Reviewer #1 (Public review):

Summary:

The manuscript presents a significant and rigorous investigation into the role of CHMP5 in regulating bone formation and cellular senescence. The study provides compelling evidence that CHMP5 is essential for maintaining endolysosomal function and controlling mitochondrial ROS levels, thereby preventing the senescence of skeletal progenitor cells.

Strengths:

The authors demonstrate that the deletion of Chmp5 results in endolysosomal dysfunction, elevated mitochondrial ROS, and ultimately enhanced bone formation through both autonomous and paracrine mechanisms. The innovative use of senolytic drugs to ameliorate musculoskeletal abnormalities in Chmp5-deficient mice is a novel and critical finding, suggesting potential therapeutic strategies for musculoskeletal disorders linked to endolysosomal dysfunction.

Weaknesses:

The manuscript requires a deeper discussion or exploration of CHMP5's roles and a more refined analysis of senolytic drug specificity and effects. This would greatly enhance the comprehensiveness and clarity of the manuscript.

Reviewer #2 (Public review):

Summary:

The authors try to show the importance of CHMP5 for skeletal development.

Strengths:

The findings of this manuscript are interesting. The mouse phenotypes are well done and are of interest to a broader (bone) field.

Weaknesses:

The mechanistic insights are mediocre, and the cellular senescence aspect poor.

In total, it has not been shown that there are actual senescent cells that are reduced after D+Q-treatment. These statements need to be scaled back substantially.

Reviewer #3 (Public review):

Summary:

In this study, Zhang et al. reported that CHMP5 restricts bone formation by controlling endolysosome-mitochondrion-mediated cell senescence. The effects of CHMP5 on osteoclastic bone resorption and bone turnover have been reported previously (PMID: 26195726), in which study the aberrant bone phenotype was observed in the CHMP5-ctsk-CKO mouse model, using the same mouse model, Zhang et al., report a novel role of CHMP5 on osteogenesis through affecting cell senescence. Overall, it is an interesting study and provides new insights in the field of cell senescence and bone.

Strengths:

Analyzed the bone phenotype OF CHMP5-periskeletal progenitor-CKO mouse model and found the novel role of senescent cells on osteogenesis and migration.

Weaknesses:

(1) There are a lot of papers that have reported that senescence impairs osteogenesis of skeletal stem cells. In this study, the author claimed that Chmp5 deficiency induces skeletal progennitor cell senescence and enhanced osteogenesis. Can the authors explain the controversial results?

(2) Co-culture of Chmp5-KO periskeletal progenitors with WT ones should be conducted to detect the migration and osteogenesis of WT cells in response to Chmp5-KO-induced senescent cells. In addition, the co-culture of WT periskeletal progenitors with senescent cells induced by H2O2, radiation, or from aged mice would provide more information.

(3) Many EVs were secreted from Chmp5-deleted periskeletal progenitors, compared to the rarely detected EVs around WT cells. Since EVs of BMSCs or osteoprogenitors show strong effects of promoting osteogenesis, did the EVs contribute to the enhanced osteogenesis induced by Chmp5-defeciency?

(4) EVs secreted from senescent cells propagate senescence and impair osteogenesis, why do EVs secreted from senescent cells induced by Chmp5-defeciency have opposite effects on osteogenesis?

(5) The Chmp5-ctsk mice show accelerated aging-related phenotypes, such as hair loss and joint stiffness. Did Ctsk also label cells in hair follicles or joint tissue?

(6) Fifteen proteins were found to increase and five proteins to decrease in the cell supernatant of Chmp5Ctsk periskeletal progenitors. How about SASP factors in the secretory profile?

(7) D+Q treatment mitigates musculoskeletal pathologies in Chmp5 conditional knockout mice. In the previously published paper (CHMP5 controls bone turnover rates by dampening NF-κB activity in osteoclasts), inhibition of osteoclastic bone resorption rescues the aberrant bone phenotype of the Chmp5 conditional knockout mice. Whether the effects of D+Q on bone overgrowth is because of the inhibition of bone resorption?

(8) The role of VPS4A in cell senescence should be measured to support the conclusion that CHMP5 regulates osteogenesis by affecting cell senescence.

(9) Cell senescence with markers, such as p21 and H2AX, co-stained with GFP should be performed in the mouse models to indicate the effects of Chmp5 on cell senescence in vivo.

(10) ADTC5 cell as osteochondromas cells line, is not a good cell model of periskeletal progenitors. Maybe primary periskeletal progenitor cell is a better choice.

Author response:

Reviewer #1 (Public review): 

Summary: 

The manuscript presents a significant and rigorous investigation into the role of CHMP5 in regulating bone formation and cellular senescence. The study provides compelling evidence that CHMP5 is essential for maintaining endolysosomal function and controlling mitochondrial ROS levels, thereby preventing the senescence of skeletal progenitor cells. 

Strengths: 

The authors demonstrate that the deletion of Chmp5 results in endolysosomal dysfunction, elevated mitochondrial ROS, and ultimately enhanced bone formation through both autonomous and paracrine mechanisms. The innovative use of senolytic drugs to ameliorate musculoskeletal abnormalities in Chmp5-deficient mice is a novel and critical finding, suggesting potential therapeutic strategies for musculoskeletal disorders linked to endolysosomal dysfunction. 

Weaknesses: 

The manuscript requires a deeper discussion or exploration of CHMP5's roles and a more refined analysis of senolytic drug specificity and effects. This would greatly enhance the comprehensiveness and clarity of the manuscript. 

We thank the reviewer for these insightful comments. The tissue-specific roles of CHMP5 and the specificity of quercetin and dasatinib treatments in Chmp5-deficient mice will be further discussed and clarified in the revised manuscript. 

Reviewer #2 (Public review): 

Summary: 

The authors try to show the importance of CHMP5 for skeletal development. 

Strengths: 

The findings of this manuscript are interesting. The mouse phenotypes are well done and are of interest to a broader (bone) field. 

Weaknesses: 

The mechanistic insights are mediocre, and the cellular senescence aspect poor. 

In total, it has not been shown that there are actual senescent cells that are reduced after D+Q-treatment. These statements need to be scaled back substantially. 

We thank the reviewer for these suggestive comments. Although multiple hallmarks of cell senescence were shown in CHMP5-deficient skeletal progenitors, we will detect and add additional markers of cell senescence in the revised manuscript. 

In addition, the effects and specificity of the Q+D treatment will be further discussed and clarified with the revision.

Reviewer #3 (Public review): 

Summary: 

In this study, Zhang et al. reported that CHMP5 restricts bone formation by controlling endolysosome-mitochondrion-mediated cell senescence. The effects of CHMP5 on osteoclastic bone resorption and bone turnover have been reported previously (PMID: 26195726), in which study the aberrant bone phenotype was observed in the CHMP5ctsk-CKO mouse model, using the same mouse model, Zhang et al., report a novel role of CHMP5 on osteogenesis through affecting cell senescence. Overall, it is an interesting study and provides new insights in the field of cell senescence and bone. 

Strengths: 

Analyzed the bone phenotype OF CHMP5-periskeletal progenitor-CKO mouse model and found the novel role of senescent cells on osteogenesis and migration. 

Weaknesses: 

(1) There are a lot of papers that have reported that senescence impairs osteogenesis of skeletal stem cells. In this study, the author claimed that Chmp5 deficiency induces skeletal progennitor cell senescence and enhanced osteogenesis. Can the authors explain the controversial results? 

Different skeletal stem cell populations in time and space have been identified and reported. This study shows that Chmp5 deficiency in periskeletal and endosteal skeletal progenitors causes cell senescence and aberrant bone formation. Although cell senescence during aging can impair osteogenesis of certain skeletal stem cells, which contributes to diseases with low bone mass such as osteoporosis, aging can also increase heterotopic mineralization/calcification in musculoskeletal soft tissues such as ligaments and tendons, which is consistent with our results in this study. These reflect out-of-order musculoskeletal mineralization during aging. We will expand the discussion and clarify the results of CHMP5-regulated cell senescence in osteogenesis in the revised manuscript.

(2) Co-culture of Chmp5-KO periskeletal progenitors with WT ones should be conducted to detect the migration and osteogenesis of WT cells in response to Chmp5-KO-induced senescent cells. In addition, the co-culture of WT periskeletal progenitors with senescent cells induced by H2O2, radiation, or from aged mice would provide more information.

Increased osteogenesis of WT skeletal progenitors in the periskeletal lesion was shown to be a paracrine mechanism of abnormal bone formation in Chmp5Ctsk mice. The coculture experiment will help confirm the effect of Chmp5-deficient skeletal progenitors on the osteogenesis of neighboring WT skeletal progenitors.

Notably, the cause and outcome of cell senescence are highly heterogeneous, and different causes of cell senescence can cause significantly different outcomes. Although the coculture of WT periskeletal progenitors with senescent cells induced by H2O2, radiation, or from aged mice would be very interesting, these are beyond the scope of the current study.

(3) Many EVs were secreted from Chmp5-deleted periskeletal progenitors, compared to the rarely detected EVs around WT cells. Since EVs of BMSCs or osteoprogenitors show strong effects of promoting osteogenesis, did the EVs contribute to the enhanced osteogenesis induced by Chmp5-defeciency? 

The WT skeletal progenitor cells from Chmp5Ctsk mice have an increased capacity of osteogenesis compared to the corresponding cells from control animals, suggesting that the EVs of the Chmp5-deleted periskeletal progenitors could promote osteogenesis of the WT skeletal progenitors, which represents a paracrine mechanism of abnormal bone formation in Chmp5 deficient animals. We will discuss and clarify these results in the revised manuscript.

(4) EVs secreted from senescent cells propagate senescence and impair osteogenesis, why do EVs secreted from senescent cells induced by Chmp5-defeciency have opposite effects on osteogenesis? 

The question is similar to comment #1. The functional heterogeneity of cellular senescence will be discussed in further detail and clarified in the revised manuscript.

(5) The Chmp5-ctsk mice show accelerated aging-related phenotypes, such as hair loss and joint stiffness. Did Ctsk also label cells in hair follicles or joint tissue? 

This is an interesting question. Although we did not check the expression of CHMP5 in hair follicles, which is outside the scope of the present study, the result in Fig. 1E showed the expression of CHMP5 in joint ligaments. Notably, abnormal periskeletal bone formation occurs predominantly at the joint ligament insertion site in Chmp5Ctsk mice, which will be elucidated and discussed in the revised manuscript.

(6) Fifteen proteins were found to increase and five proteins to decrease in the cell supernatant of Chmp5Ctsk periskeletal progenitors. How about SASP factors in the secretory profile? 

As mentioned above, the SASP phenotype and related factors of senescent cells could be highly heterogeneous depending on inducers, cell types, and timing of senescence. Most of the proteins we identified in the secretome analysis have previously been reported in the secretory profile of osteoblasts. Although we were also interested in the change of some common SASP factors, such as inflammatory cytokines, the experiment did not detect these factors because of their small molecular weights and the technical limitations of mass spec analysis. 

(7) D+Q treatment mitigates musculoskeletal pathologies in Chmp5 conditional knockout mice. In the previously published paper (CHMP5 controls bone turnover rates by dampening NF-κB activity in osteoclasts), inhibition of osteoclastic bone resorption rescues the aberrant bone phenotype of the Chmp5 conditional knockout mice. Whether the effects of D+Q on bone overgrowth is because of the inhibition of bone resorption? 

Although in Chmp5Ctsk mice we cannot exclude the effect of D+Q on osteoclasts, the effect of D+Q on osteoblast lineage cells, which is the focus of the current study, was verified in Chmp5Dmp1 mice. We will expand the discussion and make these results clearer with the revision.

(8) The role of VPS4A in cell senescence should be measured to support the conclusion that CHMP5 regulates osteogenesis by affecting cell senescence. 

We agree that additional experiments examining the role of VPS4A in cell senescence will provide more mechanistic insights. The focus of the current study is to report that CHMP5 restricts abnormal bone formation by preventing endolysosome-mitochondrion-mediated cell senescence. The roles of VPS4A in cell senescence and skeletal biology will be explored in separate studies.

(9) Cell senescence with markers, such as p21 and H2AX, co-stained with GFP should be performed in the mouse models to indicate the effects of Chmp5 on cell senescence in vivo. 

We will examine additional markers of cell senescence, as the reviewers suggest, in the revised manuscript.

(10) ADTC5 cell as osteochondromas cells line, is not a good cell model of periskeletal progenitors. Maybe primary periskeletal progenitor cell is a better choice. 

We were aware that ATDC5 cells are typically used as a chondrocyte progenitor cell line. However, our previous study showed that ATDC5 cells could also be used as a reasonable cell model for periskeletal progenitors. Furthermore, the corresponding results from primary periskeletal progenitors were shown. We will further clarify this in the revision.

In general, the comments of these reviewers will help clarify our results and further strengthen our conclusion. We will address these comments and questions point to point in more detail in the revised manuscript.

eLife Assessment

In this potentially valuable computational study, the authors conducted atomistic and coarse-grained simulations to probe the temperature-dependent phase behaviors of ELF3, a disordered component of the evening complex in plant. The results aim to highlight the role of polyQ tracts in modulating the temperature sensitivity. The level of evidence is considered incomplete, due to the lack of systematic calibration of the coarse-grained model and limited statistical uncertainty analysis, especially considering the relatively subtle nature of the differences due to temperature change.

Reviewer #1 (Public review):

Summary:

This manuscript explores the role of the Evening Complex (EC), specifically focusing on ELF3, a disordered protein component of the EC, and its temperature-dependent phase behavior. The study highlights the role of polyQ tracts in modulating temperature-sensitive condensate formation and provides a combination of computational approaches, including REST2 simulations and coarse-grained Martini simulations, to investigate how polyQ tract length and sequence context influence this behavior.

Strengths:

The study addresses a key question in plant biology - how temperature influences circadian clock-mediated growth regulation through protein phase behavior. The manuscript introduces the novel finding that polyQ tract length modulates the temperature-dependent formation of helices and condensates.

Weaknesses:

(1) Coarse-Grained Simulation Results Not Supported by Data:<br /> The results presented in Figure 6A of the manuscript do not seem to show a clear trend in the number of clusters formed as a function of polyQ tract length. This is particularly evident in the comparison between 0Q and 7Q polyQ lengths, which display statistically similar values in terms of the number of clusters. The lack of distinction between these values raises questions about the sensitivity of the coarse-grained simulations to polyQ tract length, which the authors claim as a key modulator of condensate formation. This discrepancy weakens the argument that polyQ length directly impacts the clustering behavior in the simulations.<br /> Suggested Analysis:<br /> - A more detailed statistical analysis should be performed to assess whether the observed differences between polyQ lengths are significant. This could involve hypothesis testing or the use of error bars in the graphs to better communicate the variability in the data.<br /> - Additionally, the authors should examine whether there are other features, such as cluster shape or internal structure, that might differentiate between different polyQ lengths, even if the total number of clusters is similar.

(2) Inconsistency in Cluster Size Across Temperatures (Figure 6B):<br /> The results in Figure 6B show a striking difference in the size of the largest cluster between temperatures of 290K and 300K. This abrupt shift in behavior lacks a clear mechanistic explanation. Typically, phase transitions driven by temperature are more gradual, unless there is some underlying structural or chemical shift that the authors have not accounted for. Without a clear explanation, this sudden change in behavior reduces confidence in the simulation results.<br /> Suggested Analysis:<br /> - The authors should explore possible explanations for the dramatic difference in cluster size between 290K and 300K. For example, they could investigate whether specific interactions (such as the breaking or formation of hydrogen bonds or hydrophobic contacts) might explain the behavior at higher temperatures.<br /> - It is important to check whether the coarse-grained simulation model has been adequately parameterized and scaled for accurate temperature dependence. Atomistic simulations of monomers and dimers with varying polyQ tract lengths could be used to fine-tune the coarse-grained model, ensuring it accurately reflects molecular behavior. The gross estimate of a 10% scaling factor might be insufficient and could lead to inaccurate representations of cluster formation.

(3) Scaling of Coarse-Grained Model with Atomistic Simulations:<br /> As mentioned, the coarse-grained model used in the study may not have been properly scaled against atomistic data. A simple scaling factor of 10% may not be appropriate for accurately capturing the behavior of polyQ tracts across different lengths, especially considering their sensitivity to subtle changes in temperature. Without rigorous validation against atomistic simulations, the coarse-grained model's predictions could be skewed.<br /> Suggested Analysis:

(4) To address this, the authors should compare the coarse-grained model with atomistic simulations of monomeric and dimeric forms of ELF3 with different polyQ tract lengths. By comparing key structural parameters (e.g., radius of gyration, contact maps, and clustering propensity), the authors could adjust the coarse-grained model to more accurately reflect the atomistic behavior. The authors have wealth of atomistic simulation data that could afford such benchmarking and identification of scaling factor<br /> o Additionally, the authors should investigate whether the assumed scaling factor of 10% is appropriate for each polyQ length or whether it needs to be refined based on specific properties, such as the number of hydrophobic interactions or secondary structure stability.

(5) Lack of Analysis for Liquid-Like Behavior in Phase Separation:<br /> The simulations presented in the manuscript do not analyze the liquid-like behavior of ELF3 condensates, which is a key characteristic of liquid-liquid phase separation (LLPS). In LLPS systems, condensates are often dynamic, with chains exchanging between clusters, indicating liquid-like rather than solid-like behavior. The authors fail to probe this crucial aspect, which is necessary to support the claim that ELF3 undergoes phase separation.<br /> Suggested Analysis:<br /> - The authors should conduct additional analyses to probe the liquid-like nature of the clusters formed by ELF3. One approach would be to analyze the dynamics of chain exchange between clusters, measuring how frequently chains leave one cluster and join another over time. This analysis would reveal whether the condensates behave as liquid-like, dynamic structures or more static, solid-like aggregates.<br /> - Additionally, the temperature dependence of these exchange dynamics should be investigated. In true liquid-liquid phase separation, the rate of chain exchange is often sensitive to temperature. Observing how this rate changes between 290K and 300K, for instance, could help explain the abrupt shift in cluster size seen in Figure 6B.<br /> - The authors should also analyze whether the internal structures of the condensates are consistent with a liquid-like phase. For example, radial distribution functions and contact lifetimes could be calculated to reveal whether the clusters exhibit liquid-like organization.

(6) Lack of justification of polydispersity of polyQ:<br /> The authors don't provide any rationale for choice of different copies of polyQ used in the manuscript for their chain-growth simulation studies. It will be more apt if it can be motivated via some precedent experimental observations.

(7) Lack of initiative to connect to Experiments:<br /> While the computational models and simulations provide robust theoretical insights, the absence of direct experimental validation weakens the overall impact of the manuscript. For example, experimental data on how specific mutations in the polyQ tract influence ELF3 behavior in vivo would significantly bolster the authors' claims. The manuscript would benefit from either citing existing experimental studies that corroborate these findings or from suggesting future experimental directions.

Reviewer #2 (Public review):

Summary:

The authors aimed to explore how a key protein in the circadian clock of plants, ELF3, responds to temperature changes by forming molecular condensates. They focused on understanding the role of a specific region of the protein, a polyQ tract, in promoting temperature-sensitive structural changes and regulating the formation of condensates. Through a series of computational simulations, they sought to uncover the molecular basis for ELF3's temperature responsiveness and its broader implications for plant growth and adaptation to environmental conditions.

Strengths:

The study's strength lies in its focus on an important biological question: how plants sense and respond to temperature changes at the molecular level. The authors employed a variety of computational techniques, including coarse-grained simulations, to explore the role of specific molecular features in this process. These methods provide a multi-scale view of protein behavior and offer valuable insights into how molecular structures may influence biological function.

Weaknesses:

However, there are notable weaknesses in the evidence provided. While the authors present trends in molecular changes, such as shifts in helical propensity and the formation of condensates, these results seem subtle and are not strongly substantiated by statistical analysis. The lack of error bars in the figures makes it difficult to distinguish between meaningful signals and potential noise in the data. Furthermore, the temperature-sensitive behavior appears to be influenced more by chain length than by sequence-specific effects of the polyQ region, raising questions about whether the findings truly capture the molecular mechanisms responsible for temperature sensing. Additionally, some simulations, particularly those related to the formation of condensates, do not appear fully converged, which casts further doubt on the robustness of the results.

Additional Context for Readers:

Readers should interpret the results with caution, especially regarding the molecular mechanisms proposed for temperature sensing. While the study presents interesting trends, the evidence is not definitive, and the findings may be more reflective of general protein behavior (such as the effect of chain length on condensate formation) than specific sequence-driven responses to temperature. Further experimental studies and more converged simulations will be necessary to fully understand the role of ELF3 in temperature regulation.

Author response:

We sincerely thank the reviewers for their constructive feedback and the editor for facilitating this thorough review. We found the suggestions insightful and valuable for refining our manuscript.  We would like to clarify a few points in an initial response before presenting the fully updated manuscript. First of all, we would like to emphasize the multi-scale nature of our approach, where we derived insights from both atomistic and coarse-grained simulations. Reviewers focused mostly on the coarse-grained simulations, the drawbacks of which we are aware and were a strong motivation for starting with the atomistic approach. Reviewer 1 mentioned a lack of a proposed mechanism for the increased condensate forming propensity at 300K vs. 290K, and we feel we had clearly pointed to the aromatic contacts as a mechanism for this, but we will make sure to clarify this further in the revision. Furthermore, reviewer 1 was critical of our use of the 10% adjustment to Martini protein-water interactions, which has previously been thoroughly presented and assessed in the literature (see for example Tesei et al JCTC 2022). Furthermore, for our specific system we were encouraged by the favorable comparison of our Martini simulations to the atomistic simulations, e.g. for radius of gyration, contact propensity, and solvent accessibility. We will make sure to emphasize this more clearly in the revision. Finally, we are grateful for the feedback from both reviewers and will use their comments as a guide to incorporate additional analyses and extended simulations to strengthen our conclusions in an upcoming revision.

eLife Assessment

This important study identifies species- and sex-specific neuronal cell types and gene expression in the medial preoptic area (MPOA) to help understand the evolutionary divergence of social behaviors. The evidence from single-nucleus RNA sequencing and immunostaining is convincing and suggests that cellular differences in the MPOA may contribute to behavioral variations such as mating and parental care that are apparent in two closely related deer mouse species. These rich observations provide an entry point for future hypothesis-driven experiments to demonstrate a causal role for these populations in sex- or species-variable behaviors in vertebrates. These data will be a resource that is of value to behavioral neuroscientists.

Reviewer #1 (Public review):

(1) Summary of the Paper:

This paper by Chen et al. examines the cellular composition and gene expression of the hypothalamic medial preoptic area (MPOA) in two closely related deer mouse species (P. maniculatus and P. polionotus) that exhibit distinct social behaviors. Through single-nucleus RNA sequencing (snRNA-seq), Chen et al., identify sex- and species-specific neuronal cell types that likely contribute to differences in mating and parental care. By comparing monogamous and promiscuous species, the study provides insights into how neuronal diversity and gene expression changes in the MPOA might underlie the evolution of social behaviors.

(2) Strengths of the Paper:

The paper excels in several areas. First, the data presentation is clear and well-organized, making the complex findings easy to follow. The writing is straightforward and highly accessible, which enhances the overall readability. The experimental design is innovative, particularly in how they combined samples from different species into the same dataset and then used post-hoc identification to distinguish cell types by species. This dramatically controls for potential batch effects in my opinion. Additionally, the authors contextualize their findings within the framework of previously published studies on Mus musculus, providing a strong comparative analysis that enhances the significance of their work.

3) Weaknesses of the Paper:

The major limitation of the study is the absence of causal experiments linking the observed changes in MPOA cell types to species-specific social behaviors. While the study provides valuable correlational data, it lacks functional experiments that would demonstrate a direct relationship between the neuronal differences and behavior. For instance, manipulating these cell types or gene expressions in vivo and observing their effects on behavior would have strengthened the conclusions, although I certainly appreciate the difficulty in this, especially in non-musculus mice. Without such experiments, the study remains speculative about how these neuronal differences contribute to the evolution of social behaviors.

Reviewer #2 (Public review):

Summary:

The authors report several interesting species and sex differences in cell type expression that may relate to species differences in behavior. The differential cell type abundance findings build on previously observed species/sex differences in behavior and brain anatomy. These data will be a valuable resource for behavioral neuroscientists. These findings are important but the manuscript goes too far in attributing causal influences to differences in behavior. A second important problem is that dissections used for the sequencing data include other neuropeptide-rich areas of the hypothalamus like the PVN. Although histology is included, the results in the main manuscript often do not include the mPOA making it hard to know if species/sex differences are consistent across different hypothalamic regions. The manuscript would benefit from more precise language.

Strengths:

The data are novel because cell-type atlases are available for only a few species.

The authors have clearly defined appropriate steps taken to obtain trustworthy estimations of cell type abundance. Furthermore, the criteria for each cell type assignment were described in a way for readers to easily replicate. The rigor in comparing cell abundance provides convincing evidence that these species have differences in MPOA cellular composition.

The authors have a good explanation for why 19 of the 53 neuron clusters were not classified (possible Mus/Peromyscus anatomical differences, some cell types don't have well-defined transcriptional profiles).

Validated findings with histology

Weaknesses:

Some methodology could be further explained, like the decision of a 15% cutoff value for cell type assignment per cluster, or the necessity of a multi-step analysis pipeline for gene enrichment studies.

The authors should exercise strong caution in making inferences about these differences being the basis of parental behavior. It is possible, given connections to relevant research, but without direct intervention, direct claims should be avoided. There should be clear distinctions of what to conclude and what to propose as possibilities for future research.

Histology is not performed on all regions included in the sequencing analysis.

Reviewer #3 (Public review):

Summary:

The authors performed snRNA-seq in the pre-optic area (POA), a heterogeneous brain region implicated in multiple innate behaviors, comparing two species of Peromyscus mice that possess strikingly different parenting behaviors. P. polionotus shows high levels of parental care from both sexes of parent, and P. maniculatus shows lower levels of care, predominantly displayed by dams rather than sires. The overall goal of understanding the genomic basis of behavioral variation is significant and of broad interest and comparative studies in POA in these two species is an excellent approach to tackle this question. The authors correctly point out that existing studies largely compare species that are highly divergent, such as mice and humans, which confounds the association of specific neuronal populations or gene expression patterns with distinct behaviors. They identify neuronal populations with differential abundance between species and sexes and additionally report sex and species differences in gene expression within each transcriptomic cell type. Their cell type classification is aided by mapping their Peromyscus cells onto a previously existing POA single-cell dataset generated in lab mice. However, a significant fraction of the cells cannot be assigned to Mus types, which confounds their analysis. The detection and validation of previously observed sex differences in the Gal/Moxd1 cell type and species differences in Avp expression provide additional support that their data are solid. This study provides an important resource for comparative single-cell studies in the brain.

Strengths:

This is a pioneering comparative snRNA-seq study that provides a roadmap for similar approaches in non-traditional model organisms.

The authors have identified populations that may underlie sex- and species- differences in parenting behavior in rodents.

A significant strength of the manuscript is the histological validation of their most robust marker genes.

Weaknesses:

My primary concern is that the dataset is limited: 52,121 neuronal nuclei across 24 samples, which does not provide many cells per cluster to analyze comparatively across sex and species, particularly given the heterogeneity of the region dissected. The Supplementary table reports lower UMIs/genes per cell than is typically seen as well. Perhaps additional information could be obtained from the data by not restricting the analyses to cells that can be assigned to Mus types. A direct comparison of the two Peromyscus species could be valuable as would a more complete Peromyscus POA atlas.

In Supplement 7, it appears that most neurons can be assigned as excitatory or inhibitory, but then so many of these cells remain in the unassigned "gray blob" seen in panel 1E. Clustering of excitatory and inhibitory neurons separately, as in in prior cited work in Mus POA (refs 31 and 57) may boost statistical power to detect sex and species differences in cell types. Perhaps the cells that cannot be assigned to Mus contain too few reads to be useful, in which case they should be filtered out in the QC. The technical challenges of a comparative single-cell approach are considerable, so it benefits the scientific community to provide transparency about them.

The Calb1 dimorphism as observed by immunostaining, appears much more extensive in P. maniculatus compared to P. polionotus (Figures 3 E and F). This finding is not reflected in the counts of the i20:Gal/Moxd1 cluster. The use of Calb1 staining as a proxy for the Gal/Moxd1 cluster would be strengthened if the number of POA Calb1+ neurons that are found in each cluster was apparent. There may be additional Calb+ neurons in the cells that are not annotated to a Mus cluster. This clarification would add support to the overall conclusion that there is reduced sexual dimorphism in P. polionotus.

The relationship between the sex steroid receptor expression and the sex bias in gene expression would be improved if the sex bias in sex steroid receptor expression was included in Supplementary Figure 10.

There is no explanation for the finding that there is a female bias in gene expression across all cell types in P. polionotus.

Author response:

We thank the reviewers for their thoughtful comments. 

Based on their suggestions we will: 

(1) Use more accurate language to describe the hypothalamus regions under investigation in this study. While we aimed to primarily investigate the medial preoptic area (MPOA), our dissections and sequencing data in fact capture several regions of the anterior hypothalamus including the anteroventral periventricular (AVPV), paraventricular (PVN), supraoptic (SON), suprachiasmatic nuclei (SCN), and more. We will revise the language in our manuscript to reflect that our study in fact investigates the cellular evolution of the anterior hypothalamus across behaviorally divergent deer mice.

(2) Revise our language to clarify that while our study provides a rich dataset for generating hypotheses about which cell types may contribute to behavioral differences, it does not provide any evidence of causal relationships. We hope to investigate this further in future work.

(3) Clarify specific methodological choices for which reviewers had questions, especially about the hypothalamic regions for which we did histology to validate cell abundance differences and methodological choices related to mapping our cell clusters to Mus cell types.

Our responses to each reviewer’s specific comments are below.

Reviewer #1:

The major limitation of the study is the absence of causal experiments linking the observed changes in MPOA cell types to species-specific social behaviors. While the study provides valuable correlational data, it lacks functional experiments that would demonstrate a direct relationship between the neuronal differences and behavior. For instance, manipulating these cell types or gene expressions in vivo and observing their effects on behavior would have strengthened the conclusions, although I certainly appreciate the difficulty in this, especially in non-musculus mice. Without such experiments, the study remains speculative about how these neuronal differences contribute to the evolution of social behaviors.

Yes, we agree the study lacks functional experiments. We hope that the dataset is of value for generating hypotheses about how hypothalamic neuronal cell types may govern species-specific social behaviors, and for these hypotheses to be functionally tested by us and others in future work.

Reviewer #2:

Some methodology could be further explained, like the decision of a 15% cutoff value for cell type assignment per cluster, or the necessity of a multi-step analysis pipeline for gene enrichment studies.

A 15% cutoff value for cell type assignment was chosen to include all known homology correspondences between our dataset and the Mus atlas. For example, i14:Avp/Cck cells from the Mus atlas represent Avp cells from the suprachiasmatic nuclei (SCN). Though only 17.3% of cluster 15 maps to i14:Avp/Cck, we know these two clusters correspond based on the expression of Avp and additional SCN marker genes in cluster 15 (Supp Fig 6). We will further explain this cutoff in the revised manuscript.

Our gene enrichment study includes a multi-step analysis pipeline because we wanted to control for confounders that may be introduced because of gene expression level. Genes that are more highly expressed are more accurately quantified and thus more likely to be identified as differentially expressed. Therefore, we wanted to test for gene enrichments in our set of DE genes against a background of genes with similar expression levels. We will clarify this motivation in the revised manuscript.

The authors should exercise strong caution in making inferences about these differences being the basis of parental behavior. It is possible, given connections to relevant research, but without direct intervention, direct claims should be avoided. There should be clear distinctions of what to conclude and what to propose as possibilities for future research.

Yes, we agree that we are unable to make direct claims about neuronal differences being the basis of parental behavior. We will revise our language to be clearer about which relationships we are hypothesizing and what we propose as possibilities for future research.

Histology is not performed on all regions included in the sequencing analysis.

We apologize that our language describing the hypothalamic regions included in the sequencing analysis and those included in the histology is unclear. We aimed to dissect the medial preoptic region for the sequencing analysis, but additionally captured parts of the anterior hypothalamus including the paraventricular (PVN), supraoptic (SON), and suprachiasmatic nuclei (SCN), and more.  Our histology was performed across the entire hypothalamus and includes all regions included in the sequencing data. We will revise the manuscript to more accurately describe the hypothalamic regions for which we investigated.

Reviewer #3:

My primary concern is that the dataset is limited: 52,121 neuronal nuclei across 24 samples, which does not provide many cells per cluster to analyze comparatively across sex and species, particularly given the heterogeneity of the region dissected. The Supplementary table reports lower UMIs/genes per cell than is typically seen as well. Perhaps additional information could be obtained from the data by not restricting the analyses to cells that can be assigned to Mus types. A direct comparison of the two Peromyscus species could be valuable as would a more complete Peromyscus POA atlas.

Our dataset reports ~1,500 genes and ~1,000 UMIs per nuclei which is indeed lower than is typically reported in other single nuclei datasets. Some of this discrepancy is due to a lower quality genome and annotated transcriptome available for Peromyscus compared to Mus musculus, which results in a lower mapping rate than is typically reported in Mus studies. However, our dataset was sufficient to identify known peptidergic cell types (Supp Fig 6) and to map homology to Mus cell types for 34 (64%) of our 53 clusters. Additionally, although some of our clusters contain small numbers of cells, our differential abundance analysis accounts for the variance in cell numbers observed across samples and should be robust against any increase in variance due to small numbers. In fact, even differential abundance of very small cell clusters such as oxytocin neurons (cell type 40) was validated by histology. 

We would like to clarify that all analyses were performed on all cell clusters, regardless of whether or not they could be assigned homology to a Mus cell type. All the cell types that we identified as differentially abundant or contained significant sex differences happened to be cell types for which homology to a Mus cell type could be defined. This may arise for a relatively uninteresting reason: cell types that have more distinct transcriptional signatures will be more accurately clustered, leading to more accurate identification of homology as well as more accurate measurements of differential abundance / expression. We will revise language to make this more clear in our manuscript.

In Supplement 7, it appears that most neurons can be assigned as excitatory or inhibitory, but then so many of these cells remain in the unassigned "gray blob" seen in panel 1E. Clustering of excitatory and inhibitory neurons separately, as in prior cited work in Mus POA (refs 31 and 57) may boost statistical power to detect sex and species differences in cell types. Perhaps the cells that cannot be assigned to Mus contain too few reads to be useful, in which case they should be filtered out in the QC. The technical challenges of a comparative single-cell approach are considerable, so it benefits the scientific community to provide transparency about them.

We are not certain about why we are unable to cluster and assign homology to many of our cells (i.e. cells in the unassigned “gray blob”). However, we note that even in the Mus atlas, many cells did not belong to obvious clusters by UMAP visualization and that several clusters lacked notable marker genes and were designated simply as “Gaba” and “Glut” clusters. Therefore, it is unsurprising that our own dataset also contains cells that lack the transcriptional signatures needed to be clustered and/or mapped to Mus cell types. We do know, however, that the median number of reads/nuclei is uniform across cell clusters and does not explain why some clusters could not be assigned to Mus. We will add this information to our revised manuscript. 

We do not think that a two-stage clustering (i.e. clustering first by excitatory vs. inhibitory neurons) is expected to gain power to resolve cell types in this case. Excitatory vs. inhibitory neurons are clearly separable on our UMAP (Supp Fig 7) so that information is already being used by our clustering procedure. However, we will explore this further in our revised manuscript to see if doing so will boost statistical power.

The Calb1 dimorphism as observed by immunostaining, appears much more extensive in P. maniculatus compared to P. polionotus (Figures 3 E and F). This finding is not reflected in the counts of the i20:Gal/Moxd1 cluster. The use of Calb1 staining as a proxy for the Gal/Moxd1 cluster would be strengthened if the number of POA Calb1+ neurons that are found in each cluster was apparent. There may be additional Calb+ neurons in the cells that are not annotated to a Mus cluster. This clarification would add support to the overall conclusion that there is reduced sexual dimorphism in P. polionotus.

