several aspects of the modelling, assumptions, and claims warrant further consideration.

      Thank you for your feedback. Please see below for a response to each concern. 

      Major Concerns:

      Oversimplification of Modelling Assumptions:

      The model simplifies nucleoid organisation by focusing on the axial (long-axis) dimension of the cell while neglecting the radial dimension (cell width). While this approach simplifies the model, it fails to explain key experimental observations, such as:

      (1) Inconsistencies with Experimental Evidence:

      The simplified model presented in this study predicts that translation-inhibiting drugs like chloramphenicol would maintain separated nucleoids due to increased polysome fractions. However, experimental evidence shows the opposite-separated nucleoids condense into a single lobe post-treatment (Bakshi et al 2014), indicating limitations in the model's assumptions/predictions. For the nucleoids to coalesce into a single lobe, polysomes must cross the nucleoid zones via the radial shells around the nucleoid lobes.

      We do not think that the results from chloramphenicol-treated cells are inconsistent with our model. Our proposed mechanism predicts that nucleoids will condense in the presence of chloramphenicol, consistent with experiments. It also predicts that nucleoids that were still relatively close at the time of chloramphenicol treatment could fuse if they eventually touched through diffusion (thermal fluctuation) to reduce their interaction with the polysomes and minimize their conformational energy. Fusion is, however, not expected for well-separated nucleoids since their diffusion is slow in the crowded cytoplasm. This is consistent with our experimental observations: In the presence of a growth-inhibitory concentration of chloramphenicol (70 μg/mL), nucleoids in relatively close proximity can fuse, but well-separated nucleoids condense and do not fuse. Since the growth rate inhibition is not immediate upon chloramphenicol treatment, many cells with well-separated condensed nucleoids divide during the first hour. As a result, the non-fusion phenotype is more obvious in non-dividing cells, achieved by pre-treating cells with the cell division inhibitor cephalexin (50μg/mL). In these polyploid elongated cells, well-separated nucleoids condensed but did not fuse, not even after an hour in the presence of chloramphenicol (as illustrated in Author response image 2).

      In Bakshi et al, 2014, nucleoid fusion was shown for a single cell in which the sister nucleoids were relatively close to each other at the time of chloramphenicol treatment. Population statistics were provided for the relative length and width of the nucleoids, but not for the fusion events. So, it is unclear whether the illustrated fusion was universal or not. Also, we note that Bakshi et al (2014) used a chloramphenicol concentration of 300 μg/mL, which is 20-fold higher than the minimal inhibitory concentration for growth, compared to 70 μg/mL in our experiments.

      Author response image 2.

      Effects of chloramphenicol treatment on the intracellular distribution of ribosomes and nucleoids in non-dividing cells. Exponentially growing cells (M9glyCAAT at 30°C) were pre-treated with cephalexin for one hour before being spotted on an 1% agarose pad for time-lapse imaging. The agarose pad contained M9glyCAAT, cephalexin, and chloramphenicol.  (A) Phase contrast, RplA-GFP fluorescence and HupA-mCherry fluorescence images of a representative single cell. Three timepoints are shown, including the first image after spotting on the agarose pad (at 0 min), 30 minutes and one hour of chloramphenicol treatment. (B) One-dimensional profiles of the ribosomal (RplA-GFP) and nucleoid (HupA-mCherry) fluorescence from the cells shown in panel A. These intensity profiles correspond to the average fluorescence along the medial axis of the cell considering a 6-pixel region (0.4 μm) centered on the central line of the cell. The fluorescence intensity is plotted along the relative cell length, scaled from 0 to 100% between the two poles, illustrating the relative nucleoid length (L<sub>DNA</sub>/L<sub>cell</sub>) that was plotted by Bakshi et al in 2014 (PMID: 25250841).

      (2) The peripheral localisation of nucleoids observed after A22 treatment in this study and others (e.g., Japaridze et al., 2020; Wu et al., 2019), which conflicts with the model's assumptions and predictions. The assumption of radial confinement would predict nucleoids to fill up the volume or ribosomes to go near the cell wall, not the nucleoid, as seen in the data.

      The reviewer makes a good point that DNA attachment to the membrane through transertion likely contributes to the nucleoid being peripherally localized in A22 cells. We will revise the text to add this point. However, we do not think that this contradicts the proposed nucleoid segregation mechanism based on phase separation and out-of-equilibrium dynamics described in our model. On the contrary, by attaching the nucleoid to the cytoplasmic membrane along the cell width, transertion might help reduce the diffusion and thus exchange of polysomes across nucleoids. We will revise the text to discuss transertion over radial confinement.

      (3) The radial compaction of the nucleoid upon rifampicin or chloramphenicol treatment, as reported by Bakshi et al. (2014) and Spahn et al. (2023), also contradicts the model's predictions. This is not expected if the nucleoid is already radially confined.

      We originally evoked radial confinement to explain the observation that polysome accumulations do not equilibrate between DNA-free regions. We agree that transertion is an alternative explanation. Thank you for bringing it to our attention. However, please note that this does not contradict the model. In our view, it actually supports the 1D model by providing a reasonable explanation for the slow exchange of polysomes across DNA-free regions. The attachment of the nucleoid to the membrane along the cell width may act as diffusion barrier. We will revise the text and the title of the manuscript accordingly.

      (4) Radial Distribution of Nucleoid and Ribosomal Shell:

      The study does not account for well-documented features such as the membrane attachment of chromosomes and the ribosomal shell surrounding the nucleoid, observed in super-resolution studies (Bakshi et al., 2012; Sanamrad et al., 2014). These features are critical for understanding nucleoid dynamics, particularly under conditions of transcription-translation coupling or drug-induced detachment. Work by Yongren et al. (2014) has also shown that the radial organisation of the nucleoid is highly sensitive to growth and the multifork nature of DNA replication in bacteria.

      We will discuss the membrane attachment. Please see the previous response.

      The omission of organisation in the radial dimension and the entropic effects it entails, such as ribosome localisation near the membrane and nucleoid centralisation in expanded cells, undermines the model's explanatory power and predictive ability. Some observations have been previously explained by the membrane attachment of nucleoids (a hypothesis proposed by Rabinovitch et al., 2003, and supported by experiments from Bakshi et al., 2014, and recent super-resolution measurements by Spahn et al.).

      We agree—we will add a discussion about membrane attachment in the radial dimension. See previous responses.

      Ignoring the radial dimension and membrane attachment of nucleoid (which might coordinate cell growth with nucleoid expansion and segregation) presents a simplistic but potentially misleading picture of the underlying factors.

      As mentioned above, we will discuss membrane attachment in the revised manuscript.

      This reviewer suggests that the authors consider an alternative mechanism, supported by strong experimental evidence, as a potential explanation for the observed phenomena:

      Nucleoids may transiently attach to the cell membrane, possibly through transertion, allowing for coordinated increases in nucleoid volume and length alongside cell growth and DNA replication. Polysomes likely occupy cellular spaces devoid of the nucleoid, contributing to nucleoid compaction due to mutual exclusion effects. After the nucleoids separate following ter separation, axial expansion of the cell membrane could lead to their spatial separation.

      This “membrane attachment/cell elongation” model is reminiscent to the hypothesis proposed by Jacob et al in 1963 (doi:10.1101/SQB.1963.028.01.048). There are several lines of evidence arguing against it as the major driver of nucleoid segregation:

      (Below is a slightly modified version of our response to a comment from Reviewer 1—see page 3)

      (1) For this alternative model to work, axial membrane expansion (i.e., cell elongation) would have to be localized at the middle of the splitting nucleoids (i.e., midcell position for slow growth and ¼ and ¾ cell positions for fast growth) to create a directional motion. To our knowledge, there is no evidence of such localized membrane incorporation.  Furthermore, even if membrane growth was localized at the right places, the fluidity of the cytoplasmic membrane (PMID: 6996724, 20159151, 24735432, 27705775) would be problematic. To go around this fluidity issue, one could potentially evoke a potential connection to the rigid peptidoglycan, but then again, peptidoglycan growth would have to be localized at the middle of the splitting nucleoid to “push” the sister nucleoid apart from each other. However, peptidoglycan growth is dispersed prior to cell constriction (PMID: 35705811, 36097171, 2656655).

      (2) Even if we ignore the aforementioned caveats, Paul Wiggins’s group ruled out the cell elongation/transertion model by showing that the rate of cell elongation is slower than the rate of chromosome segregation (PMID: 23775792). In the revised manuscript, we will provide additional data showing that the cell elongation rate is indeed slower than the nucleoid segregation rate.

      (3) Furthermore, our correlation analysis comparing the rate of nucleoid segregation to the rate of either cell elongation or polysome accumulation argues that polysome accumulation plays a larger role than cell elongation in nucleoid segregation. These data were already shown in the original manuscript (Figure 1I and Figure 1 – figure supplement 3) but were not highlighted in this context. We will revise the text to clarify this point.

      (4) The membrane attachment/cell elongation model does not explain the nucleoid asymmetries described in our paper (Figure 3), whereas they can be recapitulated by our model.

      (5) The cell elongation/transertion model cannot predict the aberrant nucleoid dynamics observed when chromosomal expression is largely redirected to plasmid expression. In the revised manuscript, we will add simulation results showing that these nucleoid dynamics are predicted by our model.

      In line of these arguments, we do not believe that a mechanism based on membrane attachment and cell elongation is the major driver of nucleoid segregations. However, we do believe that it may play a complementary role (see “Nucleoid segregation likely involves multiple factors” in the Discussion). We will revise this section to clarify our thoughts and mention the potential role of transertion.

      Incorporating this perspective into the discussion or future iterations of the model may provide a more comprehensive framework that aligns with the experimental observations in this study and previous work.

      As noted above, we will revise the text to mention about transertion.

      Simplification of Ribosome States:

      Combining monomeric and translating ribosomes into a single 'polysome' category may overlook spatial variations in these states, particularly during ribosome accumulation at the mid-cell. Without validating uniform mRNA distribution or conducting experimental controls such as FRAP or single-molecule measurements to estimate the proportions of ribosome states based on diffusion, this assumption remains speculative.

      Indeed, for simplicity, we adopt an average description of all polysomes with an average diffusion coefficient and interaction parameters, which is sufficient for capturing the fundamental mechanism underlying nucleoid segregation. To illustrate that considering multiple polysome species does not change the physical picture, we consider an extension of our model, which contains three polysome species, each with a different diffusion coefficient (D<SUB>P</SUB> = 0.018, 0.023, or 0.028 μm<sup>2</sup>/s), reflecting that polysomes with more ribosomes will have a lower diffusion coefficient. Simulation of this model reveals that the different polysome species have essentially the same concentration distribution, suggesting that the average description in our minimal model is sufficient for our purposes. We will present these new simulation results in the revised manuscript.

    1. eLife Assessment

      This study provides valuable scRNA-seq and scATAC-seq data for testicular tissues from patients with spermatogenesis disorders. By examining the transcriptomic and epigenetic changes in Sertoli cells, the authors uncovered key regulatory mechanisms underlying male infertility and identified potential therapeutic targets. While some of the cellular profiling results are convincing, the analyses for differential profiling of NOA cases and epigenomics data remain incomplete.

    2. Reviewer #1 (Public review):

      Summary:

      In this study, Wang and colleagues generate single-cell transcriptome and chromatin accessibility data from testicular tissues of two OA and three NOA cases. The authors analyze this dataset to identify novel cellular populations, marker genes, and inter-population interactions that may contribute to proper spermatogenesis. Then they propose a role of specific Sertoli cell subtypes and their interactions via Notch signaling in germ cell development. However, I remain skeptical of their central argument (also highlighted in the title) that stage-specific interactions between Sertoli and germ cells are a key component in NOA development, as my initial concerns regarding potential data misrepresentation, lack of statistical testing, and the rationale behind some of the analyses have not been sufficiently addressed.

      (1) As noted in my previous comments, the analysis of Sertoli cell subtypes is potentially misleading and lacks proper statistical support. The authors claim a significant loss of Sertoli subpopulations in NOA cases, and provide the absolute number of cells in Figure 6B. However, this observation could easily be driven by the total number of cells captured during the experiment and the anatomical location of the specimens. There is no statistical basis to make the claim that this loss is significant. Furthermore, the same analysis should be performed on scATAC-seq cells and presented alongside.

      (2) As pointed out in my initial concerns, some parts of the analyses require additional explanation to clarify their logical flow. For example, the logic of using between-sample correlations to assess colocalization of Sertoli and germ cells is lost on me. How can this be used to infer the important role of specific Sertoli cell populations in spermatogenesis, other than the fact that some of the genes are more co-expressed in the sub-populations? And how is this related to the claim that these cell populations are actually co-localized in the tissue? The authors then dedicate nearly a page describing the pathways enriched in Sertoli and germ cells, but the relevance is unclear, and the argument that these subtypes are functionally related is not convincing enough.

      (3) The statement regarding Notch signaling as a critical component in Sertoli and germ cell interaction is not supported by actual evidence. The inference based on CellphoneDB and an epigenome snapshot that shows not much difference are insufficient to justify this claim.

      (4) The manuscript is overly wordy and descriptive, making it difficult to read and understand the points. The main text needs to be more concise and on point, with unnecessary details removed to sharpen the key points. Non-essential results (e.g. Figure S10 and S11) unrelated to the main argument should be removed.

    3. Reviewer #2 (Public review):

      Summary:

      Shimin Wang et al. investigated the role of Sertoli cells in mediating spermatogenesis disorders in non-obstructive azoospermia (NOA) through stage-specific communications. The authors utilized scRNA-seq and scATAC-seq to analyze the molecular and epigenetic profiles of germ cells and Sertoli cells at different stages of spermatogenesis.

      Strengths:

      By understanding the gene expression patterns and chromatin accessibility changes in Sertoli cells, the authors sought to uncover key regulatory mechanisms underlying male infertility and identify potential targets for therapeutic interventions. They emphasized that the absence of the SC3 subtype would be a major factor contributing to NOA.

      Comments on revisions:

      The authors have addressed my concerns. I have no further comments.

    4. Reviewer #3 (Public review):

      Summary:

      This study profiled the single-cell transcriptome of human spermatogenesis and provided many potentials molecular markers for developing testicular puncture specific marker kits for NOA patients.

      Strengths:

      Perform single-cell RNA sequencing (scRNA-seq) and single-cell assay for transposase-accessible chromatin sequencing (scATAC-seq) on testicular tissues from two OA patients and three NOA patients

      Weaknesses:

      Most results are analytical and lack specific experiments to support these analytical results and hypotheses.

      Comments on revisions:

      In the revised version of the manuscript, the authors made some effort to revise their manuscript according to reviewers' comments and addressed the problems that I had raised before.

      I have no other serious criticisms regarding the revised manuscript.

    5. Author response:

      The following is the authors’ response to the original reviews.

      Public Reviews:

      Reviewer #1 (Public Review):

      Summary:

      The manuscript is dedicated heavily to cell type mapping and identification of sub-type markers in the human testis but does not present enough results from cross-investigation between NOA cases versus control. Their findings are mostly based on transcriptome and the authors do not make enough use of the scATAC-seq data in their analyses as they put forward in the title. Overall, the authors should do more to include the differential profile of NOA cases at the molecular level - specific gene expression, chromatin accessibility, TF binding, pathway, and signaling that are perturbed in NOA patients that may be associated with azoospermia.

      Strengths:

      (1) The establishment of single-cell data (both RNA and ATAC) from the human testicular tissues is noteworthy.

      (2) The manuscript includes extensive mapping of sub-cell populations with some claimed as novel, and reports marker gene expression.

      (3) The authors present inter-cellular cross-talks in human testicular tissues that may be important in adequate sperm cell differentiation.

      Weaknesses:

      (1) A low sample size (2 OA and 3 NOA cases). There are no control samples from healthy individuals.

      Thank you for your comments. We recognize that the small sample size in this study somewhat limits its generalizability. However, in transcriptomic research, limited sample sizes are a common issue due to the complexities involved in acquiring samples, particularly in studies about the reproductive system. Healthy testicular tissue samples are difficult to obtain, and studies (doi: 10.18632/aging.203675) have used obstructive azoospermia as a control group in which spermatogenesis and development are normal.

      (2) Their argument about interactions between germ and Sertoli cells is not based on statistical testing.

      Thank you for your comments. Due to limited funding, we have not yet fully and deeply conducted validation experiments, but we plan to carry out related experiments in the later stage. We hope that the publication of this study will help to obtain more financial support to further investigate the interactions between germ cells and Sertoli cells.

      (3) Rationale/logic of the study. This study, in its present form, seems to be more about the role of sub-Sertoli population interactions in sperm cell development and does not provide enough insights about NOA.

      Thank you for your comments. In Figure 6, we conducted an in-depth analysis and comparison of the differences between the Sertoli cell subtypes and the germ cell subtypes involved in spermatogenesis in the OA and NOA groups. The results revealed that in the NOA group, especially in the NOA3 group, which has a lower sperm count compared to NOA2 and NOA1, there is a significant loss of Sertoli cell subtypes including SC3, SC4, SC5, SC6, and SC8. The NOA1 group, with a sperm count close to that of the OA group, also had a Sertoli cell profile similar to the OA group. The NOA2 group, with a sperm count between that of NOA1 and NOA3, also exhibited an intermediate profile of Sertoli cell subtypes. Therefore, we suggest that change in Sertoli cell subtypes is a key factor affecting sperm count, rather than just the total number of Sertoli cells. We believe that through these analyses, we can provide in-depth insights into NOA, and we hope that the publication of this study will help obtain more funding support to further validate and expand on these findings.

      (4) The authors do not make full use of the scATAC-seq data.

      Thank you for your comments.We have added analysis of the scATAC-seq data and shown in the revised manuscript.

      Reviewer #2 (Public Review):

      Summary:

      Shimin Wang et al. investigated the role of Sertoli cells in mediating spermatogenesis disorders in non-obstructive azoospermia (NOA) through stage-specific communications. The authors utilized scRNA-seq and scATAC-seq to analyze the molecular and epigenetic profiles of germ cells and Sertoli cells at different stages of spermatogenesis.

      Strengths:

      By understanding the gene expression patterns and chromatin accessibility changes in Sertoli cells, the authors sought to uncover key regulatory mechanisms underlying male infertility and identify potential targets for therapeutic interventions. They emphasized that the absence of the SC3 subtype would be a major factor contributing to NOA.

      Weaknesses:

      Although the authors used cutting-edge techniques to support their arguments, it is difficult to find conceptual and scientific advances compared to Zeng S et al.'s paper (Zeng S, Chen L, Liu X, Tang H, Wu H, and Liu C (2023) Single-cell multi-omics analysis reveals dysfunctional Wnt signaling of spermatogonia in non-obstructive azoospermia. Front. Endocrinol. 14:1138386.). Overall, the authors need to improve their manuscript to demonstrate the novelty of their findings in a more logical way.

      Thank you for your detailed review of our work. We greatly appreciate your feedback and have made revisions to our manuscript accordingly.

      Regarding the novelty of our research, we believe our study offers conceptual and scientific advances in several ways:

      We have systematically revealed the stage-specific roles of Sertoli cell subtypes in different stages of spermatogenesis, particularly emphasizing the crucial role of the SC3 subtype in non-obstructive azoospermia (NOA). Additionally, we identified that other Sertoli cell subtypes (SC1, SC2, SC3...SC8, etc.) also collaborate in a stage-specific manner with different subpopulations of spermatogenic cells (SSC0, SSC1/SSC2/Diffed, Pa...SPT3). These findings provide new insights into the understanding of spermatogenesis disorders.

      Compared to the study by Zeng S et al., our research not only focuses on the functional alterations in Sertoli cells but also comprehensively analyzes the interaction patterns between Sertoli cells and spermatogenic cells using scRNA-seq and scATAC-seq technologies. We uncovered several novel regulatory networks that could serve as potential targets for the diagnosis and treatment of NOA.

      We sincerely appreciate your constructive comments and will continue to explore this area further, aiming to make a more significant contribution to the understanding of NOA mechanisms.

      Reviewer #3 (Public Review):

      Summary:

      This study profiled the single-cell transcriptome of human spermatogenesis and provided many potential molecular markers for developing testicular puncture-specific marker kits for NOA patients.

      Strengths:

      Perform single-cell RNA sequencing (scRNA-seq) and single-cell assay for transposase-accessible chromatin sequencing (scATAC-seq) on testicular tissues from two OA patients and three NOA patients.

      Weaknesses:

      Most results are analytical and lack specific experiments to support these analytical results and hypotheses.

      Thank you for your thorough review of our work. We highly value your feedback and have made revisions to our manuscript accordingly. Indeed, we have conducted immunofluorescence (IF) experiments to validate the data obtained from single-cell sequencing and have expanded the sample size to enhance the reliability of our results. To better present these validation experiments, we have reorganized and renamed the sample information, making it easier for you to understand which samples were used in the specific experiments. Following the publication of this paper, we plan to secure additional funding to deepen our research, particularly in the area of experimental validation. We sincerely appreciate your support and insightful suggestions, which have greatly helped guide our future research directions.

      Reviewer #1 (Recommendations For The Authors):

      (1) The authors should include results from cross-investigation comparing NOA/OA patients versus controls.

      Thank you for your comments. In this study, OA was the control group. Healthy testicular tissue samples are difficult to obtain, and studies (doi: 10.18632/aging.203675) have used OA as a control group in which spermatogenesis and development are normal.

      (2) In Table S1, the authors should also include the metric for scATAC-seq, and do more to show the findings the authors obtained in RNA is replicated with chromatin accessibility.

      Thank you for your comments. We have added Table S2, which includes the metric for scATAC-seq.

      (3) A single sample from each OA and NOA group may not be enough to confirm colocalization. The authors should include results from all available samples and use quantitative measures.

      Thank you for your comments. I apologize that the sample size in this study was less than three and we could not conduct quantitative analysis. We will increase the sample size and conduct corresponding experiments in subsequent research.

      (4) The Methods section does not include enough description to follow how the analyses were carried out, and is missing information on some of the key procedures such as velocity and cell cycle analyses.

      Thank you for your comments. The method about velocity and cell cycle analyses was added in the revised manuscript. The description is as follows:

      “Velocity analysis

      RNA velocity analysis was conducted using scVelo's (version 0.2.1) generalized dynamical model. The spliced and unspliced mRNA was quantified by Velocity (version 0.17.17).”

      “Cell cycle analysis

      To quantify the cell cycle phases for individual cell, we employed the CellCycleScoring function from the Seurat package. This function computes cell cycle scores using established marker genes for cell cycle phases as described in a previous study by Nestorowa et al. (2016). Cells showing a strong expression of G2/M-phase or S-phase markers were designated as G2/M-phase or S-phase cells, respectively. Cells that did not exhibit significant expression of markers from either category were classified as G1-phase cells.”

      (5) For the purpose of transparency, the authors should upload codes used for analyses so that each figure can be reproduced. All raw and processed data should be made publicly available.

      Thank you for your comments. We have deposited scRNA-seq and scATAC-seq data in NCBI. ScRNA-seq data have been deposited in the NCBI Gene Expression Omnibus with the accession number GSE202647, and scATAC-seq data have been deposited in the NCBI database with the accession number PRJNA1177103.

      Reviewer #2 (Recommendations For The Authors):

      The detailed points the authors need to improve are attached below.

      The results presented in the study have several weaknesses:

      In Figure 1A, it's required to show HE staining results of all patients who underwent single-cell analysis were provided.

      Thank you very much for your valuable suggestions. In Figure 1, we present the HE staining results paired with the single-cell data, covering all patients involved in the single-cell analysis.

      - Saying "identification of novel potential molecular markers for distinct cell types" seems unsupported by the data.

      Thank you for your comments. I'm sorry for the inaccuracy of my description. We have revised this sentence. The description is as follows: These findings indicate that the scRNA-seq data from this study can serve for cellular classification.

      - The methods suggest an integrated analysis of scRNA-seq and scATAC-seq, but from the figures, it seems like separate analyses were performed. It's necessary to have data showing the integrated analysis.

      Thank you for your comments. We have added an integrated analysis of scRNA-seq and scATAC-seq. The results were shown in Figure S2.

      Figure 2 does not seem to well cover the diversity of germ cell subtypes. The main content appears to be about the differentiation process, and it seems more focused on SSCs (stem cell types), but the intended message is not clearly conveyed.

      Thank you for your comments. Figure S1 revealed the diversity of germ cell subtypes. The second part of the results described the integrated findings from Figures 2 and S1.

      - In Figure 2B, pseudotime could be shown, and I wonder if the pseudotime in this analysis shows a similar pattern as in Figure 2D.

      Thank you for your comments. Figure 2B revealed the pseudotime analysis of 12 germ cell subpopulation. Figure 2D revealed RNA velocity of 12 germ cell subpopulation. The two methods are both used for cell trajectory analysis. The pseudotime in Figure 2B showed a similar pattern as in Figure 2D.

      - While staining occurs within one tissue, saying they are co-expressed seems inaccurate as the staining locations are clearly distinct. For example, the staining patterns of A2M and DDX4 (a classical marker) are quite different, so it's hard to claim A2M as a new potential marker just because it's expressed. Also, TSSK6 was separately described as having a similar expression pattern to DDX4, but from the IF results, it doesn't seem similar.

      Thank you for your comments. We have revised the Figure.

      - It was described that A2M (expressed in SSC0-1), and ASB9 (expressed in SSC2) have open promoter sites in SSC0, SSC2, and Diffing_SPG, but it doesn't seem like they are only open in the promoters of those cell types. For example, there doesn't seem to be a peak in Diffing for either gene. The promoter region of the tracks is not very clear, so overall figure modification seems necessary.

      Thank you for your comments. We have revised the Figure.

      - The ATAC signal scale for each genomic region should be included, and clear markings for the TSS location and direction of the genes are needed.

      Thank you for your comments. We have revised the figure and shown in the revised manuscript.

      Figure 3A mostly shows the SSC2 in the G2/M phase, so it seems questionable to call SSC0/1 quiescent. Also, I wonder if the expression of EOMES and GFRA1 is well distinguished in the SSC subtypes as expected.

      Thank you for your comments. We will validate in subsequent experiments whether the expression of EOMES and GFRA1 is clearly distinguished in the SSC subtypes.

      - In Figure 3C, it would be good to have labels indicating what the x and y axes represent. The figure seems complex, and the description does not seem to fully support it.

      Thank you for your comments. We have added labels indicating what the x and y axes represent in the Figure 3C. The x and y axes represent spliced and unspliced mRNA ratios, respectively.

      - While TFs are the central focus, it's disappointing that scATAC-seq was not used.

      Thank you for your comments. TFs analysis using scATAC-seq will be carried out in the future.

      Figure 4: It would be good to have a more detailed discussion of the differences between subtypes, such as through GO analysis. The track images need modification like marking the peaks of interest and focusing more on the promoter region, similar to the previous figures.

      Thank you for your comments. GO analysis results were put in Figure S5. The description is as follows:

      As shown in Figure S5, SC1 were mainly involved in cell differentiation, cell adhesion and cell communication; SC2 were involved in cell migration, and cell adhesion; SC3 were involved in spermatogenesis, and meiotic cell cycle; SC4 were involved in meiotic cell cycle, and positive regulation of stem cell proliferation; SC5 were involved in cell cycle, and cell division; SC6 were involved in obsolete oxidation−reduction process, and glutathione derivative biosynthetic process; SC7 were involved in viral transcription and translational initiation; SC8 were involved in spermatogenesis and sperm capacitation.

      In Figure 5, it would be good to have criteria for the novel Sertoli cell subtype presented. CCDC62 is presented as a representative marker for the SC8 cluster, but from Figure 4C, it seems to be quite expressed in the SC3 cluster as well. Therefore, in Figure 5E's protein-level check, it's unclear if this truly represents a novel SC8 subtype.

      Thank you for your comments. CCDC62 expression was higher in SC8 cluster than in SC3. Since some molecular markers were not commercially available in the market, CCDC62 was selected as SC8 marker for immunofluorescence verification. Immunofluorescence results showed that CCDC62 is a novel SC8 marker.

      - It might have been more meaningful to use SOX9 as a control and show that markers in the same subtype are expressed in the same location.

      Thank you for your comments. To determine PRAP1, BST2, and CCDC62 as new markers for the SC subtype, we co-stained them with SOX9 (a well-known SC marker).

      - Figures 4 and 5 could potentially be combined into one figure.

      Thank you for your comments. Since combining Figures 4 and 5 into a single image would cause the image to be unclear, two images are used to show it.

      In Figure 6, it would be good to support the results with more NOA patient data.

      Thank you for your comments. Patient clinical and laboratory characteristics has been presented in Table 1.

      - Rather than claiming the importance of SC3 based on 3 single-cell patient data, it would be better to validate using public data with SC3 signature genes (e.g., showing the correlation between germ cell and SC3 ratios).

      Thank you for your comments. I'm sorry I didn't find public data with SC3 signature genes. In the future, we will verify the importance of SC3 through in vivo and in vitro experiments.

      - 462: It seems to be referring to Figure 6G, not 6D.

      Thank you for your comments. We have revised it. The description is as follows: As shown in Figure 6G, State 1 SC3/4/5 were tended to associated with PreLep, SSC0/1/2, and Diffing and Diffed-SPG sperm cells (R > 0.72).

      In Figure 7, the spermatogenesis process is basically well-known, so it would be better to emphasize what novel content is being conveyed here. Additionally, emphasizing the importance of SC3 in the overall process based on GO results leaves room for a better approach.

      Thank you for your valuable suggestions. Regarding Figure 7, we recognize that the spermatogenesis process is well-known, and we will focus on highlighting the novel content, particularly the role and significance of the SC3 subtype in spermatogenesis disorders. As for the importance of SC3 in the overall process based on GO results, we have validated this in Figure 8 through co-staining experiments between Sertoli cells and spermatogenic cells in OA and NOA groups. The results demonstrate a significant correlation between the number of SC3-positive cells and SPT3 spermatogenic cells, particularly in the NOA5-P8 group, where both SC3 and SPT3 cell counts are notably lower than in the NOA4-P7 group. This further supports the critical role of SC3 in the spermatogenesis process. Your suggestions have prompted us to refine our data presentation and more clearly emphasize the novel aspects of our research. We will continue to strive to ensure that every part of our research contributes meaningfully to the academic community. Thank you again for your guidance.

      In Figure 8, only the contents of the IF-stained proteins are listed, which seems slightly insufficient to constitute a subsection on its own. It might have been better to conclude by emphasizing some subtypes.

      Thank you for your comments. We have combined this part of the results with other results into one section. The description is as follows:

      “Co-localization of subpopulations of Sertoli cells and germ cells

      To determine the interaction between Sertoli cells and spermatogenesis, we applied Cell-PhoneDB to infer cellular interactions according to ligand-receptor signalling database. As shown in Figure 6G, compared with other cell types, germ cells were mainly interacted with Sertoli cells. We futher performed Spearman correlation analysis to determine the relationship between Sertoli cells and germ cells. As shown in Figure 6H, State 1 SC3/4/5 were tended to be associated with PreLep, SSC0/1/2, and Diffing and Diffed-SPG sperm cells (R > 0.72). Interestingly, SC3 was significantly positively correlated with all sperm subpopulations (R > 0.5), suggesting an important role for SC3 in spermatogenesis and that SC3 is involved in the entire process of spermatogenesis. Subsequently, to understand whether the functions of germ cells and Sertoli cells correspond to each other, GO term enrichment analysis of germ cells and sertoli cells was carried out (Figure S3, S4). We found that the functions could be divided into 8 categories, namely, material energy metabolism, cell cycle activity, the final stage of sperm cell formation, chemical reaction, signal communication, cell adhesion and migration, stem cells and sex differentiation activity, and stress reaction. These different events were labeled with different colors in order to quickly capture the important events occurring in the cells at each stage. As shown in Figure S3, we discovered that SSC0/1/2 was involved in SRP-dependent cotranslational protein targeting to membrane, and cytoplasmic translation; Diffing SPG was involved in cell division and cell cycle; Diffied SPG was involved in cell cycle and RNA splicing; Pre-Leptotene was involved in cell cycle and meiotic cell cycle; Leptotene_Zygotene was involved in cell cycle and meiotic cell cycle; Pachytene was involved in cilium assembly and spermatogenesis; Diplotene was involved in spermatogenesis and cilium assembly; SPT1 was involved in cilium assembly and flagellated sperm motility; SPT2 was involved in spermatid development and flagellated sperm motility; SPT3 was involved in spermatid development and spermatogenesis. As shown in Figure S4, SC1 were mainly involved in cell differentiation, cell adhesion and cell communication; SC2 were involved in cell migration, and cell adhesion; SC3 were involved in spermatogenesis, and meiotic cell cycle; SC4 were involved in meiotic cell cycle, and positive regulation of stem cell proliferation; SC5 were involved in cell cycle, and cell division; SC6 were involved in obsolete oxidation−reduction process, and glutathione derivative biosynthetic process; SC7 were involved in viral transcription and translational initiation; SC8 were involved in spermatogenesis and sperm capacitation. The above analysis indicated that the functions of 8 Sertoli cell subtypes and 12 germ cell subtypes were closely related.

      To further verify that Sertoli cell subtypes have "stage specificity" for each stage of sperm development, we firstly performed HE staining using testicular tissues from OA3-P6, NOA4-P7 and NOA5-P8 samples. The results showed that the OA3-P6 group showed some sperm, with reduced spermatogenesis, thickened basement membranes, and a high number of sertoli cells without spermatogenic cells. The NOA4-P7 group had no sperm initially, but a few malformed sperm were observed after sampling, leading to the removal of affected seminiferous tubules. The NOA5-P8 group showed no sperm in situ (Figure 7A). Immunofluorescence staining in Figure 7B was performed using these tissues for validation. ASB9 (SSC2) was primarily expressed in a wreath-like pattern around the basement membrane of testicular tissue, particularly in the OA group, while ASB9 was barely detectable in the NOA group. SOX2 (SC2) was scattered around SSC2 (ASB9), with nuclear staining, while TF (SC1) expression was not prominent. In NOA patients, SPATS1 (SC3) expression was significantly reduced. C9orf57 (Pa) showed nuclear expression in testicular tissues, primarily extending along the basement membrane toward the spermatogenic center, and was positioned closer to the center than DDX4, suggesting its involvement in germ cell development or differentiation. BEND4, identified as a marker fo SC5, showed a developmental trajectory from the basement membrane toward the spermatogenic center. ST3GAL4 was expressed in the nucleus, forming a circular pattern around the basement membrane, similar to A2M (SSC1), though A2M was more concentrated around the outer edge of the basement membrane, creating a more distinct wreath-like arrangement. In cases of impaired spermatogenesis, this arrangement becomes disorganized and loses its original structure. SMCP (SC6) was concentrated in the midpiece region of the bright blue sperm cell tail. In the OA group, SSC1 (A2M) was sparsely arranged in a rosette pattern around the basement membrane, but in the NOA group, it appeared more scattered. SSC2 (ASB9) expression was not prominent. BST2 (SC7) was a transmembrane protein primarily localized on the cell membrane. In the OA group, A2M (SSC1) was distinctly arranged in a wreath-like pattern around the basement membrane, with expression levels significantly higher than ASB9 (SSC2). TSSK6 (SPT3) was primarily expressed in OA3-P6, while CCDC62 (SC8) was more abundantly expressed in NOA4-P7, with ASB9 (SCC2) showing minimal expression. Taken together, germ cells of a particular stage tended to co-localize with Sertoli cells of the corresponding stages. Germ cells and sertoli cells at each differentiation stage were functionally heterogeneous and stage-specific (Figure 8). This suggests that each stage of sperm development requires the assistance of sertoli cells to complete the corresponding stage of sperm development.”

      Reviewer #3 (Recommendations For The Authors):

      The authors revealed 11 germ cell subtypes and 8 Sertoli cell subtypes through single-cell analysis of two OA patients and three NOA patients. And found that the Sertoli cell SC3 subtype (marked by SPATS1) plays an important role in spermatogenesis. It also suggests that Notch1/2/3 signaling and integrins are involved in germ cell-Stotoli cell interactions. This is an interesting and useful article that at least gives us a comprehensive understanding of human spermatogenesis. It provides a powerful tool for further research on NOA. However, there are still some issues and questions that need to be addressed.<br /> (1) How to collect testicular tissue, please explain in detail. Extract which part of testicular tissue. It's better to make a schematic diagram.

      Thank you for your comments. The process is as follows: Testicular tissues were obtained from two OA patients (OA1-P1 and OA2-P2) and three NOA patients (NOA1-P3, NOA2-P4, NOA3-P5) using micro-dissection of testicular sperm extraction separately.

      (2) Whether the tissues of these patients are extracted simultaneously or separately, separated into single cells, and stored, and then single cell analysis is performed simultaneously. Please be specific.

      Thank you for your comments. The testicular tissues of these patients were extracted separately, then separated into single cells, and single cell analysis was performed simultaneously.

      (3) When performing single-cell analysis, cells from two OA patients were analyzed individually or combined. The same problem occurred in the cells of three NOA patients.

      Thank you for your comments. Cells from two OA patients and three NOA patients were analyzed individually.

      (4) Can you specifically point out the histological differences between OA and NOA in Figure 1A? This makes it easier for readers to understand the structure change between OA and NOA. Please also label representative supporting cells.

      Thank you for your comments. We have revised the description and it was shown in the revised manuscript.

      (5) The authors demonstrate that "We speculate that this lack of differentiation may be due to the intense morphological changes occurring in the sperm cells during this period, resulting in relatively minor differences in gene expression." Please provide some verification of this hypothesis? For example, use immunofluorescence staining to observe morphological changes in sperm cells.

      Thank you for your comments. Due to limited funds, we will verified this hypothesis in future studies.

      (6) The authors demonstrate that " As shown in Figure 5E, we discovered that PRAP1, BST2, and CCDC62 were co-expressed with SOX9 in testes tissues." The staining in Figure 5D is unclear, and it is difficult to explain that SOX9 is co-expressed with PRAP1 BST2 CCDC62 based on the current staining results. The staining patterns of SOX9 (green) and SOX9 (red) are also different. (SOX9 (red) appears as dots, while the background for SOX9 (green) is too dark to tell whether its staining is also in the form of dots.) In summary, increasing the clarity of the staining makes it more convincing. Alternatively, use high magnification to display these results.

      Thank you for your comments. I have redyed and updated this part of the immunofluorescence staining results. Please refer to the files named Figure 1, Figure 2, Figure 5, and Figure 8.

      (7) In Figure 8, the author emphasized the co-localization of Sertoli cells and Germ cells at corresponding stages and did a lot of staining, but it was difficult to distinguish the specific locations of co-localization, which was similar to Figure 5E. If possible, please mark specific colocalizations with arrows or use high magnification to display these results, in order to facilitate readers to better understand.

      Thank you for your comments. We have re-stained and updated this part of the data. Please refer to the immunofluorescence staining data in the updated Figure 8.

      (8) The authors emphasize that macrophages may play an important role in spermatogenesis. Therefore, adding relevant macrophage staining to observe the differences in macrophage expression between NOA and OA should better support this idea.

      Thank you for your comments. Macrophage-related experiments will be further explored in the future.

      (9) Notch1/2/3 signaling and integrin were discovered to be involved in germ cell-Sertoli cell interaction. However there are currently no concrete experiments to support this hypothesis. At least simple verification experiments are needed.

      Thank you for your comments. Due to limited funding, studies will be carried out in the future.

      (10) Data availability statements should not be limited to the corresponding author, especially for big data analysis. This is crucial to the credibility of this data (Have the scRNA-seq and scATAC-seq in this study been deposited in GEO or other databases, and when will they be released to the public?) The data for such big data analysis needs to be saved in GEO or other databases in advance so that more research can use it.

      Thank you for your comments. We have deposited scRNA-seq and scATAC-seq data in NCBI. “ScRNA-seq data have been deposited in the NCBI Gene Expression Omnibus with the accession number GSE202647, and scATAC-seq data have been deposited in the NCBI database with the accession number PRJNA1177103.”

    1. eLife Assessment

      This manuscript describes a fundamental investigation of the functioning of Cas9 and in particular on how variant xCas9 expands DNA targeting ability by an increase-flexibility mechanism. The authors provide compelling evidence to support their mechanistic models and the relevance of flexibility and entropy in recognition. This work can be of interest to a broad community of structural biophysicists, computational biologists, chemists, and biochemists.

    2. Joint Public Review:

      Summary:

      Hossain and coworkers investigate the mechanisms of recognition of xCas9, a variant of Cas9 with expanded targeting capability for DNA. They do so by using molecular simulations and combining different flavors of simulation techniques, ranging from long classical MD simulations, to enhanced sampling, to free energy calculations of affinity differences. Through this, the authors are able to develop a consistent model of expanded recognition based on the enhanced flexibility of the protein receptor.

      Strengths:

      The paper is solidly based on the ability of the authors to master molecular simulations of highly complex systems. In my opinion, this paper shows no major weaknesses. The simulations are carried out in a technically sound way. Comparative analyses of different systems provide valuable insights, even within the well-known limitations of MD. Plus, the authors further investigate why xCas9 exhibits improved recognition of the TGG PAM sequence compared to SpCas9 via well-tempered metadynamics simulations focusing on the binding of R1335 to the G3 nucleobase and the DNA backbone in both SpCas9 and xCas9. In this context, the authors provide a free-energy profiling that helps support their final model.

      The implementation of FEP calculations to mimic directed evolution improvement of DNA binding is also interesting, original and well-conducted.

      Overall, my assessment of this paper is that it represents a strong manuscript, competently designed and conducted, and highly valuable from a technical point of view.

      Weaknesses:

      To make their impact even more general, the authors may consider expanding their discussion on entropic binding to other recent cases that have been presented in the literature recently (such as e.g. the identification of small molecules for Abeta peptides, or the identification of "fuzzy" mechanisms of binding to protein HMGB1). The point on flexibility helping adaptability and expansion of functional properties is important, and should probably be given more evidence and more direct links with a wider picture.

      Comments on revisions:

      We have read the revised version and the response letter and I find that this manuscript is ready. There is no need for further additions/revisions.

    3. Author response:

      The following is the authors’ response to the original reviews.

      Joint Public Review:

      Strengths:

      The paper is solidly based on the ability of the authors to master molecular simulations of highly complex systems. In my opinion, this paper shows no major weaknesses. The simulations are carried out in a technically sound way. Comparative analyses of different systems provide valuable insights, even within the well-known limitations of MD. Plus, the authors further investigate why xCas9 exhibits improved recognition of the TGG PAM sequence compared to SpCas9 via well-tempered metadynamics simulations focusing on the binding of R1335 to the G3 nucleobase and the DNA backbone in both SpCas9 and xCas9. In this context, the authors provide a free-energy profiling that helps support their final model.

      The implementation of FEP calculations to mimic directed evolution improvement of DNA binding is also interesting, original and well-conducted.

      We thank the reviewer for their positive evaluation of our computational strategy. To further substantiate our findings, we have incorporated additional molecular dynamics and Free Energy Perturbation (FEP) calculations for the system bound to GAT. These results corroborate our previous observations obtained with AAG, reinforcing our conclusions.

      Overall, my assessment of this paper is that it represents a strong manuscript, competently designed and conducted, and highly valuable from a technical point of view.

      Weaknesses:

      To make their impact even more general, the authors may consider expanding their discussion on entropic binding to other recent cases that have been presented in the literature recently (such as e.g. the identification of small molecules for Abeta peptides, or the identification of "fuzzy" mechanisms of binding to protein HMGB1). The point on flexibility helping adaptability and expansion of functional properties is important, and should probably be given more evidence and more direct links with a wider picture.

      We have expanded our discussion on the role of entropy in favoring TGG binding to xCas9. To this end, we performed entropy calculations using the Quasi-Harmonic approximation (details provided in the Materials and Methods section). This analysis reveals that R1335 in xCas9 experiences an entropy increase compared to SpCas9, enhancing its adaptability and interaction with the DNA. This analysis and its explanation are detailed on pages 8-9.

      Additionally, we have enriched the Discussion section by clarifying how DNA binding is entropically favored in xCas9, thereby facilitating the recognition of alternative PAM sequences. A refined explanation is also included in the Conclusions section, where we contextualize xCas9 within a broader evolutionary framework of protein-DNA recognition. This highlights how structural flexibility can enable sequence diversity while maintaining high specificity.

      Recommendations for the authors:

      Overall, this is a very interesting and elegant manuscript with compelling results that shed light on the atomistic determinants of genetic-editing technologies.

      Since the paper proposes new findings that may be helpful for experimentalists, it would be interesting if the authors point out (in their discussion/conclusions) specific amino acids to mutate/target for future tests by the experimental community. This should just appear as an open hypothesis/proposal for new experiments.

      In the Conclusions, we have incorporated a discussion on how modifications in the PAM-binding cleft can enhance the recognition of alternative PAM sequences. As an illustrative example, we reference the recently developed SpRY Cas9 variant, which is capable of recognizing a broader range of PAMs. This variant includes mutations within the PAM-binding cleft that likely increase the flexibility of the interacting residues, as suggested by recent cryo-EM structures (Hibshman et al. Nat. Commun. 2024). The importance of fine-tuning the flexibility of the PAM-interacting cleft for engineering strategies has also been highlighted in the abstract.

      Overall, in light of the reviewer’s comments and in consideration of our findings, we revised the manuscript title in: “Flexibility in PAM Recognition Expands DNA Targeting in xCas9.” This new title better highlights the key findings from our research and contextualizes them within the broader goal of expanding DNA targeting capabilities, a critical priority for developing enhanced CRISPR-Cas systems.

    1. eLife Assessment

      This study provides important computational insights into the dynamics of PROTAC-induced degradation complexes, offering a convincing demonstration that differences in degradation efficacy can be linked to linker properties. The analyses address reproducibility considerations comprehensively, reinforcing the study's conclusions. Overall, these findings are significant for advancing cancer treatments and will be of broad interest to both biochemists and biophysicists.

    2. Reviewer #1 (Public review):

      This study by Wu et al. provides valuable computational insights into PROTAC-related protein complexes, focusing on linker roles, protein-protein interaction stability, and lysine residue accessibility. The findings are significant for PROTAC development in cancer treatment, particularly breast and prostate cancers.

      Strengths:

      (1) Comprehensive computational analysis of PROTAC-related protein complexes.<br /> (2) Focus on critical aspects: linker role, protein-protein interaction stability, and lysine accessibility.

      Weaknesses:

      (1) Limited examination of lysine accessibility despite its stated importance.<br /> (2) Use of RMSD as the primary metric for conformational assessment, which may overlook important local structural changes.

      The authors' claims about the role of PROTAC linkers and protein-protein interaction stability are generally supported by their computational data. However, the conclusions regarding lysine accessibility could be strengthened with more in-depth analysis. The use of the term "protein functional dynamics" is not fully justified by the presented work, which focuses primarily on structural dynamics rather than functional aspects.

      Comments on revisions:

      The authors have addressed the questions raised substantially.

    3. Reviewer #2 (Public review):

      Summary:

      The manuscript reports the computational study of the dynamics of PROTAC-induced degradation complexes. The research investigates how different linkers within PROTACs affect the formation and stability of ternary complexes between the target protein BRD4BD1 and Cereblon E3 ligase, and the degradation machinery. Using computational modeling, docking, and molecular dynamics simulations, the study demonstrates that although all PROTACs form ternary complexes, the linkers significantly influence the dynamics and efficacy of protein degradation. The findings highlight that the flexibility and positioning of Lys residues are crucial for successful ubiquitination. The results also discussed the correlated motions between the PROTAC linker and the complex.

      Strengths:

      The field of PROTAC discovery and design, characterized by its limited research, distinguishes itself from traditional binary ligand-protein interactions by forming a ternary complex involving two proteins. The current understanding of how the structure of PROTAC influences its degradation efficacy remains insufficient. This study investigated the atomic-level dynamics of the degradation complex, offering potentially valuable insights for future research into PROTAC degradability.

      Comments on revisions:

      All my questions have been addressed.

    4. Reviewer #3 (Public review):

      The authors offer an interesting computational study on the dynamics of PROTAC-driven protein degradation. They employed a combination of protein-protein docking, structural alignment, atomistic MD simulations, and post-analysis to model a series of CRBN-dBET-BRD4 ternary complexes, as well as the entire degradation machinery complex. These degraders, with different linker properties, were all capable of forming stable ternary complexes but had been shown experimentally to exhibit different degradation capabilities. While in the initial models of the degradation machinery complex, no surface Lys residue(s) of BRD4 were exposed sufficiently for the crucial ubiquitination step, MD simulations illustrated protein functional dynamics of the entire complex and local side-chain arrangements to bring Lys residue(s) to the catalytic pocket of E2/Ub for reactions. Using these simulations, the authors were able to present a hypothesis as to how linker property affects degradation potency. They were able to roughly correlate the distance of Lys residues to the catalytic pocket of E2/Ub with observed DC50/5h values. This is an interesting and timely study that presents interesting tools that could be used to guide future PROTAC design or optimization.

    5. Author response:

      The following is the authors’ response to the original reviews.

      Reviewer #1 (Public review):

      This study by Wu et al. provides valuable computational insights into PROTAC-related protein complexes, focusing on linker roles, protein-protein interaction stability, and lysine residue accessibility. The findings are significant for PROTAC development in cancer treatment, particularly breast and prostate cancers.

      The authors' claims about the role of PROTAC linkers and protein-protein interaction stability are generally supported by their computational data. However, the conclusions regarding lysine accessibility could be strengthened with more in-depth analysis. The use of the term "protein functional dynamics" is not fully justified by the presented work, which focuses primarily on structural dynamics rather than functional aspects.

      Strengths:

      (1) Comprehensive computational analysis of PROTAC-related protein complexes.

      (2) Focus on critical aspects: linker role, protein-protein interaction stability, and lysine accessibility.

      Weaknesses:

      (1) Limited examination of lysine accessibility despite its stated importance.

      (2) Use of RMSD as the primary metric for conformational assessment, which may overlook important local structural changes.

      Reviewer #1 (Recommendations for the authors):

      (1) The authors' claims about the role of PROTAC linkers and protein-protein interaction stability are generally supported by their computational data. However, the conclusions regarding lysine accessibility could be strengthened with more in-depth analysis. Expand the analysis of lysine accessibility, potentially correlating it with other structural features such as linker length.

      We thank the reviewers for the suggestions! We performed time dependent correlation analysis to correlate the dihedral angles of the PROTACs and the Lys-Gly distance (Figures 6 and S17). We included detailed explanation on page 16:

      “To further examine the correlation between PROTAC rotation and the Lys-Gly interaction, we performed a time-dependent correlation analysis. This analysis showed that PROTAC rotation translates motion over time, leading to the Lys-Gly interaction, with a correlation peak around 60-85 ns, marking the time of the interaction (Figure 6 and Figure S17). In addition, the pseudo dihedral angles also showed a high correlation (0.85 in the case of dBET1) with Lys-Gly distance. This indicated that degradation complex undergoes structural rearrangement and drives the Lys-Gly interaction.”

      (2) The use of the term "protein functional dynamics" is not fully justified by the presented work, which focuses primarily on structural dynamics rather than functional aspects. Consider changing "protein functional dynamics" to "protein dynamics" to more accurately reflect the scope of the study.

      Thanks to the reviewer for the suggestion to use the more accurate terminology! We agreed with the reviewer that if we keep “protein functional dynamics” in the title, we should focus on how the “overall protein dynamic” links to the “function” – The function is directly related to PROTAC-induced structural dynamics which is commonly seen in “protein-structural-function” relationship, but it is not our main focus. Therefore, we changed the title to replace “functional” by “structural”.  

      (3) Incorporate more local and specific characterization methods in addition to RMSD for a more comprehensive conformational assessment.

      We thank the reviewer for the suggestion. We performed time dependent correlation analysis to understand how the rotation of PROTACs can translate to the Lys-Gly interaction. In addition, we performed dihedral entropies analysis for each dihedral angle in the linker of the PROTACs to better examine the flexibility of each PROTAC.

      We included detailed explanation at page 18: “Our dihedral entropies analysis showed that dBET57 has ~0.3 kcal/mol lower entropies than the other three linkers, suggesting dBET57 is less flexible than other PROTACs (Figure S18).”

      Reviewer #2 (Public review):

      Summary:

      The manuscript reports the computational study of the dynamics of PROTAC-induced degradation complexes. The research investigates how different linkers within PROTACs affect the formation and stability of ternary complexes between the target protein BRD4BD1 and Cereblon E3 ligase, and the degradation machinery. Using computational modeling, docking, and molecular dynamics simulations, the study demonstrates that although all PROTACs form ternary complexes, the linkers significantly influence the dynamics and efficacy of protein degradation. The findings highlight that the flexibility and positioning of Lys residues are crucial for successful ubiquitination. The results also discussed the correlated motions between the PROTAC linker and the complex.