From the Mus MPOA atlas (which includes both single-cell sequencing data and imaging-based spatial information), it is known that the i20:Gal/Moxd1 cluster comprises sexually dimorphic cells that make up both the BNST and the SDN-POA. These sexually dimorphic cells are well-studied and known to be marked by Calb1, which we used in immunostaining as a proxy for i20:Gal/Moxd1. 

However, we would like to clarify that in our study, the immunostaining of Calb1+ neurons and the sequencing counts of the i20:Gal/Moxd1 cluster are not completely reflective of each other because our sequencing dataset only captured the ventral portion of the BNST. Therefore our i20:Gal/Moxd1 counts contain a combination of some Calb1+ BNST cells and likely all Calb1+ SDN-POA cells and is difficult to interpret on its own. Our histology, however, covers the entire hypothalamus and is more reliable for identifying sex and species differences in each region. We will clarify this in the revised manuscript. 

The relationship between the sex steroid receptor expression and the sex bias in gene expression would be improved if the sex bias in sex steroid receptor expression was included in Supplementary Figure 10.

We will include this in the revised manuscript. 

There is no explanation for the finding that there is a female bias in gene expression across all cell types in P. polionotus.

We also find this observation interesting but don’t have a good explanation for why at this point. We plan to follow this up in future work.

eLife Assessment

This valuable study investigates the brain representations of Braille letters in blind participants and provides evidence using EEG and fMRI that the decoding of letter identity across the reading hand takes place in the visual cortex. The evidence supporting the claims of the authors is convincing and the work will be of interest to neuroscientists working on brain plasticity.

Reviewer #1 (Public review):

Summary:

The researchers examined how individuals who were born blind or lost their vision early in life process information, specifically focusing on the decoding of Braille characters. They explored the transition of Braille character information from tactile sensory inputs, based on which hand was used for reading, to perceptual representations that are not dependent on the reading hand.

They identified tactile sensory representations in areas responsible for touch processing and perceptual representations in brain regions typically involved in visual reading, with the lateral occipital complex serving as a pivotal "hinge" region between them.

In terms of temporal information processing, they discovered that tactile sensory representations occur prior to cognitive perceptual representations. The researchers suggest that this pattern indicates that even in situations of significant brain adaptability, there is a consistent chronological progression from sensory to cognitive processing.

Strengths:

By combining fMRI and EEG, and focusing on the diagnostic case of Braille reading, the paper provides an integrated view of the transformation processing from sensation to perception in the visually deprived brain. Such a multimodal approach is still rare in the study of human brain plasticity and allows to discern the nature of information processing in blind people early visual cortex, as well as the timecourse of information processing in a situation of significant brain adaptability.

Weaknesses:

ROI and searchlight analyses are not completely overlapping, although this might be due to the specific limits and strengths of each approach. Moreover, the conclusions regarding the behavioral relevance of the sensory and perceptual representations in the putatively reorganized brain, although important, are limited due to the behavioral measurements adopted.

Reviewer #2 (Public review):

Summary:

Haupt and colleagues performed a well-designed study to test the spatial and temporal gradient of perceiving braille letters in blind individuals. Using cross-hand decoding of the read letters, and comparing it to the decoding of the read letter for each hand, they defined perceptual and sensory responses. Then they compared where (using fMRI) and when (using EEG) these were decodable. Using fMRI, they showed that low-level tactile responses specific to each hand are decodable from the primary and secondary somatosensory cortex as well as from IPS subregions, the insula and LOC. In contrast, more abstract representations of the braille letter independent from the reading hand were decodable from several visual ROIs, LOC, VWFA and surprisingly also EVC. Using a parallel EEG design, they showed that sensory hand-specific responses emerge in time before perceptual braille letter representations. Last, they used RSA to show that the behavioral similarity of the letter pairs correlates to the neural signal of both fMRI (for the perceptual decoding, in visual and ventral ROIs) and EEG (for both sensory and perceptual decoding).

Strengths:

This is a very well-designed study and it is analyzed well. The writing clearly describes the analyses and results. Overall, the study provides convincing evidence from EEG and fMRI that the decoding of letter identity across the reading hand occurs in the visual cortex in blindness. Further, it addresses important questions about the visual cortex hierarchy in blindness (whether it parallels that of the sighted brain or is inverted) and its link to braille reading.

Author response:

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

Public Reviews: 

Reviewer #1 (Public Review): 

We thank Reviewer #1 for the relevant and insightful comments on our paper. Please find our detailed answers below in the Recommendations to the Authors section.

Summary: 

The researchers examined how individuals who were born blind or lost their vision early in life process information, specifically focusing on the decoding of Braille characters. They explored the transition of Braille character information from tactile sensory inputs, based on which hand was used for reading, to perceptual representations that are not dependent on the reading hand. 

They identified tactile sensory representations in areas responsible for touch processing and perceptual representations in brain regions typically involved in visual reading, with the lateral occipital complex serving as a pivotal "hinge" region between them.

In terms of temporal information processing, they discovered that tactile sensory representations occur prior to cognitive-perceptual representations. The researchers suggest that this pattern indicates that even in situations of significant brain adaptability, there is a consistent chronological progression from sensory to cognitive processing. 

Strengths: 

By combining fMRI and EEG, and focusing on the diagnostic case of Braille reading, the paper provides an integrated view of the transformation processing from sensation to perception in the visually deprived brain. Such a multimodal approach is still rare in the study of human brain plasticity and allows us to discern the nature of information processing in blind people's early visual cortex, as well as the time course of information processing in a situation of significant brain adaptability. 

Weaknesses: 

The lack of a sighted control group limits the interpretations of the results in terms of profound cortical reorganization, or simple unmasking of the architectural potentials already present in the normally developing brain. 

We thank the reviewer for raising this important point! We acknowledge that our claims regarding the unmasking of architectural potentials in both the normally developing and visually deprived brain are limited by the study design we employed. However, we note that defining an appropriate control group and assessing non-visual reading in sighted participants is far from straightforward. We discuss these issues in our response to the Public Review of Reviewer 2.

Moreover, the conclusions regarding the behavioral relevance of the sensory and perceptual representations in the putatively reorganized brain are limited due to the behavioral measurements adopted.

We agree with the reviewer that the relation between behavior and neural representations as established via perceived similarity judgments are task-dependent, and that a richer assessment of behavior would be valuable. Please note, however, that this limitation pertains to any experimental task used to assess behavior in the laboratory. Our major goal was to assess whether the identified neural representations are suitably formatted to be used by the brain for at least one behavior rather than being epiphenomenal. We found that the representations are suitably formatted for similarity judgments, thus establishing that they are relevant for at least this behavior. We also argue that judging similarity is a complex task that may underlie many other relevant behaviors. We discuss this point further in response to the Recommendations to the Authors.

Reviewer #2 (Public Review): 

We thank the reviewer for the considerate and thoughtful suggestions. Please find a detailed description of the implemented changes below.

Summary: 

Haupt and colleagues performed a well-designed study to test the spatial and temporal gradient of perceiving braille letters in blind individuals. Using cross-hand decoding of the read letters, and comparing it to the decoding of the read letter for each hand, they defined perceptual and sensory responses. Then they compared where (using fMRI) and when (using EEG) these were decodable. Using fMRI, they showed that low-level tactile responses specific to each hand are decodable from the primary and secondary somatosensory cortex as well as from IPS subregions, the insula, and LOC. In contrast, more abstract representations of the braille letter independent from the reading hand were decodable from several visual ROIs, LOC, VWFA, and surprisingly also EVC. Using a parallel EEG design, they showed that sensory hand-specific responses emerge in time before perceptual braille letter representations. Last, they used RSA to show that the behavioral similarity of the letter pairs correlates to the neural signal of both fMRI (for the perceptual decoding, in visual and ventral ROIs) and EEG (for both sensory and perceptual decoding). 

Strengths: 

This is a very well-designed study and it is analyzed well. The writing clearly describes the analyses and results. Overall, the study provides convincing evidence from EEG and fMRI that the decoding of letter identity across the reading hand occurs in the visual cortex in blindness. Further, it addresses important questions about the visual cortex hierarchy in blindness (whether it parallels that of the sighted brain or is inverted) and its link to braille reading. 

Weaknesses: 

Although I have some comments and requests for clarification about the details of the methods, my main comment is that the manuscript could benefit from expanding its discussion. Specifically, I'd appreciate the authors drawing clearer theoretical conclusions about what this data suggests about the direction of information flow in the reorganized visual system in blindness, the role VWFA plays in blindness (revised from the original sighted role or similar to it?), how information arrives to the visual cortex, and what the authors' predictions would be if a parallel experiment would be carried out in sighted people (is this a multisensory recruitment or reorganization?). The data has the potential to speak to a lot of questions about the scope of brain plasticity, and that would interest broad audiences. 

We thank the reviewer for the opportunity to provide clearer theoretical conclusions from our data. We elaborate on each of the points raised by the reviewer in the discussion section.

Concerning the direction of information flow in the reorganized visual system in blindness, we focus on information arrival to EVC and information flow beyond EVC.

p. 11, ll. 376-386, Discussion 4.1:

“Overall, identifying braille letter representations in widespread brain areas raises the question of how information flow is organized in the visually deprived brain. Functional connectivity studies report deprivation-driven changes of thalamo-cortical connections which could explain both arrival of information to and further flow of information beyond EVC. First, the coexistence of early thalamic connections to both S1 and V1 (Müller et al., 2019) would enable EVC to receive from different sources and at different timepoints. Second, potentially overlapping connections from both sensory cortices to other visual or parietal areas (Ioannides et al., 2013) could enable the visually deprived brain to process information in a widespread and interconnected array of brain areas. In such a network architecture, several brain areas receive and forward information at the same time. In contrast to information discretely traveling from one processing unit to the next in the sighted brain’s processing cascade, we can rather picture information flowing in a spatially and functionally more distributed and overlapping fashion.”

Regarding the role of VWFA, we propose that the functional organization of VWFA is modality-independent.

p. 10, ll. 346-348, Discussion 4.1:

“Second, we found that VWFA contains perceptual but not sensory braille letter representations. By clarifying the representational format of language representations in VWFA, our results support previous findings of the VWFA being functionally selective for letter and word stimuli in the visually deprived brain (Reich et al., 2011; Striem-Amit et al., 2012; Liu et al., 2023). Together, these findings suggest that the functional organization of the VWFA is modality-independent (Reich et al., 2011), depicting an important contribution to the ongoing debate on how visual experience shapes representations along the ventral stream (Bedny et al., 2021).” Lastly, we would like to share our thoughts about carrying out a parallel experiment in sighted people. 

In general, we agree that it seems insightful to conduct a parallel, analogous experiment in sighted participants with the aim to disentangle whether the effects seen in blind participants are due to multisensory recruitment or reorganization. However, before making predictions regarding the outcome, we would have to define an analogous experiment in sighted participants that taps into the same mechanisms. This, however, is difficult to do as it is unclear what counts as analogous. For example, if we compare braille reading to reading visually presented braille dot arrays or Roman letters, we will assess visual object processing, a different mechanism from that involved in braille reading. Alternatively, if we compare braille reading to sighted participants reading embossed Roman letters haptically or ideally even reading Braille after extensive training, we still face the inherent problem that sighted participants have visual experiences and could use visual imagery strategies in these nonvisual tasks. As we cannot experimentally ensure that sighted participants do not use visual strategies to solve a task, this would always complicate drawing conclusions about the underlying processes. More specifically, we could never pinpoint whether differences between sighted and blind participants are due to measuring different mechanisms or measuring the same mechanism and unravelling underlying changes (i.e., multisensory recruitment or reorganization). Finally, apart from potential confounds due to visual imagery, considering populations of sighted readers and Braille readers as only differing with regard to their input modality and otherwise being comparable is problematic: In general, blind populations are more heterogenous than most typical samples due to various factors such as aetiologies, onset and severity (Merabet & Pascual-Leone, 2010). Even when carrying out studies in highly specific population subsamples, such as in congenitally blind braille readers, vast within-group differences remain, e.g., the quality and quantity of their braille education, as well as across braille and print readers, e.g., different passive exposure to braille versus written letters during childhood (Englebretson et al., 2023). Hence, to fully match the groups in terms of learning experience we would, for example, have to teach sighted infants braille reading in childhood and follow them up until a comparable age. This approach does not seem feasible. 

p. 10, ll. 328-341, Discussion 4.1:

“We note that our findings contribute additional evidence but cannot conclusively distinguish between the competing hypotheses that visually deprived brains dynamically adjust to the environmental constraints versus that they undergo a profound cortical reorganization. Resolving this debate would require an analogous experiment in sighted people which taps into the same mechanisms as the present study. Defining a suitable control experiment is, however, difficult. Any other type of reading would likely tap into different mechanism than braille reading. Further, whenever sighted participants are asked to perform a haptic reading task, outcomes can be confounded by visual imagery driving visual cortex (Dijkstra et al., 2019). Thus, the results would remain ambiguous as to whether observed differences between the groups index different mechanisms or plastic changes in the same mechanisms. Last, matching groups of sighted readers and braille readers such that they only differ with regard to their input modality seems practically unfeasible: There are vast differences within the blind population in general, e.g., aetiologies, onset and severity, and the subsample of congenitally blind braille readers more specifically, e.g., the quality and quantity of their braille education, as well as across braille and print readers, e.g., different passive exposure to braille versus written letters during childhood (Englebretson et al., 2023; Merabet & Pascual-Leone, 2010).”

While we appreciate that the conclusions we can draw from our results are limited by our sample and defining an appropriate parallel experiment in sighted participants is difficult for the reasons discussed above, we would still like to share our speculations regarding the process underlying our result pattern. We think that our results, taken together with results of previous studies, suggest that EVC does not undergo fundamental reorganization in the case of visual deprivation. Rather, it can flexibly adjust to given processing requirements. This flexibility is not infinite; adjustments are limited by the area’s architectural and computational capacity. Importantly, we think that this claim refers to an unmasking of preexisting potential rather than multisensory recruitment.

To aid in drawing even more concrete conclusions about the flow of information, I suggest that the authors also add at least another early visual ROI to plot more clearly whether EVC's response to braille letters arrives there through an inverted cortical hierarchy, intermediate stages from VWFA, or directly, as found in the sighted brain for spoken language. 

We thank the reviewer for this comment. However, EVC here consists of V1 to V3, and we already also assess V4, LOC, VWFA and LFA. Thus, we assess regions at all levels of processing from mid- over low- to high-level and cannot add a further interim ROI. Our results using this ROI set do not allow us to arbitrate between the hypotheses raised by the reviewer.

Similarly, it may be informative to look specifically at the occipital electrodes' time differences between decoding for the different parameters and their correlation to behavior.

We thank the reviewer for this suggestion. However, the spatial resolution of EEG measurements is limited, and we cannot convincingly determine the neural source of signals being recorded from specific electrodes, i.e., occipital. When we reduce the number of electrodes before analysis, we primarily see comparable qualitative trends in the data albeit with a reduction in signal-to-noise-ratio.

To illustrate, we repeated the EEG time decoding and the EEG-behavior RSA with only occipital and parieto-occipital electrodes (n=8) instead of all electrodes (n=63) and added the results to the Supplementary Material (see Supplementary Figure 3 and 4). Overall, we observe a reduction in signal-to-noise-ratio. This is not surprising given that the EEG searchlight decoding results (Figure 3b) reveal sources of the decoding signals extend beyond occipital and parieto-occipital electrodes. 

In the EEG time decoding analysis, we see a comparable trend to the whole brain EEG analysis but do not find a significant difference in onsets of sensory and perceptual representation. 

In the behavior-EEG RSA, we do find that the correlations between behavior and sensory representations emerge significantly earlier than correlations between behavior and perceptual representations. (N = 11, 1,000 bootstraps, one-tailed bootstrap test against zero, P< 0.001). This result is in line with the whole brain EEG analysis.

Regarding the methods, further detail on the ability to read with both hands equally and any residual vision of the participants would be helpful.

We thank the reviewer for raising this point. We assessed participants’ letter reading capabilities in a short screening task prior to the experiment. Participants read letters with both hands separately and we used the same presentation time as in the experiment. As the result showed that average performance for recognizing letters with the left hand (89%) and right hand (88%) were comparable. We did not measure continuous reading in the present study, and we did not assess further information about participants’ ability to read equally well with both hands. 

While the information about the screening task was previously included in Methods section 5.3.2 EEG experiment, we now moved it into a separate section 5.3.3 Braille screening task to make the information better accessible. 

p. 14, ll. 529-533, Methods 5.3.3:

“Prior to the experiment, participants completed a short screening task during which each letter of the alphabet was presented for 500ms to each hand in random order. Participants were asked to verbally report the letter they had perceived to assess their reading capabilities with both hands using the same presentation time as in the experiment. The average performance for the left hand was 89% correct (SD = 10) and for the right hand it was 88% correct (SD = 13).”

We thank the reviewer for the suggestion to include information regarding participant’s residual vision. We now added information about participants’ residual light perception to Supplementary Table 1.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors): 

(1) ROI vs Searchlight Results: Figures 2 b and c do not seem to match. The ROI results (b) should be somehow consistent with the whole brain results (c), but "perceptual" decoding in the searchlight (in green) seems localized in sensorimotor areas while for the same classification, no sensorimotor ROI is significant. can the authors clarify this difference?

Similarly, perceptual decoding does not emerge in EVC with the searchlight analysis, whereas is quite strong in ROI analysis.

We agree that the results of the ROI and searchlight decoding do not show a direct match. We think that this difference is due to methodological reasons. For example, ROI decoding can be more sensitive when ROIs follow functionally relevant boundaries in the brain, in comparison to spheres used in searchlight decoding that do not. In turn, searchlight decoding may be more sensitive when information is distributed across functional boundaries that would be captured in different ROIs rather than combined, or when ROI definition is difficult (such as here in the visual system of blind participants).

However, we point out that the primary goal of our searchlight decoding was to show that no other areas beyond our hypothesized ROIs contained braille letter representations, rather than reproducing the ROI results.

Decoding accuracies are tested against chance (50% for pairwise classifications) according to methods. In the case of "sensory and perceptual" and "perceptual" classification, this is straightforward. In the case of the analysis that isolates "sensory" representations though the difference is computed between "sensory and perceptual" and "perceptual" decoding accuracies, the accuracies resulting from this difference should thus be centered around 0.

Are the accuracies tested against 0 in this case? This is not specified in the methods. Furthermore, the data reported in Figure 2 and Figure 3. seem to have 0% as a baseline and the label states "decoding accuracy". Can the authors clarify whether the reported data are the difference in accuracy with an estimated empirical baseline or an expected baseline of 50%? 

The reviewer is correct in stating that we tested “sensory and perceptual” and “perceptual” against chance level and the difference score “sensory” against 0 and that this information was missing in the methods section.

We now specify in the methods that we are testing the accuracies for the “sensory” analysis against 0.

p. 16, ll. 625-627, Methods 5.6:

“We conducted subject-specific braille letter classification in two ways. First, we classified between letter pairs presented to one reading hand, i.e., we trained and tested a classifier on brain data recorded during the presentation of braille stimuli to the same hand (either the right or the left hand). This yields a measure of hand-dependent braille letter information in neural measurements. We refer to this analysis as within-hand classification. Second, we classified between letter pairs presented to different hands in that we trained a classifier on brain data recorded during the presentation of stimuli to one hand (e.g., right), and tested it on data related to the other hand (e.g., left). This yields a measure of hand-independent braille letter information in neural measurements. We refer to this analysis as across-hand classification. We tested both within-hand and across-hand pairwise classification accuracies against a chance level of 50%. We also calculated a within-across hand classification score which we compared against 0.”

Regarding Figures 2 and 3, we plot the results as decoding accuracies minus chance level to standardize the y-axes for all three analyses, i.e., compare them to 0. We have corrected the y-axis labels accordingly. 

In our analyses, we assumed an expected baseline of 50%. But in the response below we provide evidence that our results remain stable whether using an expected or empirical baseline.

If my understanding is correct, a potential problem persists. The different analyses may not be comparable, because in the "sensory" analysis the baseline is empirically defined, being the classification accuracies of the "perceptual" decoding, while in the other two analyses, the baseline is set at 50%. There are suggestions in the literature to derive empirically defined baselines by randomly shuffling the trial labels and repeating the classification accuracies [grootswagers 2017]. In the context of the present work, its use will make the different statistical analyses more comparable. I would thus suggest the authors define the baseline empirically for all their analyses or, given the high computational demand of this analysis, provide evidence that the results are not affected by this difference in the baseline. 

We thank the reviewer for raising this point. As the reviewer correctly stated, the “sensory” analysis has an empirically defined baseline because it is a difference score while in the other two analyses the baseline is set at 50%.

To provide evidence that our results are not affected by this difference in baseline, we now re-ran the EEG time decoding. We derived null distributions from the empirical data for all three analyses, following the guidelines from Grootswagers 2017 (page 688, section “Evaluation of Classifier Performance and Group Level Statistical Testing Statistical”):

“Another popular alternative is the permutation test, which entails repeatedly shuffling the data and recomputing classifier performance on the shuffled data to obtain a null distribution, which is then compared against observed classifier performance on the original set to assess statistical significance (see, e.g., Kaiser et al., 2016; Cichy et al., 2014; Isik et al., 2014). Permutation tests are especially useful when no assumptions about the null distribution can be made (e.g., in the case of biased classifiers or unbalanced data), but they take much longer to run (e.g., repeating the analysis 10,000 times).”

Running a sign permutation test with 10,000 repetitions, we show that the results are comparable to the previously reported results based on one-sided Wilcoxon signed rank tests. We are, therefore, confident that our reported results are not affected by this difference in baseline. We now added this control analysis to the results section and supplementary material (see Supplementary Figure 5).

p. 7-8, ll. 213-215, Results 3.2: 

“Importantly, the temporal dynamics of sensory and perceptual representations differed significantly. Compared to sensory representations, the significance onset of perceptual representations was delayed by 107ms (21-167ms) (N = 11, 1,000 bootstraps, one-tailed bootstrap test against zero, P= 0.012). This results pattern was consistent when defining the analysis baseline empirically (see Supplementary Figure 5).”

(2) According to the authors, perceptual rather than sensory braille letter representations identified in space are suitably formatted to guide behavior. However, they acknowledge that this finding is likely to be task-dependent because it is based on subject similarity ratings.

Maybe they could use a more objective similarity measurement of Braille letters similarity?

For instance, they can compare letters using Jaccard similarity (See for instance: Bottini et al. 2022). 

We thank the reviewer for the opportunity to clarify. We acknowledge that our findings regarding the behavioral relevance of the identified neural representations are task-dependent. But, importantly, this is not because we use perceived similarity ratings as a measurement, but because we only use one measurement while there are infinitely many other potential tasks to assess behavior. This means that the same limitation holds when using another similarity measure like Jaccard similarity. We now clarify this in the Discussion section: 

p. 12, ll. 419-420, Discussion 4.3:

“Our results clarified that perceptual rather than sensory braille letter representations identified in space are suitably formatted to guide behavior. However, we only use one specific task to assess behavior and, therefore, acknowledge that this finding is taskdependent.”

Nevertheless, we calculated Jaccard similarity based on the definition used in Bottini et. al. There are no significant correlations for the EEG-behavior or fMRI-behavior RSA when we use the Jaccard matrix and subject-specific EEG or fMRI RDMs (see Supplementary Figure 6).

This demonstrates that braille letter similarity ratings are significantly correlated with neural representations in space and time but Jaccard similarity of braille dot overlaps is not. 

(3) If the primacy of perceptual similarity holds also with more objective measures of letter similarity, I think the authors should spend a few more words characterizing the results in fMRI and EEG that are rather divergent (concerning this analysis). Indeed, EEG analysis shows a significant correlation between similarity ratings and within-hand classification accuracy, although this correlation does not emerge in the "sensory" ROIs. I think these findings can be put together, hypothesizing that sensory-based similarity correlates with behavior but only in perceptual ROIs. However, why so? Can the authors provide a more mechanistic explanation? Am I missing something? 

We thank the reviewer for this intriguing idea. We now speculate about how we could harmonize the results from the behavior-EEG and behavior-fMRI RSAs in the discussion section. 

p. 12, ll. 438-442, Discussion 4.3:

“Similarity ratings and sensory representations as captured by EEG are correlated, and so are similarity ratings and representations in perceptual ROIs, but not sensory ROIs. This might be interpreted as suggesting a link between the sensory representations captured in EEG and the representations in perceptual ROIs. However, we do not have any evidence towards this idea. Differing signalto-noise ratios for the different ROIs and sensory versus perceptual analysis could be an alternative explanation.“

(4) In the methods they state that EEG decoding is tested against chance at each time point but these results are not reported, only latency analysis is reported. Can the authors report the significant time points of the EEG time series decoding?  

We thank the reviewer for catching this inconsistency! We have now added this information to Figure 3a.

(5) In fMRI ROI definition procedure, the top 321 voxels of each anatomical ROI that had the highest functional activation were selected. The number of voxels is based on the smaller ROI, which to my understanding means that for this ROI all the voxels were selected potentially introducing noise and impacting the comparison between ROIs. Can the authors clarify which ROI was the smallest? 

Thank you for the question! The smallest ROI was V4. This indeed means that for this ROI all voxels were selected. This could have led to our results being noisy in V4 but should not influence the results in other ROIs. We now added this information to the methods section.  p. 15, ll. 592, Methods 5.4.4:

“The smallest mask was V4 which included 321 voxels.”

(6) Finally, the author suggests that: "Importantly, higher-level computations are not limited to the EVC in visually deprived brains. Natural sound representations 41 and language activations 53 are also located in EVC of sighted participants. This suggests that EVC, in general, has the capacity to process higher-level information 54. Thus, EVC in the visually deprived brain might not be undergoing fundamental changes in brain organization 53. This promotes a view of brain plasticity in which the cortex is capable of dynamic adjustments within pre-existing computational capacity limits 4,53-55." - The presence of a sighted control group would have strengthened this claim. 

We agree with the reviewer and now discuss the limitations of our approach in the discussion section (see response to weaknesses raised by Reviewer 2 in the Public Review above).

Reviewer #2 (Recommendations For The Authors): 

(1) Can the authors comment on the reaction time of the two reading hands? Completely ambidextrous reading is not necessarily common, so any differences in ability or response time across the hands may affect the EEG results. Alternatively, do the authors have any additional behavioral data about the participants' ability to read well with both hands? 

We thank the reviewer for these questions! We did not assess reaction times and acknowledge this as a limitation. We did, however, measure accuracies and would have expected to see a speed-accuracy-trade off if reaction times would differ between hands, i.e., we would have expected lower accuracy for the hand with higher RTs. But this was not the case: our participants had comparable accuracy values when reading letters with both hands (see methods section 5.3.3 and answer to Public Review above). This measure indicated that participants recognized Braille letters presented for 500ms equally well with both index fingers.

(2) Please add information about any residual sight in the blind participants (or are they all without light perception?)

We have now added information about residual light perception in Supplementary Table 1 (see above in response to Public Review).

(3) Is active tactile exploration involved, or are the participants not moving their fingers at all over the piezo-actuators? Can the authors elaborate more on how the participants used this passive input?

We thank the reviewer for the opportunity to clarify. Our experimental setup does not involve tactile exploration or sliding motions. Instead, participants rest their index fingers on the piezo-actuators and feel the static sensation of dots pushing up against their fingertips. We assume that participants used the passive input of specific dot stimulation location on fingers to perceive a dot array which, in turn, led to the percept of a braille letter.

We now specify this information in the methods section.

p. 13, ll. 474-475, Methods 5.2:

“The modules were taped to the clothes of a participant for the fMRI experiment and on the table for the EEG and behavioral experiment. This way, participants could read in a comfortable position with their index fingers resting on the braille cells to avoid motion confounds. Importantly, our experimental setup did not involve tactile exploration or sliding motions. We instructed participants to read letters regardless of whether the pins passively stimulated their immobile right or left index finger.”

(4) I appreciated the RSA analysis, but remain curious about what the ratings were based on.

Do the authors know what parameters participants used to rate for? Were these consistent across participants? That would aid in interpreting the results.

We thank the reviewer for the interest in our representational similarity analyses linking the neural representations to behavior. 

We do not know which parameters participants explicitly used to rate the similarity between letters. We instructed participants to freely compare the similarity of pairs of braille letters without specifying which parameters they should use for the similarity assessment. We speculate that participants used a mixture of low-level features such as stimulation location on fingers and higher-level features such as linguistic similarity between letters. We now clarify the free comparison of braille letter pairs in the methods section:

p. 14, ll. 538-539, Methods 5.3.4:

“Each pair of letters was presented once, and participants compared them with the same finger. We instructed participants to freely compare the similarity of pairs of Braille letters without specifying which parameters they should use for the similarity assessment. The rating was without time constraints, meaning participants decided when they rated the stimuli. Participants were asked to verbally rate the similarity of each pair of braille letters on a scale from 1 = very similar to 7 = very different and the experimenter noted down their responses.”

(5) Can the authors provide confusion matrices for the decoding analyses in the supplementary materials? This could be informative in understanding what pairs of letters are most discernable and where. 

We have added confusion matrices for within- and between-hand decoding for all ROIs and for the time points 100ms, 200ms, 300ms and 400ms to the Supplementary Material (see Supplementary Figures 7-10).

(6) Was slice time correction done for the fMRI data? This is not reported. 

We now added this information to the methods section - our fMRI preprocessing pipeline did not include slice timing correction.  

p. 14, ll. 554, Methods 5.4.2:

“We did not apply high or low-pass temporal filters and did not perform slice time correction.”

eLife Assessment

This study presents a useful finding on the ferroptosis-mediated tumor microenvironment (TME) in triple-negative breast cancer (TNBC) using public single-cell RNA sequencing (scRNA-seq) and bulk RNA sequencing data. The evidence supporting the claims of the authors is somewhat incomplete and some data are rather questionable; the authors should clarify the relations between ferroptosis-related genes in immune cells and those genes applied in a risk factor analysis in tumor cells. Moreover, the authors should provide experimental validation for the risk score model based on ferroptosis-related genes. The work will be of interest to scientists or clinical scientists working in the field of breast cancer.

Reviewer #1 (Public review):

Summary:

Triple-negative breast cancer (TNBC) accounts for approximately 15-20% of all breast cancers. Compared to other types of breast cancer, TNBC exhibits highly aggressive clinical characteristics, a greater likelihood of metastasis, poorer clinical outcomes, and lower survival rates. Immunotherapy is an important treatment option for TNBC, but there is significant heterogeneity in treatment response. Therefore, it is crucial to accurately identify immunosuppressive patients before treatment and actively seek more effective therapeutic approaches for TNBC patients.

Strengths:

In this work, the authors collected and integrated data from single cells and large volumes of RNA sequencing and RNA-SEQ to analyze the TME landscape mediated by genes associated with iron death. On this basis, the prediction model of prognosis and treatment response of 131 patients was constructed using a machine learning algorithm, which is beneficial to provide individualized and precise treatment guidance for breast cancer patients.

Weaknesses:

However, there are still some issues that need to be clarified:

(1) The description of the research background is too brief and concise, and it is necessary to add some information about the limitations of existing methods and the differences and advantages of this study compared with other published relevant studies, so as to better highlight the necessity and research value of this study.

(2) This study is a retrospective analysis of a public data set and lacks experimental validation and prospective experiments to support the results of bioinformatics analysis. This should be added to the acknowledgment of limitations in the study.

Reviewer #2 (Public review):

Summary:

This study aims to explore the ferroptosis-related immune landscape of TNBC through the integration of single-cell and bulk RNA sequencing data, followed by the development of a risk prediction model for prognosis and drug response. The authors identified key subpopulations of immune cells within the TME, particularly focusing on T cells and macrophages. Using machine learning algorithms, the authors constructed a ferroptosis-related gene risk score that accurately predicts survival and the potential response to specific drugs in TNBC patients.

Strengths:

The study identifies distinct subpopulations of T cells and macrophages with differential expression of ferroptosis-related genes. The clustering of these subpopulations and their correlation with patient prognosis is highly insightful, especially the identification of the TREM2+ and FOLR2+ macrophage subtypes, which are linked to either favorable or poor prognoses. The risk model thus holds potential not only for prognosis but also for guiding treatment selection in personalized oncology.

Weaknesses:

The study has a relatively small sample size, with only 9 samples analyzed by scRNA-seq. Given the typically high heterogeneity of the tumor microenvironment (TME) in cancer patients, this may affect the accuracy of the conclusions. The scRNA-seq analysis focuses on the expression of ferroptosis-related genes in various cells within the TME. In contrast, bulk RNA sequencing uses data from tumor samples, and the results between the two analyses are not consistent. The bulk RNA sequencing results may not accurately capture the changes happening in the microenvironment.

eLife Assessment

This fundamental study substantially advances our understanding of how habitat fragmentation and climate change jointly influence bird community thermophilization in a fragmented island system. The authors provide convincing evidence using appropriate and validated methodologies to examine how island area and isolation affect the colonization of warm-adapted species and the extinction of cold-adapted species. This study is of high interest to ecologists and conservation biologists, as it provides insight into how ecosystems and communities respond to climate change.

Reviewer #3 (Public review):

Summary:

Juan Liu et al. investigated the interplay between habitat fragmentation and climate-driven thermophilization in birds in an island system in China. They used extensive bird monitoring data (9 surveys per year per island) across 36 islands of varying size and isolation from the mainland covering 10 years. The authors use extensive modeling frameworks to test a general increase of the occurrence and abundance of warm-dwelling species and vice versa for cold-dwelling species using the widely used Community Temperature Index (CTI), as well the relationship between island fragmentation in terms of island area and isolation from the mainland on extinction and colonization rates of cold- and warm-adapted species. They found that indeed there was thermophilization happening during the last 10 years, which was more pronounced for the CTI based on abundances and less clearly for the occurrence based metric. Generally, the authors show that this is driven by an increased colonization rate of warm-dwelling and an increased extinction rate of cold-dwelling species. Interestingly, they unravel some of the mechanisms behind this dynamic by showing that warm-adapted species increased while cold-dwelling decreased more strongly on smaller islands, which is - according to the authors - due to lowered thermal buffering on smaller islands (which was supported by air temperature monitoring done during the study period on small and large islands). They argue, that the increased extinction rate of cold-adapted species could also be due to lowered habitat heterogeneity on smaller islands. With regards to island isolation, they show that also both thermophilization processes (increase of warm and decrease of cold-adapted species) was stronger on islands closer to the mainland, due to closer sources to species populations of either group on the mainland as compared to limited dispersal (i.e. range shift potential) in more isolated islands.

The conclusions drawn in this study are sound, and mostly well supported by the results. Only few aspects leave open questions and could quite likely be further supported by the authors themselves thanks to their apparent extensive understanding of the study system.

Strengths:

The study questions and hypotheses are very well aligned with the methods used, ranging from field surveys to extensive modeling frameworks, as well as with the conclusions drawn from the results. The study addresses a complex question on the interplay between habitat fragmentation and climate-driven thermophilization which can naturally be affected by a multitude of additional factors than the ones included here. Nevertheless, the authors use a well balanced method of simplifying this to the most important factors in question (CTI change, extinction, colonization, together with habitat fragmentation metrics of isolation and island area). The interpretation of the results presents interesting mechanisms without being too bold on their findings and by providing important links to the existing literature as well as to additional data and analyses presented in the appendix.

Weaknesses:

The metric of island isolation based on distance to the mainland seems a bit too oversimplified as in real-life the study system rather represents an island network where the islands of different sizes are in varying distances to each other, such that smaller islands can potentially draw from the species pools from near-by larger islands too - rather than just from the mainland. Although the authors do explain the reason for this metric, backed up by earlier research, a network approach could be worthwhile exploring in future research done in this system. The fact, that the authors did find a signal of island isolation does support their method, but the variation in responses to this metric could hint on a more complex pattern going on in real-life than was assumed for this study.

Comments on revisions:

I'm happy with the revisions made by the authors.