      Strengths:

      The field of PROTAC discovery and design, characterized by its limited research, distinguishes itself from traditional binary ligand-protein interactions by forming a ternary complex involving two proteins. The current understanding of how the structure of PROTAC influences its degradation efficacy remains insufficient. This study investigated the atomic-level dynamics of the degradation complex, offering potentially valuable insights for future research into PROTAC degradability.

      Reviewer #2 (Recommendations for the authors):

      (1) Regarding the modeling of the ternary complex, the BRD4 structure (3MXF) is from humans, whereas the CRBN structure in 4CI3 is derived from Gallus gallus. Is there a specific reason for not using structures from the same species, especially considering that human CRBN structures are available in the Protein Data Bank (e.g., 8OIZ, 4TZ4)?

      We appreciate the reviewer’s insightful comment regarding the choice of crystal structures of BRD4 and CRBN structures from two species. Our initial selection of 4CI3 for CRBN structure was based on its high resolution and publication in Nature journal. Furthermore, the Gallus gallus CRBN structure shares high degree of sequence and structural similarity with Homo sapiens CRBN, especially in the ligand binding region. At the time of our study, we were aware of 4TZ4 as Homo sapiens CRBN, however, we did not use this structure since no publication or detailed experimental was associated with it. Additionally, PDB 8OIZ, was not publicly available yet for other researchers to use at the time.

      (2) Based on the crystal structure (PDB ID: 6BNB) discussed in Reference 6, the ternary complex of dBET57 exhibits a conformation distinct from other PROTACs, with CRBN adopting an "open" conformation. Using the same CRBN structure for dBET57 as for other PROTACs might result in inaccurate docking outcomes.

      Thank you for the reviewer’s comment! As noted by the authors in Reference 6, the observed open conformation of CRBN in the dBET57 ternary complex may result from the high salt crystallization conditions, which could drive structural rearrangement, and crystal contacts that may induce this conformation. The authors also mentioned that this open conformation could, in part, reflect CRBN’s intrinsic plasticity. However, they acknowledged that further studies are needed to determine whether this conformational flexibility is a characteristic feature of CRBN that enables it to accommodate a variety of substrates. Despite these observations, we believe that the compatibility of the observed BRD4<sup>BD1</sup> binding conformation with both open and closed CRBN states suggests that these conformational changes are all possible. Therefore, we believe using the same initial CRBN structure for dBET57 as for other PROTACs can still reasonably reveal the dynamic nature of the ternary complex and would not significantly affect the accuracy of our docking outcomes either.

      (3) Figure 2 displays only a single frame from the simulations, which might not provide a comprehensive representation. Could a contact frequency heatmap of PROTAC with the proteins be included to offer a more detailed view?

      We thank the reviewer for the suggestion! We performed the contact map analysis to observe the average distance between PROTACs and BRD4<sup>BD1</sup> over 400ns of MD simulation (new Figure S4 added).

      We included detailed explanation at page 8 and 9: “The residues contact map throughout the 400ns MD simulation also showed different pattern of protein-protein interactions, indicating that the linkers were able to adopt different conformations (Figure S4).”

      (4) The conclusions in Figure 3 and S11 are based on a single 400 ns trajectory. The reproducibility of these results is therefore uncertain.

      We thank the reviewer for the suggestion! We added one more random seed MD simulation for each PROTAC to ensure the reproducibility of the results. The Result is shown in Figure S21 and the details for each MD run are updated in Table 1.

      (5) Figure 4 indicates significant differences between the first and last 100 ns of the simulations. Does this suggest that the simulations have not converged? If so, how can the statistical analysis presented in this paper be considered reliable?

      We thank the reviewers for the question. The simulation was initiated with a 10-15A gap between BRD4 and Ub to monitor the movement of degradation machinery and Lys-Gly interaction. The significant changes in pseudo dihedral in Figure 4 shows that the large-scale movement of the degradation complex can initiate the Lys-Gly binding. It does not relate to unstable sampling because the system remains very stable when BRD4 comes close to Ub.

      (6) In Figure 5, the dihedral angle of dBET57_#9MD1 is marked on a peptide bond. Shouldn't this angle have a high energy barrier for rotation?

      We thank the reviewers for catching the error! Indeed, it was an error that the dihedral angles were marked on the peptide bond. We reworked the figure and double checked our dihedral correlation analysis. The updated correlate dihedral angle selection and the correlation coefficient is shown in Figure 5.

      (7) Given that crystal structures for dBET 70, 23, and 57 are available, why is there a need to model the complex using protein-protein docking?

      We thank the reviewer for the feedback. Only dBET23 has the ternary complex available in a crystal structure, which has the PROTAC and both proteins, while dBET1, dBET57 and dBET70 are not completed as ternary complexes. Although dBET70 has a crystal structure, its PROTAC’s conformation is not resolved, and thus we decided to still perform protein-protein docking with dBET70. 

      We includeed the explanation at page 8: “Only dBET23 crystal structure is available with the PROTAC and both proteins, while the experimentally determined ternary complexes of dBET1, dBET57 and dBET70 are not available. “

      (8) On page 9, it is mentioned that "only one of the 12 PDB files had CRBN bound to DDB1 (PDB ID 4TZ4)." However, there are numerous structures of the DDB1-CRBN complex available, including those used for docking like 4CI3, as well as 4CI1, 4CI2, 8OIZ, etc.

      We thank the reviewers for the comment! We acknowledged the existence of several DDB1-CRBN complex crystal structures, such as PDB IDs 4CI1, 4CI2, 4CI3, and the more recent 8OIZ. For our study, we chose to use 4TZ4 to maintain consistency in complex construction and to align with the methodology established in a previously published JBC paper (https://doi.org/10.1016/j.jbc.2022.101653), which successfully utilized the same structure for a similar construct. At the time our study was conducted, the 8OIZ structure had not yet been released. We appreciate your suggestion and will consider incorporating alternative structures in future studies to further investigate our findings.

      (9) Table 2 is first referenced on page 8, while Table 1 is mentioned first on page 10. The numbering of these tables should be reversed to reflect their order of appearance in the text.

      We thank the reviewer for catching the error! We switched the order of Table 1 and Table 2.

      Reviewer #3 (Public review):

      The authors offer an interesting computational study on the dynamics of PROTAC-driven protein degradation. They employed a combination of protein-protein docking, structural alignment, atomistic MD simulations, and post-analysis to model a series of CRBN-dBET-BRD4 ternary complexes, as well as the entire degradation machinery complex. These degraders, with different linker properties, were all capable of forming stable ternary complexes but had been shown experimentally to exhibit different degradation capabilities. While in the initial models of the degradation machinery complex, no surface Lys residue(s) of BRD4 were exposed sufficiently for the crucial ubiquitination step, MD simulations illustrated protein functional dynamics of the entire complex and local side-chain arrangements to bring Lys residue(s) to the catalytic pocket of E2/Ub for reactions. Using these simulations, the authors were able to present a hypothesis as to how linker property affects degradation potency. They were able to roughly correlate the distance of Lys residues to the catalytic pocket of E2/Ub with observed DC50/5h values. This is an interesting and timely study that presents interesting tools that could be used to guide future PROTAC design or optimization.

      Reviewer #3 (Recommendations for the authors):

      (1) My most important comment refers to the MM/PBSA analysis, the results of which are shown in Figure S9: binding affinities of -40 to -50 kcal/mol are unrealistic. This would correspond to a dissociation constant of 10^-37 M. This analysis needs to be removed or corrected.

      We thank the reviewer for the comment! MM/PBSA analysis indeed cannot give realistic binding free energy. It does not include the configurational entropy loss which should be a large positive value. In addition, while the implicit PBSA solvent model computes solvation free energy, the absolute values may not be very accurate. However, because this is a commonly used energy calculation, and some readers may like to see quantitative values to ensure that the systems have stable intermolecular attractions, we kept the analysis in SI. We edited the figure legend, moved the Figure S10 in SI page 19, and added sentences to clearly state that the calculations did not include configuration entropy loss “Note that the energy calculations focus on non-bonded intermolecular interactions and solvation free energy calculations using MM/PBSA, where the configuration entropy loss during protein binding was not explicitly included. “.

      (2) I think that the analysis of what in the different dBETx makes them cause different degradation potency is underdeveloped. The dihedral angle analysis (Figure 4B) did not explain the observed behavior in my opinion. Please add additional, clearer analysis as to what structural differences in the dBETx make them sample very different conformations.

      We thank the reviewer for the suggestions! Based on the suggestion, we further performed dihedral entropy analysis for each dihedral angle in the linker part of the PROTAC to examine the flexibility of each PROTAC. Because each PROTAC has a different linker, we now clearly label them in a new Figure S18 in SI page 27. Low dihedral entropies indicate a more rigid structure and thus less flexibility to make a PROTAC more difficult to rearrange and facilitate the protein structural dynamic necessary for ubiquitination.

      We added detailed explanation on page 18: “Our dihedral entropy analysis showed that dBET57 has ~0.3 kcal/mol lower configuration entropies than the other dBETs with three different linkers, suggesting that dBET57 is less flexible than the other PROTACs (Figure S18).”

      (3) "The movement of the degradation machinery correlated with rotations of specific dihedrals of the linker region in dBETs (Figure 5).": this is not sufficiently clear from the figure. Definitely not in a quantitative way.

      We thank the reviewers for the suggestions! To further understand the correlation between PROTACs dihedral angles and the movement of degradation machinery, we performed time dependent correlation analysis to correlate the dihedral angles of the PROTACs and the Lys-Gly distance (Figures 6 and S17).

      We included detailed explanation on page 16:

      “To further examine the correlation between PROTAC rotation and the Lys-Gly interaction, we performed a time-dependent correlation analysis. This analysis showed that PROTAC rotation translates motion over time, leading to the Lys-Gly interaction, with a correlation peak around 60-85 ns, marking the time of the interaction (Figure 6 and Figure S17). In addition, the pseudo dihedral angles also showed a high correlation (0.85 in the case of dBET1) with Lys-Gly distance. This indicated that degradation complex undergoes structural rearrangement and drives the Lys-Gly interaction.

      (4) Cartoons are needed at multiple stages throughout the paper to enhance the clarity of what the modeled complexes looked like (e.g. which subunits they contained).

      We thank the reviewers for the suggestions. We added and remade several Figures with cartoons to better represent the stages. We also used higher resolution and included clearer labels for each protein system.

      (5) The difference between CRL4A E3 ligase and CRBN E3 ligase is not clear to the non-expert reader.

      Thanks for the reviewer’s comment! To clarify the terms "CRL4A E3 ligase" and "CRBN E3 ligase", which refer to different levels of description for the protein complexes, we added a couple of sentences in the Figure 1 legend. As a result, the non-expert readers can clearly know the differences.

      As illustrated in Figure 1,

      • CRL4A E3 ligase refers to the full E3 ligase complex, which includes all protein components such as CRBN, DDB1, CUL4A, and RBX1.

      • CRBN E3 ligase, on the other hand, is a more colloquial term typically used to describe just the CRBN protein, often in isolation from the full CRL4A complex.

      (6) Figure 1, legend: unclear why it's E3 in A and E2 in B.

      We thank the reviewer for the question! E3 ligase in Figure 1A refers to CRBN E3 ligase, where researchers also simply term it CRBN. We have added a sentence to specify that CRBN E3 ligase is also termed CRBN for simplicity. In Figure 1B, E2 was unclear in the sentences. The full name of E2 should be E2 ubiquitin-conjugating enzyme. Because the name is a bit long, researchers also call it E2 enzyme. We have corrected it and used E2 enzyme to make it clearer. 

      (7) "Although the protein-protein binding affinities were similar, other degraders such as dBET1 and dBET57 had a DC50/5h of about 500 nM". It's unclear what experimental data supports the assertion that the protein-protein binding affinities are similar.

      We thank reviewer for the question. Indeed, the statement is unclear.

      We corrected the sentence in page 6: “Although utilizing the exact same warheads, other degraders such as dBET1 and dBET57 had a DC<sub>50/5h</sub> of about 500 nM.”

      (8) Was the construction of the degradation machinery complex guided by experimental data (maybe cryo-EM or tomography)? If not, what is the accuracy of the starting complex for MD? This may impact the reliability of the obtained results.

      Thank you for your insightful comments! Yes, the construction of the degradation machinery complex was guided by available high-resolution crystal structures, which was selected to maintain consistency and align with the methodology established in a previously published JBC paper (https://doi.org/10.1016/j.jbc.2022.101653).

      We acknowledged that static crystal structures represent only a single snapshot of the system and may not capture the full conformational flexibility of the complex. To address this limitation, we performed MD simulations using multiple starting structures. This approach allowed us to explore a broader conformational landscape and reduced the dependence on any single starting configuration, thereby enhancing the reliability of the results.

      We hope this clarifies the robustness of our methodology and the steps taken to ensure accuracy in our simulations.

      (9) "With quantitative data, we revealed the mechanism underlying dBETx-induced degradation machinery": I think this may be too strong of an assertion. The authors may have developed a mechanistic hypothesis that can be tested experimentally in the future.

      We thank the reviewer for the suggestion. This is indeed a strong assertion and needs to be modified. We edited the sentence in page 7: “With quantitative data, we revealed the importance of the structural dynamics of dBETx-induced motions, which arrange positions of surface lysine residues of BRD4<sup>BD1</sup> and the entire degradation machinery.”

      (10) Figure S2: are the RMSDs calculated over all residues? Or just the BRD4 residues? Given that the structures are aligned with respect to CRBN, the reported RMSD numbers might be artificially low since there are many more CRBN residues than there are BRD4 residues. Also, why weren't the crystal structures used for dBET 23 and 70 for the modeling? Wouldn't you want to use the most accurate possible structures? Simulations were run for 23. Why not for 70?

      We thank the reviewer for the suggestion. We added a sentence to more clearly explain the RMSD calculations in Figure S2: “The structural superposition is performed based on the backbone of CRBN and RMSD calculation is conducted based on the backbone of BRD4<sup>BD1</sup>.”

      Although dBET70 has crystal structure, its PROTAC structure is not resolved, and thus we decided to still perform protein-protein docking with dBET70.  dBET1 and dBET57 do not have a crystal structure for the ternary complexes.

      We included the explanation at page 8: “Only dBET23 crystal structure is available with the PROTACs and both proteins, while the experimentally determined ternary complexes of dBET1, PROTACs of dBET57 and dBET70 are not available. “

      a. And there are no crystal structures available for 1 and 57? If so, please clearly say that. Otherwise please report the RMSD.

      We thank the reviewer for the suggestion. We included the explanation at page 8: “Only dBET23 crystal structure is available with the PROTACs and both proteins, while the experimentally determined ternary complexes of dBET1, PROTACs of dBET57 and dBET70 are not available.”

      (11) Table 2 is referenced before Table 1.

      We thank the reviewer for catching the error! We switched the order for Table 1 and Table 2.

      (12) Figure S3 is not referenced in the main paper.

      We thank the reviewer for catching the error! We now referred Figure S3 on page7.

      (13) Minor comments on grammar and sentence structure:

      a. It should be "binding of a ternary complex"

      b. "Our shows the importance": word missing.

      c. "...providing insights into potential orientations for ubiquitination. observe whether the preferred conformations are pre-organized for ubiquitination." Word or words missing.

      We thank reviewer for catching the errors! We corrected grammatical errors and unclear sentences throughout the entire paper and revised the sentences to make them easily understandable for non-expert readers.

    1. eLife Assessment

      This study provides a valuable approach to image and analyze in vivo metabolic flux through glucose turnover kinetics in glioblastoma tumor microenvironments. The evidence for the method's validity is convincing, which establishes the dynamic Deuterium Metabolic Imaging technique as an effective tool enabling non-invasive exploration of various tumors.

    2. Reviewer #1 (Public review):

      In the resubmission Simões et al. emphasize the efficacy of their novel, non-invasive imaging methodology in mapping glucose-kinetics to predict key tumor features in two commonly used syngeneic mouse models of glioblastoma. The authors highlight that DGE-DMI has the potential to capture metabolic fluxes with greater sensitivity and acknowledge that future validation of DGE-DMI in patient-derived and spontaneous GBM models, as well as in the context of genetic manipulation of metabolism, would strengthen its clinical application. To further demonstrate the ability of DGE-DMI to predict tumor features, they included an assessment of myeloid cell infiltration along with proliferation, peritumoral invasion, and distant migration. Overall, the authors offer a novel method to the scientific community that can be further tested and adapted for interrogating GBM heterogeneity.

    3. Reviewer #3 (Public review):

      Summary:

      Simoes et al enhanced dynamic glucose-enhanced (DGE) deuterium spectroscopy with Deuterium Metabolic Imaging (DMI) to characterize the kinetics of glucose conversion in two murine models of glioblastoma (GBM). The authors combined spectroscopic imaging and noise attenuation with histological analysis and showcased the efficacy of metabolic markers determined from DGE DMI to correlate with histological features of the tumors. This approach is also potent to differentiate the two models from GL261 and CT2A.

      Strengths:

      The primary strength of this study is to highlight the significance of DGE DMI to interrogate the metabolic flux from glucose. The authors focused on glutamine/glutamate and lactate. They attempted to correlate the imaging findings with in-depth histological analysis to depict the link between metabolic features and pathological characteristics such as cell density, infiltration, and distant migration.

    4. Author response:

      The following is the authors’ response to the original reviews.

      eLife assessment

      This work describes a convincingly validated non-invasive tool for in vivo metabolic phenotyping of aggressive brain tumors in mice brains. The analysis provides a valuable technique that tackles the unmet need for patient stratification and hence for early assessment of therapeutic efficacy. However, wider clinical applicability of the findings can be attained by expanding the work to include more diverse tumor models.

      We thank the Editors for their comments. This concern was also raised by Reviewer 1 in the Public Review, where we address in more detail – please refer to comment PR-R1.C1. In brief, we agree that a more clinically relevant model should provide more translatable results to patients, and acknowledge this better in the revised manuscript: page 18 (lines 14-17), “While patient-derived xenografts and de novo models would be more suited to recapitulate human GBM heterogeneity and infiltration features, and genetic manipulation of glycolysis and mitochondrial oxidation pathways potentially relevant to ascertain DGE-DMI sensitivity for their quantification, (…)”. However, we also believe that the potential of DGE-DMI for application to different glioblastoma models or patients is demonstrated clearly enough with the two immunocompetent models we chose, extensively reported in the literature as reliable models of glioblastoma.

      Public Reviews:

      Reviewer #1 (Public Review):

      Summary:

      This work introduces a new imaging tool for profiling tumor microenvironments through glucose conversion kinetics. Using GL261 and CT2A intracranial mouse models, the authors demonstrated that tumor lactate turnover mimicked the glioblastoma phenotype, and differences in peritumoral glutamate-glutamine recycling correlated with tumor invasion capacity, aligning with histopathological characterization. This paper presents a novel method to image and quantify glucose metabolites, reducing background noise and improving the predictability of multiple tumor features. It is, therefore, a valuable tool for studying glioblastoma in mouse models and enhances the understanding of the metabolic heterogeneity of glioblastoma.

      Strengths:

      By combining novel spectroscopic imaging modalities and recent advances in noise attenuation, Simões et al. improve upon their previously published Dynamic Glucose-Enhanced deuterium metabolic imaging (DGE-DMI) method to resolve spatiotemporal glucose flux rates in two commonly used syngeneic GBM mouse models, CT2A and GL261. This method can be standardized and further enhanced by using tensor PCA for spectral denoising, which improves kinetic modeling performance. It enables the glioblastoma mouse model to be assessed and quantified with higher accuracy using imaging methods.

      The study also demonstrated the potential of DGE-DMI by providing spectroscopic imaging of glucose metabolic fluxes in both the tumor and tumor border regions. By comparing these results with histopathological characterization, the authors showed that DGE-DMI could be a powerful tool for analyzing multiple aspects of mouse glioblastoma, such as cell density and proliferation, peritumoral infiltration, and distant migration.

      Weaknesses:

      (1) Although the paper provides clear evidence that DGE-DMI is a potentially powerful tool for the mouse glioblastoma model, it fails to use this new method to discover novel features of tumors. The data presented mainly confirm tumor features that have been previously reported. While this demonstrates that DGE-DMI is a reliable imaging tool in such circumstances, it also diminishes the novelty of the study.

      PR-R1.C1 – We thank the Reviewer for the detailed analysis and reply below to each point. PR-R1.C1.1 - novelty: We thank the Reviewer for the comments and understand their perspective. While we acknowledge that our paper is more methodologically oriented, we also believe that significant methodological advances are critical for new discoveries. This was our main motivation and is demonstrated in the present work, showing the ability to map in vivo metabolic fluxes in mouse glioma, a “hot topic” and very desirable in the cancer field. 

      PR-R1.C1.2 – additional tumor features: To strengthen the biological relevance of this methodologic novelty, we have now included immune cell infiltration among the tumor features assessed, besides perfusion, histopathology, cellularity and cell proliferation. For this, we performed iba-1 immunostaining for microglia/ macrophages, now included in Fig. 2-B. These new results demonstrate significantly higher microglia/macrophage infiltration in CT2A tumors compared to GL261, particularly at the tumor border. This is very consistent with the respective tumor phenotypes, namely differences in cell density and cellularity between the 2 cohorts and across pooled cohorts, as we now report: page 9 (lines 10-18), “Such phenotype differences were reflected in the regional infiltration of microglia/macrophages: significantly higher at the CT2A peritumoral rim (PT-Rim) compared to GL261, and slightly higher in the tumor region as well (Fig 2B). Further quantitative regional analysis of Tumor-to-PT-Rim ROI ratios revealed: (i) 47% lower cell density (p=0.004) and 32% higher cell proliferation (p=0.026) in GL261 compared to CT2A (Fig 2C, Table S3); and (ii) strong negative correlations in pooled cohorts between microglia/macrophage infiltration and cellularity (R=-0.91, p=<0.001) or cell density (R=-0.77, p=0.016), suggesting more circumscribed tumor growth with higher peripheral/peritumoral infiltration of immune cells.”; and page 16 (lines 13-19), “GL261 tumors were examined earlier after induction than CT2A (17±0 vs. 30±5 days, p = 0.032), displaying similar volumes (57±6 vs. 60±14, p = 0.813) but increased vascular permeability (8.5±1.1 vs 4.3±0.5 10<sup>3</sup>/min: +98%, p=0.001),  more disrupted stromal-vascular phenotypes and infiltrative growth (5/5 vs 0/5), consistent with significantly lower tumor cell density (4.9±0.2 vs. 8.2±0.3 10<sup>-3</sup> cells/µm<sup>2</sup>: -40%, p<0.001) and lower peritumoral rim infiltration of microglia/macrophages (2.1±0.7 vs. 10.0±2.3 %: -77%, p=0.008)”.

      PR-R1.C1.3 – new tumor features and DGE-DMI: Importantly, such regional differences in cellularity/cell density and immune cell infiltration between the two cohorts were remarkably mirrored by the lactate turnover maps (Fig 3-C), as we now report in the manuscript: page 12 (lines 6-15), “GL261 tumors accumulated significantly less lactate in the core (1.60±0.25 vs 2.91±0.33 mM: -45%, p=0.013) and peritumor margin regions (0.94±0.09 vs 1.46±0.17 mM: 36%, p=0.025) than CT2A – Fig 3 A-B, Table S1. Consistently, tumor lactate accumulation correlated with tumor cellularity in pooled cohorts (R=0.74, p=0.014). Then, lower tumor lactate levels were associated with higher lactate elimination rate, k<sub>lac</sub> (0.11±0.1 vs 0.06±0.01 mM/min: +94%, p=0.006) – Fig 3B – which in turn correlated inversely with peritumoral rim infiltration of microglia/macrophages in pooled cohorts (R=-0.73, p=0.027) – Fig 3-C. Further analysis of Tumor/P-Margin metabolic ratios (Table S3) revealed: (i) +38% glucose (p=0.002) and -17% lactate (p=0.038) concentrations, and +55% higher lactate consumption rate (p=0.040) in the GL261 cohort; and (ii) lactate ratios across those regions reflected the respective cell density ratios in pooled cohorts (R=0.77, p=0.010) – Fig 3-C”. This is a novel, relevant feature compared to our previous work, as highlighted in our discussion: page 17 (lines 1-8), “Tumor vs peritumor border analyses further suggest that lactate metabolism reflects regional histologic differences:

      lactate accumulation mirrors cell density gradients between and across the two cohorts; whereas lactate consumption/elimination rate coarsely reflects cohort differences in cell proliferation, and inversely correlates with peritumoral infiltration by microglia/macrophages across both cohorts. This is consistent with GL261’s lower cell density and cohesiveness, more disrupted stromal-vascular phenotypes, and infiltrative growth pattern at the peritumor margin area, where less immune cell infiltration is detected and relatively lower cell division is expected [43]”.

      We trust that these new features recovered from DGE-DMI (Fig 2-B and Fig 3-C) show its potential for new discoveries in glioblastoma.

      (2) When using DGE-DMI to quantitatively map glycolysis and mitochondrial oxidation fluxes, there is no comparison with other methods to directly identify the changes. This makes it difficult to assess how sensitive DGE-DMI is in detecting differences in glycolysis and mitochondrial oxidation fluxes, which undermines the claim of its potential for in vivo GBM phenotyping.

      PR-R1.C2: We thank the reviewer for raising this important point. The validity of the method for mapping specific metabolic kinetics in mouse glioma was reported in our previous work, using the same animal models, as specified in the introduction (page 4, lines 10-13): “we recently (…) propose[d] Dynamic Glucose-Enhanced (DGE) 2H-MRS [31], demonstrating its ability to quantify glucose fluxes through glycolysis and mitochondrial oxidation pathways in vivo in mouse GBM (…)”. Therefore, this was not reproduced in the present work. 