Author response:

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

Public Reviews:

Reviewer #3 (Public review):

Summary:

Juan Liu et al. investigated the interplay between habitat fragmentation and climate-driven thermophilization in birds in an island system in China. They used extensive bird monitoring data (9 surveys per year per island) across 36 islands of varying size and isolation from the mainland covering 10 years. The authors use extensive modeling frameworks to test a general increase of the occurrence and abundance of warm-dwelling species and vice versa for cold-dwelling species using the widely used Community Temperature Index (CTI), as well the relationship between island fragmentation in terms of island area and isolation from the mainland on extinction and colonization rates of cold- and warm-adapted species. They found that indeed there was thermophilization happening during the last 10 years, which was more pronounced for the CTI based on abundances and less clearly for the occurrence based metric. Generally, the authors show that this is driven by an increased colonization rate of warm-dwelling and an increased extinction rate of cold-dwelling species. Interestingly, they unravel some of the mechanisms behind this dynamic by showing that warm-adapted species increased while cold-dwelling decreased more strongly on smaller islands, which is - according to the authors - due to lowered thermal buffering on smaller islands (which was supported by air temperature monitoring done during the study period on small and large islands). They argue, that the increased extinction rate of cold-adapted species could also be due to lowered habitat heterogeneity on smaller islands. With regards to island isolation, they show that also both thermophilization processes (increase of warm and decrease of cold-adapted species) was stronger on islands closer to the mainland, due to closer sources to species populations of either group on the mainland as compared to limited dispersal (i.e. range shift potential) in more isolated islands.

The conclusions drawn in this study are sound, and mostly well supported by the results. Only few aspects leave open questions and could quite likely be further supported by the authors themselves thanks to their apparent extensive understanding of the study system.

Strengths:

The study questions and hypotheses are very well aligned with the methods used, ranging from field surveys to extensive modeling frameworks, as well as with the conclusions drawn from the results. The study addresses a complex question on the interplay between habitat fragmentation and climate-driven thermophilization which can naturally be affected by a multitude of additional factors than the ones included here. Nevertheless, the authors use a well balanced method of simplifying this to the most important factors in question (CTI change, extinction, colonization, together with habitat fragmentation metrics of isolation and island area). The interpretation of the results presents interesting mechanisms without being too bold on their findings and by providing important links to the existing literature as well as to additional data and analyses presented in the appendix.

Weaknesses:

The metric of island isolation based on distance to the mainland seems a bit too oversimplified as in real-life the study system rather represents an island network where the islands of different sizes are in varying distances to each other, such that smaller islands can potentially draw from the species pools from near-by larger islands too - rather than just from the mainland. Although the authors do explain the reason for this metric, backed up by earlier research, a network approach could be worthwhile exploring in future research done in this system. The fact, that the authors did find a signal of island isolation does support their method, but the variation in responses to this metric could hint on a more complex pattern going on in real-life than was assumed for this study.

Thank you again for this suggestion. Based on the previous revision, we discussed more about the importance of taking the island network into future research. The paragraph is now on Lines 294-304:

“As a caveat, we only consider the distance to the nearest mainland as a measure of fragmentation, consistent with previous work in this system (Si et al., 2014), but we acknowledge that other distance-based metrics of isolation that incorporate inter-island connections and island size could hint on a more complex pattern going on in real-life than was assumed for this study, thus reveal additional insights on fragmentation effects. For instance, smaller islands may also potentially utilize species pools from nearby larger islands, rather than being limited solely to those from the mainland. The spatial arrangement of islands, like the arrangement of habitat, can influence niche tracking of species (Fourcade et al., 2021). Future studies should use a network approach to take these metrics into account to thoroughly understand the influence of isolation and spatial arrangement of patches in mediating the effect of climate warming on species.”

Recommendations for the authors:

Reviewer #3 (Recommendations for the authors):

Great job on the revision! The new version reads well and in my opinion all comments were addressed appropriately. A few additional comments are as follows:

Thank you very much for your further review and recognition. We have carefully modified the manuscript according to all recommendations.

(1) L 62: replace shifts with process

Done. We also added the word “transforming” to match this revision. The new sentence is now on Lines 61-63:

“Habitat fragmentation, usually defined as the process of transforming continuous habitat into spatially isolated and small patches”

(2) L 363: Your metric for habitat fragmentation is isolation and habitat area and I think this could be introduced already in the introduction, where you somewhat define fragmentation (although it could be clearer still). You could also discuss this in the discussion more, that other measures of fragmentation may be interesting to look at.

Thank you for this suggestion. We now introduced metric of habitat fragmentation in the Introduction part after habitat fragmentation was defined. The sentence is now on Lines 64-66:

“Among the various ways in which habitat fragmentation is conceptualized and measured, patch area and isolation are two of the most used measures (Fahrig, 2003).”

(3) L 384: replace for with because of

Done.

(4) L 388: "Following this filtering, 60 ...."

Done.

(5) Figure 1: In panels b-d you use different terms (fragmented, small, isolated) but aiming to describe the same thing. I would highly recommend to either use fragmented islands or isolated islands for all panels. Although I see that in your study fragmentation includes both, habitat loss and isolation. So make this clear in the figure caption too...

Thank you very much for this suggestion. It’s important to maintain consistency in using “fragmentation”. We change “fragmented, small, isolated” into “Fragmented patches” in the caption of b-d. The modified caption is now on Line 771:

(6) L 783: replace background with habitat (or landscape) and exhibit with exemplify

Done. The new sentence is now on Lines 782-784:

“The three distinct patches signify a fragmented landscape and the community in the middle of the three patches was selected to exemplify colonization-extinction dynamics in fragmented habitats.”

(7) One bigger thing is the definition of fragmentation in your study for which you used habitat area (from habitat loss process) and isolation. This could still be clarified a bit more, especially in the figures. In Fig. 1 the smaller panels b-d could all be titled fragmented islands as this is what the different terms describe in your study (small, isolated) and thus the figure would become even clearer. Otherwise I'm happy with the changes made.

Thank you for raising this important question. Yes, “habitat fragmentation” in our research includes both habitat loss and fragmentation per se. We have clarified the caption of b-d in Figure 1 as suggested by Recommendation (5). We believe this can make it clearer to the readers.

eLife Assessment

Leveraging state-of-the-art experimental and analytical approaches, this important study characterizes the recruitment and activation of large populations of human motor units during slow isometric contractions in two lower limb muscles. Evidence for the main claims is solid and advances our understanding of how humans generate and control voluntary force.

Reviewer #1 (Public review):

Summary:

The Avrillon et al. explore the neural control of muscle by decomposing the firing activity of constituent motor units from the grid of surface electromyography (EMG) in the Tibialis (TA) Anterior and Vastus Lateralis (VL) during isometric contractions. The study involves extensive samples of motor units across the broadest range of voluntary contraction intensities up to 80% of MVC. The authors examine rate coding of the population of motor units, which describes the instantaneous firing rate of each motor unit as a function of muscle force. This relationship is characterized by a natural logarithm function that delineates two distinct phases: an initial phase with a steep acceleration in firing rate, particularly pronounced in low-threshold motor units, and a subsequent modest linear increase in firing rate, more significant in high-threshold motor units.

Strengths:

The study makes a significant contribution to the field of neuromuscular physiology by providing a detailed analysis of motor unit behavior during muscle contractions in a few ways.

(1) The significance lies in its comprehensive framework of motor unit activity during isometric contractions in the broad range of intensities, providing insights into the non-linear relationship between the firing rate and the muscle force. The extensive sample of motor units across the pool confirms the observation in animal studies in which the the spinal motoneuron exhibits a discharge consists of the distinct phases in response to synaptic currents, under the influence of persistent inward currents. As such, it is now reasonable to state the human motor units across the pool are also under control of gain modulation via some neuromodulatory effects in addition to synaptic inputs arising from ionotropic effects.<br /> (2) The firing scheme across in the entire motoneuron pool revealed in this study reconciles the discrepancy in firing organization under debate; i.e., whether it is 'onion skin' like or not (Heckman and Enoka 2012). The onion skin like model states that the low threshold motor units discharge higher than high threshold motor units and has been held for long time because the firing behaviors were examined in a partial range of contraction force range due to technical limitations. This reconciliation is crucial because it is fundamental to modelling the organization of motor unit recruitment and rate coding to achieve a desired force generation to advance our understanding of motor control.<br /> (3) The extensive data collection with a novel blind source separation algorithm on the expanded number of channel of surface EMG signal provides a robust dataset that enhances the reliability and validity of findings, setting a new standard for empirical studies in the field. \par<br /> Collectively, this study fills several knowledge gaps in the field and advances our understanding the mechanism underlying the isometric force generation.

Reviewer #2 (Public review):

Avrillon et al. provides a comprehensive assessment of firing rate parameters from a large percentage of the motor unit pool, in two muscles, during voluntary isometric contractions. The authors have used new quantitative methods to extract more unique motor units across contractions than prior studies. This was achieved by recording muscle fibre action potentials from four high density surface electromyogram (HDsEMG) arrays, quantifying residual EMG comparing the recorded and data-based simulation (Fig. 1A-B), and developing a metric to compare the spatial identification for each motor unit (Fig. 1D-E). From identified motor units, the authors have provided a detailed characterization of recruitment and firing rate responses during slow voluntary isometric contractions in the vastus lateralis and tibialis anterior muscles up to 75-80% of maximum intensity. In the lower limb it is interesting how lower threshold motor units have firing rate responses that saturate, whereas higher threshold units that presumably produce higher muscle contractile forces continue to increase their firing rate. Conceptually, the authors rightly focus on the literature of intrinsic motoneurone properties, but in vivo, other possibilities (that are difficult to measure in awake human participants) are that the form of descending supraspinal drive, spinal network dynamics and afferent inputs may have different effects across motor unit sizes, muscles and types of contractions. These results from single trail contractions and with a larger sample of motor units, supports the summary rate coding profiles of motor units in the extensor digitorum communis muscle (Monster and Chan, 1977).

Reviewer #3 (Public review):

Summary:

This is an interesting manuscript which uses state of the art experimental and simulation approaches to quantify motor unit discharge patterns in the human TA and VL. The non-linear profiles of motor unit discharge were calculated and found to have an initial acceleration phase followed by an attenuation phase. Lower threshold motor units had a larger gain of the initial acceleration whereas the higher threshold motor unit had a higher gain in the attenuation phase. These data represent a technical feat and are important for understanding how humans generate and control voluntary force.

Strengths:

The authors used rigorous, state-of-the art analyses to decompose and validate their motor unit data during a wide range of voluntary efforts.

Analyses are clearly presented, applied, and visualized.

The supplemental data provides important transparency.

Weaknesses:

Number of participants and muscles tested are relatively small - particularly given the constraints on yield. It is unclear if this will translate to other motor pools. The justification for TA and VL should be provided.

While in impressive effort was made to identify and track motor units across a range of contractions, it appears that a substantial portion of muscle force was not identified. Though high intensity contractions are challenging to decompose - the authors are commended in their technical ability in recording population motor unit discharge times with recruitment thresholds up to 75% a participant's maximal voluntary contractions. However previous groups have seen substantial recruitment motor units above 80% and even 90% maximum activation in the soleus. Given the innervation ratios of higher threshold motor units, if recruitment continued to 100%, the top quartile would likely represent a substantial portion of the traditional fast-fatigable motor units. It would be highly interesting to understand the recruitment and rate coding of the highest threshold motor units, at a minimum I would suggest using terms other than "entire range" or "full spectrum of recruitment thresholds"

The quantification of hysteresis using torque appears to make self-evident the observation that lower threshold motor units demonstrate less hysteresis with respect to torque - If there was motor unit discharge there will be force. I believe this limitation goes beyond the floor effects discussed in the manuscript. Traditionally individuals have used the discharge of a lower threshold unit as the measure on which to apply hysteresis analyses to infer ion channel function in human spinal motoneurons.

The main findings are not entirely novel. See Monster and Chan 1977 and Kanosue et al 1979

Comments on revisions:

I thank the authors for their thoughtful revision.

Just to confirm, the ranges for motor unit yield are for a single contraction. So, for example, in a participant there were 71 unique and concurrently active VL motor units able to be decomposed.

Author response:

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

Reviewer #1 (Public Review): 

Summary: 

This study explores the neural control of muscle by decomposing the firing activity of constituent motor units from the grid of surface electromyography (EMG) in the Tibialis (TA) Anterior and Vastus Lateralis (VL) during isometric contractions. The study involves extensive samples of motor units across the broadest range of voluntary contraction intensities up to 80% of MVC. The authors examine the rate coding of the population of motor units, which describes the instantaneous firing rate of each motor unit as a function of muscle force. This relationship is characterized by a natural logarithm function that delineates two distinct phases: an initial phase with a steep acceleration in firing rate, particularly pronounced in low-threshold motor units, and a subsequent modest linear increase in firing rate, more significant in high-threshold motor units. 

Strengths: 

The study makes a significant contribution to the field of neuromuscular physiology by providing a detailed analysis of motor unit behavior during muscle contractions in a few ways.

(1) The significance lies in its comprehensive framework of motor unit activity during isometric contractions in a broad range of intensities, providing insights into the non-linear relationship between the firing rate and the muscle force. The extensive sample of motor units across the pool confirms the observation in animal studies in which the spinal motoneuron exhibits a discharge consisting of distinct phases in response to synaptic currents, under the influence of persistent inward currents. As such, it is now reasonable to state the human motor units across the pool are also under the control of gain modulation via some neuromodulatory effects in addition to synaptic inputs arising from ionotropic effects.

(2) The firing scheme across the entire motoneuron pool revealed in this study reconciles the discrepancy in firing organization under debate; i.e., whether it is 'onion skin' like or not (Heckman and Enoka 2012). The onion skin like model states that the low threshold motor units discharge higher than high threshold motor units and have been held for a long time because the firing behaviors were examined in a partial range of contraction force range due to technical limitations. This reconciliation is crucial because it is fundamental to modelling the organization of motor unit recruitment and rate coding to achieve a desired force generation to advance our understanding of motor control.

(3) The extensive data collection with a novel blind source separation algorithm on the expanded number of channels of surface EMG signal provides a robust dataset that enhances the reliability and validity of findings, setting a new standard for empirical studies in the field. 

Collectively, this study fills several knowledge gaps in the field and advances our understanding of the mechanism underlying the isometric force generation.

We thank the reviewer for their positive appreciation of our work.

Weaknesses: 

Although the findings and claims based on them are mostly well aligned, some accounts of the methods and claims need to be clarified.

(1) The authors examine the input-output function of a motor unit by constructing models, using force as an input and discharge rate as an output. It sounds circular, or the other way around to use the muscle force as an input variable, because the muscle force is the result of motor unit discharges, not the cause that elicits the discharges. More specifically, as a result of non-linear interactions of synchronous and/or asynchronous discharges of a population of a given motoneuron pool that give rise to transient increase/maintenance in twitch force, the gross muscle force is attained. I acknowledge that it is extremely challenging experimentally to measure synaptic currents impinging upon the spinal motoneurons in human subjects and the author has an assumption that the force could be used as a proxy of synaptic currents. However, it is necessary to explicitly provide the caveats and rationale behind that. Force could be used as the input variable for modelling.

Force is indeed used in this study as a proxy of the common excitatory synaptic currents as their direct measurement is not possible in vivo in humans. It is worth noting that this approach has been extensively used in the past by many groups to study rate coding (e.g., Monsters & Chan, De Luca’s, Heckman’s, and Fuglevand’s groups). Heckman’s, Gorassini’s, Fuglevand’s groups and others have considered the non-linearities in the relation between motor unit firing rates and muscle force in humans as an indicator of the impact of neuromodulation on motor unit behaviour and changes of the intrinsic properties of motoneurons.

One could also use the cumulative spike train as a more direct estimate of common excitatory inputs, assuming that it is possible to identify a group of motor units not influenced by PICs, as done when selecting a reference low-threshold motor neuron in the delta F method (Gorassini et al., 1998), or the cumulative spike train of low-threshold motor neurons (Afsharipour et al., 2020). However, this approach was not possible in our study as we did not have the same units across contractions to estimate cumulative spike trains. It was therefore not possible to pool the data across contractions as we did to generate force/firing rate relations on the widest range of force.

We added a sentence in the discussion to highlight this limitation (P19, L470):

‘This result must be confirmed with a more direct proxy of the net synaptic drive, such as the firing rate of a reference low-threshold motor neuron used in the delta F method (Gorassini et al., 1998), or the cumulative spike train of low-threshold motor neurons (Afsharipour et al., 2020)’.

(2) The authors examine the firing organizations in TA and VL in this study without explicit purposes and rationale for choosing these muscles. The lack of accounts makes it hard for the readers to interpret the data presented, particularly in terms of comparing the results from the different muscles.

We wanted to compare the rate coding of pools of motor units from proximal (VL) and distal (TA) muscles within the lower limb. Indeed, distal and proximal muscles exhibit differences in rate coding and spatial recruitments (De Luca et al., 1982, J Physiol), potentially due to different levels of recurrent inhibition (Cullheim & Kellerth, 1978, J Physiol; Rossi & Mazzocchio, 1991, Exp Brain Res; Edgley et al., 2021, J Neurosci) or different levels of neuromodulation depending on their involvement (or not) in postural control (Hoonsgaard et al., 1988, J Physiol; Kim et al., 2020, J Neurophysiol).

We added a paragraph at the beginning of the result section to support our muscle choice (P6; L137): ‘16 participants performed either isometric dorsiflexion (n = 8) or knee extension tasks (n = 8) while we recorded the EMG activity of the tibialis anterior (TA - dorsiflexion) or the vastus lateralis (VL – knee extension) with four arrays of 64 surface electrodes (256 electrodes per muscle). The motoneuron pools of these two muscles of the lower limb receive a large part of common input (Laine et al., 2015; Negro et al., 2016a), constraining the recruitment of their motor units in a fixed order across tasks. They are therefore good candidates for an accurate description of rate coding. Moreover, we wanted to determine whether differences in rate coding observed between proximal and distal muscles in the upper limb (De Luca et al., 1982) were also present in the lower limb.’.

Another factor that guided our muscle choice was the low risk of crosstalk. For this, we verified with ultrasound that our arrays of 256 electrodes only covered the muscle of interest, staying away from the neighbouring muscles. This was possible as superficial muscles from the leg are bulkier than those from the upper limb. Given the small diameter of each electrode (2 mm), it is unlikely that the motor units from the neighbouring muscles were in the recorded muscle volume (Farina et al., 2003, IEEE Trans Biomed Eng)

(3) In the methods, the author described the manual curation process after applying the blind source separation algorithm. For the readers to understand the whole process of decomposition and to secure rigor and robustness of the analyses, it would be necessary to provide details on what exact curation is performed with what criteria. 

The manual curation of EMG decomposition with blind source separation is different from what is classically done with intramuscular EMG and template-matching algorithms. 

In short, our decomposition algorithm uses fast independent component analysis (fastICA) to retrieve motor unit spike trains from the EMG signals. For this, it iteratively optimises a set of weights, i.e., a separation vector, for each motor unit. The projection of the EMG signals on this separation vector generates a sparse motor unit pulse train, with most of its samples close to zero and only a few samples close to one (Figure 1B). The discharge times are estimated from this motor unit pulse train using a peak detection function and a k-mean classification with two classes to separate the high peaks (spikes) from the low peaks (noise and other motor units).

The manual curation consists of inspecting the automatic detection of the peaks of the motor unit pulse train and manually add missed peaks (missed discharge times) or remove wrongly detected peaks. Then, the separation vector is updated using the correct discharge times and the motor unit pulse train recalculated. This procedure generally improves the distance between the discharge times and the noise, which confirm the accuracy of the manual curation. If that’s not the case, the motor unit is discarded from the analyses.

We added a section on manual editing in the methods (P23, L615):

‘At the end of these automatic steps, all the motor unit pulse trains and identified discharge times were visually inspected, and manual editing was performed to correct the false identification of artifacts or the missed discharge times (Del Vecchio et al., 2020; Hug et al., 2021; Avrillon et al., 2023). The manual editing consisted of i) removing the spikes causing erroneous discharge rates (outliers), ii) adding the discharge times clearly separated from the noise, iii) recalculating the separation vector, iv) reapplying the separation vector on the entire EMG signals, and v) repeating this procedure until the selection of all the discharge times is achieved. The manual editing of potential missed discharge times and falsely identified discharge times was never immediately accepted. Instead, the procedure was consistently followed by the application of the updated motor unit separation vector on the entire EMG signals to generate a new motor unit pulse train. Then, the manual editing was only accepted when the silhouette value increased or stayed well above the threshold of 0.9 quantified with the silhouette value (Negro et al., 2016b). Only these motor units were retained for further analysis.’

(4) In Figure 3, the early recruited units tend to become untraceable in the higher range of contraction. This is more pronounced in the muscle VL. This limitation would ambiguate the whole firing curve along the force axis and therefore limitation and the applicability in the different muscles needs to be discussed. 

The loss of low threshold motor units in the higher range of contractions was caused either by the decrease in signal-to-noise ratio for small motor units when many larger ones are recruited, or by the cancellation of the surface action potentials of the small units in the interference electromyographic signal, or by the recruitment of a motor unit with a very similar spatio-temporal filter (an example is shown in the figure below). In the latter case, the motor unit pulse train contains peaks that represent the discharge times of both motor units (green and red dots in the simulated example below), making them undistinguishable by the operator during manual editing.

Author response image 1.

This was discussed in the results (P7; L190):

‘On average, we tracked 67.1 ± 10.0% (25th–75th percentile: 53.9 – 80.1%) of the motor units between consecutive contraction levels (10% increments, e.g., between 10% and 20% MVC) for TA and 57.2 ± 5.1% (25th–75th percentile: 46.6 – 68.3%) of the motor units for VL (Figure S2). There are two explanations for the inability to track all motor units across consecutive contraction levels. First, some motor units are recruited at higher targets only. Second, it is challenging to track small motor units beyond a few contraction levels due to a lower signal-to-noise ratio for the small motor units when larger motor units are recruited, or signal cancellation (Keenan et al., 2005; Farina et al., 2014a).’

However, we believe that it had a limited impact on the output of the paper, as the non-linear portion of the rate coding/force relation due to the persistent inward currents occurs during the first seconds after recruitment, before plateauing (for a review see Binder et al., 2020, Physiology).

(5) It is unclear how commonly the notion "the long-held belief that rate coding is similar across motor units from the same pool" is held among the community without a reference. Different firing organizations have been modelled and discussed in the seminal paper by Fuglevand et al. (1993) and as far as I understand, the debate has not converged to a specific consensus. As such, any reference would be required to support the claim the notion is widely recognized.

In the paper of Fuglevand et al., (1993, J Neurophysiol), all the motor units had the same rate coding pattern relative to the excitatory input, though they changed the slope of the relations and the saturation threshold of motor units between simulations. This is similar to the paper of De Luca & Contessa (2012, J Neurophysiol), where the equation used to simulate the rate coding was non-linear, but consistent across motor units.  

We added these citations to the text:

‘Overall, we found that motor units within a pool exhibit distinct rate coding with changes in force level (Figure 2 and 3), which contrasts with the long-held belief that rate coding is similar across motor units from the same pool (Fuglevand et al., 1993; De Luca and Contessa, 2012).’

(6) The authors claim that the firing behavior as a function of force is well characterized by a natural logarithmic function, which consists of initial steep acceleration followed by a modest increase in firing rate. Arguably the gain modulation in firing rate could be attributed to a neuromodulatory effect on the spinal motoneuron, which has been suggested by a number of animal studies. However, the complexity of the interactions between ionotropic and neuromodulatory inputs to motoneurons may require further elucidation to fully understand the mechanisms of neural control; it is possible to consider the differential acceleration among different threshold motor units as a differential combinatory effect of ionotropic and neuromodulatory inputs, but it is not trivially determined how differentially or systematically the inputs are organized. Likewise, the authors make an account for the difference in firing rate between TA and VL in terms of different amounts or balances of excitatory and inhibitory inputs to the motoneuron pool, but again this could be explained by other factors, such as a different extent of neuromodulatory effects. To determine the complexity of the interactions, further studies will be warranted.

We appreciate the reviewer’s view on this point, as we indeed only indirectly inferred the combination of neuromodulatory and ionotropic inputs to motoneurons in this study. A more direct manipulation of the sources of neuromodulatory and ionotropic inputs will be required in the future to directly highlight the mechanisms responsible for these variations in rate coding within pools. However, it is also worth noting that the acceleration in firing rate, the increase in firing rate during the ramp up, and the hysteresis between ramps up and downs have been used to infer the distribution of ionotropic and neuromodulatory inputs from the firing rate/force relations (Johnson et al., 2017; Beauchamp et al., 2023; Chardon et al., 2023). This approach has been validated with hundreds of thousands of simulations using a biophysical model of motor neurons (Chardon et al., 2023). There is also a series of studies in humans showing how the absence of neuromodulation modulated via inhibitory inputs (Revill & Fuglevand, 2017) or medication blocking serotonin receptors (Goodlich et al., 2023) impact the non-linearity of the firing rate/force relation. Therefore, we are confident that the differences observed within and between pools are linked to different distribution of excitatory/inhibitory inputs and neuromodulation.

We added a sentence in the discussion to highlight this point (P18; L435):

‘Taken together, these results show how ionotropic and neuromodulatory inputs to motoneurons uniquely combine to generate distinct rate coding across the pool, even if a more direct manipulation of the sources of neuromodulatory and ionotropic inputs will be required to directly estimate their interactions.’

(7) It is unclear with the account " ... the bandwidth of muscle force is < 10Hz during isometric contraction" in the manuscript alone, and therefore, it is difficult to understand the following claim. It appears very interesting and crucial for motor unit discharge and force generation and maintenance because it would pose a question of why the discharge rate of most motor units is higher than 10Hz, despite the bandwidth being so limited, but needs to be elaborated.

We described the slow fluctuations in smoothed firing rates associated with the variations in force observed during isometric contractions. The bandwidth of muscle force is lower than 10Hz due to the contractile properties of muscle tissues (Baldissera et al., 1998, J Physiol). Having an average firing rate higher than this bandwidth enables the pool of motor neurons to effectively transmit the common inputs (the main discriminant of muscle force) over this bandwidth without distortion (Farina et al., 2014, J Physiol). Increasing the firing rate beyond the muscle bandwidth also increases the power of the spike train at the direct current frequency (frequency equal to 0) since this power is related to the number of spikes per second. Thus, increasing the firing rate well beyond the muscle bandwidth still has a clear effect in force. To illustrate this point, note that electrical stimuli delivered at 100 Hz can lead to an increase in muscle force.

Reviewer #2 (Public Review):  

Summary: 

The motivation for this study is to provide a comprehensive assessment of motor unit firing rate responses of entire pools during isometric contractions. The authors have used new quantitative methods to extract more unique motor units across contractions than prior studies. This was achieved by recording muscle fibre action potentials from four high-density surface electromyogram (HDsEMG) arrays (Caillet et al., 2023), quantifying residual EMG comparing the recorded and data-based simulation (Figure 1A-B), and developing a metric to compare the spatial identification for each motor unit (Figure 1D-E). From identified motor units, the authors have provided a detailed characterization of recruitment and firing rate responses during slow voluntary isometric contractions in the vastus lateralis and tibialis anterior muscles up to 80% of maximum intensity. In the lower limb, it is interesting how lower threshold motor units have firing rate responses that saturate, whereas higher threshold units that presumably produce higher muscle contractile forces continue to increase their firing rate. In many ways, these results agree with the rate coding of motor units in the extensor digitorum communis muscle (Monster and Chan, 1977). The paper is detailed, and the analyses are well explained. However, there are several points that I think should be addressed to strengthen the paper.

We thank the reviewer for their positive appreciation of our work.

General comments: 

(1) The authors claim they have measured the complete rate coding profiles of motor units in the vastus lateralis and tibialis anterior muscles. However, this study quantified rate coding during slow and prolonged voluntary isometric contractions whereas the function of rate coding during movements (Grimby and Hannerz, 1977) or more complex isometric contractions (Cutsem and Duchateau, 2005; Marshall et al., 2022) remains unexplored. For example, supraspinal inputs may not scale the same way across low and higher threshold motor units, or between muscles (Devanne et al., 1997), making the response of firing rates to increasing isometric contraction force less clear. 

We agree with the reviewer that rate coding strategies may vary with the velocity and the type of contractions (Duchateau & Enoka, 2008, J Physiol). It is thus likely that the firing rate would increase during the first milliseconds of fast contractions, with the occurrence of doublets (Cutsem and Duchateau, 2005, J Physiol; Del Vecchio et al., 2019, J Physiol), or that motor unit firing rate may be lower during lengthening than shortening contractions (Duchateau & Enoka, J Physiol). 

However, the decomposition of EMG signals in non-stationary conditions remains challenging, and is still limited to slow varying patterns of force (Chen et al., 2000, Oliveira & Negro, 2021, Mendez Guerra et al., 2024, Yeung et al., 2024). Future methodological developments will be required to expand our findings to other patterns of force.

Conceptually, the authors focus on the literature on intrinsic motoneurone properties, but in vivo, other possibilities are that descending supraspinal drive, spinal network dynamics, and afferent inputs have different effects across motor unit sizes, muscles, and types of contractions. Also, the influence from local muscles that act as synergists (e.g., vastii muscles for the vastus lateralis, and peroneal muscles that evert the foot for the tibialis anterior) or antagonists (coactivation during higher contraction intensities would stiffen the joint) may provide differential forms of proprioceptive feedback across motor pools. 

The reviewer is right that differences in spinal network dynamics and afferent inputs may explain the differences in rate coding observed between the two muscles. Indeed, computational models have shown how the pattern of inhibitory inputs may affect the increase in firing rate during linear increase in force (Powers & Heckman, 2017, J Neurophysiol; Chardon et al., 2023, Elife). Specifically, the difference observed between proportional inhibitory inputs vs. a push pull pattern mirror the differences observed here between the TA (push-pull like pattern) and the VL (proportional pattern). This difference may reflect the impact of various pathways of inhibition, such as reciprocal inhibition or recurrent inhibition from homonymous motor units or motor units from synergistic muscles. 

These points have been further discussed in the manuscript (P19; L475):

‘The increase in firing rate was also significantly greater for TA motor units than for those in VL. This difference may reflect a varying balance between excitatory/inhibitory synaptic inputs and neuromodulation due to multiple spinal circuits (Heckman and Binder, 1993; Heckman et al., 2008; Johnson et al., 2017; Powers and Heckman, 2017; Chardon et al., 2023; Škarabot et al., 2023). Specifically, the strength of recurrent and reciprocal inhibitory inputs to motoneurons innervating VL and TA, and their proportional or inverse covariation with excitatory inputs, respectively, may explain the differences in rate limiting and maximal firing rates (Heckman and Binder, 1993; Heckman et al., 2008; Johnson et al., 2017; Powers and Heckman, 2017; Chardon et al., 2023; Škarabot et al., 2023). Thus, the motor units from the VL may receive more recurrent inhibition than those of distal muscles, though direct evidence of these differences remains to be found in humans (Windhorst, 1996). Interestingly, similar differences in rate coding were previously observed between proximal and distal muscles of the upper limb (De Luca et al., 1982). However, other muscles that serve different functions within the human body, such as muscles from the face, have different rate coding characteristics with much higher firing rates (Kirk et al., 2021). Future work should investigate those muscles and other to reveal the myriads of rate coding strategies in human muscles.’

(2) The evidence that the entire motor unit pool was recorded per muscle is not clear. There appears to be substantial residual EMG (Figure 1B), signal cancellation of smaller motor units (lines 172-176), some participants had fewer than 20 identified motor units, and contractions never went above 80% of MVC. Also, to my understanding, there remains no gold-standard in awake humans to estimate the total motor unit number in order to determine if the entire pool was decomposed. 

The reviewer is right that we did not decode the full pool of motor units. As indicated in the initial version of the manuscript (e.g. title, introduction), we considered that we identified an extensive sample of motor units representative of the dynamic of the pool. This claim was supported by the identification of motor units with recruitment thresholds ranging from 0 to 75% of the maximal force. 

This statement was in the introduction (P4; L109): ‘We were able to identify up to ~200 unique active motor units per muscle and per participant in two human muscles in vivo, yielding extensive samples of motor units that are representative of the entire motoneuron pools (Caillet et al., 2023a).’

Furthermore, using four HDsEMG arrays also raises questions about how some channels were placed over non-target muscles, and if motor units were decomposed from surrounding synergists.

A factor that guided our muscle choice was the low risk of crosstalk. For this, we verified with ultrasound that our arrays of 256 electrodes only covered the muscle of interest, staying away from the neighbouring muscles. This was possible as superficial muscles from the leg are bulkier than those from the upper limb. Given the small diameter of each electrode (2 mm), it is unlikely that the motor units from the neighbouring muscles were in the recorded muscle volume.

(3) The authors claim (Abstract L51; Discussion L376) that a commonly held view in the field is that rate coding is similar across motor units from the same pool. Perhaps this is in reference to some studies that have carefully assessed lower threshold motor units during lower force ramp contractions (e.g., Fuglevand et al., 2015; Revill and Fuglevand, 2017). However, a more complete integration of the literature exploring motor unit firing rate responses during rapid isometric contractions, comparing different muscles and contraction intensities would be helpful. From Figure 3, the range of rate coding in the tibialis anterior (~7-40 Hz) is greater than the vastus lateralis (~5-22 Hz) muscle across contraction levels. In agreement with other studies, the range of rate coding within some muscles is different than others (Kirk et al., 2021) and during maximal intensity (Bellemare et al., 1983) or rapid contractions (Desmedt and Godaux, 1978). Likewise, within a motor pool, there is a diversity of firing rate responses across motor units of different sizes as a function of isometric force (Monster and Chan, 1977; Desmedt and Godaux, 1977; Kukula and Clamann, 1981; Del Vecchio et al., 2019; Marshall et al., 2022). A strength of this paper is how firing rate responses are quantified across a wide range of motor unit recruitment thresholds and between two muscles. I suggest improving clarity for the general reader, especially in the motivation for testing two lower limb muscles, and elaborating on some of the functional implications.

We thank the reviewer for his input on this question. We have added references to these works and lines of research in the discussion:

(P18; L449): ‘In addition, rate coding patterns should also vary with the pattern of contractions, with fast contractions lowering the range of recruitment thresholds within motoneuron pools (Desmedt and Godaux, 1977b, 1979; van Bolhuis et al., 1997). The variability in rate coding observed here between motor units from the same pool could lead to small deviations from the size principle sometimes observed between pairs of units during isometric contractions with various patterns of force (Desmedt and Godaux, 1979; Marshall et al., 2022) or during the derecruitment phase (Bracklein et al., 2022).’ (P19; L487): ‘However, other muscles that serve different functions within the human body, such as muscles from the face, have different rate coding characteristics with much higher firing rates (Kirk et al., 2021). Future work should investigate those muscles and other to reveal the myriads of rate coding strategies in human muscles.’

In addition to the responses above, we have added a section at the beginning of the results to motivate the choice of the muscles (P6; L137):

‘16 participants performed either isometric dorsiflexion (n = 8) or knee extension tasks (n = 8) while we recorded the EMG activity of the tibialis anterior (TA - dorsiflexion) or the vastus lateralis (VL – knee extension) with four arrays of 64 surface electrodes (256 electrodes per muscle). The motoneuron pools of these two muscles of the lower limb receive a large part of common input (Laine et al., 2015; Negro et al., 2016a), constraining the recruitment of their motor units in a fixed order across tasks. They are therefore good candidates for an accurate description of rate coding. Moreover, we wanted to determine whether differences in rate coding observed between proximal and distal muscles in the upper limb (De Luca et al., 1982) were also present in the lower limb.’.

Reviewer #3 (Public Review): 

Summary: 

This is an interesting manuscript that uses state-of-the-art experimental and simulation approaches to quantify motor unit discharge patterns in the human TA and VL. The non-linear profiles of motor unit discharge were calculated and found to have an initial acceleration phase followed by an attenuation phase. Lower threshold motor units had a larger gain of the initial acceleration whereas the higher threshold motor unit had a higher gain in the attenuation phase. These data represent a technical feat and are important for understanding how humans generate and control voluntary force. 