      In brief, our DGE-DMI results are very consistent with our previous study, where DGE single voxel deuterium spectroscopy was performed in the same tumor models with higher temporal resolution and SNR (as state on page 16, lines 9-10: glycolytic lactate synthesis rate, 0.59±0.04 vs. 0.55±0.07 mM/min; glucose-derived glutamate-glutamine synthesis rate, 0.28±0.06 vs. 0.40±0.08 mM/min), which in turn matched well the values reported by others for glucose consumption rate through: 

      (i) glycolysis, in different tumor models including mouse lymphoma in vivo (0.99 mM/min, by DGE-DMI (Kreis et al. 2020), rat breast carcinoma in situ (1.43 mM/min, using a biochemical assay (Kallinowski et al. 1988), and even perfused GBM cells (1.35 fmol min<sup>−1</sup> cell<sup>−1</sup>, according to Hyperpolarized 13C-MRS (Jeong et al. 2017), very similar to our previous in vivo measurements in GL261 tumors: 0.50 ± 0.07 mM min<sup>−1</sup> = 1.25 ± 0.16 fmol min<sup>−1</sup> cell<sup>−1</sup> (Simoes et al. 2022)); 

      (ii) mitochondrial oxidation, very similar to previous in vivo measurements in mouse GBM xenografts (0.33 mM min<sup>−1</sup>, using 13C spectroscopy (Lai et al. 2018)), and particularly to our in situ measurements in cell culture for (GL261, 0.69 ± 0.09 fmol min<sup>−1</sup> cell<sup>−1</sup>; and CT2A 0.44 ± 0.08 fmol min<sup>−1</sup> cell<sup>−1</sup>), remarkably similar to the in vivo measurements in the respective tumors in vivo (Gl261, 0.32 ± 0.10 mM min<sup>−1</sup> = 0.77 ± 0.23 fmol min<sup>−1</sup> cell<sup>−1</sup>; and CT2A, 0.51 ± 0.11 mM min<sup>−1</sup> = 0.60 ± 0.12 fmol min<sup>−1</sup> cell<sup>−1</sup>) (Simoes et al. 2022)). 

      (3) The study only used intracranial injections of two mouse glioblastoma cell lines, which limits the application of DGE-DMI in detecting and characterizing de novo glioblastomas. A de novo mouse model can show tumor growth progression and is more heterogeneous than a cell line injection model. Demonstrating that DGE-DMI performs well in a more clinically relevant model would better support its claimed potential usage in patients.

      PR-R1.C3: We agree that a more clinically relevant model, such as the one suggested by the Reviewer, would in principle be better suited to provide more translatable results to patients. We however believe that the potential of DGE-DMI for application to different glioblastoma models or patients, with GBM or any other types of brain tumors for that matter, is demonstrated clearly enough with the two syngeneic models we chose, given their robustness and general acceptance in the literature as reliable immunocompetent models of GBM, and for their different histologic and metabolic properties. This way we could fully focus on the novel metabolic imaging method, as compared to our previous single-voxel approach. While both tumor cohorts (GL261 and CT2A) were studied at more advanced stages of tumor progression, the metabolic differences depicted are consistent with the histopathologic features reported, as discussed in the manuscript; namely, the lower glucose oxidation rates. We have now modified the manuscript to highlight this point: page 18 (lines 12-14), “While patient-derived xenografts and de novo models would be more suited to recapitulate human GBM heterogeneity and infiltration features, and genetic manipulation of glycolysis and mitochondrial oxidation pathways could be relevant to ascertain DGE-DMI sensitivity for their quantification, (…)”.

      Reviewer #2 (Public Review):

      Summary:

      In this work, the authors attempt to noninvasively image metabolic aspects of the tumor microenvironment in vivo, in 2 mouse models of glioblastoma. The tumor lesion and its surrounding appearance are extensively characterized using histology to validate/support any observations made with the metabolic imaging approach. The metabolic imaging method builds on a previously used approach by the authors and others to measure the kinetics of deuterated glucose metabolism using dynamic 2H magnetic resonance spectroscopic imaging (MRSI), supported by de-noising methods.

      Strengths:

      Extensive histological evaluation and characterization.

      Measurement of the time course of isotope labeling to estimate absolute flux rates of glucose metabolism.

      Weaknesses:

      (1) The de-noising method appears essential to achieve the high spatial resolution of the in vivo imaging to be compatible with the dimensions of the tumor microenvironment, here defined as the immediately adjacent rim of the mouse brain tumors. There are a few challenges with this approach. Often denoising methods applied to MR spectroscopy data have merely a cosmetic effect but the actual quantification of the peaks in the spectra is not more accurate than when applied directly to original non-denoised data. It is not clear if this concern is applicable to the denoising technique applied here. However, even if this is not an issue, no denoising method can truly increase the original spatial resolution at which data were acquired. A quick calculation estimates that the spatial resolution of the 2H MRSI used here is 30-40 times too low to capture the much smaller tumor rim volume, and therefore there is concern that normal brain tissue and tumor tissue will be the dominant metabolic signal in so-called tumor rim voxels. This means that the conclusions on metabolic features of the (much larger) tumor are much more robust than the observations attributed to the (much smaller) tumor microenvironment/tumor rim.

      PR-R2.C1: We thank the Reviewer for the constructive comments regarding resolution and tumor rim, and denoising. These issues were raised more extensively in the section Recommendations For The Authors, where they are addressed in detailed (RA-R2.C2). In summary, we agree with the Reviewer that no denoising method can increase the nominal resolution; not was that our purpose. Thus, we clarify the relevance of spectral matrix interpolation in MRSI, and how our display resolution should in principle provide a better approximation to the ground truth than the nominal resolution, relevant for ROI analysis in the tumor margin. While we further show relevant correlations between metabolic maps and histologic features in tumor core and margin, we agree with the reviewer that our observations in the tumor core are more robust than those in the margin, and acknowledge this in the Discussion: page 19, lines 6-10: “Therefore, further DGE-DMI preclinical studies aimed at detecting and quantifying relatively weak signals, such as tumor glutamate-glutamine, and/or increase the nominal spatial resolution to better correlate those metabolic results with histology findings (e.g. in the tumor margin), should improve basal SNR with higher magnetic field strengths, more sensitive RF coils, and advanced DMI pulse sequences [55]).”

      (2) To achieve their goal of high-level metabolic characterization the authors set out to measure the deuterium labeling kinetics following an intravenous bolus of deuterated glucose, instead of the easier measurement of steady-state after the labeling has leveled off. These dynamic data are then used as input for a mathematical model of glucose metabolism to derive fluxes in absolute units. While this is conceptually a well-accepted approach there are concerns about the validity of the included assumptions in the metabolic model, and some of the model's equations and/or defining of fluxes, that seem different than those used by others.

      PR-R2.C2: These concerns about the metabolic model, were also raised in more detail in the section Recommendations For The Authors, where they are addressed more extensively – please refer to RA-R2.C3 (glucose infusion protocol) and RA-R2.C4 (equations). In brief, we explain that the total volume injected (100uL/25g animal) is standard for i.v. administration in mice, and clarify this better in the manuscript (page 24, line 23); as well as the differences between our kinetic model and the original one reported by Kreis et al. (Radiology 2020), who quantified glycolysis kinetics on a subcutaneous mouse model of lymphoma, exclusively glycolytic and thus estimating the maximum glucose flux rate was from the lactate synthesis rate (Vmax = Vlac). Instead, we extended this model to account for glucose flux rates for lactate synthesis (Vlac) and also for glutamate-glutamine synthesis (Vglx) in mouse glioblastoma, where Vmax = Vlac + Vglx, also acknowledging its simplistic approach in the Discussion (page 20, lines 22-24: “(…) metabolic fluxes [estimations] through glycolysis and mitochondrial oxidation (…) could potentially benefit from an improved kinetic model simultaneously assessing cerebral glucose and oxygen metabolism, as recently demonstrated in the rat brain with a combination of 2H and 17O MR spectroscopy [62] (…)”).

      Reviewer #3 (Public Review):

      Summary:

      Simoes et al enhanced dynamic glucose-enhanced (DGE) deuterium spectroscopy with Deuterium Metabolic Imaging (DMI) to characterize the kinetics of glucose conversion in two murine models of glioblastoma (GBM). The authors combined spectroscopic imaging and noise attenuation with histological analysis and showcased the efficacy of metabolic markers determined from DGE DMI to correlate with histological features of the tumors. This approach is also potent to differentiate the two models from GL261 and CT2A.

      Strengths:

      The primary strength of this study is to highlight the significance of DGE DMI in interrogating the metabolic flux from glucose. The authors focused on glutamine/glutamate and lactate. They attempted to correlate the imaging findings with in-depth histological analysis to depict the link between metabolic features and pathological characteristics such as cell density, infiltration, and distant migration.

      Weaknesses:

      (1) A lack of genetic interrogation is a major weakness of this study. It was unclear what underlying genetic/epigenetic aberrations in GL261 and CT2A account for the metabolic difference observed with DGE DMI. A correlative metabolic confirmation using mass spectrometry of the two tumor specimens would give insight into the observed imaging findings.

      PR-R3.C1: We thank the Reviewer for the helpful comments, which we break down below.

      PR-R3.C1.1 - genetic interrogation/manipulation: While we did not have access to conditional models for key enzymes of each metabolic pathway, for their genetic manipulation, we did however assess the mitochondrial function in each cell line, showing a significantly higher respiration buffer capacity and more efficient metabolic plasticity between glycolysis and mitochondrial oxidation in GL261 cells compared to CT2A (Simoes et al. NIMG:Clin 2022). This could drive e.g. more active recycling of lactate through mitochondrial metabolism in GL261 cells, aligned with our observations of increased glucose-derived lactate consumption rate in those tumors compared to CT2A. We have now included this in the discussion (page 17, lines 812): “our results suggest increased lactate consumption rate (active recycling) in GL261 tumors with higher vascular permeability, e.g. as a metabolic substrate for oxidative metabolism [44] promoting GBM cell survival and invasion [45], aligned with the higher respiration buffer capacity and more efficient metabolic plasticity of GL261 cells than CT2A [31].”

      PR-R3.C1.2 - correlation with post-mortem metabolic assessment: implementing this validation step would require an additional equipment, also not accessible to us: focalized irradiator, to instantly halt all metabolic reactions during animal sacrifice. We do believe that DGE-DMI could guide further studies of such nature, aimed at validating the spatio-temporal dynamics of regional metabolite concentrations in mouse brain tumors. Thus, the importance of end-point validation is now stressed more clearly in the manuscript (page 20, lines 13-16): “(…) mapping pathway fluxes alongside de novo concentrations (…) may be determinant for the longitudinal assessment of GBM progression, with end-point validation (…)”.

      These concerns and recommendations were also raised by the Reviewer in the Recommendations to Authors section, where we address them more extensively – please see RA-R1.C3 and RA-R1.C2, respectively.

      (2) A better depiction of the imaging features and tumor heterogeneity would support the authors' multimodal attempt.

      PR-R3.C2: We agree with the Reviewer that including more imaging features would improve the non-invasive characterization of each tumor. Due to the RF coil design and time constraints, we did not acquire additional data, such as diffusion MRI to assess tissue microstructure. Instead, our multi-modal protocol included two dynamic MRI studies on each animal, for multiparametric assessment of tumor volume, metabolism and vascular permeability, using 1H-MRI, 2H-spectroscopy during 2H-labelled glucose injection, and 1H-imaging during Gd-DOTA injection, respectively. Rather than aiming at tumor radiomics, we focused on the dynamic assessment of tumor metabolic turnover with heteronuclear spectroscopy, which is challenging per se and particularly in mouse brain tumors, given their very small size. For such multi-modal studies we used a previously developed dual tuned RF coil: the deuterium coil (2H) positioned in the mouse head, for optimal SNR; whereas the proton coil (1H) had suboptimal performance compared a conventional single tuned coil, and was used only for basic localization and adjustments, reference imaging and tumor volumetry (T2-weighted), and DCE-T1 MRI (T1weighted). The latter was analyzed pixel-wise to assess spatial correlations between tumor permeability and metabolic metrics, as shown in Fig S3. Whereas the limited T2w MRI data collected was only analyzed for tumor volume assessment; no additional imaging features were extracted (e.g. kurtosis/skewness), since such assessment did not shown any differences between the two tumor cohorts in our previous study (Simoes et al NIMG:Clin 2022).

      (3) Integration of the various cell types in the tumor microenvironment, as allowed with the resolution of DGE DMI, will explain the observed difference between GL261 and CT2A. Is there a higher percentage of infiltrative "other cells" observed in GL261 tumor?

      PR-R3.C3: While DGE-DMI resolution is far larger than brain and brain tumor cell sizes, we now performed additional analysis to assess the percentage of microglia/macrophages in both cohorts. The results are now included in the manuscript, namely Fig. 2B, as previously explained in PR-R1.1. Interestingly though, we observed a lower percentage of infiltrative "other cells" in GL261 tumors compared to CT2A, which we discuss in the manuscript: pages 19-20 (lines 20-24 and 1-4), “Finally, our results are indicative of higher microglia/macrophage infiltration in CT2A than GL261 tumors, which is inconsistent with another study reporting higher immunogenicity of GL261 tumors than CT2A for microglia and macrophage populations [56]. Such discrepancy could be related to methodologic differences between the two studies, namely the endpointguided assessment of tumor growth (bioluminescence vs MRI, more precise volumetric estimations) and the stage when tumors were studied (GL261 at 23-28 vs 16-18 days postinjection, i.e. less time for immune cell to infiltration in our case), presence/absence of a cell transformation step (GFP-Fluc engineered vs we used original cell lines), or perhaps media conditioning effects during cell culture due to the different formulations used (DMEM vs RPMI).”

      (4) This underlying technology with DGE DMI is capable of identifying more heterogeneous GBM tumors. A validation cohort of additional in vivo models will offer additional support to the potential clinical impact of this study.

      PR-R3.C4: We agree with the Reviewer that applying DGE-DMI to more clinically-relevant models of human brain tumors will enhance its translational impact to patients, as also suggested by Reviewer 1 and addressed in PR-R1.C3. We also believe that the feasibility and potential of DGE-DMI for application to different glioblastoma models or patients, with GBM or any other primary or secondary brain tumors, is clearly demonstrated in our work, using two reliable and well-described immunocompetent models of GBM. In any case, we have now modified the manuscript to better acknowledge this point: page 18 (lines 14-16), “(…) patient-derived xenografts and de novo models would be more suited to recapitulate human GBM heterogeneity and infiltration features (…)”.

      Recommendations for the authors:

      Reviewer #1 (Recommendations For The Authors):

      (1) The authors utilize longitudinal MRI to track tumor volumes but perform DMI at endpoint with late-stage tumors. Their previous publication applied metabolic imaging in tumors before the presence of necrosis. It would be valuable to perform longitudinal DMI to examine the evolution of glucose flux metabolic profile over time in the same tumor.

      RA-R1.C1: We thank the Reviewer for the very useful comments to our manuscript. We agree – in this work, we aimed at “extending” our previous DGE-2H single-voxel methodology to multivoxel (DMI), thoroughly demonstrating (1) its in vivo application to the same immunocompetent models of glioblastoma and (2) the ability to depict their phenotypic differences, and therefore (3) the potential for the metabolic characterization of more advanced models of GBM and/or their progression stages. We believe these objectives were achieved. Our results indeed open several possibilities, from longitudinal assessment of the spatio-temporal metabolic changes during GBM progression (and treatment-response) to its application to other models recapitulating more closely the human disease. Now that we have comprehensively demonstrated a protocol for DGE-DMI acquisition, processing and analysis in mouse GBM (a very challenging methodology), and demonstrate it in different mouse GBM cell lines, new studies can be designed to tackle more specific questions, like the one suggested here by the Reviewer. We have modified the manuscript to make this point clearer: page 20 (lines 15-17), “This may be determinant for the longitudinal assessment of GBM progression, with end-point validation; and/or treatment-response, to help selecting among new therapeutic modalities targeting GBM metabolism (…)”; page 21 (lines 5-8), “(…) we report a DGE-DMI method for quantitative mapping of glycolysis and mitochondrial oxidation fluxes in mouse GBM, highlighting its importance for metabolic characterization and potential for in vivo GBM phenotyping in different models and progression stages.”.

      (2) The authors demonstrate a promising correlation between metabolic phenotypes in vivo and key histopathological features of GBM at the endpoint. Directly assessing metabolites involved in glucose fluxes on endpoint tumor samples would strengthen this correlation.

      RA-R1.C2: While we acknowledge the Reviewer’s point, there were two main limitations to implementing such validation step in our protocol: 

      (1) Since we performed dynamic experiments, at the end of each study most 2H-glucose-derived metabolites were already below their maximum concentration (or barely detectable in some cases), as depicted by the respective kinetic curves (Fig 1-D and Fig S7), and thus no longer detectable in the tissues. Importantly, DGE-DMI could guide further studies towards selecting the ideally time-point for validating different metabolite concentrations in specific brain regions.

      (2) Such validation would require sacrificing the animals with a focalized irradiator (which we did not have), to instantly halt all metabolic reactions. Only then we could collect and analyze the metabolic profile of specific brain regions, either by in vitro MS or high-resolution NMR following extraction, or by ex vivo HRMAS analysis of the intact tissue, as reported previously by some of the authors for validation of glucose accumulation in different regions of mouse GL261 tumors (Simões et al. NMRB 2010: https://doi.org/10.1002/nbm.1421). Importantly, even if we did have access to a focalized irradiator, such protocols for metabolic characterization would compromise tissue integrity and thus the histopathologic analysis performed in this study. 

      We do agree with the importance of end-point validation and therefore stress it more clearly in the revised manuscript (page 20, lines 14-16): “(…) mapping pathway fluxes alongside de novo concentrations (…) may be determinant for the longitudinal assessment of GBM progression, with end-point validation (…)”.

      (3) Genetic manipulation of key players in the metabolic pathways studied in this paper (glycolysis and mitochondrial oxidation) would offer a strong validation for the sensitivity of DGE-DMI in accurately distinguishing metabolites (lactate, glutamate-glutamine) and their dynamics.

      RA-R1.C3: Thank you for this comment, we agree. This would be particularly relevant in the context of treatment-response monitoring. While such models were not available to us (conditional spatio-temporal manipulation of metabolic pathway fluxes), we believe our results can still demonstrate this point: We previously used in vivo DGE 2H-MRS to show evidence of decreased glucose oxidation fraction (Vglx/Vlac) in GL261 tumors under acute hypoxia (FiO2=12 %) compared to regular anesthesia conditions (FiO2=31 %), consistent with the inhibition of OXPHOS due to lower oxygens tensions (Simoes et al. NIMG:Clin 2022). In the present work, enhanced glycolysis in tumors vs peritumoral brain regions was clearly observed in all the animals studied, from both cohorts, as shown in Fig 1-B and Fig S4. Moreover, the spectral background (before glucose injection) is limited to a single peak in all the voxels: basal DHO, used as internal reference for spatio-temporal quantification of glucose, glutamine-glutamate, and lactate, all de novo and extensively characterized in healthy and glioma-bearing rodent brain (Lu et al. JCBFM 2018; Zhang et al. NMR Biomed 2024, de Feyter et al. SciAdv 2018; Batsios et al ClinCancerRes 2022;  Simoes et al. NIMG:Clin 2022) and other rodent tumors (Kreis et al. Radiology 2020, Montrazi et al. SciRep 2023). We have modified the manuscript to clarify this point (page 18, lines 14-17) “(…) patient-derived xenografts and de novo models would be more suited to recapitulate human GBM heterogeneity and infiltration features, and genetic manipulation of glycolysis and mitochondrial oxidation pathways could be relevant to ascertain DGE-DMI sensitivity for their quantification (…)”.

      (4) Please explain more why DEG-DMI can distinguish different glucose metabolites and how accurate it is.

      RA-R1.C4: DGE-DMI is the imaging extension of our previous work based on single-voxel deuterium spectroscopy, therefore relying on the same fundamental technique and analysis pipeline but moving from a temporal analysis to a spatio-temporal analysis for each metabolite, and thus dealing with more data. Unlike conventional proton spectroscopy (1H), only metabolites carrying the deuterium label (2H) will be detected in this case, including the natural abundance DHO (~0.03%), the deuterated glucose injected and its metabolic derivatives, namely deuterated lactate and deuterated glutamate-glutamine. Due to their different molecular structures, the deuterium atoms will resonate at specific frequencies (chemical shifts, ppm) during a 2H magnetic resonance spectroscopy experiment, as illustrated in Fig 1-A. The method is fully reproducible and accurate, and has been extensively reported in the literature from high-resolution NMR spectroscopy to in vivo spectroscopic imaging of different nuclei, such as proton (1H), deuterium (2H), carbon (13C), phosphorous (31P), and fluorine (19F). Since the fundamental principles of DMI and its application to brain tumors have been very well described in the flagship article by de Feyter et al., we have now highlighted this in the manuscript: page 4 (lines 4-7), “Deuterium metabolic imaging (DMI) has been (…) demonstrated in GBM patients, with an extensive rationale of the technique and its clinical translation [18], and more recently in mouse models of patient-derived GBM subtypes (…)”.

      (5) When mapping glycolysis and mitochondrial oxidation fluxes, add a control method to compare the reliability of DEG-DMI.

      RA-R1.C5: This concern (“lack of a control method”) was also raised by the Reviewer in the section Public Reviews section, where we already address it (PR-R1.2).

      (6) If using peritumoral glutamate-glutamine recycling as a marker of invasion capacity, what would be the correct rate of the presence of secondary brain lesions?

      RA-R1.C6: While our results suggest the potential of peritumoral glutamate-glutamine recycling as a marker for the presence of secondary brain lesions, this remains to be ascertained with higher sensitivity for glutamate-glutamine detection. Therefore, we cannot make further conclusions in this regard.  

      To make this point clear, we state in different sections of the discussion: page 19 (lines 1-2), “(…) recycling of the glutamate-glutamine pool may reflect a phenotype associated with secondary brain lesions.”; and page 19 (lines 6-10), “Therefore, further DGE-DMI preclinical studies aimed at detecting and quantifying relatively weak signals, such as tumor glutamateglutamine, and/or increase spatial resolution to correlate those metabolic results with histology findings (e.g in the tumor margin), should improve basal SNR with higher magnetic field strengths, more sensitive RF coils, and advanced DMI pulse sequences [55]).”).  

      (7) There are duplicated Vlac in Figure S3 B.

      RA-R1.C7: This was a typo that has now been corrected. Thank you.

      (8) Figure 4, it would be better to add a metabolic map of a tumor without secondary brain lesions to compare.

      RA-R1.C8: We fully agree and have modified Fig 4 accordingly, together with its legend.

      Particularly, we have included tumors C4 (without secondary lesions) vs G4 (with) for this “comparison”, since details of their histology, including the secondary lesions, are provided in Fig 2.

      (9) Full name of SNR and FID should be listed when first mentioned.

      RA-R1.C9: Agreed and modified accordingly, on pages 6-7 (lines 22-1), ”signal-to-noise-ratio (SNR)”, and page 19 (lines 5-6), “free induction decay (FID)”.

      (10) Page 2, Line 14: (59{plus minus}7 mm3) is not needed in the abstract.

      RA-R1.C10: As requested we have removed this specification from the Abstract.

      (11) Page 4, Line 22: Closing out the Introduction section with a statement on broader implications of the present work would enhance the effectiveness of the section.

      RA-R1.C11: We have added an additional sentence in this regard – pages 4-5 (lines 24-2): “Since DMI is already performed in humans, including glioblastoma patients [18], DGE-DMI could be relevant to improve the metabolic mapping of the disease.”

      (12) Define all acronyms to facilitate comprehension. For example, principal component analysis (PCR) and signal-to-noise ratio (SNR).

      R1.C12: Thank you for the comment. We have now defined all the acronyms when first used, including PCA (page 4 (line 11), “Marcheku-Pastur Principal Component Analysis (MP-PCA)”) and SNR (pages 6-7 (lines 22-1), as indicated above in comment R1.9).

      (13) Some elements within the figures have lower resolution, specifically bar graphs.

      RA-R1.C13: We apologize for this oversight. All the Figures have been revised accordingly, to correct this problem. Thank you.

      (14) Page 13, Line 8: "underly" should be spelled "underlie."

      RA-R1.C14: The typo has been corrected on page 15 (line 8), thank you.

      (15) Page 14, Line 13: "better vascular permeability" would be more effectively phrased as "increased vascular permeability."

      RA-R1.C15: This has also been corrected on page 16 (line 14), thank you.

      Reviewer #2 (Recommendations For The Authors):

      (1) I strongly suggest adding a scale bar in the histology figures.

      RA-R2.C1: Thank you for spotting our oversight! This has now been added as requested to Fig 2.

      (2) The 2H MRSI data were acquired at a nominal resolution of 2.25 x 2.27 x 2.25 mm^3, resulting in a nominal voxel volume of 11.5 uL. (In reality, this is larger due to the point spread function leading to signal bleeding from adjacent voxels.) If we estimate the volume of the tumor rim, as indicated by the histology slides, as (generously) ~ 50 um in width, 3.2 mm long (the diagonal of a 2.25 x 2.25 mm^2 square, and 2.27 mm high, we get a volume of 0.36 uL. Therefore the native spatial resolution of the 2H MRSI is at least 30 times larger than the volume occupied by the tumor rim/microenvironment. Normal tissue and tumor tissue will contribute the majority of the metabolic signal of that voxel. I feel an opposite approach could have been pursued: find out the spatial resolution needed to characterize the tumor rim based on the histology, then use a de-noising method to bring the SNR of those data to be acceptable. (this is just a thought experiment that assumes de-noising actually works to improve quantification for MRS data instead of merely cosmetically improve the data, so far the jury is still out on that, in my view).

      RA-R2.C2 – We thank the Reviewer for the detailed analysis and reply below to each point.

      RA-R2.C2.1 – spatial resolution and tumor rim: Our nominal voxel volume was indeed 11.5 uL, defined in-plane by the PSF which explains signal bleeding effects, as in any other imaging modality. The DMI raw data were Fourier interpolated before reconstruction, rendering a final in-plane resolution of 0.56 mm (0.72 uL voxel volume). The tumor rim (margin) analyzed was roughly 0.1 mm width (please note, not 0.05 mm), as explained in the methods section (page 28, line 16) and now more clearly defined with the scale bars in Fig 2. According to the Reviewer’s analysis, this would correspond to 0.1*3.2*2.27 = 0.73 uL, which we approximated with 1 voxel (0.72 uL), as displayed in Fig 3-A. Importantly, it has long been demonstrated that Fourier interpolation provides a better approximation to the ground truth compared to the nominal resolution, and even to more standard image interpolation performed after FT - see for instance Vikhoff-Baaz B et al. (MRI 2001. 19: 1227-1234), now citied in the Methods section: page 24, line 24 ([69]). While we do agree that both normal brain and tumor should contribute significantly to the metabolic signal in this relatively small region, we rely on extensive literature to maintain that despite its smoothing effect, the display resolution provides a better approximation to the ground truth and is therefore more suited than the nominal resolution for ROI analysis in this region. Still, we acknowledge this potential limitation in the Discussion: page 19, lines 6-10: “Therefore, further DGE-DMI preclinical studies aimed at detecting and quantifying relatively weak signals, such as tumor glutamate-glutamine, and/or increase the nominal spatial resolution to better correlate those metabolic results with histology findings (e.g. in the tumor margin), should improve basal SNR with higher magnetic field strengths, more sensitive RF coils, and advanced DMI pulse sequences [55]).”

      RA-R2.C2.2 – metabolic and histologic features at the tumor rim: Furthermore, we also performed ROI analysis of lactate metabolic maps in tumor and peritumoral rim areas closely reflected regional differences in cellularity and cell density, and immune cell infiltration between the 2 tumor cohorts and across pooled cohorts, as explained in the Public Review section - PR-R1.1 – and now report in the manuscript: page 12 (lines 6-16), “GL261 tumors accumulated significantly less lactate in the core (1.60±0.25 vs 2.91±0.33 mM: -45%, p=0.013) and peritumor margin regions (0.94±0.09 vs 1.46±0.17 mM: -36%, p=0.025) than CT2A – Fig 3 A-B, Table S1. Consistently, tumor lactate accumulation correlated with tumor cellularity in pooled cohorts (R=0.74, p=0.014). Then, lower tumor lactate levels were associated with higher lactate elimination rate, k<sub>lac</sub> (0.11±0.1 vs 0.06±0.01 mM/min: +94%, p=0.006) – Fig 3B – which in turn correlated inversely with peritumoral margin infiltration of microglia/macrophages in pooled cohorts (R=-0.73, p=0.027) - Fig 3-C. Further analysis of Tumor/P-Margin metabolic ratios (Table S3) revealed: (i) +38% glucose (p=0.002) and -17% lactate (p=0.038) concentrations, and +55% higher lactate consumption rate (p=0.040) in the GL261 cohort; and (ii) lactate ratios across those regions reflected the respective cell density ratios in pooled cohorts (R=0.77, p=0.010) – Fig 3-C”; page 17 (lines 1-8), “Tumor vs peritumor border analyses further suggest that lactate metabolism reflects regional histologic differences: lactate accumulation mirrors cell density gradients between and across the two cohorts; whereas lactate consumption/elimination rate coarsely reflects cohort differences in cell proliferation, and inversely correlates with peritumoral infiltration by microglia/macrophages across both cohorts. This is consistent with GL261’s lower cell density and cohesiveness, more disrupted stromal-vascular phenotypes, and infiltrative growth pattern at the peritumor margin area, where less immune cell infiltration is detected and relatively lower cell division is expected [43]”.

      RA-R2.C2.3 – alternative method: Regarding the alternative method suggested by the Reviewer, we have tested a similar approach in another region (tumor) and it did not work, as explained the Discussion section (page 19, lines 5-6) and Fig S11. Essentially, Tensor PCA performance improves with the number of voxels and therefore limiting it to a subregion hinders the results. In any case, if we understand correctly, the Reviewer suggests a method to further interpolate our data in the spatial dimension, which would deviate even more from the original nominal resolution and thus sounds counter-intuitive based on the Reviewer’s initial comment about the latter. More importantly, we would like to remark the importance of spectral denoising in this work, questioned by the Reviewer. There are several methods reported in the literature, most of them demonstrated only for MRI. We previously demonstrated how MPPCA denoising objectively improved the quantification of DCE-2H MRS in mouse glioma by significantly reducing the CRLBs: 19% improved fitting precision. In the present study, Tensor PCA denoising was applied to DGE-DMI, which led to an objective 63% increase in pixel detection based on the quality criteria defined, unambiguously reflecting the improved quantification performance due to higher spectral quality. 

      (3) Concerns re. the metabolic model: 2g/kg of glucose infused over 120 minutes already leads to hyperglycemia in plasma. Here this same amount is infused over 30 seconds... such a supraphysiological dose could lead to changes in metabolite pool sizes -which are assumed to not change since they are not measured, and also fractional enrichment which is not measured at all. Such assumptions seem incompatible with the used infusion protocol.

      RA-R2.C3:  We understand the concern. However, the protocol was reproduced exactly as originally reported by Kreis et al (Radiology 2020) that performed the measurements in mice and measured the fraction of deuterium enrichment (f=0.6). Since we also worked with mice, we adopted the same value for our model. The total volume injected was 100uL/25g animal, and adjusted for animal weight (96uL/24g average – Table S1), as we reported before (Simões et al. NIMG:Clin 2022), which is standard for i.v. bolus administration in mice as it corresponds to ~10% of the total blood volume. This volume is therefore easily diluted and not expected to introduce significant changes in the metabolic pool sizes. Continuous infusion protocols on the other hand will administer higher volumes, easily approaching the mL range when performed over periods as large as 120 min. This would indeed be incompatible with our bolus infusion protocol. We have now clarified this in the manuscript – page 24 (line 23): “i.v. bolus of 6,6<sup>′2</sup>H<sub>2</sub>-glucose (2 mg/g, 4 µL/g injected over 30 s (…)”.

      (4) Vmax = Vlac + Vglx. This is incorrect: Vmax = Vlac.

      RA-R2.C4: Thank you for raising this concern. As indicated in RA-R2.C3, our model (Simões et al. NIMG:Clin 2022) was adapted from the original model proposed by Kreis et al. (Radiology 2020), where the authors quantified glycolysis kinetics on a subcutaneous mouse model of lymphoma, exclusively glycolytic and thus estimating the maximum glucose flux rate was from the lactate synthesis rate (Vmax = Vlac). However, we extended this model to account for glucose flux rates for lactate synthesis (Vlac) and also for glutamate-glutamine synthesis (Vglx), where Vmax = Vlac + Vglx, as explained in our 2022 paper. While we acknowledge the rather simplistic approach of our kinetic model compared to others - reported by 13C-MRS under continuous glucose infusion in healthy mouse brain (Lai et al. JCBFM 2018) and mouse glioma (Lai et al. IJC 2018) – and acknowledge this in the Discussion (page 20, lines 22-24: “(…) metabolic fluxes [estimations] through glycolysis and mitochondrial oxidation (…) could potentially benefit from an improved kinetic model simultaneously assessing cerebral glucose and oxygen metabolism, as recently demonstrated in the rat brain with a combination of 2H and 17O MR spectroscopy [62] (…)”), our Vlac and Vglx results are consistent with our previous DGE 2H-MRS findings in the same glioma models, and very aligned with the literature, as discussed in PR-R1.C2.1.

      (5) Some other items that need attention: 0.03 % is used as the value for the natural abundance of DHO. The natural abundance of 2H in water can vary somewhat regionally, but I have never seen this value reported. The highest seen is 0.015%.

      RA-R2.C5: The Reviewers is referring to the natural abundance of deuterium in hydrogen: 1 in ~6400 is D, i.e. 0.015 %. The 2 hydrogen atoms in a water molecule makes ~3200 DHO, i.e. 0.03%. Indeed the latter can have slight variations depending on the geographical region, as nicely reported by Ge et al (Front Oncol 2022), who showed a 16.35 mM natural-abundance of DHO in the local tap water of St Luis MO, USA (55500/16.35 = 1/3364 = 0.034%).

      (6) Based on the color scale bar in Figure 1, the HDO concentration appears to go as high as 30 mM. Even if this number is off because of the previous concern (HDO), it appears to be a doubling of the HDO concentration. Is this real? What would be the origin of that? No study using [6,6'-2H2]-glucose that I'm aware of has reported such an increase in HDO.

      RA-R2.C6: As explained before (RA-R2.C3 and RA-R2.C4), we based our protocol and model on Kreis et al (Radiology 2020), who reported ~10 mM basal DHO levels raising up to ~27 mM after 90min, which are well within the ~30 mM ranges we report over a longer period (132 min).

      Similar DHO levels were mapped with DGE-DMI in mouse pancreatic tumors (Montrazi et al. SciRep 2023).

      (7) "...the central spectral matrix region selected (to discard noise regions outside the brain, as well as the olfactory bulb and cerebellum)". This reads as if k-space points correspond one-toone with imaging pixels, which is not the case.

      RA-R2.C7: We rephrased the sentence to avoid such potential misinterpretation, specifically: page 25 (lines 19-21): “Each dataset was averaged to 12 min temporal resolution and the noise regions outside the brain, as well as the olfactory bulb and cerebellum, were discarded (…)”.

      (8) The use of the term "glutamate-glutamine recycling" is not really appropriate since these metabolites are not individually detected with 2H MRS, which is a requirement to measure this neurotransmitter cycling.

      RA-R2.C8: Thank you for this comment. To avoid this misinterpretation, we have now rephrased "glutamate-glutamine recycling" to “recycling of the glutamate-glutamine pool” in all the sentences, namely: page 2 (lines 14-15); page 15 (line 8); page 15 (line 8); page 19 (line 1); page 21 (line 10).

      Reviewer #3 (Recommendations For The Authors):

      (1) One major issue is the lack of underlying genetics, and therefore it is hard for readers to put the observed difference between GL261 and CT2A into context. The authors might consider perturbing the genetic and regulatory pathways on glycolysis and glutamine metabolism, repeating DGE DMI measure, in order to enhance the robustness of their findings.

      RA-R3.C1: We thank the reviewer for the helpful revision and comments. The point made here is aligned with Reviewer 1’s, addressed in RA-R1.C3; and also with our previous reply to the Reviewer, PR-R3.C1. Thus, we agree that conditional spatio-temporal manipulation of metabolic pathway fluxes would be relevant to further demonstrate the robustness of DGEDMI, particularly for treatment-response monitoring. While such models were not available to us, our previous findings seem compelling enough to demonstrate this point. Thus, we previously showed a significantly higher respiration buffer capacity and more efficient metabolic plasticity between glycolysis and mitochondrial oxidation in GL261 cells compared to CT2A (Simoes et al. NIMG:Clin 2022), which could enhance lactate recycling through mitochondrial metabolism in GL261 cells and thus explain our observations of increased glucose-derived lactate consumption rate in those tumors compared to CT2A. We have now included this in the discussion (page 17, lines 8-12): “our results suggest increased lactate consumption rate (active recycling) in GL261 tumors with higher vascular permeability, e.g. as a metabolic substrate for oxidative metabolism [44] promoting GBM cell survival and invasion [45], aligned with the higher respiration buffer capacity and more efficient metabolic plasticity of GL261 cells than CT2A [31].” Moreover, we previously showed evidence of DGE-2H MRS’ ability to detect decreased glucose oxidation fraction (Vglx/Vlac) in GL261 tumors under acute hypoxia (FiO2=12 %) compared to regular anesthesia conditions (FiO2=31 %), consistent with the inhibition of OXPHOS due to lower oxygens tensions (Simoes et al. NIMG:Clin 2022).

      (2) Is increased resolution possible for DGE DMI to correlate with histological findings?

      RA-R3.C2: The resolution achieved with DGE DMI, or any other MRI method, is limited by the signal-to-noise ratio (SNR), which in turn depends on the equipment (magnetic field strength and radiofrequency coil), the pulse sequence used, and post-processing steps such as noiseremoval. Thus, increased resolution could be achieved with higher magnetic field strengths, more sensitive RF coils, more advanced DMI pulse sequences, and improved methods for spectral denoising if available. We have used the best configuration available to us and discussed such limitations in the manuscript, including now a few modifications to address the Reviewer’s point more clearly – page 19 (lines 6-10): “Therefore, further DGE-DMI preclinical studies aimed at detecting and quantifying relatively weak signals, such as tumor glutamateglutamine, and/or increase the nominal spatial resolution to better correlate those metabolic results with histology findings (e.g in the tumor margin), should improve basal SNR with higher magnetic field strengths, more sensitive RF coils, and advanced DMI pulse sequences [55])”.

      (3) The authors might consider measuring the contribution of stromal cells and infiltrative immune cells in the analysis of DGE DMI data, to construct a more comprehensive picture of the microenvironment.

      RA-R3.C3: Thank you for this important point. We now added additional Iba-1 stainings of infiltrating microglia/macrophages, for each tumor, as suggested by the Reviewer; stromal cells would be more difficult to detect and we did not have access to a validated staining method for doing so. Our new data and results - now included in Fig 2B – indicate significantly higher levels of Iba-1 positive cells in CT2A tumors compared to GL261, which are particularly noticeable in the periphery of CT2A tumors and consistent with their better-defined margins and lower infiltration in the brain parenchyma. This has been explained more extensively in PRR1.1.

      (4) Additional GBM models with improved understanding of the genetic markers would serve as an optimal validation cohort to support the potential clinical translation.

      RA-R3.C4: We agree with the Reviewer and direct again to RA-R1.3, where we already addressed this suggestion in detail and introduced modifications to the manuscript accordingly.

    1. eLife Assessment

      Seminal plasma is a crucial component of semen that can affect sperm capacitation. However, the role of seminal plasma components, including fatty acids, in sperm function and fertility is poorly understood. In this important study, the authors provide a solid evidence of the testosterone-induced metabolic shift in the epithelial cells of seminal vesicle to support an fatty acid synthesis and also describe the potential effect of oleic acid on sperm motility.

    2. Reviewer #1 (Public review):

      Summary:

      In this revised report, Yamanaka and colleagues investigate a proposed mechanism by which testosterone modulates seminal plasma metabolites in mice. The authors identify oleic acid as a particularly important metabolite, derived from seminal vesicle epithelium, that stimulates linear progressive motility in isolated cauda epidydimal sperm in vitro. The authors provide additional experimental evidence of a testosterone dependent mechanism of oleic acid production by the seminal vesicle epithelium.

      Strengths:

      Often, reported epidydimal sperm from mice have lower percent progressive motility compared with sperm retrieved from the uterus or by comparison with human ejaculated sperm. The findings in this report may improve in vitro conditions to overcome this problem, as well as add important physiological context to the role of reproductive tract glandular secretions in modulating sperm behaviors. The strongest observations are related to the sensitivity of seminal vesicle epithelial cells to testosterone. The revisions include addition of methodological detail, modified language to reflect the nuance of some of the measurements, as well as re-performed experiments with more appropriate control groups. The findings are likely to be of general interest to the field by providing context for follow-on studies regarding the relationship between fatty acid beta oxidation and sperm motility pattern.

      Weaknesses:

      Support for the proposed mechanism is stronger in this revised report than in the previous report, but there are many challenges in measuring sperm metabolism and its direct relationship with motility patterns. This study is no exception and largely relies on correlations between various experiments in lieu of direct testing. Additionally, the discussion is framed from a human pre-clinical perspective, and it should be noted that the reproductive physiology between mice and humans is very different.

    3. Reviewer #2 (Public review):

      Using a combination of in vivo studies with testosterone-inhibited and aged mice with lower testosterone levels as well as isolated mouse and human seminal vesicle epithelial cells the authors show that testosterone induces an increase in glucose uptake. They find that testosterone induces a difference in gene expression with a focus on metabolic enzymes. Specifically, they identify increased expression of enzymes regulating cholesterol and fatty acid synthesis, leading to increased production of 18:1 oleic acid. The revised version strengthens the role of ACLY as the main regulator of seminal vesicle epithelial cell metabolic programming. 18:1 oleic acid is secreted by seminal vesicle epithelial cells and taken up by sperm, inducing an increase in mitochondrial respiration. The difference in sperm motility and in vivo fertilization in the presence of 18:1 oleic acid and the absence of testosterone, however, is small. Additional experiments should be included to further support that oleic acid positively affects sperm function.

    4. Author response:

      The following is the authors’ response to the original reviews.

      Public Reviews:

      Reviewer #1 (Public Review):

      Summary:

      In this report, the authors investigated the effects of reproductive secretions on sperm function in mice. The authors attempt to weave together an interesting mechanism whereby a testosterone-dependent shift in metabolic flux patterns in the seminal vesicle epithelium supports fatty acid synthesis, which they suggest is an essential component of seminal plasma that modulates sperm function by supporting linear motility patterns.

      Strengths:

      The topic is interesting and of general interest to the field. The study employs an impressive array of approaches to explore the relationship between mouse endocrine physiology and sperm function mediated by seminal components from various glandular secretions of the male reproductive tract.

      Thank you for your positive evaluation of our study's topic and approach. We are pleased that you found our investigation into the effects of reproductive secretions on sperm function to be of general interest to the field. We appreciate your positive feedback on the diverse methods we employed to explore this complex relationship.

      Weaknesses:

      Unfortunately, support for the proposed mechanism is not convincingly supported by the data, and the experimental design and methodology need more rigor and details, and the presence of numerous (uncontrolled) confounding variables in almost every experimental group significantly reduce confidence in the overall conclusions of the study.

      The methodological detail as described is insufficient to support replication of the work. Many of the statistical analyses are not appropriate for the apparent designs (e.g. t-tests without corrections for multiple comparisons). This is important because the notion that different seminal secretions will affect sperm function would likely have a different conclusion if the correct controls were selected for post hoc comparison. In addition, the HTF condition was not adjusted to match the protein concentrations of the secretion-containing media, likely resulting in viscosity differences as a major confounding factor on sperm motility patterns.

      We appreciate you highlighting concerns regarding our weak points and apologize for our unclear description. We revised the manuscript to be as rigorous and detailed as possible. In addition, some experimental designs were changed to simpler direct comparisons, and additional experiments were conducted (New Figure 1A-F, lines 103-113). We have made our explanations more consistent with the provided data, which includes further experimentation with additional controls and larger sample sizes to increase the reliability of the findings.

      To address the multiple testing problem, a multiple testing correction was made by making the statistical tests more stringent (Please see Statistical analysis in the Methods section and the Figure legends). Based on different statistical methods, the analysis results did not require significant revisions of the previous conclusions.

      Because the experiments on mixing extracts from the seminal vesicles were exploratory, we planned to avoid correcting for multiple comparisons. Repeating the t-test could lead to a Type I error in some results, so we apologize for not interpreting and annotating them. In the revised version, we removed the dataset for experiments on mixing extracts from the seminal vesicles and prostate, and we changed the description to refer to the clearer dataset mentioned above.

      The viscosity of the secretion-containing medium was measured with a viscometer, confirming that secretions did not significantly affect the viscosity of the solution. In addition, as the reviewer pointed out, we addressed the issue that the HTF condition could not be used as a control because of the heterogeneity in protein concentration (New Fig.1G, lines 110-111).

      Overall, we concluded that seminal vesicle secretion improves the linear motility of sperm more than prostate secretion.

      There is ambiguity in many of the measurements due to the lack of normalization (e.g. all Seahorse Analyzer measurements are unnormalized, making cell mass and uniformity a major confounder in these measurements). This would be less of a concern if basal respiration rates were consistently similar across conditions and there were sufficient independent samples, but this was not the case in most of the experiments.

      We apologize for the many ambiguities in the first manuscript. Cell culture experiments in the paper, including the flux analysis, were performed under conditions normalized or fixed by the number of viable cells. The description has also been revised to emphasize that the measurement values are standardized by cell count (lines 183-185, 189-190, 194-197). We emphasize that testosterone affects metabolism under the same number of viable cells (New Fig.4). This change in basal respiration is thought to be due to the shift in the metabolic pathway of seminal vesicle epithelial cells to a “non-normal TCA cycle” in which testosterone suppresses mitochondrial oxygen consumption, even under aerobic conditions (New Figs.3, 4, 5).

      The observation that oleic acid is physiologically relevant to sperm function is not strongly supported. The cellular uptake of 10-100uM labeled oleic acid is presumably due to the detergent effects of the oleic acid, and the authors only show functional data for nM concentrations of exogenous oleic acid. In addition, the effect sizes in the supporting data were not large enough to provide a high degree of confidence given the small sample sizes and ambiguity of the design regarding the number of biological and technical replicates in the extracellular flux analysis experiments.

      Thank you for your important critique. As you noted, the too-high oleic acid concentration did not reflect physiological conditions. Therefore, we changed the experimental design of an oleic acid uptake study and started again. We added an in vitro fertilization experiment corresponding to the functional data of exogenous oleic acid at nM concentrations (New Fig.7J,K, Lines 274-282).

      For the flux data to determine the effect of oleic acid on sperm metabolism, we have indicated in the text that the data were obtained based on eight male mice and two technical replicates. Pooled sperm isolated and cultured from multiple mice were placed in one well. The measurements were taken in three different wells, and each experiment was repeated four times. We did not use the extracellular flux analyzers XFe24 or XFe96. The measurements were also repeated because the XF HS Mini was used in an 8-well plate (only a maximum of 6 samples at a run since 2 wells were used for calibration).

      Overall, the most confident conclusion of the study was that testosterone affects the distribution of metabolic fluxes in a cultured human seminal vesicle epithelial cell line, although the physiological relevance of this observation is not clear.

      We thank the comments that this finding is one of the more robust conclusions of our study. Below we have written our thoughts on the physiological relevance of the observation results and our proposed revisions. In the mouse experiments, when the action of androgens was inhibited by flutamide, oleic acid was no longer synthesized in the seminal vesicles. The results of the experiments using cultured seminal vesicle epithelial cells showed that oleic acid was not being synthesized because of a change in metabolism dependent on testosterone. We have also added IVF data on the effects of oleic acid on sperm function (New Fig.7 and Supplementary Fig. 5, lines 274-282).<br /> As you can see, we have obtained consistent data in vitro and in vivo in mice. Our data also showed that the effects of testosterone on metabolic fluxes in vitro are similar in mouse and human seminal vesicle epithelial cells (New Fig.9). Therefore, it can be assumed that a decrease in testosterone levels causes abnormalities in the components of human semen. However, the conclusion was overestimated in the original manuscript, so we changed the wording as follows: It could be assumed that a decrease in testosterone levels causes abnormalities in the components of human semen. (lines 422-423)

      In the introduction, the authors suggest that their analyses "reveal the pathways by which seminal vesicles synthesize seminal plasma, ensure sperm fertility, and provide new therapeutic and preventive strategies for male infertility." These conclusions need stronger or more complete data to support them.

      We appreciate your comments about the suggestion presented in the introduction.

      We also removed our conclusions regarding treatment and prevention strategies for male infertility (lines 96-98). We wanted to discuss our findings not conclusively but as future applications that could result from further research based on our initial findings.

      The last sentence of the introduction has been revised to tone down these assertions as follows: These analyses revealed that testosterone promotes the synthesis of oleic acid in seminal vesicle epithelial cells and its secretion into seminal plasma, and the oleic acid ensures the linear motility and fertilization ability of sperm.

      We are grateful for your suggestions, which have prompted us to refine our manuscript.

      Reviewer #2 (Public Review):

      Summary:

      Using a combination of in vivo studies with testosterone-inhibited and aged mice with lower testosterone levels, as well as isolated mouse and human seminal vesicle epithelial cells, the authors show that testosterone induces an increase in glucose uptake. They find that testosterone induces differential gene expression with a focus on metabolic enzymes. Specifically, they identify increased expression of enzymes that regulate cholesterol and fatty acid synthesis, leading to increased production of 18:1 oleic acid.

      Strength:

      Oleic acid is secreted by seminal vesicle epithelial cells and taken up by sperm, inducing an increase in mitochondrial respiration. The difference in sperm motility and in vivo fertilization in the presence of 18:1 oleic acid and the absence of testosterone is small but significant, suggesting that the authors have identified one of the fertilization-supporting factors in seminal plasma.

      Thank you for your positive comments regarding our work on the role of testosterone in regulating metabolic enzymes and the subsequent production of 18:1 oleic acid in seminal vesicle epithelial cells. We are pleased that the strength of our findings, particularly identifying oleic acid as a factor influencing sperm motility and mitochondrial respiration, has been recognized.

      Weaknesses:

      Further studies are required to investigate the effect of other seminal vesicle components on sperm capacitation to support the author's conclusions. The author's experiments focused on potential testosterone-induced changes in the rate of seminal vesicle epithelial cell glycolysis and oxphos, however, provide conflicting results and a potential correlation with seminal vesicle epithelial cell proliferation should be confirmed by additional experiments.

      Thank you very much for your valuable criticism. Although we fully agree with your comment, conducting experiments to investigate the effects of other seminal vesicle components on the fertilization potential of sperm would be a great challenge for us. This is because it has taken us the last three years to identify oleic acid as a key factor in seminal plasma. We are considering a follow-up study to explore the effect of other seminal vesicle components on sperm capacitation. Therefore, we have revised the Introduction and conclusions to tone down our assertions .

      The revised manuscript also includes additional data showing a correlation between changes in metabolic flux and the proliferation of seminal vesicle epithelial cells using shRNA. As a result, it was shown that cell proliferation is promoted when mitochondrial oxidative phosphorylation is promoted by ACLY knockdown (New Fig.8D, lines 303-305). This shows a close relationship between the metabolic shift in seminal vesicle epithelial cells and cell proliferation. The revised manuscript includes an interpretation and discussion of these results (lines 369-379).

      We are grateful for your suggestions, which have prompted us to refine our manuscript.

      Reviewer #3 (Public Review):

      Summary:

      Male fertility depends on both sperm and seminal plasma, but the functional effect of seminal plasma on sperm has been relatively understudied. The authors investigate the testosterone-dependent synthesis of seminal plasma and identify oleic acid as a key factor in enhancing sperm fertility.

      Strengths:

      The evidence for changes in cell proliferation and metabolism of seminal vesicle epithelial cells and the identification of oleic acid as a key factor in seminal plasma is solid.

      Weaknesses:

      The evidence that oleic acids enhance sperm fertility in vivo needs more experimental support, as the main phenotypic effect in vitro provided by the authors remains simply as an increase in the linearity of sperm motility, which does not necessarily correlate with enhanced sperm fertility.

      We appreciate the positive feedback on the solid evidence of cell proliferation and metabolic changes in seminal vesicle epithelial cells and the identification of oleic acid as an important factor in seminal plasma. We fully agree with the assessment that the evidence linking oleic acid and increased sperm fertility in vivo needs further experimental support. To address this concern, we changed the experimental design of an oleic acid study and started again to be more physiological regarding the effect of oleic acid on fertility outcomes, increased the replicates of artificial insemination, and added in vitro fertilization assessments (New Fig.7 and supplementary Fig.5, lines 274-282). The revised manuscript describes these experiments and discusses the association between oleic acid and fertility.

      We are grateful for your suggestions, which have prompted us to refine our manuscript.

      Recommendations for the authors:

      Reviewing Editor's note:

      As you can see from the three reviewers' comments, the reviewers agree that this study can be potentially important if major concerns are adequately addressed. The major concern common to all the reviewers is the incomplete mechanistic link between the physiological androgen effect on the production of oleic acid and its effect on sperm function. Statistical analyses need more rigor and consideration of other important capacitation parameters are needed to address these concerns and to improve the manuscript to support the current conclusions.

      Thank you for summarizing the reviewers' feedback and for your insights regarding the major concerns raised. We appreciate the reviewers' understanding of the potential importance of our work and have addressed the issues highlighted to strengthen the manuscript. We believe these changes will improve the quality of the manuscript and provide a clearer and more complete understanding of the role of androgens and oleic acid in sperm function.