Strengths: 

The authors used rigorous, state-of-the-art analyses to decompose and validate their motor unit data during a wide range of voluntary efforts.

The analyses are clearly presented, applied, and visualized. 

The supplemental data provides important transparency. 

We thank the reviewer for their positive appreciation of our work.

Weaknesses: 

The number of participants and muscles tested are quite small - particularly given the constraints on yield. It is unclear if this will translate to other motor pools. The justification for TA and VL should be provided.

One strength of our study is to provide relations between key-parameters of rate coding (acceleration in firing rate, increase in firing rate, hysteresis) and the recruitment thresholds of motor units within two different pools, and for each individual participant. These relations were consistent across all the participants (Figures 2 to 4), making us confident that increasing the sample size would not change the conclusions of the study.

It is likely that the differences observed here between the VL and TA will also appear between other muscles of the leg, due to differences in the arrays of excitatory and inhibitory inputs they receive, the pattern of inhibitory inputs during increases in force (recurrent/reciprocal inhibition), and different levels of neuromodulation (Johnson et al., 2017, J Neurophysiol; Beauchamp et al., 2023; J Neural Eng). We have added a paragraph in the results to motivate our choice of muscles (P6; L137):

‘16 participants performed either isometric dorsiflexion (n = 8) or knee extension tasks (n = 8) while we recorded the EMG activity of the tibialis anterior (TA - dorsiflexion) or the vastus lateralis (VL – knee extension) with four arrays of 64 surface electrodes (256 electrodes per muscle). The motoneuron pools of these two muscles of the lower limb receive a large part of common input (Laine et al., 2015; Negro et al., 2016a), constraining the recruitment of their motor units in a fixed order across tasks. They are therefore good candidates for an accurate description of rate coding. Moreover, we wanted to determine whether differences in rate coding observed between proximal and distal muscles in the upper limb (De Luca et al., 1982) were also present in the lower limb.’.

While an impressive effort was made to identify and track motor units across a range of contractions, it appears that a substantial portion of muscle force was not identified. Though high-intensity contractions are challenging to decompose - the authors are commended for their technical ability to record population motor unit discharge times with recruitment thresholds up to 75% of a participant's maximal voluntary contractions. However previous groups have seen substantial recruitment of motor units above 80% and even 90% maximum activation in the soleus. Given the innervation ratios of higher threshold motor units, if recruitment continued to 100%, the top quartile would likely represent a substantial portion of the traditional fast-fatigable motor units. It would be highly interesting to understand the recruitment and rate coding of the highest threshold motor units, at a minimum I would suggest using terms other than "entire range" or "full spectrum of recruitment thresholds"

Motor units were indeed identified between 0 and 80% of the maximal force in this study. This is due to the requirements of the decomposition algorithm that needs sustained and stable contraction to converge toward a set of separation vectors that generate sparse spike trains. Thus, it was not possible for our participants to sustain contractions above 80%MVC without generating fatigue.

However, it is important to note that only a few motor units are recruited above 80% of the maximal force in the TA (Van Cutsem et al., 1998, J Physiol), as well as in other muscles of the lower limb (Oya et al., 2009, J Physiol; Aeles et al., 2020, J Neurophysiol). Thus, we may have only missed a few motor units recruited above 80% of the maximal force. Nevertheless, we removed the terms ‘full spectrum of recruitment thresholds’ and ‘entire range’ from the manuscript to now read ‘most of the spectrum of recruitment thresholds observed in humans.’.

The quantification of hysteresis using torque appears to make self-evident the observation that lower threshold motor units demonstrate less hysteresis with respect to torque. If there is motor unit discharge there will be force. I believe this limitation goes beyond the floor effects discussed in the manuscript. Traditionally, individuals have used the discharge of a lower threshold unit as the measure on which to apply hysteresis analyses to infer ion channel function in human spinal motoneurons.

We agree with the reviewer that the hysteresis is classically estimated using the firing rate of a ‘reporter unit’ with the delta F method (introduced in humans by Gorassini et al..), or most recently with the advances in motor unit identification using the cumulative spike train of the identified motor unit. The researchers use this data as a proxy of the synaptic drive, and compare their values at recruitment and derecruitment thresholds of the ‘test unit’. 

As mentioned above in response to reviewer 1, this approach was not possible in our study as we did not have the same units across contractions to estimate cumulative spike trains. It was therefore not possible to pool the data across contractions as we did here to generate force/firing rate relations on the widest range of force. This limitation is now highlighted in the discussion section (P19; L470): ‘This result must be confirmed with a more direct proxy of the net synaptic drive, such as the firing rate of a reference low-threshold motor neuron used in the delta F method (Gorassini et al., 1998), or the cumulative spike train of low-threshold motor neurons (Afsharipour et al., 2020).’.

The main findings are not entirely novel. See Monster and Chan 1977 and Kanosue et al 1979. 

We agree with the reviewer that the results of the paper are remarkably aligned with previous experimental findings in humans, in animals, or with in vitro and in silico models. However, we believe that our study shows in humans the incredible variety of rate coding patterns within a pool of motor units that span most of the spectrum of recruitment thresholds observed in humans. It also highlights the variability of rate coding patterns between motor neurons that have a similar recruitment threshold. Finally, we observe differences between pools of motor neurons innervating two different muscles in the lower limb, mirroring what has been done in the past in the upper limb muscle. 

Recommendations for the authors:  

Reviewer #1 (Recommendations For The Authors): 

The wording 'decode' across the manuscript may sound somewhat unsuitable for the context, because 'decode' would involve interpreting the signals and activities to understand how they relate to specific variables or proxies of behavior. Here in this study it does not necessarily involve the interpretation, but sounds to be used for decomposing the signal into the constituent motor units. As such, it might be appropriate to use other words such as decompose, read out, or extract.

‘Decode’ was removed from the manuscript to now read motor unit ‘identification’

Reviewer #2 (Recommendations For The Authors): 

Figures 1 and 2 are informative and interesting. Figures 3 and 4 are harder to interpret. For example, in Figure 4, data plotted along the diagonal is overplotted and not as informative.

For the sake of clarity, we separated the lines of the fits and the scatter plots in in the right panels in Figure 3. In Figure 4, we remove the scatter plots and only reported the lines of the fits for each participant. 

Do you think the different durations of the isometric plateau across contraction intensities influenced motor unit derecruitment? Longer duration in lower threshold motor units would have resulted in a larger effect of PICs?

We did not find an effect of the duration of the plateau on the derecruitment threshold. Notably, a computational study found that the duration of the plateau may impact the delta F, due to the combination of PICs, spike threshold accommodation and spike frequency adaptation (Revill & Fuglevand, 2011, J Neurophysiol). However, we did not use the delta F value here to estimate the effect of PICs on the hysteresis. 

L703. For the measure of firing rate hysteresis the difference between recruitment and derecruitment was calculated, but why not use the delta-F method? This is more commonly used to assess hysteresis as a rough estimate of intrinsic dynamics.

As further discussed above, this approach was not possible in our study as we did not have the same units across contractions to estimate cumulative spike trains. It was therefore not possible to pool the data across contractions as we did here to generate force/firing rate relations on the widest range of force.

This was mentioned in the discussion (P19; L470):

‘This result must be confirmed with a more direct proxy of the net synaptic drive, such as the firing rate of a reference low-threshold motor neuron used in the delta F method (Gorassini et al., 1998), or the cumulative spike train of low-threshold motor neurons (Afsharipour et al., 2020).’

L144. The standard deviation seems high. Some participants had fewer than 20 motor units and your number of participants per muscle was eight, could you state the complete range?

A table was added in the results section to indicate the yields of the decomposition per contraction.

If other studies are able to randomly sample motor units with intramuscular electrodes does this also represent an estimate of rate coding from the 'entire' pool? One criticism of HDsEMG arrays is that they are biased towards decomposing superficial larger motor units and in the male sex. 

The decomposition of EMG signals recorded with arrays of surface electrodes is indeed biased toward the identification of motor units with the larger action potentials in the signal (large and superficial; Farina & Holobar, 2016, Proceedings of IEEE). We took advantage of the latter limitation by performing successive contractions at different levels of force with the objective to identify the last recruited motor units (larger units according to the size principle), while tracking the smaller ones. In that way, we were able to sequentially identify motor units recruited from 0% to 75% of the maximal force. A similar approach could be applied to selective intramuscular electrodes. However, because identifying motor units up to maximal force requires a highly selective pair of fine wires or needle electrodes, the procedure described above should be repeated hundreds of times to reach the same samples as those obtained in our study.

L151-161. The ratio between simulated and decomposed surface EMG reached 55% for the TA and 70% for the VL. How does this provide support that the "entire" MU pool was sampled?

As said above, we do not identify all the motor units during each contraction, but rather the larger ones with the larger action potentials within the EMG signals. However, we used here a sequential approach to identify new motor units during each trial while tracking smaller units. In that way, we were able to sequentially identify on average 130 motor units per muscle.

To avoid any confusion, we removed the references to ‘entire’ pools in the manuscript.  

L266. How is it possible that in some participants no motor units were recruited below 5% of MVC? Do the authors suspect they produced force from synergist muscles or that the decomposition failed to identify these presumably smaller and deeper motor units?

This mostly results from the limitations of the decomposition algorithm. In these participants, it is likely that the decomposition was biased toward motor units only active during the plateau of force or recruited at the end of the ramp.

Figure 2B. Do the higher threshold motor units with linear responses receive more inhibitory input (coactivation) or are devoid of large PIC effects?

Were antagonist muscles recorded? During higher contraction intensities, greater antagonist coactivation in some trials or participants may have linearized the firing rate profiles (e.g., Revill and Fuglevand, 2017).

L427. This is a neat finding that higher threshold motor units are less likely to have the functional  hallmark of a strong PIC effect and may therefore be more representative of extrinsic inputs. Could this be an advantage to increase the precision of stronger contractions or reduce the fatigability of muscle fibres during repeated strong contractions?

Synaptic contacts with Renshaw cells (Fyffe, 1991, J Neurophysiol) and Ia inhibitory interneurons (Heckman & Binder, 1991, J Neurophysiol) are widespread within pools of motor units, which induces homogeneously distributed inhibitory inputs. However, the amplitude of these inhibitory inputs can increase with muscle force. We found that the EMG amplitude of the soleus and the gastrocnemius medialis recorded with bipolar EMG during the dorsiflexion increased with the force. Therefore, the higher inhibitory at higher force may also contribute to the linearisation of the force/firing rate relations observed with high threshold motor neurons, as suggested by Revill and Fuglevand (2017, J Physiol). 

We discussed this point in the new version of the manuscript (P17; L415):

‘The level of recurrent and reciprocal inhibition has also probably increased with the increase in force during the ramp up, progressively blunting the effect of persistent inward currents for late-recruited motor units (Kuo et al., 2003; Hyngstrom et al., 2007; Revill and Fuglevand, 2017). This may also explain the larger percentage of high-threshold motor units with a linear fit for the firing rate/force relation (Figure 2), as the integration of larger inhibitory inputs should linearise the firing rate/force relation (Revill and Fuglevand, 2017).’. 

In Figure 2B, it makes sense that linear firing rate responses occur later in the ramp contraction when myotendinous slack is lower. Do the authors think contractile dynamics are matched to the firing rate profiles?

To our knowledge, there is no direct data on the link between the linearity of the force/firing rate relation and the stiffness of the tendon. A recent work from Mazzo et al. (2021, J Physiol) has shown that repeated stretches of calf muscles, which induce a decrease in their stiffness, induced an increase in motor unit firing rate at low levels of forces. This indicates that the contractile properties of the muscle may potentially also impact the profile of rate coding when considered as function of force. 

We added this point in the discussion (P20; L512):

‘On a different note, the steep increase in firing rate over the first percentages of the ramp-up may also enable the motor units to produce the required level of force despite having a more compliant muscletendon unit (Mazzo et al., 2021).’

L371. It is likely that Marshall et al., 2022, recorded over 100 unique motor units from the same animal.

The reviewer is right that Marshall may have identified hundreds of motor units across sessions in one non-human primate. However, there is no ways to verify this statement as they used fine wire electrodes inserted in different locations in each session, which made it impossible to verify the uniqueness of each identified unit. Conversely, we verified in our study that all the motor units were unique using the distribution of their surface action potentials across the 236 surface electrodes.

L378. What do the authors mean by "rate coding is similar"? I find this statement confusing. Is this regarding the absolute firing rate range, response to force increases, hysteresis, or how they scale with contraction intensity?

This statement was removed from the discussion to avoid any confusion.

Reviewer #3 (Recommendations For The Authors): 

The authors may want to consider other mechanisms of the linearization of discharge rates of medium and high threshold motor units. Monica's work may suggest that, over time, there is a subthreshold activation of the PIC, which serves to linearize the eventual suprathreshold activation underlying repetitive discharge. Additionally, Andy has shown that inhibitory drive from cutaneous inputs can linearize the initial acceleration of low threshold motor units - cutaneous inputs, or even Ib inputs, may be greater later in the contraction and serve to linearize discharge rates. 

We thank the reviewer for their input on the discussion, where we now discuss this point:

‘The level of recurrent and reciprocal inhibition has also probably increased with the increase in force during the ramp up, progressively blunting the effect of persistent inward currents for late-recruited motor units (Kuo et al., 2003; Hyngstrom et al., 2007; Revill and Fuglevand, 2017). This may also explain the larger percentage of high-threshold motor units with a linear fit for the firing rate/force relation (Figure 2), as the integration of larger inhibitory inputs should linearise the firing rate/force relation (Revill and Fuglevand, 2017).’. 

Lines 433 - intrinsic properties, in particular the afterhyperpolarization, will likely influence maximal discharge rate and provide a ceiling to the change in firing rate.

This point is now discussed in the draft (P17; L428):

‘This difference may be explained by smaller excitatory synaptic inputs onto low- than high-threshold motoneurons (Powers and Binder, 2001; Heckman and Enoka, 2012), lower synaptic driving potential of the dendritic membrane (Powers and Binder, 2000; Cushing et al., 2005; Fuglevand et al., 2015), and longer and larger afterhyperpolarisation phase in low- than high-threshold motoneurons (Bakels and Kernell, 1993; Gardiner, 1993; Deardorff et al., 2013; Caillet et al., 2022).’

The actual yield per contraction is not entirely clear. Figure S2 is quite nice in this regard, but a table with this and other information on it may be helpful. This would help with the beginning of the abstract and discussion when it is stated that, on average over 100 motor units were identified per person. 

We added a table in the results to give the number of motor units identified per contraction.

Are the thin film units represented in S2 and S3?

Only motor units identified from signals recorded with arrays of surface electrodes are presented in figures S2 and S3.

eLife Assessment

This study provides valuable advances in our understanding of how inputs from multiple sources can impact the physiology of motor neurons during the process of multisensory integration. Specifically, the authors show how streams of auditory and principally visual information modulate the physiology of Mauthner neurons in goldfish, thus allowing the different senses to influence escape behavior. Supporting evidence is generally convincing, although material reporting the direct control of behavior is less representative of the data.

Reviewer #2 (Public review):

Summary:

In this manuscript, Otero-Coronel and colleagues use a combination of acoustic stimuli and electrical stimulation of the tectum to study MSI in the M-cells of adult goldfish. They first perform a necessary piece of groundwork in calibrating tectal stimulation for maximal M-cell MSI, and then characterize this MSI with slightly varying tectal and acoustic inputs. Next, they quantify the magnitude and timing of FFI that each type of input has on the M-cell, finding that both the tectum and the auditory system drive FFI, but that FFI decays more slowly for auditory signals. These are novel results that would be of interest to a broader sensory neuroscience community. By then providing pairs of stimuli separated by 50ms, they assess the ability of the first stimulus to suppress responses to the second, finding that acoustic stimuli strongly suppress subsequent acoustic responses in the M-cell, that they weakly suppress subsequent tectal stimulation, and that tectal stimulation does not appreciably inhibit subsequent stimuli of either type. Finally, they show that M-cell physiology mirrors previously reported behavioural data in which stronger stimuli underwent less integration.

The manuscript is generally well written and clear. The discussion of results is appropriately broad and open-ended. It's a good document. Our major concerns regarding the study's validity are captured in the individual comments below. In terms of impact, the most compelling new observation is the quantification of the FFI from the two sources and the logical extension of these FFI dynamics to M-cell physiology during MSI. It is also nice, but unsurprising, to see that the relationship between stimulus strength that MSI is similar for M-cell physiology to what has previously been shown for behavior. While we find the results interesting, we think that they will be of greatest interest to those specifically interested in M-cell physiology and function.

Strengths:

The methods applied are challenging and appropriate and appear to be well executed. Open questions about the physiological underpinnings of M-cell function are addressed using sound experimental design and methodology, and convincing results are provided that advance our understanding of how two streams of sensory information can interact to control behavior.

Weaknesses:

Our concerns about the manuscript are captured in the following specific comments, which we hope will provide a useful perspective for readers and actionable suggestions for the authors.

Comments relevant to the revised manuscript:

Our general assessment (above) stands unchanged from the original version. All of our comments and concerns about the original manuscript have been addressed except for two, one very minor and one quite important:

Original Comment 1 (Minor):<br /> "Line 124. Direct stimulation of the tectum to drive M-cell-projecting tectal neurons not only bypasses the retina, it also bypasses intra-tectal processing and inputs to the tectum from other sources (notably the thalamus). This is not an issue with the interpretation of the results, but this description gives the (false) impression that bypassing the retina is sufficient to prevent adaptation. Adding a sentence or two to accurately reflect the complexity of the upstream circuitry (beyond the retina) would be welcome."

The authors have replied:<br /> "The reviewer is right in that direct tectal stimulation bypasses all neural processing upstream, not only that produced in the retina and that the tectum does not exclusively process visual information. The revised version now acknowledges (lines 245-252, revised manuscript) the complexity of the system."

We think that this is sufficient to address our concern. Some citations may be in order to underpin the new text.

Original Comment 5 (Major):<br /> Figure 4C and lines 398-410.<br /> "These are beautiful examples of M-cell firing, but the text suggests that they occurred rarely and nowhere close to significantly above events observed from single modalities. We do not see this a valid result to report because there is insufficient evidence that the phenomenon shown is consistent or representative of your data."

The authors have replied:<br /> "Our experimental conditions required anesthesia and paralysis, conditions designed to reduce neuronal firing and suppress motor output. We think it is valuable to report that we still see that simultaneous presentation subthreshold unisensory stimuli can add up to become suprathreshold, paralleling behavioral observations. We do not claim and acknowledge that those examples are representative of our recording conditions, but are likely to be more representative of the multisensory integration process taking place in freely moving fish. The revised manuscript adds context to these example traces to justify their inclusion (lines 420-426)."

We do not feel that this important concern has been addressed. The stats are definitively negative. There is no statistical evidence from these data that multisensory integration is occurring in this assay. The aesthesia, paralysis, and low n may provide explanations for this negative result, but it is still a negative result (p=0.5269). To show two examples of multisensory integration for subthreshold stimuli fits the narrative, but this result is not supported. Examples where individual stimuli caused APs (and combined stimuli did not) also occurred, presumably, and at a rate that is statistically indistinguishable to the examples shown in Figure 5. As such, if results from this assay are going to be in the manuscript, acoustic-only and tectum-only examples should be shown as well, although they would not fit the narrative. To be meaningful, this experiment would have to show that multisensory integration is happening in this circuit. Frustrating though it must be, the experiment has given a negative result to that question.

Author response:

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

Reviewer #1 (Public Review):

Summary:

Otero-Coronel et al. address an important question for neuroscience - how does a premotor neuron capable of directly controlling behavior integrate multiple sources of sensory inputs to inform action selection? For this, they focused on the teleost Mauthner cell, long known to be at the core of a fast escape circuit. What is particularly interesting in this work is the naturalistic approach they took. Classically, the M-cell was characterized, both behaviorally and physiologically, using an unimodal sensory space. Here the authors make the effort (substantial!) to study the physiology of the M-cell taking into account both the visual and auditory inputs. They performed well-informed electrophysiological approaches to decipher how the M-cell integrates the information of two sensory modalities depending on the strength and temporal relation between them.

Strengths:

The empirical results are convincing and well-supported. The manuscript is well-written and organized. The experimental approaches and the selection of stimulus parameters are clear and informed by the bibliography. The major finding is that multisensory integration increases the certainty of environmental information in an inherently noisy environment.

Weaknesses:

Even though the manuscript and figures are well organized, I found myself struggling to understand key points of the figures.

For example, in Figure 1 it is not clear what are actually the Tonic and Phasic components. The figure will benefit from more details on this matter. Then, in Figure 4 the label for the traces in panel A is needed since I was not able to pick up that they were coming from different sensory pathways.

We added an inset to Figure 1 showing how the tonic and phasic components are measured. We now use solid colors instead of transparencies, and the color scheme was modified for consistency. We added labels to the traces used as examples in Figure 4 panel A.

In line 338 it should be optic tectum and not "optical tectum".

We replaced two instances of the term “optical tectum” with “optic tectum”.

Reviewer #2 (Public Review):

Summary:

In this manuscript, Otero-Coronel and colleagues use a combination of acoustic stimuli and electrical stimulation of the tectum to study MSI in the M-cells of adult goldfish. They first perform a necessary piece of groundwork in calibrating tectal stimulation for maximal M-cell MSI, and then characterize this MSI with slightly varying tectal and acoustic inputs. Next, they quantify the magnitude and timing of FFI that each type of input has on the M-cell, finding that both the tectum and the auditory system drive FFI, but that FFI decays more slowly for auditory signals. These are novel results that would be of interest to a broader sensory neuroscience community. By then providing pairs of stimuli separated by 50ms, they assess the ability of the first stimulus to suppress responses to the second, finding that acoustic stimuli strongly suppress subsequent acoustic responses in the M-cell, that they weakly suppress subsequent tectal stimulation, and that tectal stimulation does not appreciably inhibit subsequent stimuli of either type. Finally, they show that M-cell physiology mirrors previously reported behavioural data in which stronger stimuli underwent less integration.

The manuscript is generally well-written and clear. The discussion of results is appropriately broad and open-ended. It's a good document. Our major concerns regarding the study's validity are captured in the individual comments below. In terms of impact, the most compelling new observation is the quantification of the FFI from the two sources and the logical extension of these FFI dynamics to M-cell physiology during MSI. It is also nice, but unsurprising, to see that the relationship between stimulus strength and MSI is similar for M-cell physiology to what has previously been shown for behavior. While we find the results interesting, we think that they will be of greatest interest to those specifically interested in M-cell physiology and function.

Strengths:

The methods applied are challenging and appropriate and appear to be well executed. Open questions about the physiological underpinnings of M-cell function are addressed using sound experimental design and methodology, and convincing results are provided that advance our understanding of how two streams of sensory information can interact to control behavior.

Weaknesses:

Our concerns about the manuscript are captured in the following specific comments, which we hope will provide a useful perspective for readers and actionable suggestions for the authors.

Comment 1 (Minor):

Line 124. Direct stimulation of the tectum to drive M-cell-projecting tectal neurons not only bypasses the retina, it also bypasses intra-tectal processing and inputs to the tectum from other sources (notably the thalamus). This is not an issue with the interpretation of the results, but this description gives the (false) impression that bypassing the retina is sufficient to prevent adaptation. Adding a sentence or two to accurately reflect the complexity of the upstream circuitry (beyond the retina) would be welcome.

The reviewer is right in that direct tectal stimulation bypasses all neural processing upstream, not only that produced in the retina and that the tectum does not exclusively process visual information. The revised version now acknowledges (lines 245-252, revised manuscript) the complexity of the system.

Comment 2 (Major): The premise is that stimulation of the tectum is a proxy for a visual stimulus, but the tectum also carries the auditory, lateral line, and vestibular information. This seems like a confound in the interpretation of this preparation as a simple audio-visual paradigm. Minimally, this confound should be noted and addressed. The first heading of the Results should not refer to "visual tectal stimuli".

We changed the heading of the corresponding section of the Results section as requested and also omitted the term “optic” when we did not specifically refer to tectal circuits that process optic information.  

Comment 3 (Major): Figure 1 and associated text.

It is unclear and not mentioned in the Methods section how phasic and tonic responses were calculated. It is clear from the example traces that there is a change in tonic responses and the accumulation of subthreshold responses. Depending on how tonic responses were calculated, perhaps the authors could overlay a low-passed filtered trace and/or show calculations based on the filtered trace at each tectal train duration.

The revised version of the manuscript now includes a description of how the phasic and tonic components were calculated (lines 163-172). We also modified the color scheme and the inset of Figure 1A to clarify how these two components were defined. Since we quantified the response in a 12 ms window, we did not include an overlayed low-pass filtered trace since it might be confusing with respect to the metric used.

Comment 4 (Minor): Figure 3 and associated text.

This is a lovely experiment. Although it is not written in text, it provides logic for the next experiment in choosing a 50ms time interval. It would be great if the authors calculated the first timepoint at which the percentage of shunting inhibition is not significantly different from zero. This would provide a convincing basis for picking 50ms for the next experiment. That said, I suspect that this time point would be earlier than 50 ms. This may explain and add further complexity to why the authors found mostly linear or sublinear integration, and perhaps the basis for future experiments to test different stimulus time intervals. Please move calculations to Methods.

We moved calculations to the Methods section (lines 201-208). We mention the rationale for selecting the 50 ms interval in the next experiment (Figure 4, lines 369-371) and discuss in detail the potential contribution of FFI to the complexity of the integration taking place in the M-cell circuit (Discussion, lines 512-535).

Comment 5 (Major): Figure 4C and lines 398-410.

These are beautiful examples of M-cell firing, but the text suggests that they occurred rarely and nowhere close to significantly above events observed from single modalities. We do not see this as a valid result to report because there is insufficient evidence that the phenomenon shown is consistent or representative of your data.

Our experimental conditions required anesthesia and paralysis, conditions designed to reduce neuronal firing and suppress motor output. We think it is valuable to report that we still see that simultaneous presentation subthreshold unisensory stimuli can add up to become suprathreshold, paralleling behavioral observations. We do not claim and acknowledge that those examples are representative of our recording conditions, but are likely to be more representative of the multisensory integration process taking place in freely moving fish. The revised manuscript adds context to these example traces to justify their inclusion (lines 420-426).

Reviewer #2 (Recommendations For The Authors):

Methods

The Methods section on "Auditory stimuli" contains a long background on the biophysics of the M-cell and its inputs. This does not belong in Methods. The same is true, to a lesser degree, in the next heading. The argument that direct stimulation of the tectum is necessary to bypass adaptation should be in Results, not Methods.

Following the reviewer recommendation, we have moved both paragraphs to the Results section.

Figure 1 and associated text.

Visually, the use of transparency to differentiate phasic and tonic calculations is difficult to read. Example traces are also cut off at the top and bottom at random sizes.

We changed the color scheme to avoid the use of transparency and modified the inset of Figure 1A to clarify how the phasic and tonic components were calculated. We also modified the dimensions of the clipping mask used to trim the stimulation artifacts of sample traces to make them more similar while still enabling clear observation of the phasic and tonic components of the response.

Line 338 "optical tectum" is not correct. "optic tectum" is more common, or better still, just "tectum".

We apologize for the error. The two instances of “optical tectum” were replaced by the correct term (“optic tectum”).

eLife Assessment

This important study highlights the use of siderophores as antibacterials, and the authors also discuss the consequences and efficacy of 'siderophore therapy' in more complex communities/environments. The evidence supporting the overall hypotheses ranges is largely convincing. The work will be of broad interest to people working in the fields of evolutionary ecology, microbiology and medical sciences.

eLife Assessment

This work is an important contribution to the development of a biologically plausible theory of statistical modeling of spiking activity. The authors convincingly implemented the statistical inference of input likelihood in a simple neural circuit, demonstrating the relationship between synaptic homeostasis, neural representations, and computational accuracy. This work will be of interest to neuroscientists, both theoretical and experimental, who are exploring how statistical computation is implemented in neural networks.

Reviewer #1 (Public review):

Summary

A novel statistical model of neural population activity called the Random Projection model has been recently proposed. Not only is this model accurate, efficient, and scalable, but also is naturally implemented as a shallow neural network. This work proposes a new class of RP model called the reshaped RP model. Inheriting the virtue of the original RP model, the proposed model is more accurate in terms of data fitting and efficient in terms of lower firing rate than the original, as well as compatible with various biological constraints. In particular, the authors have demonstrated that normalizing the total synaptic input in the reshaped model has a homeostatic effect on the firing rates of the neurons, resulting in even more efficient representations with equivalent accuracy. These results suggest that synaptic normalization contributes to synaptic homeostasis as well as efficiency in neural encoding.

Strength

This paper demonstrates that the accuracy and efficiency of the random projection models can be improved by extending the model with reshaped projections. Furthermore, it broadens the applicability of the model under biological constraints of synaptic regularization. It also suggests the advantage of the sparse connectivity structure over the fully connected model for modeling spiking statistics. In summary, this work successfully integrates two different elements, statistical modeling of the spikes and synaptic homeostasis in a single biologically plausible neural network model. The authors logically demonstrate their arguments with clear visual presentations and well-structured text, facilitating an unambiguous understanding for readers.

Discussions

The authors have clearly responded to most of our questions in the revised manuscript and we are happy to recommend publishing the final version of the article as it is. Below, we would like to present some alternative interpretations of the results. These comments are not exclusive with the claims made in the articles; it is rather intended to enhance the understanding of readers by providing additional perspectives.

As summarized above, the main contribution of the work consists of two parts; (1) the reshaped RP model achieved higher performance in explaining the statistics of the spiking activity of cortical neurons with more efficient representations (=lower firing rate), (2) synaptic homeostatic normalization in the reshaped RP model yields even more efficient representations without impairing the fitting performance.

For part (1),<br /> Suppl. Fig. 1B compares reshaped RP models with RP and RP with pruning and replacement (R&P). The better performance of RP with P&R might imply the advantage of pruning over gradient descent in this setting, reflecting the non-convexities of the problem. Alternatively, it might suggest the benefit of the increased number of parameters, since pruning allows the network to explore the broader parameter space during the learning process. This latter view might partially explain the superiority of the reshaped RP model over the original RP model.<br /> It is interesting that the backprop model has higher firing rate than the reshaped model (Fig. 1D). This trend is unchanged when optimization of the neural threshold is also allowed (Supp. Fig. 2A). Since backprop model overperforms reshaped model slightly but robustly, high firing rates in the backprop model might be a genuine feature of the spike statistics.

For part (2),<br /> We note that λ regulates the average firing rate, since maximizing the entropy <-ln p(x)> with a regularization term -λ <\sum _i f(x_i)> results in λ_i = λ for all i in the Boltzmann distribution of eq. 2. Suppl. Fig. 2B could be understood as demonstrating this "homeostatic" effect of λ.<br /> Suppl. Fig. 3 could be interpreted as demonstrating the interaction of two different homeostatic mechanisms: one at the firing-rate level (as regulated by λ) and the other at the synaptic level (as regulated by φ). It shows that the range of synaptic constraints where the fitting performance is not impaired differs by the value of λ. For example, if lambda is small (\lambda = 0.25), synaptic constraint can easily deteriorate the performance; on the other hand, if lambda is large (\lambda = 4), performance remains unchanged under strict synaptic constraint. Considering that practically we are most interested in the regime where the model performs best (λ = 2.0), an advantageous feature of the homeostatic model is that homeostatic constraint is harmless at λ=2.0 for the wide range of constraints.

Author response:

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

Public Comments:

(1) We find it interesting that the reshaped model showed decreased firing rates of the projection neurons. We note that maximizing the entropy <-ln p(x)> with a regularizing term -\lambda <\sum _i f(x_i)>, which reflects the mean firing rate, results in \lambda _i = \lambda for all i in the Boltzmann distribution. In other words, in addition to the homeostatic effect of synaptic normalization which is shown in Figures 3B-D, setting all \lambda_i = 1 itself might have a homeostatic effect on the firing rates. It would be better if the contribution of these two homeostatic effects be separated. One suggestion is to verify the homeostatic effect of synaptic normalization by changing the value of \lambda.

This is an interesting question and we, therefore, explored the effects of different values of $\lambda$ on the performance of unconstrained reshaped RP models and their firing rates. The new supp. Figure 2B presents the results of this exploration: We found that for models with a small set of projections, a high value of $\lambda$ results in better performance than models with low ones, while for models with a large set of projections we find the opposite relation. The mean firing rates of the projection neurons for models with different values of $\lambda$ show a clear trend, where higher $\lambda$ values results in lower mean firing rates.

Thus, these results suggest an interplay between the optimal size of the projection set and the value of $\lambda$ one should pick. For the population sizes and projection sets we have used here, $\lambda=1$ is a good choice, but, for different population sizes or data sets a different value of $\lambda$ might be better.

Thus, in addition to supp. Figure 2B, we therefore added the following to the main text:

“An additional set of parameters that might affect the Reshaped RP models are the coefficients $\lambda$, that weigh each of the projections. Above, we used $\lambda=1$ for all projections, here we investigated the effect of the value of $\lambda$ on the performance of the Reshaped RP models (supp. Figure 2B). We find that for models with a small projection set, high $\lambda$ values result in better performance than models with low values. We find an opposite relation for models with large number projection sets. (We submit that the performance decrease of Reshaped RP models with high value of $\lambda$, as the number of projections grows, is a reflection of the non-convex nature of the Reshaped RP optimization problem).

The mean firing rates of the projection neurons for models with different values of $\lambda$ show a clear trend, higher $\lambda$ values results in lower mean firing rates. Thus, we conclude that there is an interplay between the number of projections and the value of $\lambda$ we should pick. For the sizes of projection sets we have used here, $\lambda=1$ is a good choice, but, we note that in general, one should probably seek the appropriate value of $\lambda$ for different population sizes or data sets.”

In addition, we explored the effect of synaptic normalization on models with different values of $\lambda$ (supp. Figure 3). We found that homeostatic Reshaped RP models are superior to the non-homeostatic Reshaped RP models: For low values of $\lambda$, the homeostatic and Reshaped RP models show similar performance in terms of log-likelihood, whereas the homeostatic models are more efficient. For high values of $\lambda_i$ homeostatic models are not only more efficient but also show better performance. These results indicate that the benefit of the homeostatic model is insensitive to the specific choice of $\lambda$.

In addition to supp. Figure 3, we added the following to the main text:

“Exploring the effect of synaptic normalization on models with different values of $\lambda$ (supp. Figure 3), we find that homeostatic Reshaped RP models are superior to the non-homeostatic Reshaped RP models: For low values of $\lambda$, the homeostatic and Reshaped RP models show similar performance in terms of log-likelihood, whereas the homeostatic models are more efficient. Importantly, for high values of $\lambda_i$ homeostatic models are not only more efficient but also show better performance. We conclude that the benefit of the homeostatic model is insensitive to the specific choice of $\lambda$.”

(2) As far as we understand, \theta_i (thresholds of the neurons) are fixed to 1 in the article. Optimizing the neural threshold as well as synaptic weights is a natural procedure (both biologically and engineeringly), and can easily be computed by a similar expression to that of a_ij (equation 3). Do the results still hold when changing \theta _i is allowed as well? For example,

a. If \theta _i becomes larger, the mean firing rates will decrease. Does the backprop model still have higher firing rates than the reshaped model when \theta _i are also optimized?

b. Changing \theta _i affects the dynamic range of the projection neurons, thus could modify the effect of synaptic constraints. In particular, does it affect the performance of the bounded model (relative to the homeostatic input models)?

We followed the referee’s suggestion, and extended our current analysis, and added threshold optimization to the Reshape and Backpropagation models, which is now shown in supp. Figure 2A.  Comparing the performance and properties of these models to ones with fixed thresholds, we found that this addition had a small effect on the performance of the models in terms of their likelihood. (supp. Figure 2A). We further find that backpropagation models with tuned thresholds show lower firing rates compared to backpropagation models with fixed threshold, while reshaped RP models with optimized thresholds show higher firing rates compared to models with fixed threshold. These differences are, again, rather small, and both versions of the reshaped RP models show lower firing rates compared to both versions of the backpropagation models.