      Reviewer #1 (Recommendations For The Authors):

      The following comments are provided with the hope of aiding the authors in improving the alignment between the data and their interpretations.

      Thank you for allowing us to strengthen our manuscript with your valuable comments and queries. We have made our best efforts to reflect your feedback.

      Major Comments:

      (1) The methodological detail is not sufficient to reproduce the work. For example:<br /> a. Manufacturer protocols are referred to extensively. These protocols are neither curated nor version-controlled. Please consider describing the underlying components of the assays. If information is not available, please consider providing catalog numbers and lot numbers in the methods (if appropriate for journal style requirements).

      We appreciate this suggestion, which we believe is important to ensure reproducibility. We described the catalog number in our Methodology and included as much information as possible.

      b. Please consider describing the analyses in full, with consideration given to whether blinding was part of the design. For example- line 492: "apoptotic cells were quantified using ImageJ". How was this quantified? How were images pre-processed? Etc.

      Although blinding was not performed, experiments and analyses based on Fisher's three principles were conducted to eliminate bias (lines 549-552). In order to avoid false-positive or false-negative results, it is clearly stated that tissue sections treated with DNAse were used as positive controls, and tissue sections without TdT were used as negative controls for apoptosis. We have added detailed quantification methods (lines 544-546).

      c. Please consider providing versions of all acquisition and analysis software used.

      We have added software version information in Materials and Methods.

      (2) Please consider revisiting the statistical analyses. Many of the analyses don't seem appropriate for the design. For example, the use of a t-test with multiple comparisons for repeated measures design in Figure 2 and the use of t-test for two-factor design in Figure 8. etc.

      To address the multiple testing issues, the statistical methodology was changed to a more rigorous one. Details are given in the Statistical analysis in the Methods section and the Figure legends.

      (3) The increase in % LIN in Figure 1 may be confounded by differences in viscosity between HTF and the fluid secretion mixtures. For this reason, HTF may not be an appropriate control for the ANOVA post hoc analysis. HTF protein was not adjusted to the same concentration as the secretion mixtures, correct? Ultimately, it does not appear that there would be a significant statistical effect of the different fluid mixtures if appropriate statistical comparisons were made. This detracts from the notion that the secretions impact sperm function.

      (4) Figure 1, the statistical analysis in the legend suggests that the experiments were analyzed with a t-test. Were corrections made for multiple comparisons in B-D? An ANOVA would probably be more appropriate.

      We used a viscometer to measure the viscosity of a solution of prostate and seminal vesicle secretions adjusted to a protein concentration of 10 mg/mL. The results showed that the secretions did not cause any significant viscosity changes (New Fig.1G, Lines 110-111).

      As you pointed out, the protein levels in the HTF medium and the secretion mixture are not adjusted to the same concentration. In addition, the original manuscript was not a controlled experiment because the two factors, seminal vesicle and prostate extracts, were modified. Therefore, to investigate the effect of prostate and seminal vesicle secretions on sperm motility, we modified the experimental design to directly compare the effects of the two groups: seminal vesicle and prostate extracts (New Fig.1A-G, lines 101-113). To show the sperm quality used in this study, motility data from sperm cultured in the HTF medium are presented independently in New Supplemental Fig.1A.

      (5) Additionally in Figure 1, there is no baseline quality control data to show that there are no intrinsic differences between sperm sampled from the two treatment groups. So baseline differences in sperm quality/viability remain a potential confounder.

      We thank you for this important point. Epididymal sperm were collected from healthy mice. We recovered only the seminal vesicle secretions from the flutamide-treated mice to pursue its role in the accessory reproductive glands, since testosterone targets the testes and accessory reproductive organs. So, there was no qualitative difference between the epididymal sperm before treatment. Nevertheless, incubation with seminal vesicle secretion for one hour altered the sperm motility pattern and in vivo fertilization results. Sperm function was altered by seminal vesicle secretion in a short period of culture time. We apologize for the confusion, and we have revised the text and figure to carry a clearer message (lines 128-132).

      (6) Figure 1E, did the authors confirm that flutamide-treated mice had decreased serum androgens? How often were mice treated with flutamide? This is important because flutamide has a relatively short half-life and is rapidly metabolized to inert hydroxyflutamide.

      Serum testosterone levels were unchanged. Flutamide was administered every 24 hours for 7 consecutive days. Although there was no change in blood testosterone levels (New Supplemental Fig.1B), a decrease in the weight of the seminal vesicles, prostate, and epididymis was confirmed. This is thought to be due to the pharmacological activity of flutamide.

      (7) Figure 1H, the meaning of 'relative activity of mitochondria' isn't clear. JC-1 does not measure 'activity'. A decreased average voltage potential across the inner mitochondrial membrane may indicate that more of the sperm from the flutamide group were dead. Additionally, J-aggregates are slow to form, generally requiring long incubation periods of at least 90 minutes or more. Additional positive and negative controls for predictable mitochondrial transmembrane voltage potential polarization states would have improved the quality of this experiment.

      Thank you for pointing this out. We have replaced the relative activity of mitochondria with high mitochondrial membrane potential (New Fig.1M, lines 125-128). Actually, it is thought that the sperm cultured in seminal vesicle secretions from mice that had been administered flutamide died because the motility of the sperm was also significantly reduced. Since antimycin reduces mitochondrial membrane potential, we have added an experiment in which 10 µM antimycin-treated sperm were used as a control to confirm that the JC-1 reaction is sensitive to changes in membrane potential.

      (8) Figure 4, the extracellular flux data appear to be unnormalized. The Seahorse instruments are extremely sensitive to the mass and uniformity of the cells at the bottom of the well. This may be a significant confounder in these results. For example, all of the observed differences between groups could simply be a product of differential cell mass, which is in line with the reduced growth potential of testosterone-treated cells indicated by the authors in the results section.

      We thank you for this important point. After correcting for cell viability, we seeded the same number of viable cultured cells into wells between experimental groups before measuring them in the flux analyzer. There were no significant differences in survival rates in all experiments. As a result, an increase in glucose-induced ECAR and a suppression of mitochondrial respiration were observed. We would like to emphasize that this difference based on metabolic data does not imply a reduction in the growth potential of the cells due to testosterone treatment.

      We described that these measurements are normalized based on cell count and viability (lines 184, 190, 195).

      (9) How did the authors know that the isolated mouse primary cells were epithelial cells? Was this confirmed? What was the relative sample purity?

      The cells were labeled with multiple epithelial cell markers (cytokeratin) and confirmed using immunostaining and flow cytometry. The percentage of cells positive for epithelial cell markers was approximately 80%. A stromal cell marker (vimentin) was also used to confirm purity, but only a few percent of cells were positive. The contaminating cell type was considered to be mainly muscle cells because the gene expression levels of muscle cell markers verified by RNA-seq were relatively high.

      (10) It is misleading to include the lactate/pyruvate media measurements in the middle of the figure in Figure 4 D and E because it seems at first glance like these measurements were made in the seahorse media but they are completely unrelated. Additionally, these measures are not normalized and are sensitive to confounding differences in cell viability, seeding density, mass, etc.

      Thank you for pointing this out. We have placed the lactate and pyruvate measurement graphs after the flux data of ECAR. We noted that these measurements are normalized based on cell count and viability (lines 189-190). The doubling time of seminal vesicle epithelial cells was approximately 3 days, and testosterone inhibited cell proliferation. Therefore, the seeding concentration of cells was increased 4-fold in the testosterone-treated group compared to the control, and experiments were conducted to ensure that the confluency at the time of measurement after 7 days of culture was comparable between groups.

      (11) The flux analyzer assays sold by Agilent have many ambiguities and problems of interpretation. Unfortunately, Agilent's interest in marketing/sales has outpaced their interest in scientific rigor. Please consider revising some of the language regarding the measurements. For example, 'ATP production rate' is not directly measured. Rather, oligomycin-sensitive respiration rate is measured. The conversion of OCR to ATP production rate is an estimation that depends on complex assumptions often requiring additional testing and validation. The same is true for other ambiguous terms such as 'maximal respiration' referring to FCCP uncoupled respiration, and glycolytic rate- which is also not measured directly. If the authors are interested in a more detailed description of the problems with Agilent's interpretation of these assays please see the following reference (PMID: 34461088).

      Thank you for your critical criticism and thoughtful advice, as well as for sharing the excellent reference. We agree with you on the flux analyzer ambiguities and data interpretation problems. The description of the measured values has been revised as follows.

      We have replaced the “ATP production rate” with the “oligomycin-sensitive respiratory rate.” Similarly, we have replaced “maximal respiration” with “FCCP-induced unbound respiration.” (lines 197-202) We chose not to deal with the conversion of OCR to ATP production rate because it is outside the scope of interest in our study.

      Avoid using the term "glycolytic capacity". We use “Oligomycin-sensitive ECAR.” (line 186) We recognize that the ECARs measured in this study reflect experimental conditions and may not fully represent physiological glycolytic flux in vivo. So, the main section includes a data set of glucose uptake studies to emphasize the significance of the changes obtained with the flux analyzer assay. (New Fig.6, lines 230-254)

      Figure 6, it's not surprising to see the accumulation of labeled oleic acid in the cells, however, this does not mean that oleic acid is participating in normal metabolic processes. Oleic acid will have detergent effects at high (uM) concentrations. The observation that sperm 'take up' OA at 10-100 uM concentrations should also be validated against sperm function the health of the cells is very likely to be negatively impacted. Additionally, no apparent accumulation is noted in the fluorescence imaging at 1uM, but the authors insinuate that uptake occurs at low nM concentrations. The effects in Figure 6D-F are nominal at best and are likely a result of the small sample sizes.

      Thank you for your good suggestion. We agree with the reviewer that high concentrations of oleic acid had a detergent effect. To improve the consistency of functional data and observations, oleic acid uptake tests were performed under the same concentration range as the sperm motility tests (New Fig.7A-C). The oleic acid concentration at this time was calculated regarding the oleic acid concentration in seminal fluid recovered from mice as detected by GCMS to reflect in vivo conditions.

      Epididymal sperm were incubated with fluorescently labeled oleic acid and observed after quenching of extracellular fluorescence. Fluorescent signals were detected selectively in the midpiece of the sperm. The fluorescence intensity of sperm quantified by flow cytometry increased significantly in a dose-dependent manner (New Fig.7A-C, lines 261-264).

      Furthermore, increasing the sample size did not change the trend of the sperm motility data. Although the effect size of oleic acid on sperm motility was small (New Fig.7D-G, lines 265-268), an improvement in fertilization ability was observed both in vitro (IVF) and in vivo (AI) (New Fig.7J-L, lines 274-282, 286-291). We conclude that the effect of oleic acid on sperm is of substantial significance. These data and interpretations have been revised in the text in the Results section.

      (12) Figure 6H, I applaud the authors for attempting intrauterine insemination experiments to test their previous findings. That said, there is no supporting data included to show that the sperm from the treatment groups had comparable starting viability/quality. Additionally, it is difficult to tell if the results are due to the small sample sizes and particularly the apparent outlier in the flutamide-only group.

      Thanks for the praise and comments for improvement. As we answered in your comment #5 above, the epididymal sperm was collected from healthy mice. Therefore, there is no qualitative difference in the epididymal sperm before treatment. This is described in the figure legend (lines 1130-1131). We apologize again for this complication. We also more than doubled the number of replications of the experiment. The impact of the outlier would have been minimal.

      (13) One final question related to Figure 6H: how did the authors know they were retrieving all of the possible 2-cell embryos from the uterus? Perhaps the authors could provide the raw counts of unfertilized eggs and 2-cell embryos so we can see if there were differences between the mice.

      We retrieved the pronuclear stage embryos from the fallopian tubes. It is not certain whether all embryos were recovered. Therefore, we added the number of embryos in the graph and in the supplementary data.

      (14) Figure 7 has the same seahorse assay normalization problem as mentioned earlier. Without normalization, it is difficult to tell if the effects are simply due to differences in cell mass. Were the replicates indicated in the graphs run on the same plate? If so, it would be much more convincing to see a nested design, with technical replicates within plates, and additional replicates run on separate plates.

      As we answered in your comment #8 above, these measurements were normalized based on sperm count. This has been corrected to be noted in the text and the figure legend (lines 1123-1124).

      Pooled sperm isolated and cultured from multiple mice were placed in one well. The measurements were taken in three different wells, and each experiment was repeated four times. We did not use the extracellular flux analyzers XFe24 or XFe96. The measurements were also repeated because the XF HS Mini was used in an 8-well plate (only a maximum of 6 samples at a run since 2 wells were used for calibration).

      (15) The statistical test in Figures 8E and F described in the legend is inappropriate (t-test), this appears to be a two-factor design.

      Thank you for pointing this out. Differences between groups were assessed using a two-way analysis of variance (ANOVA). When the two-way ANOVA was significant, differences among values were analyzed using Tukey's honest significant difference test for multiple comparisons.

      (16) The data in Figure 8 are interesting, and the effects appear to be a little more consistent compared with the mouse primary cells, potentially due to cell uniformity. However, the data are unnormalized, causing significant ambiguity, and there are no measures of cell viability to determine if the effects are due to cell death (or at least relative cell mass).

      As we answered in your comments #8 and #14 above, these measurements were normalized based on cell count and viability. This has been corrected to be noted in the figure legend (lines 1185-1186).

      Minor Comments:

      (1) The section title indicating the beginning of the results section is missing.

      A section title has been added to indicate the beginning of the results section.

      (2) There were several typos and confusingly worded statements throughout. Please consider additional editing.

      We used a proofreading service and corrected as much as possible.

      (3) In the introduction, a brief description of seminal fluid physiology is provided, but the reference is directed toward human physiology. Given that the research is performed solely in the mouse, a brief comparative description of mouse physiology would be helpful. For example, what is the role of mouse seminal fluid in the formation of the mating plug? What are the implications of the relative size disparity in seminal vesicles in mice versus humans? Etc.

      The third paragraph of the introduction has been revised (lines 57-60).

      Reviewer #2 (Recommendations For The Authors):

      Thank you for allowing us to strengthen our manuscript with your valuable comments and queries. We have made our best efforts to reflect your feedback.

      (1) The abstract is confusing and partly misleading and should be revised to more clearly and accurately summarize the study.

      The abstract was revised to be clearer and more accurate (lines 20-34).

      (2) The introduction should be revised to more accurately describe the sperm life cycle. Spermatogenesis, per definition, for example, exclusively takes place in the testis, sperm do not gain fertilization competence in the epididymis, sperm isolated from the epididymis cannot fertilize an oocyte unless in vitro capacitated, etc. In the last paragraph the connection between changes in fructose and citrate concentration, sperm metabolism and testicular-derived testosterone and AR remain unclear.

      The introduction was revised to be clearer and more accurate (lines 44-45).

      Citric acid and fructose are chemical components that are the subject of biochemical testing and are commonly used as semen testing items for humans and livestock. This is because the secretory function of the prostate and seminal vesicles is dependent on androgens. The measurement of citric acid and fructose concentrations in semen is routinely used to indicate testicular androgen production function (ISBN: 978-1-4471-1300-3, 978 92 4 0030787).

      (3) Throughout the manuscript the concept of (in vitro) capacitation is missing. Mixing sperm with seminal plasma is not the only way to achieve sperm that can fertilize the oocyte. Since media containing bicarbonate and albumin is the standard procedure in the field to capacitate epididymal mouse sperm rein vitro, the manuscript would gain value from a comparison between the effect of seminal plasma and in vitro capacitating media. Interesting readouts in addition to motility would i.e. be sAC activation, PKA-substrate phosphorylation, and acrosomal exocytosis.

      Thank you for pointing out this important point. As the reviewer points out, fertilization can be achieved in artificial insemination and in vitro fertilization using epididymal sperm which have not been exposed to seminal plasma. This has historically led to an underestimation of the role of accessory reproductive glands, such as the prostate and seminal vesicles. However, it has been reported that the removal of seminal vesicles in rodents decreases the fertilization rate after natural mating. This has been shown to be due to multiple factors affecting sperm motility rather than factors involved in plug formation (PMID: 3397934), but details of these factors and the whole picture of the role of the accessory glands were not known. This led us to become interested in the effects of sperm plasma on sperm other than fertilization and led us to begin research on the role of the accessory glands that synthesize sperm plasma.

      Early in our study, we found that simply exposing sperm to seminal vesicle extracts for 1 hour before IVF dramatically reduced fertilization rates, even in HTF medium containing bicarbonate and albumin. The experiment was designed on the assumption that seminal plasma contains factors that inhibit sperm from acquiring fertilizing ability. Therefore, we conducted experiments using modified HTF without albumin to avoid unintended motility patterns.

      However, we also respect the reviewer's opinion, and we have added our preliminary data related to IVF (New supplementary Fig.5).

      (4) In the introduction and throughout the manuscript it is unclear what the authors mean by "linear motility". An increase in VSL doesn't mean that the sperm swim in a more linear or straight way, or even that the sperm are 'straightened', it means that they swim faster from point A to point B. Do the authors mean progressive or hyperactivated motility? Please clarify.

      For all conditions tested the authors should follow the standard in the field and include the % of motile, progressively motile, and hyperactivated sperm.

      Thank you for pointing this out. We appreciate your feedback regarding the terminology. In our manuscript, "linear motility" refers to the degree to which sperm move in a straight line. We have clarified this by explaining that VSL (Straight-Line Velocity) and LIN (Linearity) are used to quantify and describe linear motility in sperm analysis: Higher VSL values indicate more direct, linear movement. A higher LIN value indicates a straighter path, thus representing greater linear motility. These terms have been standardized, and explanations have been added to the main text (lines 111-113).

      In response to your suggestion, we have included the percentage of motility and progressive motility for all conditions tested. However, since the experiment was performed using modified HTF without albumin, we have decided not to report the percentage of hyperactivation to avoid confusion.

      (5) Did the authors confirm that the injection of flutamide decreases androgen levels? That control needs to be included in the experiment to validate the conclusion.

      Injection of flutamide did not reduce androgen levels (see reviewer #1, comment 6). This is because flutamide's mechanism of action is based on antagonizing androgen and inhibiting its binding to the androgen receptor (New Fig.2A).

      (6) The role of mitochondrial activity in sperm progressive motility is still under investigation. PMID: 37440924 i.e. showed that inhibition of the ETC does not affect progressive but hyperactivated motility. The authors should either include additional experiments to confirm the correlation between mitochondrial activity and sperm progressive motility or tone down that conclusion.

      We have previously shown that treatment with D-chloramphenicol, an inhibitor of mitochondrial translation, significantly reduced sperm mitochondrial membrane potential, ATP levels, and linear motility (PMID: 31212063). Also, in the previous manuscript, we did not address progressive motility or hyperactivated motility in our analysis. We have chosen to discuss the effect of mitochondrial activity on linear motility rather than on progressive motility and hyperactivation of sperm.

      Was mitochondrial activity also altered in epididymal sperm incubated with and without seminal plasma or in aged mice?

      The mitochondrial membrane potential of epididymal sperm cultured with seminal vesicle extract (SV) was higher than that of epididymal sperm cultured without seminal vesicle extract (without SV: 67.3 ± 0.8%, with SV: 83.4 ± 1.8%). On the other hand, the mitochondrial membrane potential of epididymal sperm cultured with seminal vesicle extract recovered from aged mice was decreased (SV from aged: 60.3 ± 2.7%). It should be noted that the epididymal spermatozoa used in these experiments were healthy individuals, different from those from which seminal vesicle extracts were collected. (See also the response to reviewer 1's comment #5.)

      (7) The quality of the provided images showing AR, Ki67, and TUNEL staining should be improved or additional images should be included. Especially the AR staining is hard to detect in the provided images. The authors should also include a co-staining between AR and vesicle epithelial cells. That epithelial cells are multilayered does not come across in the pictures provided.

      We apologize for any inconvenience caused. The image has been replaced with one of higher resolution. The multilayered structure of the epithelial cells will also be seen.

      For the 12-month-old mice, an age-matched control should be included to support the authors' conclusion.

      To clarify the seminal vesicle changes associated with aging, we included images of 3-month-old mice as controls (New Supplementary Fig.2D).

      Overall, the rationale for the experiment does not become clear. How are the amount of seminal vesicle epithelial cells, testosterone, and AR expression connected to seminal plasma secretions? Why is it a disadvantage to have proliferating seminal vesicle epithelial cells? How is proliferation connected to the proposed switch in metabolic pathway activity?

      We have added some explanations and supporting data to the manuscript (New Fig.8D, lines 303-305, 315-319, 369-379). Cell proliferation stopped when the metabolic shift occurred, redirecting glucose toward fatty acid synthesis. Fatty acid synthesis is an important function of the seminal vesicle, and in the presence of testosterone, fatty acid synthesis enhancement and arrest of proliferation occur simultaneously. The connection between metabolism and cell proliferation was further demonstrated when ACLY was knocked down by shRNA, which stopped fatty acid synthesis and released the proliferative arrest induced by testosterone, allowing the cells to proliferate again. However, we do not know what effects occur when cell proliferation is stopped.

      (8) The experiments provided for glycolysis and oxphos are inconsistent and insufficient to support the authors' conclusion that testosterone shifts glycolytic and oxphos activity of seminal vesicle epithelial cells. Multiple groups (PMID 37440924, 37655160, 32823893) have shown that the increased flux through central carbon metabolism during capacitation is accompanied by an accumulation of intracellular lactate and increased secretion of lactate into the surrounding media. How do the authors explain that they see an increase in glucose uptake and ECAR but not in lactate and a decrease in pyruvate? Did the authors additionally quantify intracellular pyruvate and lactate? Since pyruvate and lactate are in constant equilibrium, it is odd that one metabolite is changing and the other one is not.

      Thank you for pointing this out. Since ECAR is often used as an alternative to lactate production but does not directly measure lactate levels, we measured changes in lactate and pyruvate concentrations in the culture medium. Under our experimental conditions, glucose appeared to be directed primarily towards anabolic processes, such as fatty acid synthesis, rather than the OXPHOS pathway, which may explain the lack of lactate production. The observed decrease in pyruvate might indicate its conversion to acetyl-CoA in the mitochondria, supporting both fatty acid synthesis and the TCA cycle. This shift would be consistent with the metabolic reprogramming toward anabolic activity.

      What do the authors mean by "the glycolytic pathway was not enhanced despite the activation of glycolysis" Seahorse, especially using a series of pathway inhibitors, only provides an indirect measurement of glycolysis and oxphos since the instrument does not provide a distinction from which pathways the detected protons are originating. The authors should consider a more optimized experimental design, i.e. the authors could monitor ECAR and OCR in the presence of glucose over time with and without the addition of testosterone. That would be less invasive since the sperm are not starved at the beginning of the experiment and would provide a more direct read-out. Did the authors normalize cell numbers in their experiment? Alternatively, the authors could consider performing metabolomics experiments.

      I agree with the reviewer. Buzzwords such as “glycolytic capacity” simply do not make sense, so we have removed them from the phrases noted by the reviewer. Please refer to the response to some of reviewer 1's points regarding the ambiguity of the data measured by the flux analyzer. Nevertheless, the assay design of the flux analysis could be used as a good “starting point” and provide information on the glycolytic system and respiratory control. Therefore, the interpretation of the flux analysis is supported by subsequent data sets.

      (9) The authors would strengthen their results by confirming their gene expression data by quantifying the expression of the respective proteins.

      Does testosterone treatment increase GLUT4 protein levels in isolated seminal vesicle epithelial cells? Or does it change the localization of the transporter? Are GLUT4 gene and protein levels altered in flutamide-treated cells? How do the authors explain that testosterone increases glucose uptake without changing Glut gene expression?

      We performed Western blot analysis to measure GLUT4 protein levels in seminal vesicle epithelial cells after testosterone treatment. The results showed that testosterone does not alter the expression of GLUT4 protein but simply changes its subcellular localization (New Fig.6C,D, lines 238-244).

      The discussion includes the interpretation of the observation that testosterone increases glucose uptake by altering localization without altering GLUT4 gene expression, a phenomenon commonly seen in other cells, such as cardiomyocytes (lines 362-364). The revised main figure also includes a data set of changes in GLUT4 localization, including flutamide-treated data. See also Reviewer 3's main comment #1.

      (10) Considering that the authors claim that SV secretions are crucial for sperm fertilization capacity, how do they explain that fertilization rates are still at 40 % when sperm are treated with flutamide?

      It is actually about 50% fertilized with HTF because it is fertilized without SV. Considering this baseline, we found that seminal vesicle secretions positively affect sperm in vivo fertilization. On the other hand, seminal plasma from flutamide-treated mice reduced the fertilization ability of healthy sperm. These are described in the text (lines 283-294).

      (11) It would be beneficial for the reader to include a schematic summarizing the results.

      Thank you for your advice from the reader's point of view. We have visualized the summaries of this study and added them to the manuscript (New Fig.10).

      Minor comments:

      Line 38: Male fertility, no article, please revise.

      I have changed “The male fertility” to “Male fertility” and added some references (lines 42-43).

      Line 55: Seminal plasma or TGFb? Please clarify.

      Corrected as follows. “TGFβ, a component of seminal plasma, increases antigen-specific Treg cells in the uterus of mice and humans, which induces immune tolerance, resulting in pregnancy.” (lines 60-62)

      Line 63: Why do the authors find it surprising that blood and seminal plasma have different compositions?

      This is because seminal plasma contains unique biochemical components that are not normally found in blood or only in small quantities. The intention was to emphasize the unique function of seminal plasma in supporting the physiological functions of sperm and to highlight its complex role by comparing it to blood. We clarified these intentions and reflected them in the revised text (lines 62-67).

      Line 94: The headline causes confusion. Seminal plasma does not induce sperm motility, it increases progressive sperm motility.

      Corrected as follows. “The effect of androgen-dependent changes in mouse seminal vesicle secretions on the linear motility of sperm” (lines 101-102)

      Reviewer #3 (Recommendations For The Authors):

      Thank you for allowing us to strengthen our manuscript with your valuable comments and queries. We have made our best efforts to reflect your feedback.

      Major:

      Figure 4 and Figure 5: The trend shows that GLUT3 is up-regulated and GLUT4 is downregulated although both of them are not statistically significant. However, GLUT4 is picked for all the following experiments based on protein localization. Providing other evidence/discussion why not to further consider other GLUTs will help to justify. Also, this reviewer suggests including GLUT4 localization data in the main figure as it is important data for the logical flow to link the following figures.

      We focused on GLUT4 because it was known that testosterone increases glucose uptake by changing the localization of GLUT4 without changing its expression (lines 230-231). In the revised manuscript, the increasing trend in Glut3 gene expression was also mentioned in the discussion, in addition to GLUT4 (lines 360-362). In any case, the results showed that testosterone increased glucose uptake by regulating the function of glucose transporters.