In addition to supp. Figure 2A, we added the following to the main text:

“The projections' threshold $\theta_i$, which is analogous to the spiking threshold of the projection neurons, strongly affects the projections' firing rates. We asked how, in addition to reshaping the coefficients of each projection, we can also change $\theta_i$ to optimize the reshaped RP and backpropagation models.

We find that this addition has a small effect on the performance of the models in terms of their likelihood (supp. Figure 2A).

We also find that this has a small effect on the firing rates of the projection neurons: backpropagation models with tuned thresholds show lower firing rates compared to backpropagation models with fixed threshold, whereas reshaped RP models with optimized thresholds show higher firing rates compared to models with fixed threshold. Yet, both versions of the reshaped RP models show lower firing rates compared to both versions of the backpropagation models. Given the small effect of tuning threshold on models' performance and their internal properties, we will, henceforth, focus on Reshaped RP models with fixed thresholds.”

(3) In Figure 1, the authors claim that the reshaped RP model outperforms the RP model. This improved performance might be partly because the reshaped RP model has more parameters to be optimized than the RP model. Indeed, let the number of projections N and the in-degree of the projections K, then the RP model and the reshaped RP model have N and KN parameters, respectively. Does the reshaped model still outperform the original one when only (randomly chosen) N weights (out of a_ij) are allowed to be optimized and the rest is fixed? (or, does it still outperform the original model with the same number of optimized parameters (i.e. N/K neurons)?)

Indeed, the number of tuned parameters in the reshaped RP model is much larger compared to the number of tuned parameters in an RP model with the same projection set size. Yet, we submit that the larger number of tuned parameters is not the reason for the improved performance of the reshaped RP model: Maoz et al [30] have already shown that by optimizing an RP model with a small projection set using the pruning and replacement of projections (P&R), one can reach high accuracy with an almost order of magnitude fewer projections. Thus, we argue that the improved performance stems from the properties of the projections in the model.

Accordingly, we therefore added supp. Figure 2B that shows the performance of P&R sigmoid RP model compared to RP and reshaped RP models. We added the following to the main text:

“Because reshaping may change all the existing synapses of each projection, the number of parameters is the number of projections times the projections in-degree. While this is much larger than the number of parameters that we learn for the RP model (one for each projection), we suggest that the performance of the reshaped models is not a naive result of having more parameters. In particular, we have seen that RP models that use a small set of projections can be very accurate when the projections are optimized using the pruning and replacement process [30] (see also supp. Figure 1B). Thus, it is really the nature of the projections that shapes the performance. Indeed, our results here show that a small fixed connectivity projection set with weight tuning is enough for accurate performance which is on par or better than an RP model with more projections.”

(4) In Figure 2, the authors have demonstrated that the homeostatic synaptic normalization outperforms the bounded model when the allowed synaptic cost is small. One possible hypothesis for explaining this fact is that the optimal solution lies in the region where only a small number of |a_ij| is large and the rest is near 0. If it is possible to verify this idea by, for example, exhibiting the distribution of a_ij after optimization, it would help the readers to better understand the mechanism behind the superiority of the homeostatic input model.

We modified supp. Figure 4 and made the following change in the relevant part in the main text to address the reviewer comment about the distribution of the $a_{ij}$ values:

“Figure 5E shows the mean rotation angle over 100 homeostatic models as a function of synaptic cost -- reflecting that the different forms of homeostatic regulation results in different reshaped projections. We show in Supp. Figure 4C the histogram of the rotation angles of several different homeostatic models, as well as the unconstrained Reshape model.

Analyzing the distribution of the synaptic weights $a_{ij}$ after learning leads to a similar conclusion (supp. Figure 4D): The peak of the histograms is at $a_{ij} = 0$, implying that during reshaping most synapses are effectively pruned. While the distribution is broader for models with higher synaptic budget, it is asymmetric, showing local maxima at different values of $a_{ij}$.

The diversity of solutions that the different model classes and parameters show imply a form of redundancy in model choice or learning procedure. This reflects a multiplicity of ways to learn or optimize such networks that biology could use to shape or tune neural population codes.“

(5) In Figures 5D and 5E, the authors present how different reshaping constraints result in different learning processes ("rotation"). We find these results quite intriguing, but it would help the readers understand them if there is more explanation or interpretation. For example,

a. In the "Reshape - Hom. circuit 4.0" plot (Fig 5D, upper-left), the rotation angle between the two models is almost always the same. This is reasonable since the Homeostatic Circuit model is the least constrained model and could be almost irrelevant to the optimization process. Is there any similar interpretation to the other 3 plots of Figure 5D?

We added a short discussion of this difference to the main text, but do not have a geometric or other intuitive explanation for the nature of these differences.

b. In Figure 5E, is there any intuitive explanation for why the three models take minimum rotation angle at similar global synaptic cost (~0.3)?

We added discussion of this issue to the main text, and the histogram of the rotation angles in Supp Figure 4c shows that they are not identical. But, we don’t have an intuitive explanation for why the mean values are so similar.

Recommendations for the authors:

(1) Some claims on the effect of synaptic normalization on the reshaped model sound a little overstated since the presented evidence does not clearly show the improvement of the computational performance (in comparison to the vanilla reshaped model) in terms of maximizing the likelihood of the inputs. Here are some examples of such claims: "Incorporating more biological features and utilizing synaptic normalization in the learning process, results in even more efficient and accurate models." (in Abstract), "Thus, our new scalable, efficient, and highly accurate population code models are not only biologically-plausible but are actually optimized due to their biological features." (in Abstract), or "in our Reshaped RP models, homeostatic plasticity optimizes the performance of network models" (in Discussion).

We changed the wording according to the reviewers’ suggestions.

(2) In equation (1) and the following sentence, \theta _j (threshold) should be \theta _i.

Fixed

(3) While the authors mention that "reshaping with normalization or without it drives the projection neurons to converge to similar average firing rate values (Figure 3B)", they also claim that "reshaping with normalization implies lower firing rates as well as... (Figure 3E)". These two claims look a little inconsistent to us. Besides, it is not very clear from Figure 3E that the normalization decreases the firing rate (it is clear from Figure 3B, though). How about just deleting "lower firing rates as well as"?

We changed the wording according to the reviewers’ suggestion.

(4) The captions of Figures 4D and 4E should be exchanged.

Fixed

(5) Typo in In Figure 4F: "normalized in-dgreree".

Fixed

(6) In Figure 5D (upper left plot) the choice of "Reshape" and "Bounded3.0" looks a bit weird. Is this the typo of "Hom. cicruit 4.0"?

There is no typo in the figure labels. We discussed the results of figure 5D in our response to point (5) in the public comments list and addressed the upper left panel of figure 5D in the main text.

(7) In the paper, the letter \theta represents (1) the threshold of the projection neurons (eq. 1), (2) the "ceiling" value of the bounded model, and (3) the rotation angle of projections (Figure 5). We find this notation a bit confusing and recommend using different notations for different entities.

Thanks for the suggestion, we changed the confusing notations: (1) The threshold of each projection neuron is still $\theta$, following the notation of the original RP model formulation [30]. (2) The notation of the “ceiling” value of the bounded model is now $\omega$. (3) The rotation angle of the projections during reshape is now marked by $\alpha$.

Author response:

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

We thank you for the time you took to review our work and for your feedback! The main changes to the manuscript are: 

(1) We have added additional analysis of running onsets in closed and open loop conditions for audiomotor (Figure 2H) and visuomotor (Figure 3H) coupling.  

(2) We have also added analysis of running speed and pupil dilation upon mismatch presentation (Figures S2A and S2B, S4A and S4B, and S5A and S5B).

(3) We have expanded on the discussion of the nature of differences between audiomotor and visuomotor mismatches.

Reviewer #1:

The manuscript presents a short report investigating mismatch responses in the auditory cortex, following previous studies focused on the visual cortex. By correlating the mouse locomotion speed with acoustic feedback levels, the authors demonstrate excitatory responses in a subset of neurons to halts in expected acoustic feedback. They show a lack of responses to mismatch in the visual modality. A subset of neurons show enhanced mismatch responses when both auditory and visual modalities are coupled to the animal's locomotion. 

While the study is well-designed and addresses a timely question, several concerns exist regarding the quantification of animal behavior, potential alternative explanations for recorded signals, correlation between excitatory responses and animal velocity, discrepancies in reported values, and clarity regarding the identity of certain neurons. 

Strengths: 

(1) Well-designed study addressing a timely question in the field. 

(2) Successful transition from previous work focused on the visual cortex to the auditory cortex, demonstrating generic principles in mismatch responses. 

(3) The correlation between mouse locomotion speed and acoustic feedback levels provides evidence for a prediction signal in the auditory cortex. 

(4) Coupling of visual and auditory feedback shows putative multimodal integration in the auditory cortex. 

Weaknesses: 

(1) Lack of quantification of animal behavior upon mismatches, potentially leading to alternative interpretations of recorded signals. 

(2) Unclear correlation between excitatory responses and animal velocity during halts, particularly in closed-loop versus playback conditions. 

(3) Discrepancies in reported values in a few figure panels raise questions about data consistency and interpretation. 

(4) Ambiguity regarding the identity of the [AM+VM] MM neurons. 

The manuscript is a short report following up on a series of papers focusing on mismatch responses between sensory inputs and predicted signals. While previous studies focused on the visual modality, here the authors moved to the auditory modality. By pairing mouse locomotion speed to the sound level of the acoustic feedback, they show that a subpopulation of neurons displays excitatory responses to halts in the (expected) acoustic feedback. These responses were lower in the open-loop state, when the feedback was uncorrelated to the animal locomotion. 

Overall it is a well-designed study, with a timely and well-posed question. I have several concerns regarding the nature of the MM responses and their interpretations. 

- One lacks quantification of the animal behavior upon mismatches. Behavioral responses may trigger responses in the mouse auditory cortex, and this would be an alternative explanation to the recorded signals. 

What is the animal speed following closed-loop halts (we only have these data for the playback condition)? 

We have quantified the running speed of the mouse following audiomotor and visuomotor mismatches. We found no evidence of a change in running speed. We have added this to Figures S2A and S4A, respectively.

Is there any pupillometry to quantify possible changes in internal states upon halts (both closed-loop and playback)?

The term 'internal state' may be somewhat ambiguous in this context. We assume the reviewer is asking whether we have any evidence for possible neuromodulatory changes. We know that there are noradrenergic responses in visual cortex to visuomotor mismatches (Jordan and Keller, 2023), but no cholinergic responses (Yogesh and Keller, 2023). Pupillometry, however, is likely not always sensitive enough to pick up these responses. With very strong neuromodulatory responses (e.g. to air puffs, or other startling stimuli), pupil dilation is of course detected, but this effect is likely at best threshold linear. Looking at changes in pupil size following audiomotor and visuomotor mismatch responses, we found no evidence of a change. We have added this to Figures S2B and S4B, respectively. Note, we suspect this is also strongly experience-dependent. The first audio- or visuomotor mismatch the mouse encounters is likely a more salient stimulus (to the rest of the brain, not necessarily to auditory or visual cortex), than the following ones.  

These quantifications must be provided for the auditory mismatches but also for the VM or [AM+VM] mismatches.  

During the presentation of multimodal mismatches [AM + VM], mice did not exhibit significant changes in running speed or pupil diameter. These data have been now added to Figures S5A and S5B.

- AM MM neurons supposedly receive a (excitatory) locomotion-driven prediction signal. Therefore the magnitude of the excitation should depend on the actual animal velocity. Does the halt-evoked response in a closed loop correlate with the animal speed during the halt? Is the correlation less in the playback condition? 

This is indeed what one would expect. We fear, however, that we don’t have sufficient data to address this question properly. Moreover, there is an important experimental caveat that makes the interpretation of the results difficult. In addition to the sound we experimentally couple to the locomotion speed of the mouse, the mouse self-generates sound by running (the treadmill rotating, changes to the airflow of the air-supported treadmill, footsteps, etc.). These sources of sound all also correlate in intensity with running speed. Thus, it is not entirely clear how our increase in sound amplitude with increasing running speed relates to the increase in self-generated sounds on the treadmill. This is one of the key reasons we usually do this type of experiment in the visual system where experimental control of visual flow feedback (in a given retinotopic location) is straightforward. 

Having said that, if we look at the how mismatch responses change as a function of locomotion speed across the entire population of neurons, there appears to be no systematic change with running speed (and the effects are highly dependent on speed bins we choose). However, just looking at the most audiomotor mismatch responsive neurons, we find a trend for increased responses with increasing running speed (Author response image 1). We analyzed the top 5% of cells that showed the strongest response to mismatch (MM) and divided the MM trials into three groups based on running speed: slow (10-20 cm/s), middle (20-30 cm/s), and fast (>30 cm/s). Given the fact that we have on average 14 mismatch events in total per neuron, we don’t have sufficient data to analyze this. 

Author response image 1.

The average response of strongest AM MM responders to AM mismatches as a function of running speed (data are from 51 cells, 11 fields of view, 6 mice). 

Values in Figure 2H are way higher than what can be observed in Figures 2C, and D. Could you explain the mismatch in values? Same for 3H and 4F. 

In Figure 2H (now Figure S2F), we display responses from 4 755 individual neurons. Since most recorded neurons did not exhibit significant responses to mismatch presentations, their responses cluster around zero, significantly contributing to the final average shown in panel D. To clarify how individual neurons contribute to the overall population activity, we have added a histogram showing the distribution of neurons responding to audiomotor mismatch and sound playback halts. We hope this addition clarifies how individual neuron responses affect the final population activity. 

Furthermore, neurons exhibiting suppression upon closed-loop halts (Figure 2C) show changes in deltaF/F of the same order of magnitude as the AM MM neurons (with excitatory responses). I cannot picture where these neurons are found in the scatter plot of Figure 2H. 

This is caused by a ceiling effect. While we could adjust the scale of the heat map to capture neurons with very high responses (e.g. [-50 50], Author response image 2), doing so would obscure the response dynamics of most neurons. Note that the number of neurons on the y-axis far exceeds the resolution of this figure and thus there are also aliasing issues that mask the strong responses. 

Author response image 2.

Responses of all L2/3 ACx neurons to audiomotor mismatches. Same as Figure 2C with different color scale [-50 50] which does not capture most of the neural activity.  

- Are [AM+VM] MM neurons AM neurons? 

Many of [AM + VM] and [AM] neurons overlap but it is not exactly the same population. This is partially visible in Figure 4F. There is a subset of neurons (13.7%; red dots, Figure 4F) that selectively responded to the concurrent [AM+VM] mismatch, while a different subset of neurons (11.2%; yellow dots, Figure 4F) selectively responded to the mismatch responses in isolation. The [VM] response contributes only little to the sum of the two responses [AM] + [VM]. 

Please do not use orange in Figure 4F, it is perceptually too similar to red. 

We have now changed it to yellow. 

Reviewer #2 (Public Review): 

In this study, Solyga and Keller use multimodal closed-loop paradigms in conjunction with multiphoton imaging of cortical responses to assess whether and how sensorimotor prediction errors in one modality influence the computation of prediction errors in another modality. Their work addresses an important open question pertaining to the relevance of non-hierarchical (lateral cortico-cortical) interactions in predictive processing within the neocortex. 

Specifically, they monitor GCaMP6f responses of layer 2/3 neurons in the auditory cortex of head-fixed mice engaged in VR paradigms where running is coupled to auditory, visual, or audio-visual sensory feedback. The authors find strong auditory and motor responses in the auditory cortex, as well as weak responses to visual stimuli. Further, in agreement with previous work, they find that the auditory cortex responds to audiomotor mismatches in a manner similar to that observed in visual cortex for visuomotor mismatches. Most importantly, while visuomotor mismatches by themselves do not trigger significant responses in the auditory cortex, simultaneous coupling of audio-visual inputs to movement non-linearly enhances mismatch responses in the auditory cortex. 

Their results thus suggest that prediction errors within a given sensory modality are non-trivially influenced by prediction errors from another modality. These findings are novel, interesting, and important, especially in the context of understanding the role of lateral cortico-cortical interactions and in outlining predictive processing as a general theory of cortical function. 

In its current form, the manuscript lacks sufficient description of methodological details pertaining to the closed-loop training and the overall experimental design. In several scenarios, while the results per se are convincing and interesting, their exact interpretation is challenging given the uncertainty about the actual experimental protocols (more on this below). Second, the authors are laser-focused on sensorimotor errors (mismatch responses) and focus almost exclusively on what happens when stimuli deviate from the animal's expectations. 

While the authors consistently report strong running-onset responses (during open-loop) in the auditory cortex in both auditory and visual versions of the task, they do not discuss their interpretation in the different task settings (see below), nor do they analyze how these responses change during closed-loop i.e. when predictions align with sensory evidence. 

However, I believe all my concerns can be easily addressed by additional analyses and incorporation of methodological details in the text. 

Major concerns: 

(1) Insufficient analysis of audiomotor mismatches in the auditory cortex: 

Lack of analysis of the dependence of audiomotor mismatches on the running speed: it would be helpful if the authors could clarify whether the observed audiomotor mismatch responses are just binary or scale with the degree of mismatch (i.e. running speed). Along the same lines, how should one interpret the lack of dependence of the playback halt responses on the running speed? Shouldn't we expect that during playback, the responses of mismatch neurons scale with the running speed? 

Regarding the scaling of AM mismatch responses with running speed, please see our response to reviewer 1 above to the same question. 

Regarding the playback halt response and dependence on running speed, we would not expect there to be a dependence. The playback halt response (by design) measures the strength of the sensory response to a cessation of a stimulus (think OFF response). These typically are less strong in cortex than the corresponding ON responses but need to be controlled for (else a mismatch response might just be an OFF response – the prediction error is quantified as the difference between AM mismatch response and playback halt response). Given that sound onset responses only have a small dependence on running state, we would similarly expect sound offset (playback halt) responses to exhibit only minimal dependence on running state. 

Slow temporal dynamics of audiomotor mismatches: despite the transient nature of the mismatches (1s), auditory mismatch responses last for several seconds. They appear significantly slower than previous reports for analogous visuomotor mismatches in V1 (by the same group, using the same methods) and even in comparison to the multimodal mismatches within this study (Figure 4C). What might explain this sustained activity? Is it due to a sustained change in the animal's running in response to the auditory mismatch? 

This is correct, neither AM or AM+VM mismatch return to baseline in the 3 seconds following onset. VM mismatch response in visual cortex also do not return to baseline in that time window (see e.g.

Figure 1E in (Attinger et al., 2017), or Figure 1F in (Zmarz and Keller, 2016). What the origin or computation significance of this sustained calcium response is we do not know. In intracellular signals, we do not see this sustained response (Jordan and Keller, 2020). Also peculiar is indeed the fact that in the case of AM mismatch the sustained response is similar in strength to the initial response. But also here, why this would be the case, we do not know. It is conceivable that the initial and the sustained calcium response have different origins, if the sustained response amplitude is all or nothing, the fact that the AM mismatch response is the smallest of the three could explain why sustained and initial responses are closer than for [AM+VM] or VM (in visual cortex) mismatch responses. All sustained responses appear to be roughly 1% dF/F. There are no apparent changes in running speed or pupil dilation that would correlate with the sustained activity (new panel A in Figure S2). 

(2) Insufficient analysis and discussion of running onset responses during audiomotor sessions: The authors report strong running-onset responses during open-loop in identified mismatch neurons. They also highlight that these responses are in agreement with their model of subtractive prediction error, which relies on subtracting the bottom-up sensory evidence from top-down motor-related predictions. I agree, and, thus, assume that running-onset responses during the open loop in identified 'mismatch' neurons reflect the motor-related predictions of sensory input that the animal has learned to expect. If this is true, one would expect that such running-onset responses should dampen during closed-loop, when sensory evidence matches expectations and therefore cancels out this prediction. It would be nice if the authors test this explicitly by analyzing the running-related activity of the same neurons during closed-loop sessions. 

Thank you for the suggestion. We now show running onset responses in both closed and open loop conditions for audiomotor and visuomotor coupling (new Figures 2H and 3H). In closed loop, we observe only a transient running onset response. In the open loop condition, running onset responses are sustained. For the visuomotor coupling, running onset responses are sustained in both closed and open loop conditions. This would be consistent with a slightly delayed cancellation of sound and motor related inputs in the audiomotor closed loop condition but not otherwise. 

(3) Ambiguity in the interpretation of responses in visuomotor sessions. 

Unlike for auditory stimuli, the authors show that there are no obvious responses to visuomotor mismatches or playback halts in the auditory cortex. However, the interpretation of these results is somewhat complicated by the uncertainty related to the training history of these mice. Were these mice exclusively trained on the visuomotor version of the task or also on the auditory version? I could not find this info in the Methods. From the legend for Figure 4D, it appears that the same mice were trained on all versions of the task. Is this the case? If yes, what was the training sequence? Were the mice first trained on the auditory and then the visual version? 

The training history of the animals is important to outline the nature of the predictions and mismatch responses that one should expect to observe in the auditory cortex during visuomotor sessions.

Depending on whether the mice in Figure 3 were trained on visual only or both visual and auditory tasks, the open-loop running onset responses may have different interpretations. 

a) If the mice were trained only on the visual task, how should one interpret the strong running onset responses in the auditory cortex? Are these sensorimotor predictions (presumably of visual stimuli) that are conveyed to the auditory cortex? If so, what may be their role? 

b) If the mice were also trained on the auditory version, then a potential explanation of the running-onset responses is that they are audiomotor predictions lingering from the previously learned sensorimotor coupling. In this case, one should expect that in the visual version of the task, these audiomotor predictions (within the auditory cortex) would not get canceled out even during the closedloop periods. In other words, mismatch neurons should constantly be in an error state (more active) in the closed-loop visuomotor task. Is this the case? 

If so, how should one then interpret the lack of a 'visuomotor mismatch' aligned to the visual halts, over and above this background of continuous errors? 

As such, the manuscript would benefit from clearly stating in the main text the experimental conditions such as training history, and from discussing the relevant possible interpretations of the responses. 

Mice were not trained on either audiomotor or visuomotor coupling and were reared normally. Prior to the recording day, the mice were habituated to running on the air-supported treadmill without any coupling for up to 5 days. On the first recording day, the mice experienced all three types of sessions (audiomotor, visuomotor, or combined coupling) in a random order for the first time. We have clarified this in the methods. 

Regarding the question of how one should interpret the strong running onset responses in the auditory cortex, this is complicated by the fact that – unless mice are raised visually or auditorily deprived – they always have life-long experience with visuomotor or audiomotor coupling. The visuomotor coupling they experience in VR is geometrically matched to what they would experience by moving in the real world, for the audiomotor coupling the exact relationship is less clear, but there are a diverse set of sound sources that scale in loudness with increasing running speed. Hence running onset responses reflect either such learned associations (as the reviewer also speculates), or spurious input. Rearing mice without coupling between movement and visual feedback does not abolish movement related responses in visual cortex (Attinger et al., 2017), to the contrary, it enhances them considerably. We suspect this reflects visual cortex being recruited for other functions in the absence of visual input. But given the data we have we cannot distinguish the different possible sources of running related responses. It is very likely that any “training” related effect we could achieve in a few hours pales in comparison to the life-long experience the mouse has in the world. 

Regarding the lack of a 'visuomotor mismatch' aligned to the visual halts, we are not sure we understand. Our interpretation is that there are no (or only a very small - we speculate that any nonzero VM mismatch response is just inherited from visual cortex) VM mismatch responses in auditory cortex above chance. Our data are consistent with the interpretation that there is no opposition of bottom up visual and top down motor related input in auditory cortex, hence no VM mismatch responses (independent of how strong the top-down motor related input is). This is of course not surprising – this is more of a sanity check and becomes relevant in the context of interpreting AM+VM responses. 

(4) Ambiguity in the interpretation of responses in multimodal versus unimodal sessions. 

The authors show that multimodal (auditory + visual) mismatches trigger stronger responses than unimodal mismatches presented in isolation (auditory only or visual only). Further, they find that even though visual mismatches by themselves do not evoke a significant response, co-presentation of visual and auditory stimuli non-linearly augments the mismatch responses suggesting the presence of nonhierarchical interactions between various predictive processing streams. 

In my opinion, this is an important result, but its interpretation is nuanced given insufficient details about the experimental design. It appears that responses to unimodal mismatches are obtained from sessions in which only one stimulus is presented (unimodal closed-loop sessions). Is this actually the case? An alternative and perhaps cleaner experimental design would be to create unimodal mismatches within a multimodal closed-loop session while keeping the other stimulus still coupled to the movement. 

This is correct, unimodal mismatches were acquired in unimodal coupling. Testing unimodal mismatch responses in multimodally coupled VR is an interesting idea we had initially even pursued. However, halting visual flow in a condition of coupling of both visual flow and sound amplitude to running speed has an additional complication. Introducing an audiomotor mismatch in this coupling inherently also creates an audiovisual (AV) mismatch, and the same applies to visuomotor mismatches, which cause a concurrent visuoaudio (VA) mismatch (Figure R3). This assumes that there are cross modal predictions from visual cortex to auditory cortex as there are from auditory cortex to visual cortex (Garner and Keller, 2022). There are interesting differences between the different types of mismatches, but with the all the necessary passive controls this quickly exceeded the amount of data we could reasonably acquire for this paper. This remains an interesting question for future research. 

Author response image 3.

Rationale of unimodal mismatches introduced within multimodal paradigm. 

Given the current experiment design (if my assumption is correct), it is unclear if the multimodal potentiation of mismatch responses is a consequence of nonlinear interactions between prediction/error signals exchanged across visual and auditory modalities. Alternatively, could this result from providing visual stimuli (coupled or uncoupled to movement) on top of the auditory stimuli? If it is the latter, would the observed results still be evidence of non-hierarchical interactions between various predictive processing streams? 

Mice are not in complete darkness during the AM mismatch experiments (the VR is off, but there is low ambient light in the experimental rooms primarily from computer screens), so we can rule out the possibility that the difference comes from having “no” visual input during AM mismatch responses. Addressing the question of whether it is this particular stimulus that cause the increase would require an experiment in which we couple sound amplitude but keep visual flow open loop. We did not do this, but also think this is highly unlikely. However, as described above, we did do an experiment in which we coupled both sound amplitude and visual flow to running, and then either halted visual flow, or sound amplitude, or both. Comparing the [AM+VM] and [AM+AV] mismatch responses, we find that [AM+VM] responses are larger than [AM+AV] responses as one would expect from an interaction between [AM] and [VM] responses (Author response image 4). Finally, either way the conclusion that there are nonhierarchical interactions of prediction error computations holds either way – if any visual stimulus (either visuomotor mismatch, or visual flow responses) influences audiomotor mismatch responses, this is evidence of non-hierarchical interactions.   

Author response image 4.

Average population response of all L2/3 neurons to concurrent [AM + VM] or [AM+AV] mismatch. Gray shading indicates the duration of the stimulus.

Along the same lines, it would be interesting to analyze how the coupling of visual as well as auditory stimuli to movement influences responses in the auditory cortex in close-loop in comparison to auditoryonly sessions. Also, do running onset responses change in open-loop in multimodal vs. unimodal playback sessions? 

We agree, and why we started out doing the experiments described above. We stopped with this however, because it quickly became a combinatorial nightmare. We will leave addressing the question of how different types of coupling influences responses in auditory cortex to brave future neuroscientists. 

Regarding the question of running onset responses, in both the multimodal and auditory only paradigms, running onset responses are transient; bottom-up sensory evidence is quickly subtracted from top-down motor-related prediction (Author response image 5). While there appears to be a small difference in the dynamics of running onset responses between these two paradigms, it was not significant. Note, we also have much less data than we would like here for this type of analysis. 

Author response image 5.

Running onset responses recorded in unimodal and multimodal closed loop sessions (1903 neurons, 16 fields of view, 8 mice)

We also compared running onsets in open loop sessions and did not find any significant differences between unimodal and multimodal sessions (Author response image 6). We found only six sessions in which animals performed at least two running onsets in each session type, therefore, we do not have enough data to include it in the manuscript. 

Author response image 6.

Running onset responses recorded within unimodal and multimodal open loop sessions (659 cells, 6 field of view, 5 mice).

Minor concerns and comments:

(1) Rapid learning of audiomotor mismatches: It is interesting that auditory mismatches are present even on day 1 and do not appear to get stronger with learning (same on day 2). The authors comment that this could be because the coupling is learned rapidly (line 110). How does this compare to the rate at which visuomotor coupling is learned? Is this rapid learning also observable in the animal's behavior i.e. is there a change in running speed in response to the mismatch? 

In the visual system this is a bit more complicated. If you look at visuomotor mismatch responses in a normally reared mouse, responses are present from the first mismatch (as far as we can tell given the inherently small dataset with just one response pre mouse). However, this is of course confounded by the fact that a normally reared mouse has visuomotor coupling throughout life from eye-opening. Raising mice in complete darkness, we have shown that approximately 20 min of coupling are sufficient to establish visuomotor mismatch responses (Attinger et al., 2017). 

Regarding the behavioral changes that correlate with learning, we are not sure what the reviewer would expect. We cannot detect a change in mismatch responses and hence would also not expect to see a change in behavior.

(2) The authors should clarify whether the sound and running onset responses of the auditory mismatch neurons in Figure 2E were acquired during open-loop. This is most likely the case, but explicitly stating it would be helpful. 

Both responses were measured in isolation (i.e. VR off, just sound and just running onset), not in an open-loop session. We have clarified in the figure legend that these are the same data as in Figure 1H and N. 

(3) In lines 87-88, the authors state 'Visual responses also appeared overall similar but with a small increase in strength during running ...'. This statement would benefit from clarification. From Figure S1 it appears that when the animal is sitting there are no visual responses in the auditory cortex. But when the animal is moving, small positive responses are present. Are these actually 'visual' responses - perhaps a visual prediction sent from the visual cortex to the auditory cortex that is gated by movement? If so, are they modulated by features of visual stimuli eg. contrast, intensity? Or, do these responses simply reflect motor-related activity (running)? Would they be present to the same extent in the same neurons even in the dark? 

This was wrong indeed - we have rephrased the statement as suggested. Regarding the source of visual responses, we use the term “visual response” operationally here agnostic to what pathway might be driving it (i.e. it could be a prediction triggered by visual input). 

We did not test if recorded visual responses are modulated by contrast or intensity. However, testing whether they are would not help us distinguish whether the responses are ‘visual’ or ‘visual predictions’. Finally, regarding the question about whether they are motor-related responses, this might be a misunderstanding. These are responses to visual stimuli while the mouse is already running (i.e. there is no running onset), hence we cannot test whether these responses are present in the dark (this would be the equivalent of looking at random triggers in the dark while the mouse is running).  

(4) The authors comment in the text (lines 106-107) about cessation of sound amplitude during audiomotor mismatches as being analogous to halting of visual flow in visuomotor mismatches. However, sound amplitude versus visual flow are quite different in nature. In the visuomotor paradigm, the amount of visual stimulation (photons per unit time) does not necessarily change systematically with running speed. Whereas, in the audiomotor paradigm, the SNR of the stimulus itself changes with running speed which may impact the accuracy of predictions. On a broader note, under natural settings, while the visual flow is coupled to movement, sound amplitude may vary more idiosyncratically with movement. 

This is a question of coding space. The coding space of visual cortex of the mouse is probably visual flow (or change in image) not number of photons. This already starts in the retina. The demonstration of this is quite impressive. A completely static image on the retina will fade to zero response (even though the number of photons remains constant). This is also why most visual physiologists use dynamic stimuli – e.g. drifting gratings, not static gratings – to map visual responses in visual cortex. If responses were linear in number of photons, this would make less of a difference. The correspondence we make is between visual flow (which we assume is the main coding space of mouse V1 – this is not established fact, but probably implicitly the general consensus of the field) and sound amplitude. Responses in auditory cortex are probably more linear in sound amplitude than visual cortex responses are linear in number of photons, but whether that is the correct coding space is still unclear, and as far as we can tell there is no clear consensus in the field. We did consider coupling running speed to frequency, which may work as well, but given the possible equivalence (as argued above) and the fact that we could see similar responses with sound amplitude coupling we did not explore frequency coupling. 

If visual speed is the coding space of V1, SNR should behave equivalently in both cases. 

Perhaps such differences might explain why unlike in the case of visual cortex experiments, running speed does not affect the strength of playback responses in the auditory cortex. 

Possible, but the more straightforward framing of this point is that sensory responses are enhanced by running in visual cortex while they are not in auditory cortex. A playback halt response (by design) is just a sensory response. Why running does not generally increase sensory responses in auditory cortex (L2/3 neurons), but does so in visual cortex, would be the more general version of the same question.

We fear we have no intelligent answer to this question.  

Reviewer #3 (Public Review): 

This study explores sensory prediction errors in the sensory cortex. It focuses on the question of how these signals are shaped by non-hierarchical interactions, specifically multimodal signals arising from same-level cortical areas. The authors used 2-photon imaging of mouse auditory cortex in head-fixed mice that were presented with sounds and/or visual stimuli while moving on a ball. First, responses to pure tones, visual stimuli, and movement onset were characterized. Then, the authors made the running speed of the mouse predictive of sound intensity and/or visual flow. Mismatches were created through the interruption of sound and/or visual flow for 1 second while the animal moved, disrupting the expected sensory signal given the speed of movement. As a control, the same sensory stimuli triggered by the animal's movement were presented to the animal decoupled from its movement. The authors suggest that auditory responses to the unpredicted silence reflect mismatch responses. That these mismatch responses were enhanced when the visual flow was congruently interrupted, indicates the cross-modal influence of prediction error signals. 

This study's strengths are the relevance of the question and the design of the experiment. The authors are experts in the techniques used. The analysis explores neither the full power of the experimental design nor the population activity recorded with 2-photon, leaving open the question of to what extent what the authors call mismatch responses are not sensory responses to sound interruption. The auditory system is sensitive to transitions and indeed responses to the interruption of the sound are similar in quality, if not quantity, in the predictive and the control situation. 

This study's strengths are the relevance of the question and the design of the experiment. The authors are experts in the techniques used. The analysis explores neither the full power of the experimental design nor the population activity recorded with 2-photon, leaving open the question of to what extent what the authors call mismatch responses are not sensory responses to sound interruption. The auditory system is sensitive to transitions and indeed responses to the interruption of the sound are similar in quality, if not quantity, in the predictive and the control situation. The pattern they observe is different from the visuomotor mismatch responses the authors found in V1 (Keller et al., 2012), where the interruption of visual flow did not activate neuronal activity in the decoupled condition. 

Just to add brief context to this. The reviewer is correct here, the (Keller et al., 2012) paper reports finding no responses to playback halt. However, this was likely a consequence of indicator sensitivity (these experiments were done with what now seems like a pre-historic version of GCaMP). Experiments performed with more modern indicators do find playback halt responses in visual cortex (see e.g. (Zmarz and Keller, 2016)). 

The auditory system is sensitive to transitions, also those to silence. See the work of the Linden or the Barkat labs on-off responses, and also that of the Mesgarani lab (Khalighinejad et al., 2019) on responses to transitions 'to clean' (Figure 1c) in the human auditory cortex. Since the responses described in the current work are modulated by movement and the relationship between movement and sound is more consistent during the coupled sessions, this could explain the difference in response size between coupled and uncoupled sessions. There is also the question of learning. Prediction signals develop over a period of several days and are frequency-specific (Schneider et al., 2018). From a different angle, in Keller et al. 2012, mismatch responses decrease over time as one might expect from repetition. 

Also for brief context, this might be a misconception. We don’t find a decrease of mismatch responses in the (Keller et al., 2012) paper – we assume what the reviewer is referring to is the fact that mismatch responses decrease in open-loop conditions (they normally do not in closed-loop conditions). This is the behavior one would expect if the mouse learns that movement no longer predicts visual feedback. 

It would help to see the responses to varying sound intensity as a function of previous intensity, and to plot the interruption response as a function of both transition and movement in both conditions. 