      Immunostaining of GLUT1~4 was performed to compare seminal vesicles from flutamide-treated mice with controls, and localization changes were observed only in GLUT4. Therefore, we hypothesized that GLUT4 is regulated by testosterone and performed the experiment. Fortunately, we were able to obtain a GLUT4-specific inhibitor, which dramatically inhibited the testosterone-dependent glucose uptake and subsequent lipid synthesis in seminal epithelial cells, leading us to believe that GLUT4 is a major glucose transporter.

      Increasing sperm linearity by oleic acid is observed and interpreted as enhanced sperm fertilizing potential. It is not clear why and how sperm linearity can be a determinant factor for enhancing sperm fertility in vivo. Providing an explanation of the effect of oleic acid on another key motility parameter more proven to be directly correlated with fertility (i.e., hyperactivation), and more direct evidence of oleic acid on enhancing sperm linearity indeed increasing sperm fertilization using IVF, is strongly recommended to support the author's main conclusion.

      Thank you for pointing this out. It is known that proteins derived from the seminal vesicles inhibit the hyperactivation of sperm and the acrosome reaction. Therefore, we conducted an experiment to add oleic acid, focusing on fatty acid synthesis caused by the metabolic shift of the seminal vesicles, which had not been known until now.

      Sperm were pretreated with an oleic acid-containing medium before IVF and oleic acid enhanced sperm linearity. When the sperm number was sufficient, there was no change in the cleavage rate after in vitro fertilization, but when the sperm count was reduced to one-tenth of the normal, the cleavage rate increased compared to the control (lines 274-282). In other words, the physiological role of oleic acid is to increase the probability of fertilization by keeping the sperm motility pattern linear or progressive. This increases the likelihood of the sperm passing through the female reproductive tract and environments that are unfavorable to sperm survival. Our research has uncovered significant insights into the role of seminal vesicle fluid and oleic acid in sperm fertilization. Due to the strong effect of the decapacitation factor, we found that seminal vesicle fluid reduces the fertilization rate in IVF. However, it does not interfere with the fertilization rate in in vivo during artificial insemination. This emphasizes the importance of oleic acid, along with other protein components of seminal plasma, in ensuring the in vivo fertilization ability of sperm.

      Minor:

      Please correct a typo in Line 173: sifts to shifts

      All typographical errors have been corrected.

    1. eLife Assessment

      This important study addresses the idea that defective lysosomal clearance might be causal to renal dysfunction in cystinosis. With mostly solid data, the authors observe that restoring expression of vATPase subunits and treatment with Astaxanthin ameliorate mitochondrial function in a model of renal epithelial cells, opening opportunities for translational application to humans.

    2. Reviewer #3 (Public review):

      Summary:

      In this manuscript, Sur and colleagues present insights into the potential pathways and mechanisms underlying the pathogenesis of cystinosis - a prototypical lysosomal storage disorder caused by the loss of the cystine transporter cystinosin (CTNS). This deficiency results in early dysfunction of proximal tubule (PT) cells and proximal tubulopathy, which progresses to chronic kidney disease and multisystem complications later in life. The authors utilize patient-derived cell lines and knockout (KO) strategies in immortalized PT cell systems, alongside transcriptomics and pathway enrichment analyses, to demonstrate that the loss of CTNS function reduces V-ATPase subunits (specifically V-ATP6V0A1), impairing autophagy and mitochondrial homeostasis. These findings are consistent with their prior work and follow-up studies conducted in preclinical models (mouse, rat, and zebrafish) of cystinosis and CTNS deficiency.

      Importantly, the authors highlight rescue strategies that involve correcting V-ATP6V0A1 expression or modulating redox dyshomeostasis through ATX treatment. These interventions restore cellular homeostasis in patient-derived cells, providing actionable therapeutic targets for patients in need of novel causal therapies.

      Strengths:

      The implications for health, disease, and therapeutic discovery are considerable, given the central role of autophagy and lysosome-related pathways in regulating critical cellular processes and physiological functions.

      Weaknesses:

      Despite these promising findings, further experimental research is required to strengthen the study's framework and conclusions. This includes characterizing the physiological properties of the PT cellular systems used, performing appropriate control or sentinel experiments in lysosome function assays, and further delineating disease phenotypes associated with cystinosis. Follow-up investigations into lysosome abnormalities and autophagy dysfunctions are also needed, along with a detailed exploration of the molecular mechanisms underlying the rescue of lysosomal, autophagic, and mitochondrial phenotypes through ATX treatment.

    3. Author response:

      The following is the authors’ response to the original reviews.

      Public Reviews:

      Reviewer #1 (Public Review):

      Cystinosis is a rare hereditary disease caused by biallelic loss of the CTNS gene, encoding two cystinosin protein isoforms; the main isoform is expressed in lysosomal membranes where it mediates cystine efflux whereas the minor isoform is expressed at the plasma membrane and in other subcellular organelles. Sur et al proceed from the assumption that the pathways driving the cystinosis phenotype in the kidney might be identified by comparing the transcriptome profiles of normal vs CTNS-mutant proximal tubular cell lines. They argue that key transcriptional disturbances in mutant kidney cells might not be present in non-renal cells such as CTNS-mutant fibroblasts.

      Using cluster analysis of the transcriptomes, the authors selected a single vacuolar H+ATPase (ATP6VOA1) for further study, asserting that it was the "most significantly downregulated" vacuolar H+ATPase (about 58% of control) among a group of similarly downregulated H+ATPases. They then showed that exogenous ATP6VOA1 improved CTNS(-/-) RPTEC mitochondrial respiratory chain function and decreased autophagosome LC3-II accumulation, characteristic of cystinosis. The authors then treated mutant RPTECs with 3 "antioxidant" drugs, cysteamine, vitamin E, and astaxanthin (ATX). ATX (but not the other two antioxidant drugs) appeared to improve ATP6VOA1 expression, LC3-II accumulation, and mitochondrial membrane potential. Respiratory chain function was not studied. RTPC cystine accumulation was not studied.

      In this manuscript, as an initial step, we have studied the first step in respiratory chain function by performing the Seahorse Mito Stress Test to demonstrate that the genetic manipulation (knocking out the CTNS gene and plasmid-mediated expression correction of ATP6V0A1) impacts mitochondrial energetics. We did not investigate the respirometry-based assays that can identify locations of electron transport deficiency, which we plan to address in a follow-up paper.

      We would like to draw attention to Figure 3D, where cystine accumulation has been studied. This figure demonstrates an increased intracellular accumulation of cystine.

      The major strengths of this manuscript reside in its two primary findings.

      (1) Plasmid expression of exogenous ATP6VOA1 improves mitochondrial integrity and reduces aberrant autophagosome accumulation.

      (2) Astaxanthin partially restores suboptimal endogenous ATP6VOA1 expression.

      Taken together, these observations suggest that astaxanthin might constitute a novel therapeutic strategy to ameliorate defective mitochondrial function and lysosomal clearance of autophagosomes in the cystinotic kidney. This might act synergistically with the current therapy (oral cysteamine) which facilitates defective cystine efflux from the lysosome.

      There are, however, several weaknesses in the manuscript.

      (1) The reductive approach that led from transcriptional profiling to focus on ATP6VOA1 is not transparent and weakens the argument that potential therapies should focus on correction of this one molecule vs the other H+ ATPase transcripts that were equally reduced - or transcripts among the 1925 belonging to at least 11 pathways disturbed in mutant RPTECs.

      The transcriptional profiling studies on ATP6V0A1 have been fully discussed and publicly shared. Table 2 lists the v-ATPase transcripts that are significantly downregulated in cystinosis RPTECs. We have also clarified and justified the choice of further studies on ATP6V0A1, where we state the following: "The most significantly perturbed member of the V-ATPase gene family found to be downregulated in cystinosis RPTECs is ATP6V0A1 (Table 2). Therefore, further attention was focused on characterizing the role of this particular gene in a human in vitro model of cystinosis."

      (2) A precise description of primary results is missing -- the Results section is preceded by or mixed with extensive speculation. This makes it difficult to dissect valid conclusions from those derived from less informative experiments (eg data on CDME loading, data on whole-cell pH instead of lysosomal pH, etc).

      We appreciate the reviewer highlighting areas for further improving the manuscript's readership. In our resubmission, we have revised the results section to provide a more precise description of the primary findings and restrict the inferences to the discussion section only.

      (3) Data on experimental approaches that turned out to be uninformative (eg CDME loading, or data on whole=cell pH assessment with BCECF).

      We have provided data whether it was informative or uninformative. Though lysosome-specific pH measurement would be important to measure, it was not possible to do it in our cells as they were very sick and the assay did not work. Hence we provide data on pH assessment with BCECF, which measures overall cytoplasmic and organelle pH, which is also informative for whole cell pH that is an overall pH of organelle pH and cytoplasmic pH.

      (4) The rationale for the study of ATX is unclear and the mechanism by which it improves mitochondrial integrity and autophagosome accumulation is not explored (but does not appear to depend on its anti-oxidant properties).

      We have provided rationale for the study of ATX; provided in the introduction and result section, where we mentioned the following: “correction of ATP6V0A1 in CTNS-/- RPTECs and treatment with antioxidants specifically, astaxanthin (ATX) increased the production of cellular ATP6V0A1, identified from a custom FDA-drug database generated by our group, partially rescued the nephropathic RPTEC phenotype. ATX is a xanthophyll carotenoid occurring in a wide variety of organisms. ATX is reported to have the highest known antioxidant activity and has proven to have various anti-inflammatory, anti-tumoral, immunomodulatory, anti-cancer, and cytoprotective activities both in vivo and in vitro_”._

      We are still investigating the mechanism by which ATX improves mitochondrial integrity, and this will be the focus of a follow-on manuscript.

      (5) Thoughtful discussion on the lack of effect of ATP6VOA1 correction on cystine efflux from the lysosome is warranted, since this is presumably sensitive to intralysosomal pH.

      In the revised manuscript, we have included a detailed discussion on the plausible reasons why ATP6V0A1 correction has no effect on cysteine efflux from the lysosome. We have now added to the Discussion – “However, correcting ATP6V0A1 had no effect on cellular cystine levels, likely because cystinosin is known to have multiple roles beyond cystine transport Cystinosin is demonstrated to be crucial for activating mTORC1 signaling by directly interacting with v-ATPases and other mTORC1 activators. Cystine depletion using cysteamine does not affect mTORC1 signaling. Our data, along with these observations, further supports that cystinosin has multiple functions and that its cystine transport activity is not mediated by ATP6V0A1.”

      (6) Comparisons between RPTECs and fibroblasts cannot take into account the effects of immortalization on cell phenotype (not performed in fibroblasts).

      The purpose of examining different tissue sources of primary cells in nephropathic cystinosis was to assess if any of the changes in these cells were tissue source specific. We used primary cells isolated from patients with nephropathic cystinosis—RPTECs from patients' urine and fibroblasts from patients' skin—these cells are not immortalized and can therefore be compared. This is noted in the results section - “Specific transcriptional signatures are observed in cystinotic skin-fibroblasts and RPTECs obtained from the same individual with cystinosis versus their healthy counterparts”.

      We next utilized the immortalized RPTEC cell line to create CRISPR-mediated CTNS knockout RPTECs as a resource for studying the pathophysiology of cystinosis. These cells were not compared to the primary fibroblasts.

      (7) This work will be of interest to the research community but is self-described as a pilot study. It remains to be clarified whether transient transfection of RPTECs with other H+ATPases could achieve results comparable to ATP6VOA1. Some insight into the mechanism by which ATX exerts its effects on RPTECs is needed to understand its potential for the treatment of cystinosis.

      In future studies we will further investigate the effect of ATX on RPTECs for treatment of cystinosis- this will require the conduct of Phase 1 and Phase 2 clinical studies which are beyond the scope of this current manuscript.

      Reviewer #2 (Public Review):

      Sur and colleagues investigate the role of ATP6V0A1 in mitochondrial function in cystinotic proximal tubule cells. They propose that loss of cystinosin downregulates ATP6V0A1 resulting in acidic lysosomal pH loss, and adversely modulates mitochondrial function and lifespan in cystinotic RPTECs. They further investigate the use of a novel therapeutic Astaxanthin (ATX) to upregulate ATP6V0A1 that may improve mitochondrial function in cystinotic proximal tubules.

      The new information regarding the specific proximal tubular injuries in cystinosis identifies potential molecular targets for treatment. As such, the authors are advancing the field in an experimental model for potential translational application to humans.

      Recommendations for the authors:

      Reviewer #1 (Recommendations For The Authors):

      (1) There is a lack of care with precise wording and punctuation, which negatively affects the text. Importantly, the manuscript lacks a clear description of experimental Results. This section begins with speculation, then wanders through experimentation that didn't work (could be deleted). Figure 1A and lines 94-102 could be deleted. Data from CDME loading was found to be a "poor surrogate" for cystinosis and could be deleted from the manuscript or mentioned as a minor point in the discussion. The number of individual patient cell lines used for experimentation is unclear - 8 patients are mentioned on line 109, Figure 2B shows 6 normal fibroblasts, 3 CDME-loaded fibroblasts, and an indeterminate number of normal vs CDME-loaded cells (both colored red). Cluster analysis refers to two large gene clusters - data supporting this key conclusion is not shown. It is unclear why ATP6VOA1 was selected as the most significantly reduced H+ATPase from Table II. Thus, the focus on this particular gene appears to be largely "a hunch".

      In this study, we aim to establish a new concept by using multiple cell types and various assays tailored to each affected organelle, which might be confusing. Therefore, we believe Figure 1a provides a roadmap and helps clarify what to expect from this paper.

      This study was started a decade back, when CDME-mediated lysosomal loading was regularly used as a surrogate in vitro model to study cystinosis tissue injury. That was the reason to include CDME in the study design. Since we already had the CDME-treated data and in this article we are talking about another superior in vitro cystinosis model, we would like to include it.

      In the Result and Methods section, we mentioned “8 patients” with nephropathic cystinosis from whom we collected the RPTECs and Fibroblasts. These cystinotic cells are shown in blue and purple dots, respectively in figure 2B. Normal RPTEC and fibroblast cells were purchased from company and these cells were then treated with CDME to artificially load lysosomes with cystine. Details on the cell types and its procurement can be found in the Methods section under “Study design and Samples”. Normal and CDME-loaded RPTECs are shown in red and orange dots, whereas normal and CDME-loaded fibroblasts are shown in green and yellow dots, respectively in figure 2B.

      We removed this figure from the manuscript because the data is already detailed in Tables 1 and 2. As a sub-figure, the string pathway analysis output was illegible and did not add any new information. However, for your reference, we have now provided this data below.

      Author response image 1.

      STRIG pathway analysis using the microarray transcriptomic data from normal vs.cystinotic RPTECs. Ysing K-mean clustering on the genes in these significantly enriched pathways, we identified 2 distinct clusters, red and green nodes. Red nodes are enriched in nucleus-encoded mitochondrial genes and v-ATPases family, which are crucial for lysosomes and kidney tubular acid secretion. ATP6VOA1, the topmost v-ATPase in our cystinotic transcriptome dataset is highlighted in cyan. Green nodes are enriched in genes needed for DNA replication.

      (2) It was decided to use transcriptional profiling of CTNS mutant vs wildtype renal proximal tubular cells (RPTECs) as a way to uncover defective secondary molecular pathways that might be upstream drivers of the cystinosis phenotype. Since the kidneys are the first organs to deteriorate in cystinosis, it is postulated that transcriptome differences might be more obvious in kidney cells than in non-renal tissues, such as fibroblasts. A potential pitfall is that the RPTECs were transformed cell lines whereas fibroblasts were not.

      Transcriptional profiling was done on primary cells isolated from patients with nephropathic cystinosis—RPTECs from patients' urine and fibroblasts from patients' skin—these cells are not immortalized and can therefore be compared. This is noted in the results section - “Specific transcriptional signatures are observed in cystinotic skin-fibroblasts and RPTECs obtained from the same individual with cystinosis versus their healthy counterparts”.

      We utilized the immortalized RPTEC cell line to create CRISPR-mediated CTNS knockout RPTECs as a resource for studying the pathophysiology of cystinosis. These cells were not compared to the primary fibroblasts.

      (3) The authors wanted to study intralysosomal pH but could not, so used a pH-sensitive dye that reflects whole cell pH. It would be incorrect to take this measurement as support for their hypothesis that intralysosomal pH is increased. Since these experiments cannot be interpreted, they should be deleted from the manuscript.

      We have now corrected the term to "intracellular pH." Although measuring lysosome-specific pH would be important, it was not feasible in our cells as knocking out cystinosin gene made them fragile, making the assay ineffective. Therefore, we provide data on pH assessment using BCECF, which measures the overall pH of the cytoplasm and organelles. This information is still valuable for understanding the whole cell pH, encompassing both organelle and cytoplasmic pH. We have mentioned this as one of our limitations in the Discussion section.

      (4) The choice of ATX as a potential therapy is puzzling. Its antioxidant properties seem to be irrelevant since two other antioxidants had no effect. The mechanism by which it appears to correct some aspects of the cystinosis phenotype remains unknown and this should be pointed out. A key experiment to assess whether ATX reduces lysosomal cystine accumulation is missing. While the impact of ATX on cystinosis is interesting, the mechanism is unexplored.

      A detailed study on the mechanism by which ATX corrects certain aspects of the cystinosis phenotype is currently underway and will be presented in a follow-up paper. We have measured the effect of ATX and cysteamine, both individually and combined, on cystine accumulation using HPLC, as shown in the figure below. Our results indicate a significant increase in cystine levels with ATX treatment alone, while the combined ATX and cysteamine treatment significantly reduced cystine accumulation to the normal level. This suggests that ATX addresses specific aspects of the cystinosis phenotype through a different mechanism, not by reducing the accumulated cystine levels. When co-administered with cysteamine, they have the potential to complement each other's shortcomings. We believe that the increase in cystine with ATX alone may be due to interactions between ATX's ketone or hydroxyl groups and cystine's amine or carboxylic groups. Further research on this interaction is ongoing.

      We have now added to the Discussion – “We noticed a significant increase in cystine levels with ATX treatment alone (data not shown in the manuscript), while the combined ATX and cysteamine treatment significantly reduced cystine accumulation to the normal level. This may suggest that when co-administered with cysteamine, they have the potential to complement each other's shortcomings. We believe that the increase in cystine with ATX alone could be due to interactions between ATX's ketone or hydroxyl groups and cystine's amine or carboxylic groups. Further research on this interaction is ongoing.”

      Author response image 2.

      (5) The effects of exogenous ATP6VOA1 are interesting but had no effect on lysosomal cystine efflux, a hallmark of the cystinosis cellular phenotype. A discussion of this issue would be important.

      In the revised manuscript, we have included a detailed discussion on the plausible reasons why ATP6V0A1 correction has no effect on cysteine efflux from the lysosome. We have added to the Discussion – “However, correcting ATP6V0A1 had no effect on cellular cystine levels (Figure 7C), likely because cystinosin is known to have multiple roles beyond cystine transport. Cystinosin is demonstrated to be crucial for activating mTORC1 signaling by directly interacting with v-ATPases and other mTORC1 activators. Cystine depletion using cysteamine does not affect mTORC1 signaling (47). Our data, along with these observations, further supports that cystinosin has multiple functions and that its cystine transport activity is not mediated by ATP6V0A1.”

      (6) The arguments on lines 260-273 are not comprehensible. The authors confirm that RPTC LC3-II levels are increased, a marker of active processing of autophagosome cargo, prior to delivery to lysosomes. Discussion of balfilomycin (not used), mTORC activity, and endocytosis are not directly relevant and wander from interpretation of the LC3-II observation. One possibility is that the 50% decrease in ATP6VOA1 transcript is sufficient to slow the transfer of LC3-II-tagged cargo from autophagosome to lysosome - however, it would be important to offer a plausible explanation for why decreased ATP6VOA1 expression alone does not appear to be the key limitation on lysosomal cystine efflux.

      We have now rephrased our explanation in the Discussion section – “Cystinotic cells are known to have an increased autophagy or reduced autophagosome turnover rate. Autophagic flux in a cell is typically assessed by examining the accumulation of the autophagosome or autophagy-lysosome marker LC3B-II. This accumulation can be artificially induced using bafilomycin, which targets the V-ATPase, thereby inhibiting lysosomal acidification and degradation of its contents. Taken together, the observed innate increase in LC3B-II in cystinotic RPTECs (Figure 5A) without bafilomycin treatment suggests dysfunctional lysosomal acidification and thus could be linked to inhibited v-ATPase activity”.

    1. eLife Assessment

      This is an important and convincing study of the morphological properties of Purkinje cell dendrites and dendritic spines in adult humans and mice, and the anatomical determinants of multi-innervation by climbing fibers. The data will provide a helpful resource for the field of cerebellar computation.

    2. Reviewer #1 (Public review):

      Summary:

      Busch and Hansel present a morphological and histological comparison between mouse and human Purkinje cells (PCs) in the cerebellum. The study reveals species-specific differences that have not previously been reported despite numerous observations of these species. While mouse PCs show morphological heterogeneity and occasional multi-innervation by climbing fibers (CFs), human PCs exhibit a widespread, multi-dendritic structure that exceeds expectations based on allometric scaling. Specifically, human PCs are significantly larger, and exhibit increased spine density, with a unique cluster-like morphology not found in mice.

      Strengths:

      The manuscript provides an exceptionally detailed analysis of PC morphology across species, surpassing any prior publication. Major strengths include a systematic and thorough methodology, rigorous data analysis, and clear presentation of results. This work is likely to become the go-to resource for quantitation in this field. The authors have largely achieved their aims, with the results effectively supporting their conclusions.

      Weaknesses:

      There are a few concerns that need to be addressed, specifically related to details of the methodolology as well as data interpretation based on the limits of some experimental approaches. Overall, these weaknesses are minor.

    3. Reviewer #2 (Public review):

      Summary:

      This manuscript aims to follow up on a previously published paper (Busch and Hansel 2023) which proposed that the morphological variation of dendritic bifurcation in Purkinje cells in mice and humans is indicative of the number of climbing fiber inputs, with dendritic bifurcation at the level of the soma resulting in a proportion of these neurons being multi-innervated. The functional and anatomical climbing fiber data was obtained solely from mice since all human tissue was embalmed and fixed, and the extension of these findings to human Purkinje cells was indirect. The current comparative anatomy study aims to resolve this question in human tissue more directly and to further analyse in detail the properties of adult human Purkinje cell dendritic morphology.

      Strengths:

      The authors have carried out a meticulous anatomical quantification of human Purkinje cell dendrites, in tissue preparations with a better signal-to-noise ratio than their previous study, comparing them with those from mice. Importantly, they now present immunolabelling results that trace climbing fiber axons innervating human PCs. As well as providing detailed analyses of spine properties and interesting new findings of human PC dendritic length and spine types, the work confirms that human PCs that have two clearly distinct dendritic branches have an approximately x% chance of receiving more than one CF input, segregated across the two branches. Albeit entirely observational, the data will be of widespread interest to the cerebellar field, in particular, those building computational models of Purkinje cells.

      Weaknesses:

      The work is, by necessity, purely anatomical. It remains to be seen whether there are any functional differences in ion channel expression or functional mapping of granule inputs to human PCs compared with the mouse that might mitigate the major differences in electronic properties suggested.

    4. Author response:

      We plan to submit a revised version of our manuscript eLife-RP-RA-2024-105013, in which we address all comments raised by the two expert reviewers.

      Below we describe what we like to address in this revision. We understand that the provisional response is not meant to be a point-by-point reply. Therefore, our revision plan more generally summarizes the comments of the reviewers and how we plan to address them.

      Reviewer #1:

      This reviewer is overall very positive and states that our ‘work is likely to become the go-to resource for quantification in this field’. This reviewer raises few weaknesses of the manuscript that are explicitly described as minor.

      Microscopic resolution sufficient to support quantitative spine assessments?

      In the detailed revision, we will provide quantification of microscopic resolution and will relate this to the spine comparisons offered. Where needed, we will add caveats discussing measurement limits.

      Age of the human tissue.

      Most analysis is based on the study of three brains from elderly individuals. For the analysis of dendritic spines, we added measures from a younger brain (37 years-old). We will make it more clear, which datasets contained these measures and what the results of our comparative analysis have been.

      Genetic diversity contributing to species differences?

      We provide an updated discussion on this interesting topic.

      Reviewer #2:

      This reviewer also expresses a largely positive view of the manuscript, noting that ‘..the data will be of widespread interest to the cerebellar field…’. 

      Microscopic resolution:

      see above.

      Figure panels / Fig. 3:

      We will make sure that the figures are readable and will provide a clarification of gray scales used in Fig. 3.

      Vertical vs horizontal dendrite orientation:

      This is a point that requires clarification. Per our definition, all dendrites fall either into the vertical or horizontal category. We will make sure that this is defined sufficiently well.

    1. eLife Assessment

      Combining experiments in microfluidic devices and computer simulation, this study provides a valuable analysis of the relevant parameters that determine the motility of (multicellular) magnetotactic bacteria in sediment-like environments. Despite the limitations imposed by the specific experimental design of the pores, the study presents convincing evidence that there is an optimum in the biological parameters for motile life under such conditions.

    2. Reviewer #1 (Public review):

      Summary:

      The authors track the motion of multiple consortia of Multicellular Magnetotactic Bacteria moving through an artificial network of pores and report a discovery of a simple strategy for such consortia to move fast through the network: an optimum drift speed is attained for consortia that swim a distance comparable to the pore size in the time it takes to align the with an external magnetic field. The authors rationalize their observations using dimensional analysis and numerical simulations. Finally, they argue that the proposed strategy could generalize to other species by demonstrating the positive correlation between the swimming speed and alignment time based on parameters derived from literature.

      Strengths:

      The underlying dimensional analysis and model convincingly rationalize the experimental observation of an optimal drift velocity: the optimum balances the competition between the trapping in pores at large magnetic fields and random pore exploration for weak magnetic fields.

      Weaknesses:

      The convex pore geometry studied here creates convex traps for cells, which I expect enhances their trapping. The more natural concave geometries, resulting from random packing of spheres, would create no such traps. In this case, whether a non-monotonic dependence of the drift velocity on the Scattering number would persist is unclear.