Given the large populations of neurons recorded and the diversity of the responses, from clearly negative to clearly positive, it would be interesting to understand better whether the diversity reflects the diversity of sounds used or a diversity of cell types, or both. 

Comments and questions: 

Does movement generate a sound and does this change with the speed of movement? It would be useful to have this in the methods. 

There are three ways to interpret the question – below the answers to all three:

(1) Running speed is experimentally coupled to sound amplitude of a tone played through a loudspeaker. Tone amplitude is scaled with running speed of the mouse in a closed loop fashion. We assume this is not what the reviewer meant, as this is described in the methods (and the results section). 

(2) Movements of the mouse naturally generate sounds (footsteps, legs moving against fur, etc.). Most of these sounds trivially scale with the frequency of leg movements – we assume this also not what the reviewer meant. 

(3) Finally, there are experimental sounds related to the rotation speed of the air supported treadmill that increase with running speed of the mouse. We have added this to the methods as suggested. 

Figures 1a and 2a. The mouse is very hard to see. Focus on mouse, objective, and sensory stimuli? The figures are generally very clear though. 

We have enlarged the mouse as suggested. 

1A-K was the animal running while these responses were measured? 

We did not restrict this analysis to running or sitting and pooled responses over both conditions.  We have made this more explicit in the results section.  

Data in Figure 1: Since the modulation of sensory responses by movement is relevant for the mismatch responses, I would move this analysis from S1 to Figure 1 and analyze the responses more finely in terms of running speed relative to sound and gratings. I would include here a more thorough analysis of the responses to 8kHz at varying intensities, for example in the decoupled sessions. Does the response adapt? Does it follow the intensity? 

We agree that these are interesting questions, but they do not directly pertain to our conclusions here. The key point Figure S1 addresses is whether auditory responses are generally enhanced by running (as they are e.g. in visual cortex) – the answer, on average, is no. We have tried emphasizing this more, but it changes the flow of the paper away from our main message, hence we have left the panels in the supplements. 

Regarding the 8kHz modulation, there is a general increase of the suppression of activity with increasing sound amplitude (Author response image 7 and Author response image 8). But due to the continuously varying amplitude of the stimulus, we do not have sufficient data (or do not know how to with the data we have) to address questions of adaptation. We assume there is some form of adaptation. However, either way, we don’t see how this would change our conclusions. 

Author response image 7.

Neural activity as a function of sound level in an AM open loop session. 

Author response image 8.

The average sound evoked population response of all ACx layer 2/3 neurons to 60 dB or 75 dB 8 kHz pure tones. Stimulus duration was 1 s (gray shading).

2C-D why not talk of motor modulation? Paralleling what happens in response to auditory and visual stimuli? 

This is correct, a mismatch response (we use mismatch here to operationally describe the stimulus – not the interpretation) can be described either as a prediction error (this is the interpretation) or a stimulus specific motor modulation. Note, the key here is “stimulus specific”. It is stimulus specific as there is an approximately 3x change between mismatch and playback halt (the same sensory stimulus with and without locomotion), but basically no change for sound onsets (Figure S1). Having said that, one explanation (prediction error) has predictive power (and hence is testable – see e.g. (Vasilevskaya et al., 2023) for an extensive discussion on exactly this argument for mismatch responses in visual cortex), while the other does not (a “stimulus specific” motor modulation has no predictive value or computational theory behind it and is simply a description). Thus, we choose to interpret it as a prediction error. Note, this finding does not stand in isolation and many of the testable predictions of the predictive processing interpretation have turned out to be correct (see e.g. (Keller and Mrsic-Flogel, 2018) for a review). 

Note, we try to only use the interpretation of “prediction error” when motivating why we do the experiments, and in the discussion, but not directly in the description of the results (e.g. in Figure 2).  

How does the mismatch affect the behavior of the mouse? Does it stop running? This could also influence the size of the response. 

We quantified animal behavior during audiomotor mismatches and did not find any significant acceleration or slowing down upon mismatch events. Thus, neural responses recorded during AM mismatches are unlikely to be explained by changes in animal behavior. These data have been added in Figure S2A and Figure S4A.

Figure 3. What about neurons that were positively modulated by both grating and movement? How do these neurons respond to the mismatch? 

Neurons positively modulated by both grating and movement were slightly more responsive to MM than the rest of the population, though this difference was not significant (Author response image 9). This is also visible in Figure 3G – the high VM mismatch responsive neurons are randomly distributed in regard to correlation with running speed and visual flow speed. 

Author response image 9.

Responses to visuomotor mismatches of neurons positively modulated by grating and movement and remaining of the population.

Line 176. The authors say 'Thus, in the case of a [AM + VM] mismatch both the halted visual flow and the halted sound amplitude are predicted by running speed' but the mismatch (halted flow and amplitude) is not predicted by the speed, correct? Please rephrase. 

Thank you for pointing this out – this was indeed phrased incorrectly. We have corrected this. 

How was the sound and/or visual flow interruption triggered? Did the animal have to run at a minimum speed in order for it to happen?

Sound and visual flow interruptions were triggered randomly, independent of the animal's running speed. However, for the analysis, only MM presentations during which animals were running at a speed of at least 0.3 cm/s were included. The 0.3 cm/s was simply the (arbitrary) threshold we used to determine if the mouse was running. In a completely stationary mouse a mismatch event will not have any effect (sound amplitude/visual flow speed are already at 0). This is described in the methods section.

eLife Assessment

This study provides important findings on the modulation of cortical neuronal responses to sensory stimuli by motor-driven predictive signals. The study is methodologically sound and well-designed. Solid evidence is presented for the conclusion that audiomotor mismatch responses are observed in the auditory cortex and that these are strongly modulated by crossmodal signals, though further investigation of the effects of running speed on audiomotor coupling and of sound offset effects on the observed responses would strengthen the interpretation of the results.

Reviewer #1 (Public review):

Summary:

The manuscript presents a short report investigating mismatch responses in the auditory cortex, following previous studies focused on visual cortex. By correlating mouse locomotion speed with acoustic feedback levels, the authors demonstrate excitatory responses in a subset of neurons to halts in expected acoustic feedback. They show a lack of responses to mismatch in he visual modality. A subset of neurons show enhanced mismatch responses when both auditory and visual modalities are coupled to the animal's locomotion.

While the study is well-designed and addresses a timely question, several concerns exist regarding the quantification of animal behavior, potential alternative explanations for recorded signals, correlation between excitatory responses and animal velocity, discrepancies in reported values, and clarity regarding the identity of certain neurons.

Strengths:

(1) Well-designed study addressing a timely question in the field.<br /> (2) Successful transition from previous work focused on visual cortex to auditory cortex, demonstrating generic principles in mismatch responses.<br /> (3) Correlation between mouse locomotion speed and acoustic feedback levels provides evidence for prediction signal in the auditory cortex.<br /> (4) Coupling of visual and auditory feedback show putative multimodal integration in auditory cortex.

Weaknesses:

(1) Lack of quantification of animal behavior upon mismatches, potentially leading to alternative interpretations of recorded signals.<br /> (2) Unclear correlation between excitatory responses and animal velocity during halts, particularly in closed-loop versus playback conditions.<br /> (3) Discrepancies in reported values in a few figure panels raise questions about data consistency and interpretation.<br /> (4) Ambiguity regarding the identity of the [AM+VM] MM neurons.

Comments on revisions:

I am satisfied with all clarifications and additional analyses performed by the authors.<br /> The only concern I have is about changes in running after [AM+VM] mismatches.<br /> The authors reported that they "found no evidence of a change in running speed or pupil diameter following [AM + VM] mismatch (Figures S5A)" (line 197).<br /> Nevertheless, it seems that there is a clear increase in running speed for the [AM+VM] condition (S5A). Could this be more specifically quantified? I am concerned that part of the [AM+VM] could stem from this change in running behavior. Could one factor out the running contribution?

Reviewer #2 (Public review):

In this study, Solyga and Keller use multimodal closed-loop paradigms in conjunction with multiphoton imaging of cortical responses to assess whether and how sensorimotor prediction errors in one modality influence the computation of prediction errors in another modality. Their work addresses an important open question pertaining to the relevance of non-hierarchical (lateral cortico-cortical) interactions in predictive processing within the neocortex.

Specifically, they monitor GCaMP6f responses of layer 2/3 neurons in the auditory cortex of head-fixed mice engaged in VR paradigms where running is coupled to auditory, visual, or audio-visual sensory feedback. The authors find strong auditory and motor responses in the auditory cortex, as well as weak responses to visual stimuli. Further, in agreement with previous work, they find that the auditory cortex responds to audiomotor mismatches in a manner similar to that observed in visual cortex for visuomotor mismatches. Most importantly, while visuomotor mismatches by themselves do not trigger significant responses in the auditory cortex, simultaneous coupling of audio-visual inputs to movement non-linearly enhances mismatch responses in the auditory cortex.

Their results thus suggest that prediction errors within a given sensory modality are non-trivially influenced by prediction errors from another modality. These findings are novel, interesting, and important, especially in the context of understanding the role of lateral cortico-cortical interactions and in outlining predictive processing as a general theory of cortical function.

Comments on revisions:

The authors thoroughly addressed the concerns raised. In my opinion, this has substantially strengthened the manuscript, enabling much clearer interpretation of the results reported. I commend the authors for the response to review. Overall, I find the experiments elegantly designed, and the results robust, providing compelling evidence for non-hierarchical interactions across neocortical areas and more specifically for the exchange of sensorimotor prediction error signals across modalities.

Reviewer #3 (Public review):

This study explores sensory prediction errors in sensory cortex. It focuses on the question of how these signals are shaped by non-hierarchical interactions, specifically multimodal signals arising from same level cortical areas. The authors used 2-photon imaging of mouse auditory cortex in head-fixed mice that were presented with sounds and/or visual stimuli while moving on a ball. First, responses to pure tones, visual stimuli and movement onset were characterized. Then, the authors made the running speed of the mouse predictive of sound intensity and/or visual flow (closed loop). Mismatches were created through the interruption of sound and/or visual flow for 1 second, disrupting the expected sensory signal. As a control, sensory stimuli recorded during the close loop phase were presented again decoupled from the movement (open loop). The authors suggest that auditory responses to the unpredicted interruption of the sound, which affected neither running speed nor pupil size, reflect mismatch responses. That these mismatch responses were enhanced when the visual flow was congruently interrupted, indicates cross-modal influence of prediction error signals.

This study's strengths are the relevance of the question and the design of the experiment. The authors are experts in the techniques used. The analysis explores neither the full power of the experimental design nor the population activity recorded with 2-photon, leaving open the question of to what extend what the authors call mismatch responses are not sensory responses to sound interruption (offset responses). The auditory system is sensitive to transitions and indeed responses to the interruption of the sound are similar in quality, if not quantity, in the predictive and the control situation.

Comments on revisions:

The incorporation of the analysis of the animal's running speed and the pupil size upon sound interruption improves the interpretation of the data. The authors can now conclude that responses to the mismatch are not due to behavioral effects.<br /> The issue of the relationship between mismatch responses and offset responses remains uncommented. The auditory system is sensitive to transitions, also to silence. See the work of the Linden or the Barkat labs (including the work of the first author of this manuscript) on offset responses, and also that of the Mesgarani lab (Khalighinejad et al., 2019) on responses to transitions 'to clean' (Figure 1c) in human auditory cortex. Offset responses, as the first author knows well, are modulated by intensity and stimulus length (after adaptation?). That responses to the interruption of the sound are similar in quality, if not quantity, in the closed and open loop conditions suggest that offset response might modulate the mismatch response. A mismatch response that reflects a break in predictability would presumably be less modulated by the exact details of the sensory input than an offset response. Therefore, what is the relationship between the mismatch response and the mean sound amplitude prior to the sound interruption (for example during the preceding 1 second)? And between the mismatch response and the mean firing rate over the same period?<br /> Finally, how do visual stimuli modulate sound responses in the absence of a mismatch? Is the multimodal response potentiation specific to a mismatch?

Author response:

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

(1) We agreed that there was insufficient evidence for the authors' conclusion that Myc-overexpressing clones lacking Fmi become losers. We request that the authors change the text to discuss that suppression of Myc clone growth through Fmi depletion is reminiscent of a cell acquiring loser status, although at this point in the manuscript there is no clear demonstration whether this is mostly driven by growth suppression and/or an increase in apoptosis.

We agree that at the point in the manuscript where we have only described the clone sizes, one cannot make firm conclusions about competition, so we have changed the language to reflect this. We argue that after showing our apoptosis data, those conclusions become firm. Please see the more lengthy responses to reviewers below.

(2) We agreed that the apoptosis assay, data and interpretation need to be improved. The graphs in Fig. 4O and P should be better discussed in the text and in the legend. Additionally, the graphs are lacking the red lines that are written in the text.

We regret that we did not adequately explain the data displayed in these two graphs. Supercompetition tends to cause apoptosis in both winners and losers, with the ratio between WT and super-competitor cells being critical in deciding the outcome of competition. We wanted to represent this visually but failed to properly explain our analysis. We have rewritten the figure legend and our discussion in the main text, hopefully making it clearer. 

Public Reviews:

Reviewer #1 (Public review):

Summary:

This paper is focused on the role of Cadherin Flamingo (Fmi) in cell competition in developing Drosophila tissues. A primary genetic tool is monitoring tissue overgrowths caused by making clones in the eye disc that expression activated Ras (RasV12) and that are depleted for the polarity gene scribble (scrib). The main system that they use is ey-flp, which make continuous clones in the developing eye-antennal disc beginning at the earliest stages of disc development. It should be noted that RasV12, scrib-i (or lgl-i) clones only lead to tumors/overgrowths when generated by continuous clones, which presumably creates a privileged environment that insulates them from competition. Discrete (hs-flp) RasV12, lgl-i clones are in fact out-competed (PMID: 20679206), which is something to bear in mind. They assess the role of fmi in several kinds of winners, and their data support the conclusion that fmi is required for winner status. However, they make the claim that loss of fmi from Myc winners converts them to losers, and the data supporting this conclusion is not compelling.

Strengths:

Fmi has been studied for its role in planar cell polarity, and its potential role in competition is interesting.

Weaknesses:

I have read the revised manuscript and have found issues that need to be resolved. The biggest concern is the overstatement of the results that loss of fmi from Myc-overexpressing clones turns them into losers. This is not shown in a compelling manner in the revised manuscript and the authors need to tone down their language or perform more experiments to support their claims. Additionally, the data about apoptosis is not sufficiently explained.

We take issue with this reviewer’s framing of their criticism. First, the reviewer is selectively reporting the results published in PMID: 20679206. They correctly state that those authors show that small discreet clones of RasV12 lgl are eliminated (Fig. 3B), but they omit the fact that the authors also show that larger RasV12 lgl clones induce apoptosis in the surrounding wild type cells, and therefore behave as winners (Fig. 3C). Hence, the size of the clone appears to determine its winner/loser status. Of course, lgl is not scrib, and it is not a certainty that they would behave similarly, but they also show that large RasV12 scrib clones induce considerable apoptosis of the neighboring wild type cells. 

The reviewer then discusses “continuous” clones induced by ey-flp, as we use in our manuscript. Here, the term “continuous” is probably misleading; because ey is expressed ubiquitously in the disc from early in development, it is most likely the case that the majority of cells have flipped relatively early, resulting in ~half the cells becoming clone and the other ~half twin spot. The clone cells then likely fuse to make larger clones. We show that ey-flp induced RasV12 scrib clones also behave as winners. It is logical to conclude that this is because they are large. The reviewer talks about “a privileged environment that insulates them from competition,” but if they were insulated from competition, how could they become winners? Because they occupy more territory than the wild type cells, and because they induce apoptosis in the wild type neighbors, they are winners. 

Having shown that ey-flp induced RasV12 scrib clones behave as winners, we then remove Fmi from these clones, and show that they behave as losers by the same criteria: they occupy less area than the wild type cells (our Fig. 1 and Fig. 1 Supp 2), and they induce apoptosis in the wild type cells (our Fig 4A-H). 

With respect to the comment about additional experiments are needed to support the claim that loss of Fmi from Myc winners converts them to losers, we’re not sure what additional data the reviewer would want. As for the tumor clones, we show that >>Myc clones get bigger than the twin control clones (Fig. 2), and we measure similar low levels of apoptosis in each (Fig. 4I-K, O). In contrast >>Myc fmi clones are out-grown by wild type clones, and apoptosis is higher in the >>Myc fmi clones than in the wild type clones (Fig. 4L-N, P-S). We therefore believe it is correct to say that >>Myc clones become losers when Fmi is removed.

In additional comments, the reviewer takes issue with using winner and loser language at the point in the manuscript where we have only shown the clone sizes but not yet the apoptosis data, and about this we agree. We have changed the language accordingly. 

Re explanation of the apoptosis data, see the response to reviewer #3.

Reviewer #2 (Public review):

Summary:

In this manuscript, Bosch et al. reveal Flamingo (Fmi), a planar cell polarity (PCP) protein, is essential for maintaining 'winner' cells in cell competition, using Drosophila imaginal epithelia as a model. They argue that tumor growth induced by scrib-RNAi and RasV12 competition is slowed by Fmi depletion. This effect is unique to Fmi, not seen with other PCP proteins. Additional cell competition models are applied to further confirm Fmi's role in 'winner' cells. The authors also show that Fmi's role in cell competition is separate from its function in PCP formation.

Strengths:

(1) The identification of Fmi as a potential regulator of cell competition under various conditions is interesting.

(2) The authors demonstrate that the involvement of Fmi in cell competition is distinct from its role in planar cell polarity (PCP) development.

Weaknesses:

(1) The authors provide a superficial description of the related phenotypes, lacking a mechanistic understanding of how Fmi regulates cell competition. While induction of apoptosis and JNK activation are commonly observed outcomes in various cell competition conditions, it is crucial to determine the specific mechanisms through which they are induced in fmi-depleted clones. Furthermore, it is recommended that the authors utilize the power of fly genetics to conduct a series of genetic epistasis analyses.

We agree that it is desirable to have a mechanistic understanding of Fmi’s role in competition, but that is beyond the scope of this manuscript. Here, our goal is to report the phenomenon. We understand and share with the reviewer the interest in better understanding the relationship between Fmi and JNK signaling in competition. The role of JNK in competition, tumorigenesis and cell death is infamously complex. In some preliminary experiments, we explored some epistasis experiments, but these were inconclusive so we elected to not report them here. In the future, we will continue with additional analyses to gain a better understanding of the mechanism by which Fmi affects competition.

Reviewer #3 (Public review):

Summary:

In this manuscript, Bosch and colleagues describe an unexpected function of Flamingo, a core component of the planar cell polarity pathway, in cell competition in Drosophila wing and eye disc. While Flamingo depletion has no impact on tumour growth (upon induction of Ras and depletion of Scribble throughout the eye disc), and no impact when depleted in WT cells, it specifically tunes down winner clone expansion in various genetic contexts, including the overexpression of Myc, the combination of Scribble depletion with activation of Ras in clones or the early clonal depletion of Scribble in eye disc. Flamingo depletion reduces proliferation rate and increases the rate of apoptosis in the winner clones, hence reducing their competitiveness up to forcing their full elimination (hence becoming now "loser"). This function of Flamingo in cell competition is specific of Flamingo as it cannot be recapitulated with other components of the PCP pathway, does not rely on interaction of Flamingo in trans, nor on the presence of its cadherin domain. Thus, this function is likely to rely on a non-canonical function of Flamingo which may rely on downstream GPCR signaling.

This unexpected function of Flamingo is by itself very interesting. In the framework of cell competition, these results are also important as they describe, to my knowledge, one of the only genetic conditions that specifically affect the winner cells without any impact when depleted in the loser cells. Moreover, Flamingo do not just suppress the competitive advantage of winner clones, but even turn them in putative losers. This specificity, while not clearly understood at this stage, opens a lot of exciting mechanistic questions, but also a very interesting long term avenue for therapeutic purpose as targeting Flamingo should then affect very specifically the putative winner/oncogenic clones without any impact in WT cells.

The data and the demonstration are very clean and compelling, with all the appropriate controls, proper quantifications and backed-up by observations in various tissues and genetic backgrounds. I don't see any weakness in the demonstration and all the points raised and claimed by the authors are all very well substantiated by the data. As such, I don't have any suggestions to reinforce the demonstration.

While not necessary for the demonstration, documenting the subcellular localisation and levels of Flamingo in these different competition scenarios may have been relevant and provide some hints on a putative mechanism (specifically by comparing its localisation in winner and loser cells).

While we did not perform a thorough analysis, our current revision of the manuscript shows Fmi staining results that do not support a change in subcellular localization of Fmi. In our images, Fmi seemed to localize similarly along the winner-loser clone boundaries, and inside and outside the clones. We cannot rule out that a subtle change in localization is taking place that could perhaps be detected with higher resolution imaging.

Also, on a more interpretative note, the absence of impact of Flamingo depletion on JNK activation does not exclude some interesting genetic interactions. JNK output can be very contextual (for instance depending on Hippo pathway status), and it would be interesting in the future to check if Flamingo depletion could somehow alter the effect of JNK in the winner cells and promote downstream activation of apoptosis (which might normally be suppressed). It would be interesting to check if Flamingo depletion could have an impact in other contexts involving JNK activation or upon mild activation of JNK in clones.

See our comment to Reviewer 2 regarding JNK.

Strengths:

A clean and compelling demonstration of the function of Flamingo in winner cells during cell competition

One of the rare genetic conditions that affects very specifically winner cells without any impact in losers, and then can completely switch the outcome of competition (which opens an interesting therapeutic perspective on the long term) Weaknesses:

The mechanistic understanding obviously remains quite limited at this stage especially since the signaling does not go through the PCP pathway.

We agree that in the future, it will be desirable to gain a mechanistic understanding of Fmi’s role in competition.

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

I have read the revised manuscript and have found issues that need to be resolved. The biggest concern is the overstatement of the results that loss of fmi from Myc-overexpressing clones turns them into losers. This is not shown in a compelling manner in the revised manuscript and the authors need to tone down their language or perform more experiments to support their claims.

(1) I do not agree with the language used by the authors last paragraph of p. 4 stating loss of fmi from Myc supercompetitors (Fig. 2) makes them losers. At this point in the paper, they only use clone size as a readout. By definition, losers in imaginal discs die by apoptosis, which is not measured in this figure. As such, the authors do not prove that fmi-mutant Myc over-expressing clones are now losers at this point in the manuscript. The authors should discuss this in the results section regarding Fig. 2.

We have modified the language in text and figure legend to acknowledge that the clone size data alone do not demonstrate competition.

(2) Related to point #1, I do not agree with the language in the legend of Fig. 2H that the graph is measuring "supercompetition". They are only measuring clone ratios, not apoptosis. Growing to a smaller size does not make a clone have loser status without also assessing cell death.

(a) I suggest that the authors remove the sentence "A ratio over 0 indicates supercompetition of nGFP+ clones, and below 0 indicates nGFP+ cells are losers." in the legend to Fig. 2H. Instead, they should describe the assay in times of clone ratios.

The reviewer raises a valid point, as at this point in the manuscript we did not quantify cell death and proliferation. However, based on decades of knowledge of supercompetiton, Myc clones are classified as super-competitors in every instance they’ve been studied. (Myc clones show apoptosis when competing with WT cells, while at the same time they eliminate WT neighbors by apoptosis to become winners. Their faster proliferation rate may be what ultimately makes them winners.) We changed the language to address this distinction. 

(3) In Fig. 4, they do attempt to monitor apoptosis, which is the fate of bona fide losers in imaginal tissue. However, I have several concerns about these data (panels 4I-K, O and P have been added to the revised manuscript.)

(a) In Fig. 4I-K, why is there no death of WT cells which would be expected based on de la Cova Cell 2004? The authors need to comment on this.

(b) Cell death should also be observed in the Myc over-expressing clones but none is seen in this disc (see de la Cova 2004 and PMID: 18257071 Fig. 4). The authors need to comment on this.

We do not understand why the reviewer raises these two points. We see some cell death in >Myc eye discs both in winners and losers, as displayed in the graph. In our hands, the levels were on average very low. The example shown is representative of the analysis and shows apoptosis both in WT and >Myc cells, highlighted by the arrows in 4J. We added a mention to the arrows in the figure legend to make it clearer. In the main text, we already compared our observations to the same publication the reviewer mentions (De la Cova 2004). 

(c) The data in panel 4O is not explained sufficiently in the legend or results section. What do the lines between the data points in the left side of the panel mean? Why is there a bunch of clustered data points in the right part of the Fig. 4O, when two different genotypes are listed below? I would have expected two clusters of points. The authors need to comment on this.

We intended to convey as much information as possible in an informative manner in these graphs, and we regret not explaining better the analysis shown. We modified the legends for the apoptosis analysis to better explain the displayed data.

(d) What is the sample size (n) for the genotypes listed in this figure? The authors need to comment on this and explicitly list the sample size in the legend.

We added the n for both conditions to the figure. 

(e) In panels 4L-N, why is the death occurring in the apparent center of the fmiE59>>Myc clone. If these clones are truly losers as the authors claim, then apoptosis should be seen at the boundaries between the fmiE59>>Myc clone and the WT clones. The results in this figure are not compelling, yet this is the critical piece of data to support their claim that fmiE59>>Myc clone are losers. The authors need to comment on this.

The majority of cell death in this example is observed 1-3 cells away from the clone boundary. In some cases, we observe cell death farther from the boundary, but those cells were not counted in our analyses. As described in our methods, we only considered for the analysis cells at the clone boundary or in the vicinity, as those are the ones that most probably have apoptosis triggered by the neighboring clone.

(f) There is no red line in Fig. 4O and 4P, in contrast to what is written in the legend in the revised manuscript. This should be corrected.

We thank the reviewer for catching the error about the line. We have now simplified the graph by removing the line at Y=0 and just leave one dashed line, representing the mean difference between WT and >>Myc cells.

(4) On p. 10, the reference Harvey and Tapon 2007 to support hpo-/- supercompetitor status is incorrect. The references are Ziosi 2010 and Neto-Silva 2010. This should be changed.

We thank the reviewer for the correction. While the review we provided discusses the role of the Hpo pathway in proliferation and cancer, it does not discuss competition. The reference we intended to include here was Ziosi 2010. We now cite both in the revised manuscript.

(5) The legend for Fig. 3A-H is missing from the revised manuscript. This needs to be added.

This was likely a copy-edit glitch. The missing parts of the legend have been restored.

(6) Material and methods is missing details on the hs-induced clones. The authors need to specifically state when the clones were generated and when they were analyzed in hours after egg laying.

The timing of the heat-shock and analysis was described in the methods: “Heat-shock was performed on late first instar and early second instar larvae, 48 hrs after egg laying (AEL). Vials were kept at 25ºC after heat-shock until larvae were dissected”. And additionally, in the dissection methods: “Third instar wandering larvae (120 hrs AEL) were dissected…” We have included in this revision the length of the heat-shock (15 min). 

I have read the rebuttal and some of my concerns are not sufficiently addressed.

(8) I raised the point of continuously-generated clones becoming large enough to evade competition, and I disagree with the authors' reply. I think that competition of RasV12, scrib (or lgl) competition largely depends the size of the clone, which is de facto larger when generated by continuous expression of flp (such as eyeless or tubulin promoters used in this study). I think that at that point, we are at an impasse with respect to this issue, but I wanted to register my disagreement for the record. Related to this, one possible reason for the fragmentation of the fmimutant Myc overexpressing clones in the wing disc is because they were not continuously generated and hence did not merge with other clones.

Please see the discussion above in the public comments. We remain unclear about what, exactly, the reviewer disagrees. As stated above, we think they are correct that the size of the clone is critical in determining winner vs loser status.

Reviewer #2 (Recommendations for the authors):

Although the authors have addressed some of my concerns, I still feel that a detailed mechanistic understanding is essential. I hope the authors will conduct additional experiments to solve this issue.

We also consider the mechanism of interest and will pursue this in the future. To test our hypotheses we require a set of genetic mutants that are still in the making that will help us dissect the function and potential partners of Fmi, and we hope to have these results in a future publication.

Reviewer #3 (Recommendations for the authors):

- There is no clear demonstration that the relative decrease of clone size in UASMyc/Fmi mutant is mostly driven by either a context dependant suppression of growth and/or an increase of apoptosis (the latter being the more classic feature of loser phenotype).

We believe that it is driven by both, and refrain from making assumptions about the magnitude of contribution from each. This question is something that we will be interested to explore in the future.

The distribution of cell death in Fmi/UAS-Myc mutant is somehow surprising and may not fit with most of the competition scenarios where death is mostly restricted to clone periphery (although this may be quite variable and would require much more quantification to be clear).

While we observe some cell death far from clone boundaries, most of the dying cells are a few cells away from a clone boundary. In other publications quantifying cell death, examples of cell death farther from the boundary are not rare (See for example Moreno and Basler 2004 Fig 6, De la Cova et al. Fig 2, Meyer et al 2014 Fig 2). We did not count cells dying far from clone boundaries in our analysis.

I just noticed a few mistakes in the legend :

Figure 3M legend is missing (it would be useful to know at which stage the quantification is performed)

Another reviewer brought to our attention the problems with Fig 3 legend. We restored the missing parts.

It would be good to give an estimate of the number of larvae observed when showing the representative cases in Figure 1 .

This is a good point. We now include these numbers in the figure legend.

eLife Assessment

This study investigates the role of the Cadherin Flamingo (Fmi) in cell competition in developing tissues in Drosophila melanogaster. The findings are valuable in that they show that Fmi is required in winning cells in several competitive contexts. The evidence supporting the conclusions is solid, as the authors identify Fmi as a potential new regulator of cell competition, however, they don't delve into a mechanistic understanding of how this occurs.

Reviewer #1 (Public review):

Summary:

This paper is focused on the role of Cadherin Flamingo (Fmi) in cell competition in developing Drosophila tissues. A primary genetic tool is monitoring tissue overgrowths caused by making clones in the eye disc that expression activated Ras (RasV12) and that are depleted for the polarity gene scribble (scrib). The main system that they use is ey-flp, which make continuous clones in the developing eye-antennal disc beginning at the earliest stages of disc development. It should be noted that RasV12, scrib-i (or lgl-i) clones only lead to tumors/overgrowths when generated by continuous clones, which presumably creates a privileged environment that insulates them from competition. Discrete (hs-flp) RasV12, lgl-i clones are in fact out-competed (PMID: 20679206), which is something to bear in mind. They assess the role of fmi in several kinds of winners, and their data support the conclusion that fmi is required for winner status. However, they make the claim that loss of fmi from Myc winners converts them to losers, and the data supporting this conclusion is not compelling.

Strengths:

Fmi has been studied for its role in planar cell polarity, and its potential role in competition is interesting.

Reviewer #2 (Public review):

Summary:

In this manuscript, Bosch et al. reveal Flamingo (Fmi), a planar cell polarity (PCP) protein, is essential for maintaining 'winner' cells in cell competition, using Drosophila imaginal epithelia as a model. They argue that tumor growth induced by scrib-RNAi and RasV12 competition is slowed by Fmi depletion. This effect is unique to Fmi, not seen with other PCP proteins. Additional cell competition models are applied to further confirm Fmi's role in 'winner' cells. The authors also show that Fmi's role in cell competition is separate from its function in PCP formation.

Strengths:

(1) The identification of Fmi as a potential regulator of cell competition under various conditions is interesting.<br /> (2) The authors demonstrate that the involvement of Fmi in cell competition is distinct from its role in planar cell polarity (PCP) development.

Reviewer #3 (Public review):

Summary:

In this manuscript, Bosch and colleagues describe an unexpected function of Flamingo, a core component of the planar cell polarity pathway, in cell competition in Drosophila wing and eye disc. While Flamingo depletion has no impact on tumour growth (upon induction of Ras and depletion of Scribble throughout the eye disc), and no impact when depleted in WT cells, it specifically tunes down winner clone expansion in various genetic contexts, including the overexpression of Myc, the combination of Scribble depletion with activation of Ras in clones or the early clonal depletion of Scribble in eye disc. Flamingo depletion reduces proliferation rate and increases the rate of apoptosis in the winner clones, hence reducing their competitiveness up to forcing their full elimination (hence becoming now "loser"). This function of Flamingo in cell competition is specific of Flamingo as it cannot be recapitulated with other components of the PCP pathway, does not rely on interaction of Flamingo in trans, nor on the presence of its cadherin domain. Thus, this function is likely to rely on a non-canonical function of Flamingo which may rely on downstream GPCR signaling.

This unexpected function of Flamingo is by itself very interesting. In the framework of cell competition, these results are also important as they describe, to my knowledge, one of the only genetic conditions that specifically affect the winner cells without any impact when depleted in the loser cells. Moreover, Flamingo do not just suppress the competitive advantage of winner clones, but even turn them in putative losers. This specificity, while not clearly understood at this stage, opens a lot of exciting mechanistic questions, but also a very interesting long term avenue for therapeutic purpose as targeting Flamingo should then affect very specifically the putative winner/oncogenic clones without any impact in WT cells.

The data and the demonstration are very clean and compelling, with all the appropriate controls, proper quantifications and backed-up by observations in various tissues and genetic backgrounds. I don't see any weakness in the demonstration and all the points raised and claimed by the authors are all very well substantiated by the data. As such, I don't have any suggestions to reinforce the demonstration.

While not necessary for the demonstration, documenting the subcellular localisation and levels of Flamingo in these different competition scenarios may have been relevant and provide some hints on a putative mechanism (specifically by comparing its localisation in winner and loser cells).

Also, on a more interpretative note, the absence of impact of Flamingo depletion on JNK activation does not exclude some interesting genetic interactions. JNK output can be very contextual (for instance depending on Hippo pathway status), and it would be interesting in the future to check if Flamingo depletion could somehow alter the effect of JNK in the winner cells and promote downstream activation of apoptosis (which might normally be suppressed). It would be interesting to check if Flamingo depletion could have an impact in other contexts involving JNK activation or upon mild activation of JNK in clones.

Strengths:

- A clean and compelling demonstration of the function of Flamingo in winner cells during cell competition

- One of the rare genetic conditions that affects very specifically winner cells without any impact in losers, and then can completely switch the outcome of competition (which opens an interesting therapeutic perspective on the long term)

eLife Assessment

This fundamental study advances substantially our understanding of sound encoding at synapses between single inner hair cells of the mouse cochlea and spiral ganglion neurons. Dual patch-clamp recordings-a technical tour-de force-and careful data analysis provide compelling evidence that the functional heterogeneity of these synapses contributes to the diversity of spontaneous and sound-evoked firing by the neurons. The work will be of broad interest to scientists in the field of auditory neuroscience.

Reviewer #1 (Public review):

Summary:

Tobón and Moser reveal a remarkable amount of presynaptic diversity in the fundamental Ca dependent exocytosis of synaptic vesicles at the afferent fiber bouton synapse onto the pilar or mediolar sides of single inner hair cells of mice. These are landmark findings with profound implications for understanding acoustic signal encoding and presynaptic mechanisms of synaptic diversity at inner hair cell ribbon synapses. The paper will have an immediate and long-lasting impact in the field of auditory neuroscience.

Main findings: 1) Synaptic delays and jitter of masker responses are significantly shorter (synaptic delay: 1.19 ms) at high SR fibers (pilar) than at low SR fibers (mediolar; 2.57 ms). 2) Masked evoked EPSC are significantly larger in high SR than in low SR. 3) Quantal content and RRP size are 14 vesicles in both high and low SR fibers. 4) Depression is faster in high SR synapses suggesting they have a higher release probability and tighter Ca nanodomain coupling to docked vesicles. 5) Recovery of master-EPSCs from depletion is similar for high and low SR synapses, although there is a slightly faster rate for low SR synapses that have bigger synaptic ribbons, which is very interesting. 6) High SR synapses had larger and more compact (monophasic) sEPSCs, well suited to trigger rapidly and faithfully spikes. 7) High SR synapses exhibit lower voltage (~sound pressure in vivo) dependent thresholds of exocytosis.

Great care was taken to use physiological external pH buffers and physiological external Ca concentrations. Paired recordings were also performed at higher temperatures with IHCs at physiological resting membrane potentials and in more mature animals than previously done for paired recordings. This is extremely challenging because it becomes increasingly difficult to visualize bouton terminals when myelination becomes more prominent in the cochlear afferents. In addition, perforated patch recordings were used in the IHC to preserve its intracellular milieu intact and thus extend the viability of the IHCs. The experiments are tour-de-force and reveal several novel aspects of IHC ribbon synapses. The data set is rich and extensive. The analysis is detailed and compelling.

Reviewer #2 (Public review):

Summary:

The study by Jaime-Tobon & Moser is a truly major effort to bridge the gap between classical observations on how auditory neurons respond to sounds and the synaptic basis of these phenomena. The so-called spiral ganglion neurons (SGNs) are the primary auditory neurons connecting the brain with hair cells in the cochlea. They all respond to sounds increasing their firing rates, but also present multiple heterogeneities. For instance, some present a low threshold to sound intensity, whereas others have high threshold. This property inversely correlates with the spontaneous rate, i.e., the rate at which each neuron fires in the absence of any acoustic input. These characteristics, along with others, have been studied by many reports over years. However, the mechanisms that allow the hair cells-SGN synapses to drive these behaviors are not fully understood.

The level of experimental complexity described in this manuscript is unparalleled, producing data that is hardly found elsewhere. The authors provide strong proof for heterogeneity in transmitter release thresholds at individual synapses and they do so in an extremely complex experimental settings. In addition, the authors found other specific differences such as in synaptic latency and max EPSCs. A reasonable effort is put in bridging these observations with those extensively reported in in vivo SGNs recordings. Similarities are many and differences are not particularly worrying as experimental conditions cannot be perfectly matched, despite the authors' efforts in minimizing them.

Reviewer #3 (Public review):

Summary:

The manuscript by Jaime Tobon and Moser uses patch-clamp electrophysiology in cochlear preparations to probe the pre- and post-synaptic specializations that give rise to diverse activity of spiral ganglion afferent neurons (SGN). The experiments are quite an achievement! They use paired recordings from pre-synaptic cochlear inner hair cells (IHC) that allow precise control of voltage and therefore calcium influx, with post-synaptic recordings from type I SGN boutons directly opposed to the IHC for both presynaptic control of membrane voltage and post-synaptic measurement of synaptic function with great temporal resolution.

Any of these techniques by themselves are challenging, but the authors do them in pairs, at physiological temperatures, and in hearing animals, all of which combined make these experiments a real tour de force. The data is carefully analyzed and presented, and the results are convincing. In particular, the authors demonstrate that post-synaptic features that contribute to the spontaneous rate (SR) of predominantly monophasic post-synaptic currents (PSCs), shorter EPSC latency, and higher PSC rates are directly paired with pre-synaptic features such as a lower IHC voltage activation and tighter calcium channel coupling for release to give a higher probability of release and subsequent increase in synaptic depression. Importantly, IHCs paired with Low and High SR afferent fibers had the same total calcium currents, indicating that the same IHC can connect to both low and high SR fibers. These fibers also followed expected organizational patterns, with high SR fibers primarily contacting the pillar IHC face and low SR fibers primarily contacting the modiolar face. The authors also use in vivo-like stimulation paradigms to show different RRP and release dynamics that are similar to results from SGN in vivo recordings. Overall, this work systematically examines many features giving rise to specializations and diversity of SGN neurons.

Author response:

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

Reviewer #2 (Recommendations for the authors): 

Discussion, page 28. The argument that the authors put forward justifying the (small) size of the spontaneous EPSCs seems reasonable. Nonetheless, it would be good to have an amplitude distribution constructed with voltage-evoked EPSCs to compare with that of spontaneous EPSCs. Not the large initial EPSC, obtained upon IHC depolarization but rather EPSCs occurring later during the longer pulses (figure 4). The authors made the claim that upon IHC depolarization, EPSCs sizes increased, but this is not backed with data. 

Following the reviewer recommendation, we have analyzed the voltage-evoked EPSCs occurring during the last 20 ms of the Masker stimulus. We compared the cumulative distribution of the amplitude of these eEPSCs to the cumulative distribution of the amplitude of the sEPSCs (Figure 1-figure supplement 1, panel G) from the same synapses. The two distributions are significantly different (p < 0.0001, Kolmogorov-Smirnov test), with evoked EPSCs having larger amplitudes (average sEPSC amplitude of -97.28 ± 2.22 pA [median 82.10 pA] vs average eEPSC amplitude of 135.8 ± 3.24 pA [median 120.0 pA]).

eLife Assessment

This useful study employs AlphaFold2 to predict interactions among 20 nuage proteins, identifying five novel interaction candidates, three of which are validated experimentally through co-immunoprecipitation. Expanding the analysis to 430 oogenesis-related proteins and screening ~12,000 Drosophila proteins for interactions with Piwi, the study identifies 164 potential binding partners, demonstrating how computational predictions can streamline experimental validation. This study provides a solid basis for further investigations into eukaryotic protein interaction networks.

Reviewer #1 (Public review):

Summary:

The study investigates protein-protein interactions (PPIs) within the nuage, a germline-specific organelle essential for piRNA biogenesis in Drosophila melanogaster, using AlphaFold2 to predict interactions among 20 nuage-localizing proteins. The authors identify five novel interaction candidates and experimentally validate three of them, including Spindle-E and Squash, through co-immunoprecipitation assays. They confirm the functional significance of these interactions by disrupting salt bridges at the Spn-E_Squ interface. The study further expands its scope to analyze approximately 430 oogenesis-related proteins, validating three additional interaction pairs. A comprehensive screen of around 12,000 Drosophila proteins for interactions with the key piRNA pathway player, Piwi, identifies 164 potential binding partners. Overall, the research demonstrates that in silico approaches using AlphaFold2 can link bioinformatics predictions with experimental validation, streamlining the identification of novel protein interactions and reducing the reliance on extensive experimental efforts. The manuscript is commendably clear and easy to follow; however, areas for improvement should be addressed to enhance its clarity and rigor.

Major Concerns:

(1) While AlphaFold2 was developed and trained primarily for predicting protein structures and their interactions, applying it to predict protein-protein interactions is an extrapolation of its intended use. This introduces several important considerations and risks. First, it assumes that AlphaFold's accuracy in structure prediction extends to interactions, despite not being explicitly trained for this task. Additionally, the assumption that high-scoring models with structural complementarity imply biologically relevant interactions is not always valid. Experimental validation is essential to address these uncertainties, as over-reliance on computational predictions without such validation can lead to false positives and inaccurate conclusions. The authors should expand on the assumptions, limitations, and risks associated with using AlphaFold2 for predicting protein-protein interactions.

(2) The authors experimentally validated three interactions, out of five predicted interactions, using co-immunoprecipitation (co-IP). They attributed the lack of validation for the other two predictions to the limitations of the co-IP method. However, further clarification on the potential limitations of the co-immunoprecipitation behind the negative results would strengthen the conclusions. While co-IP is a widely used technique, it may not detect weak or transient interactions, which could explain the failure to validate some predictions. Suggesting alternative validation methods such as FRET or mass spectrometry could further substantiate the results. On the other hand, AlphaFold2 predictions are not infallible and may generate false positives, particularly when dealing with structurally plausible but biologically irrelevant interactions. By acknowledging both the potential limitations of co-IP and the possibility of false positives from AlphaFold2, the authors can provide a more balanced interpretation of their findings.

(3) In line 143, the authors state that "This approach identified 13 pairs; seven of these were already known to form complexes, confirming the effectiveness of AlphaFold2 in predicting complex formations (Table 2). The highest pcScore pair was the Zuc homodimer, possibly because AlphaFold2 had learned from Zuc homodimer's crystal structure registered in the database." While the authors mentioned the presence of the Zuc homodimer's crystal structure, they do not provide a systematic bioinformatics analysis to evaluate pairwise sequence identity or check for the presence of existing structures for all the proteins or protein pairs (or their homologs) in databases such as the Protein Data Bank (PDB) or Swiss-Model. Conducting such an analysis is critical, as it significantly impacts the novelty and reliability of AlphaFold2 predictions. For instance, high sequence identity between the query proteins could lead to high-scoring models for biologically irrelevant interactions. Including this information would strengthen the conclusions regarding the accuracy and utility of the predictions.

(4) While the manuscript successfully identifies novel protein interactions, the broader biological significance of these interactions remains underexplored. The manuscript could benefit from elaborating on how these findings may contribute to understanding the piRNA pathway and its implications on germline development, transposon repression, and oogenesis.

Reviewer #2 (Public review):

Summary:

In this paper, the authors use AlphaFold2 to identify potential binding partners of nuage localizing proteins.

Strengths:

The main strength of the paper is that the authors experimentally verify a subset of the predicted interactions.

Many studies have been performed to predict protein-protein interactions in various subsets of proteins. The interesting story here is that the authors (i) focus on an organelle that contains quite some intrinsically disordered proteins and (ii) experimentally verify some (but not all) predictions.

Weaknesses:

Identification of pairwise interactions is only a first step towards understanding complex interactions. It is pretty clear from the predictions that some (but certainly not all) of the pairs could be used to build larger complexes. AlphaFold easily handles proteins up to 4-5000 residues, so this should be possible. I suggest that the authors do this to provide more biological insights.

Another weakness is the use of a non-standard name for "ranking confidence" - the author calls it the pcScore - while the name used in AlphaFold (and many other publications) is ranking confidence.

Author response:

Public Reviews:

Reviewer #1 (Public review):

Summary:

The study investigates protein-protein interactions (PPIs) within the nuage, a germline-specific organelle essential for piRNA biogenesis in Drosophila melanogaster, using AlphaFold2 to predict interactions among 20 nuage-localizing proteins. The authors identify five novel interaction candidates and experimentally validate three of them, including Spindle-E and Squash, through co-immunoprecipitation assays. They confirm the functional significance of these interactions by disrupting salt bridges at the Spn-E_Squ interface. The study further expands its scope to analyze approximately 430 oogenesis-related proteins, validating three additional interaction pairs. A comprehensive screen of around 12,000 Drosophila proteins for interactions with the key piRNA pathway player, Piwi, identifies 164 potential binding partners. Overall, the research demonstrates that in silico approaches using AlphaFold2 can link bioinformatics predictions with experimental validation, streamlining the identification of novel protein interactions and reducing the reliance on extensive experimental efforts. The manuscript is commendably clear and easy to follow; however, areas for improvement should be addressed to enhance its clarity and rigor.

Major Concerns:

(1) While AlphaFold2 was developed and trained primarily for predicting protein structures and their interactions, applying it to predict protein-protein interactions is an extrapolation of its intended use. This introduces several important considerations and risks. First, it assumes that AlphaFold's accuracy in structure prediction extends to interactions, despite not being explicitly trained for this task. Additionally, the assumption that high-scoring models with structural complementarity imply biologically relevant interactions is not always valid. Experimental validation is essential to address these uncertainties, as over-reliance on computational predictions without such validation can lead to false positives and inaccurate conclusions. The authors should expand on the assumptions, limitations, and risks associated with using AlphaFold2 for predicting protein-protein interactions.

We appreciate the reviewer's point. The prediction of protein-protein interactions using AlphaFold2 relies on the number of conserved homologous sequences and previous conformational data. We shall add limitations and risks to the AlphaFold2 prediction method in the revised manuscript.

(2) The authors experimentally validated three interactions, out of five predicted interactions, using co-immunoprecipitation (co-IP). They attributed the lack of validation for the other two predictions to the limitations of the co-IP method. However, further clarification on the potential limitations of the co-immunoprecipitation behind the negative results would strengthen the conclusions. While co-IP is a widely used technique, it may not detect weak or transient interactions, which could explain the failure to validate some predictions. Suggesting alternative validation methods such as FRET or mass spectrometry could further substantiate the results. On the other hand, AlphaFold2 predictions are not infallible and may generate false positives, particularly when dealing with structurally plausible but biologically irrelevant interactions. By acknowledging both the potential limitations of co-IP and the possibility of false positives from AlphaFold2, the authors can provide a more balanced interpretation of their findings.

We appreciate the reviewer's point of view. We have used the co-IP method to detect interactions in this study. However, as the reviewer pointed out, it is likely that weak and transient interactions may not be detected. We plan to add a note on the detection limits of the co-IP method and the possibility that AlphaFold2 method produces false positives in the revised manuscript.

(3) In line 143, the authors state that "This approach identified 13 pairs; seven of these were already known to form complexes, confirming the effectiveness of AlphaFold2 in predicting complex formations (Table 2). The highest pcScore pair was the Zuc homodimer, possibly because AlphaFold2 had learned from Zuc homodimer's crystal structure registered in the database." While the authors mentioned the presence of the Zuc homodimer's crystal structure, they do not provide a systematic bioinformatics analysis to evaluate pairwise sequence identity or check for the presence of existing structures for all the proteins or protein pairs (or their homologs) in databases such as the Protein Data Bank (PDB) or Swiss-Model. Conducting such an analysis is critical, as it significantly impacts the novelty and reliability of AlphaFold2 predictions. For instance, high sequence identity between the query proteins could lead to high-scoring models for biologically irrelevant interactions. Including this information would strengthen the conclusions regarding the accuracy and utility of the predictions.

We appreciate the reviewer's critical point. The AlphaFold2 method generates a high confidence score when the 3D structure of the protein of interest, or of proteins with very similar sequences, is solved. We will investigate whether the proteins used in this study are included in the 3D structure database and add the information to the revised manuscript.

(4) While the manuscript successfully identifies novel protein interactions, the broader biological significance of these interactions remains underexplored. The manuscript could benefit from elaborating on how these findings may contribute to understanding the piRNA pathway and its implications on germline development, transposon repression, and oogenesis.

We plan to add to the revise manuscript the potential biological significance of the novel protein-protein interactions presented in this manuscript.

Reviewer #2 (Public review):

Summary:

In this paper, the authors use AlphaFold2 to identify potential binding partners of nuage localizing proteins.

Strengths:

The main strength of the paper is that the authors experimentally verify a subset of the predicted interactions.

Many studies have been performed to predict protein-protein interactions in various subsets of proteins. The interesting story here is that the authors (i) focus on an organelle that contains quite some intrinsically disordered proteins and (ii) experimentally verify some (but not all) predictions.

Weaknesses:

Identification of pairwise interactions is only a first step towards understanding complex interactions. It is pretty clear from the predictions that some (but certainly not all) of the pairs could be used to build larger complexes. AlphaFold easily handles proteins up to 4-5000 residues, so this should be possible. I suggest that the authors do this to provide more biological insights.

We thank the reviewer for his kind suggestions. Although dimer structure predictions were made in this manuscript, if a protein is predicted to interact with two other proteins, it is possible that three proteins could interact. We plan to add such trimer predictions to the revise manuscript.

Another weakness is the use of a non-standard name for "ranking confidence" - the author calls it the pcScore - while the name used in AlphaFold (and many other publications) is ranking confidence.

We take the reviewer’s point and will revise the text accordingly.

eLife Assessment

This study presents important findings on cold tolerance shared between hibernating and non-hibernating mammals, identifying a key molecule, GPX4, through multi-species genome-wide CRISPR screens. The evidence supporting these conclusions is compelling, combining multi-species CRISPR screening with rigorous pharmacological assays. This work will be of significant interest to biologists studying hibernation physiology and medical researchers interested in cold tolerance.

Reviewer #1 (Public review):

Summary:

Through a series of CRISPR-Cas9 screens, the GPX4 antioxidant pathway was identified as a critical suppressor of cold-induced cell death in hibernator-derived cells. Hamster BHK-21 cells exposed to repeated cold and rewarming cycles revealed five genes (Gpx4, Eefsec, Pstk, Secisbp2, and Sepsecs) as critical components of the GPX4 pathway, which protects against cold-induced ferroptosis. A second screen with continuous cold exposure confirmed the essential role of GPX4 in prolonged cold tolerance. GPX4 knockout lines exhibited complete cell death within four days of cold exposure, and pharmacological inhibition of GPX4 further increased cell death, underscoring the necessity of GPX4's catalytic activity in cold conditions.

An additional CRISPR screen in human cold-sensitive K562 cells identified 176 genes for cold survival. The GPX4 pathway was found to confer significant resistance to cold in hibernators and human cells, with GPX4 loss significantly increasing cold-induced cell death.

Comparing hamster and human GPX4, overexpression of GPX4 in human K562 cells, whether hamster or human GPX4, dramatically improved cold tolerance, while catalytically dead mutants showed no such effect. These findings suggest that GPX4 abundance is a key limiting factor for cold tolerance in human cells, and primary cell types show strong sensitivity to GPX4 loss, highlighting that differences in cold tolerance across species may be due to varying GPX4-mediated protection.

Strengths:

(1) Innovative Approach: The study employs a series of unbiased genome-wide CRISPR-Cas9 screens in both hibernator- and non-hibernator-derived cells to investigate the mechanisms controlling cellular cold tolerance. Notably, this is the first genome-scale CRISPR-Cas9 screen conducted in cells derived from a hibernator, the Syrian hamster.

(2) Identification of the GPX4 Pathway: Identifying glutathione peroxidase 4 (GPX4) as a critical suppressor of cold-induced cell death significantly contributes to the field. Recently, GPX4 was also reported as a potent regulator of cold tolerance through overexpression screening (Sone et al.) in hamsters, which further supports this finding.

(3) Improved Cold Viability Assessment: The study identifies an important technical artifact in using trypan blue to assess cell viability following cold exposure. It reveals that cells stained immediately after cold exposure retain the dye, inaccurately indicating cell death. By introducing a brief rewarming period before viability assessment, the authors significantly improve the accuracy of detecting cold-induced cell death. This refinement in methodology ensures more reliable results and sets a new standard for future research on cold stress in cells.

Weaknesses:

(1) Mechanisms Regulating GPX4 Levels: While the study highlights GPX4 levels as a major determinant of cellular cold tolerance, it does not discuss how these levels are regulated or why they differ between hibernators and non-hibernators. This omission leaves an important aspect of GPX4's role in cold tolerance unexplored.

(2) Generalizability Across Species: Although the study demonstrates the role of GPX4 in several mammalian species, it does not investigate whether this mechanism extends to other vertebrates (e.g., fish and amphibians) that also face cold challenges. This limitation could restrict the broader evolutionary claims made by the study.

(3) Variability in Cold Sensitivity Across Human Cell Lines: The study observes significant variability in cold tolerance among different human cell lines but does not explain these differences clearly. This leaves a key aspect of human cell cold sensitivity insufficiently addressed.

Reviewer #2 (Public review):

Summary:

Lam et al., present a very intriguing whole genome CRISPR screen in Syrian Hamster cells as well as K562 cells to identify key genes involved in hypothermia-rewarming tolerance. Survival screens were performed by exposing cells to 4C in a cooled CO2 incubator followed by a rewarming period of 30 minutes prior to survival analysis. In this paradigm, Syrian hamster-derived cell lines exhibit more robust survival than human cell lines (BHK-21 and HaK vs HT1080, HeLa, RPE1, and K562). A genome-wide Syrian hamster CRISPR library was created targeting all annotated genes with 10 guides/gene. LV transduction of the library was performed in BHK-21 cells and the survival screen procedures involved 3 cycles of 4C cold exposure x4 days followed by 2 days of re-warming.

When compared to controls maintained at 37C, 9 genes were required for BHK-21 survival of cold cycling conditions and 5 of these 9 are known components of the GPX4 antioxidant pathway. GPX4 KO BHK-21 cells had reduced cell growth at 37C and profoundly worse cold tolerance which could be reduced by GPX4 expression. GPX4 inhibitors also reduced survival in cold. CRISPR KO screens and GPX4 KO in K562 cells revealed comparable results (though intriguingly glutathione biosynthesis genes were more critical to K562 cells than BHK-21 cells). Human or Syrian hamster GPX4 overexpression improved cold tolerance.

Strengths:

This is a very nicely written paper that clearly communicates in figures and text complicated experimental manipulations and in vitro genetic screening and cell survival data. The focus on GPX4 is interesting and relatively novel. The converging pharmacologic, loss-of-function, and gain-of-function experiments are also a strength.

Weaknesses:

A recently published article (Reference 43, Sone et al.) also independently explored the role of GPX4 in Syrian hamster cold tolerance through gain-of-function screening. Further exploration of the GPX4 species-specific mechanisms would be of great interest, but this is considered a minor weakness given the already very comprehensive and compelling data presented.

Reviewer #3 (Public review):

Summary:

This work aims to address a fundamental biological question: how do mammalian cells achieve/lose tolerance to cold exposure? The authors first tried to establish an experimental system for cell cold exposure and evaluation of cell death and then performed genome-scale CRISPR-Cas9 screening on immortalized cell lines from Syrian Hamster (BHK-21) and human (K562) for key genes that are associated with cell survival during prolonged cold exposure. From these screenings, they focused on glutathione peroxidase 4 (GPX4). Using genetic modifications or pharmacological interventions, and multiple cell models including primary cells from various mammalian species, they showed that GPX4 proteins are likely to retain their activities at 4 {degree sign}C, functioning to prevent cold-induced cell ferroptosis.

Strengths:

(1) This paper is neatly written and hence easy to follow.

(2) Experiments are well designed.

(3) The data showing the overall good cell survival after a prolonged cold exposure or repeated cold-warm cycles are helpful to show the advantages of the experimental instruments and methods the authors used, and hence the validity of their results.

(4) The CRISPR-Cas9 screening is a great attempt.

(5) Multiple cell types from hibernating mammals (cold tolerant) and cold-intolerant species are used to test their findings.

(6) Although some may argue that other labs have published works with different approaches that have pointed out the importance of GPX4 and ferroptosis in hamster cell survival from anoxia-reoxygenation or cold exposure models, hence hurting the novelty of this work, this reviewer thinks that it is highly valuable to have independent research groups and different methods/systems to validate an important concept.

Weaknesses:

(1) Only cell death was robustly surveyed; though cell proliferation was evaluated too in some experiments, other cellular functions, such as mitochondrial ATP production vs. glycolysis, and the extent of lipid peroxidation, could have been measured to reflect cellular physiology.

Validations on complex tissues or in vivo systems would have further strengthened the work and its impact.

CRISPR-Cas9 screening may have technical limitations as knock-out of some essential genes/pathways may lead to cell lethality during screening, and hence the relevance of these genes/pathways to cell cold tolerance may not be noted. From the data presented in this study, this reviewer thinks that the GPX4 pathway is likely a conserved mechanism for long-term cold survival, but not for cold sensitivity or acute cell death from cold exposure. In line with my such speculation, their CRISPR-Cas9 screening revealed genes in the GPX4 pathway from a relatively cold-sensitive human cell line, but the endogenous GPX4 pathway is seemingly operational in this cold-sensitive cell line. Also, these cells are viable after GPX4 knock-out. Dead cells from the acute cold exposure phase may detached, or their genomic DNAs have been severely damaged by the time of sample collection, hence not giving any meaningful sequencing reads. Crippling other factors/pathways such as FOXO1 (PMID: 38570500) or 5-aminolevulinic acid (ALA) metabolism (PMID: 35401816) have been shown to severely aggravate cold-induced cell death, including TUNEL-revealed DNA damage, within a much shorter time scale, whilst loss-function knockouts of FOXO1 or ALA Synthase 1 (ALAS1) are usually cell lethal. Thus, they and other possible essential genes may not be screenable from the current experimental protocol. These important points need to be taken into consideration by the authors.

Author response:

Reviewer #1 (Public review):

Summary:

Through a series of CRISPR-Cas9 screens, the GPX4 antioxidant pathway was identified as a critical suppressor of cold-induced cell death in hibernator-derived cells. Hamster BHK-21 cells exposed to repeated cold and rewarming cycles revealed five genes (Gpx4, Eefsec, Pstk, Secisbp2, and Sepsecs) as critical components of the GPX4 pathway, which protects against cold-induced ferroptosis. A second screen with continuous cold exposure confirmed the essential role of GPX4 in prolonged cold tolerance. GPX4 knockout lines exhibited complete cell death within four days of cold exposure, and pharmacological inhibition of GPX4 further increased cell death, underscoring the necessity of GPX4's catalytic activity in cold conditions.

An additional CRISPR screen in human cold-sensitive K562 cells identified 176 genes for cold survival. The GPX4 pathway was found to confer significant resistance to cold in hibernators and human cells, with GPX4 loss significantly increasing cold-induced cell death.

Comparing hamster and human GPX4, overexpression of GPX4 in human K562 cells, whether hamster or human GPX4, dramatically improved cold tolerance, while catalytically dead mutants showed no such effect. These findings suggest that GPX4 abundance is a key limiting factor for cold tolerance in human cells, and primary cell types show strong sensitivity to GPX4 loss, highlighting that differences in cold tolerance across species may be due to varying GPX4-mediated protection.

Strengths:

(1) Innovative Approach: The study employs a series of unbiased genome-wide CRISPR-Cas9 screens in both hibernator- and non-hibernator-derived cells to investigate the mechanisms controlling cellular cold tolerance. Notably, this is the first genome-scale CRISPR-Cas9 screen conducted in cells derived from a hibernator, the Syrian hamster.

(2) Identification of the GPX4 Pathway: Identifying glutathione peroxidase 4 (GPX4) as a critical suppressor of cold-induced cell death significantly contributes to the field. Recently, GPX4 was also reported as a potent regulator of cold tolerance through overexpression screening (Sone et al.) in hamsters, which further supports this finding.

(3) Improved Cold Viability Assessment: The study identifies an important technical artifact in using trypan blue to assess cell viability following cold exposure. It reveals that cells stained immediately after cold exposure retain the dye, inaccurately indicating cell death. By introducing a brief rewarming period before viability assessment, the authors significantly improve the accuracy of detecting cold-induced cell death. This refinement in methodology ensures more reliable results and sets a new standard for future research on cold stress in cells.

Weaknesses:

(1) Mechanisms Regulating GPX4 Levels: While the study highlights GPX4 levels as a major determinant of cellular cold tolerance, it does not discuss how these levels are regulated or why they differ between hibernators and non-hibernators. This omission leaves an important aspect of GPX4's role in cold tolerance unexplored.

(2) Generalizability Across Species: Although the study demonstrates the role of GPX4 in several mammalian species, it does not investigate whether this mechanism extends to other vertebrates (e.g., fish and amphibians) that also face cold challenges. This limitation could restrict the broader evolutionary claims made by the study.

(3) Variability in Cold Sensitivity Across Human Cell Lines: The study observes significant variability in cold tolerance among different human cell lines but does not explain these differences clearly. This leaves a key aspect of human cell cold sensitivity insufficiently addressed.

We thank the reviewer for the positive evaluation and thoughtful comments on the manuscript. We acknowledge that our study does not delve into the mechanisms regulating GPX4 levels, including differences between hibernators and non-hibernators, differences between cell types, or the possibility that GPX4 levels are dynamically regulated by environmental conditions. We consider these as interesting open questions that could be addressed in future studies.

While our study focused entirely on mammalian species, we agree that examining cold tolerance mechanisms across a broader range of vertebrates, including fish and amphibians, could enhance our evolutionary perspective. Interestingly, previous work has indicated that C.elegans adapt to cold temperatures through ferritin mediated Fe2+ detoxification. This suggests that cold induces Fe2+-mediated toxicity in C.elegans as well as mammalian cells, but that the mechanisms through which distantly related species counteract cold-mediated cell death may vary. 

Finally, we agree that the variability in cold sensitivity across human cell lines could be further explored, and we will strongly consider conducting follow up experiments to examine the extent to which this variability is driven by levels of GPX4.

We are grateful for these insightful comments, as they highlight important avenues for future research. Addressing these questions will enable a more comprehensive understanding of GPX4's role in cold tolerance and its evolutionary significance across diverse organisms.

Reviewer #2 (Public review):

Summary:

Lam et al., present a very intriguing whole genome CRISPR screen in Syrian Hamster cells as well as K562 cells to identify key genes involved in hypothermia-rewarming tolerance. Survival screens were performed by exposing cells to 4C in a cooled CO2 incubator followed by a rewarming period of 30 minutes prior to survival analysis. In this paradigm, Syrian hamster-derived cell lines exhibit more robust survival than human cell lines (BHK-21 and HaK vs HT1080, HeLa, RPE1, and K562). A genome-wide Syrian hamster CRISPR library was created targeting all annotated genes with 10 guides/gene. LV transduction of the library was performed in BHK-21 cells and the survival screen procedures involved 3 cycles of 4C cold exposure x4 days followed by 2 days of re-warming.

When compared to controls maintained at 37C, 9 genes were required for BHK-21 survival of cold cycling conditions and 5 of these 9 are known components of the GPX4 antioxidant pathway. GPX4 KO BHK-21 cells had reduced cell growth at 37C and profoundly worse cold tolerance which could be reduced by GPX4 expression. GPX4 inhibitors also reduced survival in cold. CRISPR KO screens and GPX4 KO in K562 cells revealed comparable results (though intriguingly glutathione biosynthesis genes were more critical to K562 cells than BHK-21 cells). Human or Syrian hamster GPX4 overexpression improved cold tolerance.

Strengths:

This is a very nicely written paper that clearly communicates in figures and text complicated experimental manipulations and in vitro genetic screening and cell survival data. The focus on GPX4 is interesting and relatively novel. The converging pharmacologic, loss-of-function, and gain-of-function experiments are also a strength.

Weaknesses:

A recently published article (Reference 43, Sone et al.) also independently explored the role of GPX4 in Syrian hamster cold tolerance through gain-of-function screening. Further exploration of the GPX4 species-specific mechanisms would be of great interest, but this is considered a minor weakness given the already very comprehensive and compelling data presented.

We greatly appreciate the reviewer’s compliments and thoughtful comments on our manuscript. We agree with the reviewer that our approach (dual unbiased genome-scale screens in human and hamster cells) and the recent investigation by Sone et al (gain-of-function screening involving the insertion of hamster cDNA into human cells) mutually strengthen the importance of GPX4 in cold tolerance across cell types and species.

Reviewer #3 (Public review):

Summary:

This work aims to address a fundamental biological question: how do mammalian cells achieve/lose tolerance to cold exposure? The authors first tried to establish an experimental system for cell cold exposure and evaluation of cell death and then performed genome-scale CRISPR-Cas9 screening on immortalized cell lines from Syrian Hamster (BHK-21) and human (K562) for key genes that are associated with cell survival during prolonged cold exposure. From these screenings, they focused on glutathione peroxidase 4 (GPX4). Using genetic modifications or pharmacological interventions, and multiple cell models including primary cells from various mammalian species, they showed that GPX4 proteins are likely to retain their activities at 4 {degree sign}C, functioning to prevent cold-induced cell ferroptosis.

Strengths:

(1) This paper is neatly written and hence easy to follow.

(2) Experiments are well designed.

(3) The data showing the overall good cell survival after a prolonged cold exposure or repeated cold-warm cycles are helpful to show the advantages of the experimental instruments and methods the authors used, and hence the validity of their results.

(4) The CRISPR-Cas9 screening is a great attempt.

(5) Multiple cell types from hibernating mammals (cold tolerant) and cold-intolerant species are used to test their findings.

(6) Although some may argue that other labs have published works with different approaches that have pointed out the importance of GPX4 and ferroptosis in hamster cell survival from anoxia-reoxygenation or cold exposure models, hence hurting the novelty of this work, this reviewer thinks that it is highly valuable to have independent research groups and different methods/systems to validate an important concept.

Weaknesses:

(1) Only cell death was robustly surveyed; though cell proliferation was evaluated too in some experiments, other cellular functions, such as mitochondrial ATP production vs. glycolysis, and the extent of lipid peroxidation, could have been measured to reflect cellular physiology.

Validations on complex tissues or in vivo systems would have further strengthened the work and its impact.

CRISPR-Cas9 screening may have technical limitations as knock-out of some essential genes/pathways may lead to cell lethality during screening, and hence the relevance of these genes/pathways to cell cold tolerance may not be noted. From the data presented in this study, this reviewer thinks that the GPX4 pathway is likely a conserved mechanism for long-term cold survival, but not for cold sensitivity or acute cell death from cold exposure. In line with my such speculation, their CRISPR-Cas9 screening revealed genes in the GPX4 pathway from a relatively cold-sensitive human cell line, but the endogenous GPX4 pathway is seemingly operational in this cold-sensitive cell line. Also, these cells are viable after GPX4 knock-out. Dead cells from the acute cold exposure phase may detached, or their genomic DNAs have been severely damaged by the time of sample collection, hence not giving any meaningful sequencing reads. Crippling other factors/pathways such as FOXO1 (PMID: 38570500) or 5-aminolevulinic acid (ALA) metabolism (PMID: 35401816) have been shown to severely aggravate cold-induced cell death, including TUNEL-revealed DNA damage, within a much shorter time scale, whilst loss-function knockouts of FOXO1 or ALA Synthase 1 (ALAS1) are usually cell lethal. Thus, they and other possible essential genes may not be screenable from the current experimental protocol. These important points need to be taken into consideration by the authors.

We thank the reviewer for highlighting the novelty of using genome-scale CRISPR-Cas9 screens and the validation of GPX4 function across cell types and mammalian species. 

We acknowledge that our study primarily focused on measuring cell death using Trypan Blue dye exclusion. To validate the Trypan Blue assay, cell survival data was orthogonally measured using the LDH release assays (Fig. 1g). The proliferation potential of putatively live cells was assessed by counting the increase in live cells following 24 h at 37°C (Fig. 1b). Prompted by your question, we will add additional data to the final version of the manuscript in which we show that following 1 day at 4°C, K562 cells rapidly restarted their cell cycle and double in numbers every 21 hours (Author response image 1). This rate is indistinguishable from the replication rate of cells that were not previously exposed to 4°C, suggesting that the cells following cold exposure are both alive and functionally capable of replicating.

Author response image 1.

Population doubling time of K562 cells cultured at 37°C (pink) and cells that are rewarmed to 37°C following 1 day of 4°C exposure

We agree that assessing additional cellular functions, such as mitochondrial ATP production, glycolysis, lipid metabolism and peroxidation could provide a more comprehensive understanding of cellular physiology under cold stress and would be valuable future studies. Similarly, we appreciate the suggestion to validate our findings in complex tissues or in vivo models. We recognize that such validation could strengthen the implications of our study and enhance its translational potential; however, due to their complexity, we believe that these additional studies are beyond the scope of our current study.

We agree with the reviewer that CRISPR-Cas9 screens have limitations. For example our screen was designed to identify genes that are preferentially required for cellular fitness at 4°C versus 37°C. There are many genes that are required for cellular survival at 4°C as well as 37°C that are not discussed (Table S2, S5). Also, given that the screen is designed to disrupt a single gene per cell, genes that have redundant functions in cold-tolerance will likely be missed. Given the reviewer’s questions, we will expand the discussion of the paper to highlight limitations of the screen.

We apologize for any lack of clarity about the methods we employed during the screen and will expand the methods section to provide further details. For example, for the BHK-21 screen we eliminated dead cells by sequencing cells that reattached after rewarming to 37°C for either 30 minutes (15 day cold exposure screen) or 24 hours (4°C cycling screen). Indeed, at the point of cell collection for both BHK-21 and K562 screens, the fraction of live cells was greater than 92% and 95%, respectively.  We respectfully disagree with the reviewer that our screens would miss genes that affect acute cold tolerance. Any cells that would have died either early or late during cold exposure would have not been sequenced, and thus the sgRNAs targeting a specific gene in those cells would appear depleted, regardless of whether these cells died early/acutely or later during cold exposure. 

We thank the reviewer for pointing out two additionally highly relevant studies. Interestingly, the genes implicated in cold tolerance in these studies, FOXO1 and ALAS1, did not appear essential for survival at 37°C or 4°C  in BHK-21 or K562 cells. There are several possibilities that could explain this finding: 1) our screen may not have successfully knocked out these genes, 2) other proteins may have compensated for their loss, or 3) these pathways may regulate cold tolerance in some but not all cell types. We apologize that in the current version of the manuscript we did not reflect on these recent studies. We will expand our discussion to include their findings. 