    3. Reviewer #2 (Public review):

      Summary:

      The authors have made microfluidic arrays of pores and obstacles with a complex shape and studied the swimming of multicellular magnetotactic bacteria through this system. They provide a comprehensive discussion of the relevant parameters of this system and identify one dimensionless parameter, which they call the scattering number and which depends on the swimming speed and magnetic moment of the bacteria as well as the magnetic field and the size of the pores, as the most relevant. They measure the effective speed through the array of pores and obstacles as a function of that parameter, both in their microfluidic experiments and in simulations, and find an optimal scattering number, which they estimate to reflect the parameters of the studied multicellular bacteria in their natural environment. They finally use this knowledge to compare different species to test the generality of this idea.

      Strengths:

      This is a beautiful experimental approach and the observation of an optimal scattering number (likely reflecting an optimal magnetic moment) is very convincing. The results here improve on similar previous work in two respects: On the one hand, the tracking of bacteria does not have the limitations of previous work, and on the other hand, the effective motility is quantified. Both features are enabled by choices of the experimental system: the use the multicellular bacteria which are larger than the usual single-celled magnetotactic bacteria and the design of the obstacle array which allows the quantification of transition rates due to the regular organization as well as the controlled release of bacteria into this array through a clever mechanism.

      Weaknesses:

      Some of the reported results are not as new as the authors suggest, specifically trapping by obstacles and the detrimental effect of a strong magnetic field have been reported before as has the hypothesis that the magnetic moment may be optimized for swimming in a sediment environment where there is a competition of directed swimming and trapping. Other than that, some of the key experimental choices on which the strength of the approach is based also come at a price and impose some limitations, namely the use of a non-culturable organism and the regular, somewhat unrealistic artificial obstacle array.

    4. Author response:

      Response to Referee 1

      We agree that convex walls increase the time that consortia remain trapped in pores at high magnetic fields. Since the non-monotonic behavior of the drift velocity with the Scattering number arises largely due to these long trapping times, we agree that experiments using concave pores are likely to show a peak drift velocity that is diminished or erased.

      However, we disagree that a random packing of spheres or similar particles provides an appropriate model for natural sediment, which is not composed exclusively of hard particles in a pure fluid. Pore geometry is also influenced by clogging. Biofilms growing within a network of convex pillars in two-dimensional microfluidic devices have been observed to connect neighboring pillars, thereby forming convex pores. Similar pore structures appear in simulations of biofilm growth between spherical particles in three dimensions. Moreover, the salt marsh sediment in which MMB live is more complex than simple sand grains, as cohesive organic particles are abundant. Experiments in microfluidic channels show that cohesive particles clog narrow passageways and form pores similar to those analyzed here. Thus, we expect convex pores to be present and even common in natural sediment where clogging plays a role.

      The concentration of convex pores in the experiments presented here is almost certainly much higher than in nature. Nonetheless, since magnetotactic bacteria continuously swim through the pore space, they are likely to regularly encounter such convexities. Efficient navigation of the pore space thus requires that magnetotactic bacteria be able to escape these traps. In the original version of this manuscript, this reasoning was reduced to only one or two sentences. That was a mistake, and we thank the reviewer for prompting us to expand on this point. As the reviewer notes, this reasoning is central to the analysis and should have been featured more prominently. In the final version, we will devote considerable space to this hypothesis and provide references to support the claims made above.

      The reviewer suggests that the generality of this work depends on our finding a "positive correlation between the swimming speed and alignment [rate] based on parameters derived from literature." We wish to emphasize that, in addition to predicting this correlation, our theory also predicts the function that describes it. The black line in Figure 3 is not fitted to the parameters found in the literature review; it is a pure prediction.

      Response to Referee 2

      In the "Recommendations for the Authors," this reviewer drew our attention to a manuscript that absolutely should have been prominently cited. As the reviewer notes, our manuscript meaningfully expands upon this work. We are pleased to learn that the phenomena discussed here are more general than we initially understood. It was an oversight not to have found this paper earlier. The final version will better contextualize our work and give due credit to the authors. We sincerely appreciate the reviewer for bringing this work to our attention.

      We disagree that the use of non-culturable organisms and our unrealistic array should be considered serious weaknesses. While any methodological choice comes with trade-offs, we believe these choices best advance our aims. First, the goal of our research, both within and beyond this manuscript, is to understand the phenotypes of magnetotactic bacteria in nature. While using pure cultures enables many useful techniques, phenotypic traits may drift as strains undergo domestication. We therefore prioritize studying environmental enrichments.

      Clearly, an array of obstacles does not fully represent natural heterogeneity. However, using regular pore shapes allows us to average over enough consortium-wall collisions to enable a parameter-free comparison between theory and experiment. Conducting an analysis like this with randomly arranged obstacles would require averaging over an ensemble of random environments, which is practically challenging given the experimental constraints. Since we find good agreement between theory and experiment in simple geometries, we are now in a position to justify extending our theory to more realistic geometries. Additionally, we note that a microfluidic device composed of a random arrangement of obstacles would also be a poor representation of environmental heterogeneity, as pore shape and network topology differ between two and three dimensions.

    1. eLife Assessment

      This valuable study assesses epigenetic clocks across ancestries, including in the context of accelerated aging in Alzheimer's Disease patients. It provides convincing evidence for population differences in age estimation accuracy across a variety of epigenetic clocks, but the degree to which these differences reflect continuous variation in ancestry, and/or are confounded by environmental or power differences is not entirely clear; consequently, the evidence that reduced portability is rooted in genetics is incomplete. Given the accelerating use of epigenetic clocks across fields, this study is nevertheless likely to be of interest to researchers working on human genetic and epigenetic variation or who apply epigenetic clocks to diverse human populations.

    2. Reviewer #1 (Public review):

      Summary:

      Cruz-González and colleagues draw on DNA methylation and paired genetic data from 621 participants (n=308 controls; n=313 participants with Alzheimer's Disease). The authors generate a panel of epigenetic biomarkers of aging with a primary focus on the Horvath multi-tissue clock. The authors find weaker correlations between predicted epigenetic age and chronological age in subgroups with higher African ancestry than within a subgroup identified as White. The authors then examine genetic variation as a potential source for between-group differences in epigenetic clock performance. The authors draw on a large collection of publicly available methylation quantitative trait loci datasets and find evidence for substantial overlap between clock CpGs located within the Horvath clock and methQTLs. Going further, the authors show that methQTLs that overlap with Horvath clock CpGs show greater allelic variation in African ancestral groups pointing to a potential explanation for poorer clock performance within this group.

      Strengths:

      This is an interesting dataset and an important research question. The authors cite issues of portability regarding polygenic risk scores as a motivation to examine between-group differences in the performance of a panel of epigenetic clocks. The authors benefit from a diverse cohort of individuals with paired genetic data and focus on a clinical phenotype, Alzheimer's disease, of clear relevance for studies evaluating age-related biomarkers.

      Weaknesses:

      While the authors tackle an important question using a diverse cohort the current manuscript is lacking some detail that may diminish the potential impact of this paper. For example:

      (1) Information on chronological ages across groups should be reported to ensure there are no systematic differences in ages or age ranges between groups (see point below).

      (2) The authors compare correlations between chronological age and epigenetic age in sub-groups within to correlations reported by Horvath (2013). Attempting to draw comparisons between these two datasets is problematic. The current study has a much smaller N (particularly for sub-group analyses) and has a more restricted age range (60-90yrs versus 0-100 yrs). Thus, is an alternative explanation simply that any weaker correlations observed in this study are driven by sample size and a restricted age range? Reporting the chronological ages (and ranges) across subgroups in the current study would help in this regard. Similarly, given the lack of association between AD status and epigenetic age (and very small effect in the white group), it may be of interest to examine the correlation between chronological age and epigenetic age in each group including the AD participants: would the between-group differences in correlations between chronological age and epigenetic be altered by increasing the sample size?

      (3) The correlation between chronological age and epigenetic age, while helpful is not the most informative estimate of accuracy. Median absolute error (and an analysis of MAE across subgroups) would be a helpful addition.

      (4) More information should be provided about how DNAm data were generated. Were samples from each ancestral group randomized across plates/slides to ensure ancestry and batch are not associated? How were batch effects considered? Given the relatively small sample sizes, it would be important to consider the impact of technical variation on measures of epigenetic age used in the current study. The use of principal Component-based versions of these clocks (Higgins Chen et al., 2023; Nature Aging https://doi.org/10.1038/s43587-022-00248-2) may help address concerns such concerns.

      (5) Marioni et al., (2015) found a very weak cross-sectional association between DNAm Age and cognitive function (r~0.07) in a cohort of >900 participants. Given these effect sizes, I would not interpret the absence of an effect in the current study to reflect issues of portability of epigenetic biomarkers.

      6) The methQTL analyses presented are suggestive of potential genetic influence on DNAm at some Horvath CpGs. Do authors see differences in DNAm across ancestral groups at these potentially affected CpGs? This seems to be a missing piece together (e.g., estimating the likely impact of methQTL on clock CpG DNAm).

    3. Reviewer #2 (Public review):

      Summary:

      This paper seeks to characterize the portability of methylation clocks across groups. Methylation clocks are trained to predict biological aging from DNA methylation but have largely been developed in datasets of individuals with primarily European ancestries. Given that genetic variation can influence DNA methylation, the authors hypothesize that methylation clocks might have reduced accuracy in non-European ancestries.

      Strengths:

      The authors evaluate five methylation clocks in 621 individuals from the MAGENTA study. This includes approximately 280 individuals sampled in Puerto Rico, Cuba, and Peru, as well as approximately 200 self-identified African American individuals sampled in the US. To understand how methylation clock accuracy varies with proportion of non-European ancestry, the authors inferred local ancestry for the Puerto Rican, Cuban, Peruvian, and African American cohorts. Overall, this paper presents solid evidence that methylation clocks have reduced accuracy in individuals with non-European ancestries, relative to individuals with primarily European ancestries. This should be of great interest to those researchers who seek to use methylation clocks as predictors of age-related, late-onset diseases and other health outcomes.

      Weaknesses:

      One clear strength of this paper is the ability to do more sophisticated analyses using the local ancestry calls for the MAGENTA study. It would be valuable to capitalize on this strength and assess portability across the genetic ancestry spectrum, as was recently advocated by Ding et al. in Nature (2023). For example, the authors could regress non-European local ancestry fraction on measures of prediction accuracy. This could paint a clearer picture of the relationship between genetic ancestry and clock accuracy, compared to looking at overall correlations within each cohort.

      The authors present two possible reasons that methylation clocks might have reduced accuracy in individuals with non-European ancestries: genetic variants disrupting methylation sites (i.e. "disruptive variants"), and genetic variants influencing methylation sites (i.e. meQTLs). The authors conclude disruptive variants do not contribute to poor methylation clock portability, but the evidence in support of this conclusion is incomplete. The site frequency spectrum of disruptive variants in Figure 4 is estimated from all gnomAD individuals, and gnomAD is comprised of primarily European individuals. Thus, the observation that disruptive variants are generally rare in gnomAD does not rule them out as a source of poor clock portability in admixed individuals with non-European ancestries.

      It is also unclear to what extent meQTLs impact methylation clock portability. The authors find that the frequency of meQTLs is higher in African ancestry populations, but this could reflect the fact that some of the analyzed meQTLs were ascertained in African Americans. The number of meQTL-affected methylation sites also varies widely between clocks, ranging from 6 to 271; thus, meQTLs likely impact the portability of different clocks in different ways. Overall, the paper would benefit from a more quantitative assessment of the extent to which meQTLs influence clock portability.

      The paper implies that methylation clocks have an inferior ability to predict AD risk in admixed populations relative to white individuals, but the difference between white AD patients and controls is not significant when correcting for multiple testing. This nuance should be made more explicit.

      Finally, this paper overlooks the possibility that environmental exposures co-vary with genetic ancestry and play a role in decreasing the accuracy of methylation clocks in genetically admixed individuals. Quantifying the impact of environmental factors is almost certainly outside of the scope of this paper. However, it is worth acknowledging the role of environmental factors to provide the field with a more comprehensive overview of factors influencing methylation clock portability. It is also essential to avoid the assumption that correlations with genetic ancestry necessarily arise from genetic causes.

    4. Reviewer #3 (Public review):

      This manuscript examines the accuracy of DNA methylation-based epigenetic clocks across multiple cohorts of varying genetic ancestry. The authors find that clocks were generally less accurate at predicting age in cohorts with large proportions of non-European (especially African) ancestry, compared to cohorts with high European ancestry proportions. They suggest that some of this effect might be explained by meQTLs that occur near CpG sites included in clocks, because these variants may be at higher frequencies (or at least different frequencies) in cohorts with high proportions of non-European ancestry relative to the training set. They also provide discussions of potential paths forward to alleviate bias and improve portability for future clock algorithms.

      The topic is timely due to the increasing popularity of DNA methylation-based clocks and the acknowledgment that many algorithms (e.g., polygenic risk scores) lack portability when applied to cohorts that substantially differ in ancestry or other characteristics from the training set. This has been discussed to some degree for DNA methylation-based clocks, but could of course use more discussion and empirical attention which the authors nicely provide using an impressive and diverse collection of data.

      The manuscript is clear and well-written, however, some key background was missing (e.g., what we know already about the ancestry composition of clock training sets) and most importantly several analyses would benefit from being taken one step further. For example, the main argument of the paper is that ancestry impacts clock predictions, but this is determined by subsetting the data by recruitment cohort rather than analyzing ancestry as a continuous variable. Extending some of the analyses could really help the authors nail down their hypothesized sources of lack of portability, which is critical for making recommendations to the community and understanding the best paths forward.

    5. Author response:

      Public Reviews:

      Reviewer #1 (Public review):

      Summary: 

      Cruz-González and colleagues draw on DNA methylation and paired genetic data from 621 participants (n=308 controls; n=313 participants with Alzheimer's Disease). The authors generate a panel of epigenetic biomarkers of aging with a primary focus on the Horvath multi-tissue clock. The authors find weaker correlations between predicted epigenetic age and chronological age in subgroups with higher African ancestry than within a subgroup identified as White. The authors then examine genetic variation as a potential source for between-group differences in epigenetic clock performance. The authors draw on a large collection of publicly available methylation quantitative trait loci datasets and find evidence for substantial overlap between clock CpGs located within the Horvath clock and methQTLs. Going further, the authors show that methQTLs that overlap with Horvath clock CpGs show greater allelic variation in African ancestral groups pointing to a potential explanation for poorer clock performance within this group. 

      Thank you for this summary.

      Strengths:  

      This is an interesting dataset and an important research question. The authors cite issues of portability regarding polygenic risk scores as a motivation to examine between-group differences in the performance of a panel of epigenetic clocks. The authors benefit from a diverse cohort of individuals with paired genetic data and focus on a clinical phenotype, Alzheimer's disease, of clear relevance for studies evaluating age-related biomarkers.  

      Weaknesses:  

      While the authors tackle an important question using a diverse cohort the current manuscript is lacking some detail that may diminish the potential impact of this paper. For example:  

      (1) Information on chronological ages across groups should be reported to ensure there are no systematic differences in ages or age ranges between groups (see point below).  

      Thank you for pointing out this omission. The age ranges are similar across cohorts. No individuals under 60 were considered, and the average ages per cohort ranged from 72 to 76. Neither average age nor age range was consistently higher or lower in the admixed cohorts for which the clocks had lower performance compared to the White cohort. We will report the age distributions in supplementary material in the revision.

      (2) The authors compare correlations between chronological age and epigenetic age in sub-groups within to correlations reported by Horvath (2013). Attempting to draw comparisons between these two datasets is problematic. The current study has a much smaller N (particularly for sub-group analyses) and has a more restricted age range (6090yrs versus 0-100 yrs). Thus, is an alternative explanation simply that any weaker correlations observed in this study are driven by sample size and a restricted age range? Reporting the chronological ages (and ranges) across subgroups in the current study would help in this regard. Similarly, given the lack of association between AD status and epigenetic age (and very small effect in the white group), it may be of interest to examine the correlation between chronological age and epigenetic age in each group including the AD participants: would the between-group differences in correlations between chronological age and epigenetic be altered by increasing the sample size?

      Our conclusions about the reduced accuracy of the clocks in admixed individuals are based on comparisons within the MAGENTA cohorts, not on the comparisons to previous reports. We show significantly reduced accuracy on African American and Puerto Rican cohorts in MAGENTA compared to the White MAGENTA cohort. The reviewer is correct that the lower correlation in each of the cohorts compared to those in the Horvath study is due to the older age range of our cohort. Indeed, other studies applying the Horvath clock have seen similar correlations to those observed on the White MAGENTA cohort (Marioni et al., 2015, Horvath 2013, and Shireby et al., 2020). Following the suggestion to increase sample size, we conducted the chronological age vs. epigenetic age correlation analysis with the inclusion of AD cases. The significantly lower performance of the clock on Puerto Ricans and African Americans relative to White individuals remains after including all individuals in each cohort. We will include these results on the full cohorts in MAGENTA in the revision.

      (3) The correlation between chronological age and epigenetic age, while helpful is not the most informative estimate of accuracy. Median absolute error (and an analysis of MAE across subgroups) would be a helpful addition.  

      We used correlation because this is commonly used to evaluate the performance of epigenetic age clocks, but we agree that direct error quantification provides a complementary perspective. We confirm that the African American and Puerto Rican cohorts have higher error than the White cohort, and we will report these comparisons in the revision.

      (4) More information should be provided about how DNAm data were generated. Were samples from each ancestral group randomized across plates/slides to ensure ancestry and batch are not associated? How were batch effects considered? Given the relatively small sample sizes, it would be important to consider the impact of technical variation on measures of epigenetic age used in the current study. The use of principal Component-based versions of these clocks (Higgins Chen et al., 2023; Nature Aging https://doi.org/10.1038/s43587-022-00248-2) may help address concerns such concerns.  

      Thank you for pointing out the need for additional context on data generation. All omics data from the MAGENTA study were generated using protocols that aim to minimize technical artifacts and batch effects. We will add detailed protocol information will be detailed in the revision. We also thank the reviewer for their suggestion on applying the principal component clock to account for potential technical variation. We are planning to perform these analyses and include them in the revision.

      (5) Marioni et al., (2015) found a very weak cross-sectional association between DNAm Age and cognitive function (r~0.07) in a cohort of >900 participants. Given these effect sizes, I would not interpret the absence of an effect in the current study to reflect issues of portability of epigenetic biomarkers. 

      We agree that previous links between DNAm Age and AD/cognitive function have been small in magnitude. For example, the PhenoAge paper (Levine et al., 2018) and a study using the Horvath clock (Levine et al., 2015) found age acceleration of less than a year in AD patients relative to non-demented individuals. These effects have been detected in studies with relatively small sample sizes (e.g., 700 for Levine et al. 2015 and 604 for Levine et al. 2018). Our study is of similar size, but the cohort-specific analyses have lower power. Nonetheless, we replicate the modest, but significant association with AD in the white MAGENTA cohort. We have performed power calculations and find that we have 26% power to detect an effect of this size in the Cubans, 46% for the Peruvians, 66% for the Whites, 74% for the Puerto Ricans, and 84% for the African Americans. Given the relatively high power in the Puerto Rican and African American cohorts, we suggest that the reduced accuracy of the clocks contributes to the lack of association. We will also add caveats about power and the small sample size in the revision.

      6) The methQTL analyses presented are suggestive of potential genetic influence on DNAm at some Horvath CpGs. Do authors see differences in DNAm across ancestral groups at these potentially affected CpGs? This seems to be a missing piece together (e.g., estimating the likely impact of methQTL on clock CpG DNAm). 

      Thank you for this excellent suggestion. We will add this analysis in the revision. This will enable us to test for further evidence for our hypothesis about the role of ancestryspecific meQTL on clock accuracy.  

      Reviewer #2 (Public review):

      Summary:  

      This paper seeks to characterize the portability of methylation clocks across groups. Methylation clocks are trained to predict biological aging from DNA methylation but have largely been developed in datasets of individuals with primarily European ancestries. Given that genetic variation can influence DNA methylation, the authors hypothesize that methylation clocks might have reduced accuracy in non-European ancestries.  

      Strengths:  

      The authors evaluate five methylation clocks in 621 individuals from the MAGENTA study. This includes approximately 280 individuals sampled in Puerto Rico, Cuba, and Peru, as well as approximately 200 self-identified African American individuals sampled in the US. To understand how methylation clock accuracy varies with proportion of nonEuropean ancestry, the authors inferred local ancestry for the Puerto Rican, Cuban, Peruvian, and African American cohorts. Overall, this paper presents solid evidence that methylation clocks have reduced accuracy in individuals with non-European ancestries, relative to individuals with primarily European ancestries. This should be of great interest to those researchers who seek to use methylation clocks as predictors of agerelated, late-onset diseases and other health outcomes.

      Thank you for this summary.

      Weaknesses:  

      One clear strength of this paper is the ability to do more sophisticated analyses using the local ancestry calls for the MAGENTA study. It would be valuable to capitalize on this strength and assess portability across the genetic ancestry spectrum, as was recently advocated by Ding et al. in Nature (2023). For example, the authors could regress non-European local ancestry fraction on measures of prediction accuracy. This could paint a clearer picture of the relationship between genetic ancestry and clock accuracy, compared to looking at overall correlations within each cohort. 

      Thank you for this excellent suggestion. We agree that modeling portability across genetic ancestry as a spectrum would help support our conclusions. We will add this to the revision.

      The authors present two possible reasons that methylation clocks might have reduced accuracy in individuals with non-European ancestries: genetic variants disrupting methylation sites (i.e., "disruptive variants") and genetic variants influencing methylation sites (i.e., meQTLs). The authors conclude disruptive variants do not contribute to poor methylation clock portability, but the evidence in support of this conclusion is incomplete. The site frequency spectrum of disruptive variants in Figure 4 is estimated from all gnomAD individuals, and gnomAD is comprised of primarily European individuals. Thus, the observation that disruptive variants are generally rare in gnomAD does not rule them out as a source of poor clock portability in admixed individuals with non-European ancestries. 

      Thank you for this question. The allele frequencies were so low that even if they all occurred in individuals of non-European ancestries, they would still be incredibly rare. Nonetheless, in the revision, we will make this clear by reporting ancestry-specific allele frequencies.

      It is also unclear to what extent meQTLs impact methylation clock portability. The authors find that the frequency of meQTLs is higher in African ancestry populations, but this could reflect the fact that some of the analyzed meQTLs were ascertained in African Americans. The number of meQTL-affected methylation sites also varies widely between clocks, ranging from 6 to 271; thus, meQTLs likely impact the portability of different clocks in different ways. Overall, the paper would benefit from a more quantitative assessment of the extent to which meQTLs influence clock portability. 

      We agree that the meQTL likely influence the clocks in different ways and that the ascertainment of the meQTLs in different populations makes direct comparisons challenging. To provide mechanistic insights into the ways that meQTL influence the methylation clocks, we plan to leverage the individual-level genetic data generated for the MAGENTA individuals. This will allow us to explore whether the individuals who have the specified clock-influencing meQTL receive less accurate predictions from the methylation clocks. In addition, the new analysis of whether individuals from different cohorts have different methylation levels at clock CpGs with ancestry-variable meQTLs will help establish the differences between groups (see response to Reviewer #1 point 6). Finally, to resolve potential bias due to ascertaining some of the meQTL in African Americans, we will conduct the same analyses from the manuscript, holding out the set of meQTL from African Americans. These results will be included in the revision.

      The paper implies that methylation clocks have an inferior ability to predict AD risk in admixed populations relative to white individuals, but the difference between white AD patients and controls is not significant when correcting for multiple testing. This nuance should be made more explicit. 

      We agree that the signal is not particularly strong in the white cohort, but the effect size is in line with previous studies. We will add power calculations and discussion to help the interpretation of these results (see response to Reviewer #1 point 5).  

      Finally, this paper overlooks the possibility that environmental exposures co-vary with genetic ancestry and play a role in decreasing the accuracy of methylation clocks in genetically admixed individuals. Quantifying the impact of environmental factors is almost certainly outside of the scope of this paper. However, it is worth acknowledging the role of environmental factors to provide the field with a more comprehensive overview of factors influencing methylation clock portability. It is also essential to avoid the assumption that correlations with genetic ancestry necessarily arise from genetic causes.  

      We entirely agree about the importance of discussing environmental exposures. We did not intend to discount them in our manuscript. We will clarify their potential role and the scope of our analyses in the revision. We expect that environmental factors certainly contribute to differences between groups. The revisions outlined above may help us better quantify the genetic contribution.

      Reviewer #3 (Public review):

      This manuscript examines the accuracy of DNA methylation-based epigenetic clocks across multiple cohorts of varying genetic ancestry. The authors find that clocks were generally less accurate at predicting age in cohorts with large proportions of nonEuropean (especially African) ancestry, compared to cohorts with high European ancestry proportions. They suggest that some of this effect might be explained by meQTLs that occur near CpG sites included in clocks, because these variants may be at higher frequencies (or at least different frequencies) in cohorts with high proportions of non-European ancestry relative to the training set. They also provide discussions of potential paths forward to alleviate bias and improve portability for future clock algorithms.  

      The topic is timely due to the increasing popularity of DNA methylation-based clocks and the acknowledgment that many algorithms (e.g., polygenic risk scores) lack portability when applied to cohorts that substantially differ in ancestry or other characteristics from the training set. This has been discussed to some degree for DNA methylation-based clocks, but could of course use more discussion and empirical attention which the authors nicely provide using an impressive and diverse collection of data.  

      The manuscript is clear and well-written, however, some key background was missing (e.g., what we know already about the ancestry composition of clock training sets) and most importantly several analyses would benefit from being taken one step further. For example, the main argument of the paper is that ancestry impacts clock predictions, but this is determined by subsetting the data by recruitment cohort rather than analyzing ancestry as a continuous variable. Extending some of the analyses could really help the authors nail down their hypothesized sources of lack of portability, which is critical for making recommendations to the community and understanding the best paths forward.  

      Thank you for these suggestions. As noted in our response to reviewer #2, we will analyze ancestry as a continuous variable in the revision. We will also add details on the training of previous clocks and previous work on clock accuracy.

    1. eLife Assessment

      Using several hundreds of samples and cutting-edge genomic methods, including BioNano, PacBio HiFi, and advanced bioinformatic pipelines, the authors identify six large chromosomal inversions segregating in over 100 species of Lake Malawi cichlids. This important study provides compelling evidence for the presence of these six inversions, their differential distribution among populations, and the association of chromosome 10 inversion with a sex-determination locus. This work also provides a starting point for further investigating the role of these inversions with respect to local adaptation, speciation, sex determination, hybridization, and incomplete lineage sorting in cichlids, which represent ~5% of the extant vertebrate species and are one of the most prominent examples of adaptive radiations.