Once again, we are grateful for the reviewer’s insights, which have highlighted key areas for further exploration as well as pointed to specific ways to improve our manuscript.

eLife Assessment

The presented evidence is compelling given a range of complementary and mutually supportive studies. Experiments are generally robustly conducted and well-presented, supporting the claims regarding miRNA mechanisms converging on EMC10 overexpression with 22q11 Del. This is an important study that works to establish a novel antisense oligonucleotide-based approach to treating 22q11.2 deletion syndrome; the findings are likely to advance therapeutic efforts. The authors provide evidence both in vitro in patient-derived iPSCs and in vivo in a 22q11 Del mouse supporting the knockdown (KD) of EMC10 as an effective strategy for the amelioration of neuronal and behavioral deficits.

Reviewer #1 (Public review):

Summary:

This is an important and very well-presented set of experiments following up on prior work from the lab investigating knock-down (KD) of EMC10 in the restoration of neuronal and cognitive deficits in 22q11.2 Del models, including now both human iPSCs and a mouse model in vivo now with ASOs.

The valuable progress in this current manuscript is the development of ASOs, and the proof of efficacy in vivo in mice of the ASO in knock-down of EMC10 and amelioration of in vivo behavioral phenotypes.

The experiments include iPSC studies demonstrating elevations of EMC10 in a solid collection of paired iPSC lines. These studies also provide evidence of manipulation of EMC10 by overexpression and inhibition of miRNAs that exist in the 22q11 interval. The iPSC studies also nicely demonstrate the rescue of impairments with KD of EMC10 in neuronal arborization as well as KCl-induced neuronal activity. The major in vivo contributions reflect an impressive demonstration of the efficacy of two ASOs in vivo on both KD of EMC10 in vivo and through improvement in behavioral abnormalities in the 22q11 mouse in a range of different behaviors, including social behavior and learning behaviors.

Overall, there are many strengths reflected in this study, including in particular the synergy between in vitro studies in human cell models and in vivo studies in the well-characterized mouse model. The experiments are generally rigorously performed, well-powered, and nicely presented. The claims with regard to the mechanisms of EMC10 elevations and the importance of restoration of EMC10 expression to neuronal morphology and behavior are well supported by the data. The work may be further supported in future studies, by investigation of rescue by ASOs of circuit dysfunction in vivo or ex vivo through electrophysiology in the mouse model. Also, in future studies, investigation of the mechanism by which EMC10, an ER protein involved in protein processing, may function in the observed neuronal abnormalities; however, these studies are clearly for future investigations.

The potential impact of the work is found in the potential value of the ASO approach to the treatment of 22q11, or the pre-clinical evidence that knock-down of this protein may lead to some amelioration of cognitive symptoms. Overall, a very convincing and complementary set of experiments to support EMC10 KD as a therapeutic strategy.

Reviewer #2 (Public review):

Summary:

The manuscript by Thakur et. al seeks to establish a novel ASO-based approach to treat 22q11.2 deletion syndrome. Central to this thesis is that an ER membrane complex member called EMC10 is significantly increased in the disorder, which is largely attributed to the loss of miRNA-mediated repression. The authors generated three new iPSC cell lines for the disorder and showed that deletion of EMC10 rescues morphology and Ca-flux deficits. They go on to show that post-symptomatic deletion of Emc10 in mice using a conditional-off tamoxifen allele reverses social memory phenotypes. Finally, in collaboration with Ionis, they developed two new ASOs to knock down EMC10 and show that social and spatial memory phenotypes are rescued, even two months after injection.

Strengths:

In general, this represents a substantial undertaking and an impressive body of work. The experiments follow a logical progression and in most cases are well-controlled. The isolation of EMC10 effects relative to the broader miRNA disruption is viewed as impactful. The use of both genetic and ASO approaches to validate the therapeutic strategy is also viewed as highly positive. The authors' contention that EMC10 can be targeted at post-symptomatic time points to reverse 22q11.2 deletion syndrome is supported by the data. Further, they have provided a therapeutic mechanism to do so. These findings are likely to be impactful and lead to further development efforts.

Weaknesses:

The primary weaknesses of the manuscript lie in incomplete or inappropriate data analysis, as well as a failure to validate key experiments. For example, both genetic and ASO-mediated EMC10-mediated reductions are assessed at the level of mRNA, but only one experiment, in one brain region, is validated at the protein level. This brain region is the PFC, which is problematic when many of the phenotypes used have a strong hippocampal component. Likewise, the iPSC experiments make the case that excitatory neurons are central to the phenotype, but no effort is made to show that the ASOs are entering that type of neuron, or even any quantification of what percentage of cells in the target brain regions (HPC, PFC, etc.) are positive for the ASO. There is only a single image provided of staining with a phosphorothioate antibody and a claim of robust uptake, which cannot be assumed. The iPSC transcriptomics work would also benefit from a more comprehensive comparison between the EMC10 knockout lines and their parent 22q11 deletion lines. Further, there are other examples where the statistics used are either wrong (Figure 3 t-test vs ANOVA) or missing (Figure S2). These technical and analytical shortcomings make it challenging to fully interpret the data and detract from an otherwise exciting manuscript.

eLife Assessment

This is an important study that aims to investigate the behavioral relevance of multisensory responses recorded in the auditory cortex. The experiments are elegant and well-designed, and are supported by appropriate analyses of the data. However, the evidence presented for learning-dependent encoding of visual information is incomplete and it is possible that the surprisingly short-latency increases in activity are actually motor-related signals. Demonstrating that they really are visual responses is necessary in order to draw definitive conclusions from this study.

Reviewer #1 (Public review):

Summary:

Chang and colleagues used tetrode recordings in behaving rats to study how learning an audiovisual discrimination task shapes multisensory interactions in the auditory cortex. They found that a significant fraction of neurons in the auditory cortex responded to visual (crossmodal) and audiovisual stimuli. Both auditory-responsive and visually-responsive neurons preferentially responded to the cue signaling the contralateral choice in the two-alternative forced choice task. Importantly, multisensory interactions were similarly specific for the congruent audiovisual pairing for the contralateral side.

Strengths:

The experiments were conducted in a rigorous manner. Particularly thorough are the comparisons across cohorts of rats trained in a control task, in a unisensory auditory discrimination task, and the multisensory task, while also varying the recording hemisphere and behavioral state (engaged vs. anesthesia). The resulting contrasts strengthen the authors' findings and rule out important alternative explanations. Through the comparisons, they show that the enhancements of multisensory responses in the auditory cortex are specific to the paired audiovisual stimulus and specific to contralateral choices in correct trials and thus dependent on learned associations in a task-engaged state.

Weaknesses:

The main result is that multisensory interactions are specific for contralateral paired audiovisual stimuli, which is consistent across experiments and interpretable as a learned task-dependent effect. However, the alternative interpretation of behavioral signals is crucial to rule out, which would also be specific to contralateral, correct trials in trained animals. Although the authors focus on the first 150 ms after cue onset, some of the temporal profiles of activity suggest that choice-related activity could confound some of the results.

The auditory stimuli appear to be encoded by short transient activity (in line with much of what we know about the auditory system), likely with onset latencies (not reported) of 15-30 ms. Stimulus identity can be decoded (Figure 2j) apparently with an onset latency around 50-75 ms (only the difference between A and AV groups is reported) and can be decoded near perfectly for an extended time window, without a dip in decoding performance that is observed in the mean activity Figure 2e. The dynamics of the response of the example neurons presented in Figures 2c and d and the average in 2e therefore do not entirely match the population decoding profile in 2j. Population decoding uses the population activity distribution, rather than the mean, so this is not inherently problematic. It suggests however that the stimulus identity can be decoded from later (choice-related?) activity. The dynamics of the population decoding accuracy are in line with the dynamics one could expect based on choice-related activity. Also the results in Figures S2e,f suggest differences between the two learned stimuli can be in the late phase of the response, not in the early phase.

First, it would help to have the same time axis across panels 2,c,d,e,j,k. Second, a careful temporal dissociation of when the central result of multisensory enhancements occurs in time would discriminate better early sensory processing-related effects versus later decision-related modulations.

In the abstract, the authors mention "a unique integration model", "selective multisensory enhancement for specific auditory-visual pairings", and "using this distinct integrative mechanisms". I would strongly recommend that the authors try to phrase their results more concretely, which I believe would benefit many readers, i.e. selective how (which neurons) and specific for which pairings?

Reviewer #2 (Public review):

Summary

In this study, rats were trained to discriminate auditory frequency and visual form/orientation for both unisensory and coherently presented AV stimuli. Recordings were made in the auditory cortex during behaviour and compared to those obtained in various control animals/conditions. The central finding is that AC neurons preferentially represent the contralateral-conditioned stimulus - for the main animal cohort this was a 10k tone and a vertically oriented bar. Over 1/3rd of neurons in AC were either AV/V/A+V and while a variety of multisensory neurons were recorded, the dominant response was excitation by the correctly oriented visual stimulus (interestingly this preference was absent in the visual-only neurons). Animals performing a simple version of the task in which responses were contingent on the presence of a stimulus rather than its identity showed a smaller proportion of AV stimuli and did not exhibit a preference for contralateral conditioned stimuli. The contralateral conditioned dominance was substantially less under anesthesia in the trained animals and was present in a cohort of animals trained with the reverse left/right contingency. Population decoding showed that visual cues did not increase the performance of the decoder but accelerated the rate at which it saturated. Rats trained on auditory and then visual stimuli (rather than simultaneously with A/V/AV) showed many fewer integrative neurons.

Strengths

There is a lot that I like about this paper - the study is well-powered with multiple groups (free choice, reversed contingency, unisensory trained, anesthesia) which provides a lot of strength to their conclusions and there are many interesting details within the paper itself. Surprisingly few studies have attempted to address whether multisensory responses in the unisensory cortex contribute to behaviour - and the main one that attempted to address this question (Lemus et al., 2010, uncited by this study) showed that while present in AC, somatosensory responses did not appear to contribute to perception. The present manuscript suggests otherwise and critically does so in the context of a task in which animals exhibit a multisensory advantage (this was lacking in Lemus et al.,). The behaviour is robust, with AV stimuli eliciting superior performance to either auditory or visual unisensory stimuli (visual were slightly worse than auditory but both were well above chance).

Weaknesses

I have a number of points that in my opinion require clarification and I have suggestions for ways in which the paper could be strengthened. In addition to these points, I admit to being slightly baffled by the response latencies; while I am not an expert in the rat, usually in the early sensory cortex auditory responses are significantly faster than visual ones (mirroring the relative first spike latencies of A1 and V1 and the different transduction mechanisms in the cochlea and retina). Yet here, the latencies look identical - if I draw a line down the pdf on the population level responses the peak of the visual and auditory is indistinguishable. This makes me wonder whether these are not sensory responses - yet, they look sensory (very tightly stimulus-locked). Are these latencies a consequence of this being AuD and not A1, or ... ? Have the authors performed movement-triggered analysis to illustrate that these responses are not related to movement out of the central port, or is it possible that both sounds and visual stimuli elicit characteristic whisking movements? Lastly, has the latency of the signals been measured (i.e. you generate and play them out synchronously, but is it possible that there is a delay on the audio channel introduced by the amp, which in turn makes it appear as if the neural signals are synchronous? If the latter were the case I wouldn't see it as a problem as many studies use a temporal offset in order to give the best chance of aligning signals in the brain, but this is such an obvious difference from what we would expect in other species that it requires some sort of explanation.

Reaction times were faster in the AV condition - it would be of interest to know whether this acceleration is sufficient to violate a race model, given the arbitrary pairing of these stimuli. This would give some insight into whether the animals are really integrating the sensory information. It would also be good to clarify whether the reaction time is the time taken to leave the center port or respond at the peripheral one.

The manuscript is very vague about the origin or responses - are these in AuD, A1, AuV... ? Some attempts to separate out responses if possible by laminar depth and certainly by field are necessary. It is known from other species that multisensory responses are more numerous, and show greater behavioural modulation in non-primary areas (e.g. Atilgan et al., 2018).

Reviewer #3 (Public review):

Summary:

The manuscript by Chang et al. aims to investigate how the behavioral relevance of auditory and visual stimuli influences the way in which the primary auditory cortex encodes auditory, visual, and audiovisual information. The main result is that behavioral training induces an increase in the encoding of auditory and visual information and in multisensory enhancement that is mainly related to the choice located contralaterally with respect to the recorded hemisphere.

Strengths:

The manuscript reports the results of an elegant and well-planned experiment meant to investigate if the auditory cortex encodes visual information and how learning shapes visual responsiveness in the auditory cortex. Analyses are typically well done and properly address the questions raised

Weaknesses:

Major

(1) The authors apparently primarily focus their analyses of sensory-evoked responses in approximately the first 100 ms following stimulus onset. Even if I could not find an indication of which precise temporal range the authors used for analysis in the manuscript, this is the range where sensory-evoked responses are shown to occur in the manuscript figures. While this is a reasonable range for auditory evoked responses, the same cannot be said for visual responses, which commonly peak around 100-120 ms, in V1. In fact, the latency and overall shape of visual responses are quite different from typical visual responses, that are commonly shown to display a delay of up to 100 ms with respect to auditory responses. All traces that the authors show, instead, display visual responses strikingly overlapping with auditory ones, which is not in line with what one would expect based on our physiological understanding of cortical visually-evoked responses. Similarly, the fact that the onset of decoding accuracy (Figure 2j) anticipates during multisensory compared to auditory-only trials is hard to reconcile with the fact that visual responses have a later onset latency compared to auditory ones. The authors thus need to provide unequivocal evidence that the results they observe are truly visual in origin. This is especially important in view of the ever-growing literature showing that sensory cortices encode signals representing spontaneous motor actions, but also other forms of non-sensory information that can be taken prima facie to be of sensory origin. This is a problem that only now we realize has affected a lot of early literature, especially - but not only - in the field of multisensory processing. It is thus imperative that the authors provide evidence supporting the true visual nature of the activity reported during auditory and multisensory conditions, in both trained, free-choice, and anesthetised conditions. This could for example be achieved causally (e.g. via optogenetics) to provide the strongest evidence about the visual nature of the reported results, but it's up to the authors to identify a viable solution. This also applies to the enhancement of matched stimuli, that could potentially be explained in terms of spontaneous motor activity and/or pre-motor influences. In the absence of this evidence, I would discourage the author from drawing any conclusion about the visual nature of the observed activity in the auditory cortex.

(2) The finding that AC neurons in trained mice preferentially respond - and enhance - auditory and visual responses pertaining to the contralateral choice is interesting, but the study does not show evidence for the functional relevance of this phenomenon. As has become more and more evident over the past few years (see e.g. the literature on mouse PPC), correlated neural activity is not an indication of functional role. Therefore, in the absence of causal evidence, the functional role of the reported AC correlates should not be overstated by the authors. My opinion is that, starting from the title, the authors need to much more carefully discuss the implications of their findings.

MINOR:

(1) The manuscript is lacking what pertains to the revised interpretation of most studies about audiovisual interactions in primary sensory cortices following the recent studies revealing that most of what was considered to be crossmodal actually reflects motor aspects. In particular, recent evidence suggests that sensory-induced spontaneous motor responses may have a surprisingly fast latency (within 40 ms; Clayton et al. 2024). Such responses might also underlie the contralaterally-tuned responses observed by the authors if one assumes that mice learn a stereotypical response that is primed by the upcoming goal-directed, learned response. Given that a full exploration of this issue would require high-speed tracking of orofacial and body motions, the authors should at least revise the discussion and the possible interpretation of their results not just on the basis of the literature, but after carefully revising the literature in view of the most recent findings, that challenge earlier interpretations of experimental results.

(2) The methods section is a bit lacking in details. For instance, information about the temporal window of analysis for sensory-evoked responses is lacking. Another example: for the spike sorting procedure, limited details are given about inclusion/exclusion criteria. This makes it hard to navigate the manuscript and fully understand the experimental paradigm. I would recommend critically revising and expanding the methods section.

eLife Assessment

This study addresses a significant question in sensory ethology and active sensing in particular. It links the production of a specific signal - electrosensory chirps - to various contexts and conditions to propose that chirps may also serve an active sensing role in addition to their more well-known role in communication. The evidence supporting the role for active sensing is strong. In particular, the evidence showing increased chirping in more cluttered environments and the relationship between chirping and movement are convincing. The study provides a lot of valuable data, and is likely to stimulate follow-up behavioral and physiological studies.

Reviewer #1 (Public Review):

The authors investigate the role of chirping in a species of weakly electric fish. They subject the fish to various scenarios and correlate the production of chirps with many different factors. They find major correlations between the background beat signals (continuously present during any social interactions) or some aspects of social and environmental conditions with the propensity to produce different types of chirps. By analyzing more specifically different aspects of these correlations they conclude that chirping patterns are related to navigation purposes and the need to localize the source of the beat signal (i.e. the location of the conspecific).

The study provides a wealth of interesting observations of behavior and much of this data constitutes a useful dataset to document the patterns of social interactions in these fish. Some data, in particular the high propensity to chirp in cluttered environments, raises interesting questions. Their main hypothesis is a useful addition to the debate on the function of these chirps and is worth being considered and explored further.

After the initial reviewers' comments, the authors performed a welcome revision of the way the results are presented. Overall the study has been improved by the revisions.

Reviewer #2 (Public Review):

Studying Apteronotus leptorhynchus (the weakly electric brown ghost knifefish), the authors provide evidence that 'chirps' (brief modulations in the frequency and amplitude of the ongoing wave-like electric signal) function in active sensing (specifically homeoactive sensing) rather than communication. Chirping is a behavior that has been well studied, including numerous studies on the sensory coding of chirps and the neural mechanisms for chirp generation. Chirps are largely thought to function in communication behavior, so this alternative function is a very exciting possibility that should have a great impact on the field.

The authors provide convincing evidence that chirps may function in homeoactive sensing. In particular, the evidence showing increased chirping in more cluttered environments and a relationship between chirping and movement are especially strong and suggestive. Their evidence arguing against a role for chirps in communication is not as strong. However, based on an extensive review of the literature, the authors conclude, I think fairly, that the evidence arguing in favor of a communication function is limited and inconclusive. Thus, the real strength of this study is not that it conclusively refutes the communication hypothesis, but that it calls this hypothesis into question while also providing compelling evidence in favor of an alternative function.

In summary, although the evidence against a role for chirps in communication is not as strong as the evidence for a role in active sensing, this study presents very interesting data that is sure to stimulate discussion and follow-up studies. The authors acknowledge that chirps could function as both a communication and homeactive sensing signal, and the language arguing against a communication function is appropriately measured. A given electrical behavior could serve both communication and homeoactive sensing. I suspect this is quite common in electric fish (not just in gymnotiforms such as the species studied here, but also in the distantly related mormyrids), and perhaps in other actively sensing species such as echolocating animals.

Reviewer #3 (Public Review):

Summary:

This important paper provides the best-to-date characterization of chirping in weakly electric fish using a large number of variables. These include environment (free vs divided fish, with or without clutter), breeding state, gender, intruder vs resident, social status, locomotion state and social and environmental experience, without and with playback experiments. It applies state-of-the-art methods for reducing the dimensionality of the data and finding patterns of correlation between different kinds of variables (factor analysis, K-means). The strength of the evidence, collated from a large number of trials with many controls, leads to the conclusion that the traditionally assumed communication function of chirps may be secondary to its role in environmental assessment and exploration that takes social context into account. Based on their extensive analyses, the authors suggest that chirps are mainly used as probes that help detect beats caused by other fish as well as objects.

Strengths:

The work is based on completely novel recordings using interaction chambers. The amount of new data and associated analyses is simply staggering, and yet, well organized in presentation. The study further evaluates the electric field strength around a fish (via modelling with the boundary element method) and how its decay parallels the chirp rate, thereby relating the above variables to electric field geometry. The BEM modelling also convincingly predicts how the electric image of a receiver conspecific on a sending fish is enhanced by a chirp.

The main conclusions are that the lack of any significant behavioural correlates for chirping, and the lack of temporal patterning in chirp time series, cast doubt on a primary communication goal for most chirps. Rather, the key determinants of chirping are the difference in frequency between two interacting conspecifics as well as individual subjects' environmental and social experience. The paper concludes that there is a lack of evidence for stereotyped temporal patterning of chirp time series, as well as of sender-receiver chirp transitions beyond the known increase in chirp frequency during an interaction. The authors carefully submit that the new putative echolocation function of chirps is not mutually exclusive with a possible communication function.

These conclusions by themselves will be very useful to the field. They will also allow scientists working on other "communication" systems to perhaps reconsider and expand the goals of the probes used in those senses. A lot of data are summarized in this paper, with thorough referencing to past work.

The alternative hypotheses that arise from the work are that chirps are mainly used as environmental probes for better beat detection and processing and object localization, and in this sense are self-directed signals. This led to their prediction that environmental complexity ("clutter") should increase chirp rate, which is fact was revealed by their new experiments. The authors also argue that waveform EODs have less power across high spatial frequencies compared to pulse-type fish, with a resulting relatively impoverished power of resolution. Chirping in wave-type fish could temporarily compensate for the lower frequency resolution while still being able to resolve EOD perturbations with a good temporal definition (which pulse-type fish lack due to low pulse rates).

The authors also advance the interesting idea that the sinusoidal frequency modulations caused by chirps are the electric fish's solution to the minute (and undetectable by neural wetware) echo-delays available to it, due to the propagation of electric fields at the speed of light in water. The paper provides a number of experimental avenues to pursue in order to validate the non-communication role of chirps.

Author response:

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

eLife Assessment

This study addresses a question in sensory ethology and active sensing in particular. It links the production of a specific signal - electrosensory chirps - to various contexts and conditions to argue that the main function is to enhance conspecific localization rather than communication as previously believed. The study provides a lot of valuable data, but the methods section is incomplete making it difficult to evaluate the claims.

We have now added to the methods a new paragraph describing in better detail the analysis done to prepare the data used in figure 7. The figure itself has been substantially changed: we now show EOD fields and electric images using voltage, instead of current and we have better illustrated the comparisons between chirps and beats using statistical analysis.

Eventually, we are equally grateful to all Reviewers for the constructive criticism and for the time spent in evaluating our manuscript. It certainly helped to improve both the quality of the data presented as well as the readability of the text.

Public Reviews:

Reviewer #1 (Public Review):

The authors investigate the role of chirping in a species of weakly electric fish. They subject the fish to various scenarios and correlate the production of chirps with many different factors. They find major correlations between the background beat signals (continuously present during any social interactions) or some aspects of social and environmental conditions with the propensity to produce different types of chirps. By analyzing more specifically different aspects of these correlations they conclude that chirping patterns are related to navigation purposes and the need to localize the source of the beat signal (i.e. the location of the conspecific).

The study provides a wealth of interesting observations of behavior and much of this data constitutes a useful dataset to document the patterns of social interactions in these fish. Some data, in particular the high propensity to chirp in cluttered environments, raises interesting questions. Their main hypothesis is a useful addition to the debate on the function of these chirps and is worth being considered and explored further.

After the initial reviewers' comments, the authors performed a welcome revision of the way the results are presented. Overall the study has been improved by the revision. However, one piece of new data is perplexing to me. The new figure 7 presents the results of a model analysis of the strength of the EI caused by a second fish to localize when the focal fish is chirping. From my understanding of this type of model, EOD frequency is not a parameter in the model since it evaluates the strength of the field at a given point in time. Therefore the only thing that matters is the phase relationship and strength of the EOD. Assuming that the second fish's EOD is kept constant and the phase relationship is also the same, the only difference during a chirp that could affect the result of the calculation is the potential decrease in EOD amplitude during the chirp. It is indeed logical that if the focal fish decreased its EOD amplitude the target fish's EOD becomes relatively stronger. Where things are harder to understand is why the different types of chirps (e.g. type 1 vs type 2) lead to the same increase in signal even though they are typically associated with different levels of amplitude modulations. Also, it is hard to imagine that a type 2 chirp that is barely associated with any decrease in EOD amplitude (0-10% maybe), would cause a doubling of the EI strength. There might be something I don't understand but the authors should provide a lot more details on how this result is obtained and convince us that it makes sense.

We hope we have now resolved the Reviewer’s concerns by applying major edits to Figure 7. We now use voltage - not current - to quantify the impact of chirps on electric images. The effect of chirps is here estimated using the integral of the beat AM, as a broad measure of the potential effects chirping may have on electroreceptors. We underline in the text that this analysis does not represent proof for any type of processing occurring in the fish brain, but we only express in hypothetical terms that - based on the beat perturbations measured - additional spatial information may potentially be available in electric images, as a consequence of chirping. Whether the fish uses this information, or not, needs to be assessed through electrophysiology in future studies.

Finally, the reviewer is concerned about this sentence in the rebuttal - "The methods section has been edited to clarify the approach (not yet)". This section is unfinished, which suggests that it is difficult to explain the modeling results from a logical point of view. Thus the reviewer's major concern from the previous review remains unresolved. To summarize, the model calculates field strengths at an instant in time and integrates over time with a 500 ms window. This window is 10 times longer than the small chirps, while the longer chirps cover a much larger proportion of the window. Yet, the small chirps have a bigger impact on discriminability than the longer chirps. The authors should attempt to explain this seemingly contradictory result. This remains a major issue because this analysis was the most direct evidence that chirping could impact localization accuracy.

We added a new method section describing the new figure and hopefully it is explaining more clearly how the effect of chirps is calculated. Since most p-units are affected by the beat cyclic AMs, any change on the electric image caused by a chirp will result in changes in transcutaneous voltage - i.e. the voltage measurable at the receptor level. Overall, this added analysis is not a central point of the manuscript, it is part of an attempt to hint to physiological mechanisms implied which cannot be explored in the current study. We do not mean to propose that these estimates represent alternatives to electrophysiological recordings, rather theoretical evidences which could in fact support this type of investigation. 

Reviewer #2 (Public Review):

Studying Apteronotus leptorhynchus (the weakly electric brown ghost knifefish), the authors provide evidence that 'chirps' (brief modulations in the frequency and amplitude of the ongoing wave-like electric signal) function in active sensing (specifically homeoactive sensing) rather than communication. Chirping is a behavior that has been well studied, including numerous studies on the sensory coding of chirps and the neural mechanisms for chirp generation. Chirps are largely thought to function in communication behavior, so this alternative function is a very exciting possibility that should have a great impact on the field.

The authors provide convincing evidence that chirps may function in homeoactive sensing. In particular, the evidence showing increased chirping in more cluttered environments and a relationship between chirping and movement are especially strong and suggestive. Their evidence arguing against a role for chirps in communication is not as strong. However, based on an extensive review of the literature, the authors conclude, I think fairly, that the evidence arguing in favor of a communication function is limited and inconclusive. Thus, the real strength of this study is not that it conclusively refutes the communication hypothesis, but that it calls this hypothesis into question while also providing compelling evidence in favor of an alternative function.

In summary, although the evidence against a role for chirps in communication is not as strong as the evidence for a role in active sensing, this study presents very interesting data that is sure to stimulate discussion and follow-up studies. The authors acknowledge that chirps could function as both a communication and homeactive sensing signal, and the language arguing against a communication function is appropriately measured. A given electrical behavior could serve both communication and homeoactive sensing. I suspect this is quite common in electric fish (not just in gymnotiforms such as the species studied here, but also in the distantly related mormyrids), and perhaps in other actively sensing species such as echolocating animals.

We are grateful to the Reviewer for the kind assessment.

Reviewer #3 (Public Review):

Summary:

This important paper provides the best-to-date characterization of chirping in weakly electric fish using a large number of variables. These include environment (free vs divided fish, with or without clutter), breeding state, gender, intruder vs resident, social status, locomotion state and social and environmental experience, without and with playback experiments. It applies state-of-the-art methods for reducing the dimensionality of the data and finding patterns of correlation between different kinds of variables (factor analysis, K-means). The strength of the evidence, collated from a large number of trials with many controls, leads to the conclusion that the traditionally assumed communication function of chirps may be secondary to its role in environmental assessment and exploration that takes social context into account. Based on their extensive analyses, the authors suggest that chirps are mainly used as probes that help detect beats caused by other fish as well as objects.

Strengths:

The work is based on completely novel recordings using interaction chambers. The amount of new data and associated analyses is simply staggering, and yet, well organized in presentation. The study further evaluates the electric field strength around a fish (via modelling with the boundary element method) and how its decay parallels the chirp rate, thereby relating the above variables to electric field geometry. The BEM modelling also convincingly predicts how the electric image of a receiver conspecific on a sending fish is enhanced by a chirp.

The main conclusions are that the lack of any significant behavioural correlates for chirping, and the lack of temporal patterning in chirp time series, cast doubt on a primary communication goal for most chirps. Rather, the key determinants of chirping are the difference in frequency between two interacting conspecifics as well as individual subjects' environmental and social experience. The paper concludes that there is a lack of evidence for stereotyped temporal patterning of chirp time series, as well as of sender-receiver chirp transitions beyond the known increase in chirp frequency during an interaction. The authors carefully submit that the new putative echolocation function of chirps is not mutually exclusive with a possible communication function.

These conclusions by themselves will be very useful to the field. They will also allow scientists working on other "communication" systems to perhaps reconsider and expand the goals of the probes used in those senses. A lot of data are summarized in this paper, with thorough referencing to past work.

The alternative hypotheses that arise from the work are that chirps are mainly used as environmental probes for better beat detection and processing and object localization, and in this sense are self-directed signals. This led to their prediction that environmental complexity ("clutter") should increase chirp rate, which is fact was revealed by their new experiments. The authors also argue that waveform EODs have less power across high spatial frequencies compared to pulse-type fish, with a resulting relatively impoverished power of resolution. Chirping in wave-type fish could temporarily compensate for the lower frequency resolution while still being able to resolve EOD perturbations with a good temporal definition (which pulse-type fish lack due to low pulse rates).

The authors also advance the interesting idea that the sinusoidal frequency modulations caused by chirps are the electric fish's solution to the minute (and undetectable by neural wetware) echo-delays available to it, due to the propagation of electric fields at the speed of light in water. The paper provides a number of experimental avenues to pursue in order to validate the non-communication role of chirps.

We are grateful to the Reviewer for the kind assessment.

eLife Assessment

This work attempts to demonstrate an ATP-independent non-canonical role of proteasomal component PA28y in the promotion of oral squamous cell carcinoma growth, migration, and invasion. The evidence remains incomplete and the work would benefit from further experimental work. The authors have not adequately addressed the reviewers' comments.

Reviewer #1 (Public review):

Summary:

The manuscript the authors have tried to dissect the functions of Proteasome activator 28γ (PA28γ) which is known to activate proteosomal function in an ATP independent manner. Although there are multiple works that have highlighted the role of this protein in tumour, this study specifically tried to develop a correlate with Complement C1q binding protein (C1QBp) that is associated with immune response and energy homeostasis.

Strengths:

The observations of the authors hint that beyond PA28y association with proteasome, it might also stabilize certain proteins such as C1QBP which influences the energy metabolism.

Weaknesses:

The strength of the work also becomes its main drawback. That is, how PA28y stabilizes C1QBP or how C1QBP elicits its pro-tumourigenic role under PA28y OE.

In most of the experiments the authors have been dependent on the parallel changes in the expression of both the proteins to justify their stabilizing interaction. However, this approach is indirect at best and does not confirm the direct stabilizing effect of this interaction. IP experiments do not indicate direct interaction and have some quality issues. The upregulation of C1QBP might be indirect at best. It is quite possible that PA28y might be degrading some secondary protein/complex which is responsible for C1QBP expression. Since the core idea of the work is PA28y direct interaction with C1QBP stabilizing it, the same should be demonstrated in more convincing manner.

In all of the assays C1QBP has been detected as doublet. However, the expression pattern of the two bands vary depending on the experiment. In some cases the upper band is intensely stained and in some the lower bands. Does C1QBP isoforms exist and whether they are differentially regulated depending on experiment conditions/tissue types?

Problems with the background of the work: Line 76. This statement is far-fetched. There are presently a number of literatures that have dealt with metabolic programming of OSCC including identification of specific metabolites. Moreover, beyond estimation of OCR, the authors have not conducted any experiments related to metabolism. In the Introduction, significance of this study and how it will extend our understanding of OSCC needs to be elaborated.

Review of Revised Version:

Although the authors have partly corrected the manuscript by removing the mislabeling in their Co-IP experiments, my primary concern on the actual functional connotations and direct interaction between PA28y and C1QBP still remains unaddressed. As already mentioned in my previous review, since the core idea of the work is PA28y's direct interaction with C1QBP, stabilizing it, the same should be demonstrated in a more convincing manner.

My other observation on the detection of C1QBP as a doublet has been addressed by usage of anti-C1QBP Monoclonal antibody against the polyclonal one used before. C1QBP doublets have not been observed in the present case.

The authors have also worked on the presentation of the background by suitably modifying the statements and incorporating appropriate citations.

However, the authors are requested to follow the recommendations provided to them by the reviewers to address the major concerns.

Reviewer #2 (Public review):

Summary:

The authors tried to determine how PA28g functions in oral squamous cell carcinoma (OSCC) cells. They hypothesized it may act through metabolic reprogramming in the mitochondria.

Strengths:

They found that the genes of PA28g and C1QBP are in an overlapping interaction network after an analysis of a genome database. They also found that the two proteins interact in coimmunoprecipitation and pull-down assays using the lysate from OSCC cells with or without expression of the exogenous genes. They used truncated C1QBP proteins to map the interaction site to the N-terminal 167 residues of C1QBP protein. They observed the levels of the two proteins are positively correlated in the cells. They provided evidence for the colocalization of the two proteins in the mitochondria and the effect on mitochondrial form and function in vitro and in vivo OSCC models, and the correlation of the protein expression with the prognosis of cancer patients.

Weaknesses:

Many data sets are shown in figures that cannot be understood without more descriptions either in the text or the legend, e.g., Fig. 1A. Similarly, many abbreviations are not defined.

The revision addressed these issues.

Some of the pull-down and coimmunoprecipitation data do not support the conclusion about the PA28g-C1QBP interaction. For example, in Appendix Fig. 1B the Flag-C1QBP was detected in the Myc beads pull-down when the protein was expressed in the 293T cells without the Myc-PA28g, suggesting that the pull-down was not due to the interaction of the C1QBP and PA28g proteins. In Appendix Fig. 1C, assume the SFB stands for a biotin tag, then the SFB-PA28g should be detected in the cells expressing this protein after pull-down by streptavidin; however, it was not. The Western blot data in Fig. 1E and many other figures must be quantified before any conclusions about the levels of proteins can be drawn.

The revision addressed these problems.

The immunoprecipitation method is flawed as it is described. The antigen (PA28g or C1QBP) should bind to the respective antibody that in turn should binds to Protein G beads. The resulting immunocomplex should end up in the pellet fraction after centrifugation, and analyzed further by Western blot for coprecipitates. However, the method in the Appendix states that the supernatant was used for the Western blot.

The revision corrected this method.

To conclude that PA28g stabilizes C1QBP through their physical interaction in the cells, one must show whether a protease inhibitor can substitute PA28q and prevent C1QBP degradation, and also show whether a mutation that disrupt the PA28g-C1QBP interaction can reduce the stability of C1QBP. In Fig. 1F, all cells expressed Myc-PA28g. Therefore, the conclusion that PA28g prevented C1QBP degradation cannot be reached. Instead, since more Myc-PA28g was detected in the cells expressing Flag-C1QBP compared to the cells not expressing this protein, a conclusion would be that the C1QBP stabilized the PA28g. Fig. 1G is a quantification of a Western blot data that should be shown.

The binding site for PA28g in C1QBP was mapped to the N-terminal 167 residues using truncated proteins. One caveat would be that some truncated proteins did not fold correctly in the absence of the sequence that was removed. Thus, the C-terminal region of the C1QBP with residues 168-283 may still bind to the PA29g in the context of full-length protein. In Fig. 1I, more Flag-C1QBP 1-167 was pull-down by Myc-PA28g than the full-length protein or the Flag-C1QBP 1-213. Why?

The interaction site in PA28g for C1QBP was not mapped, which prevents further analysis of the interaction. Also, if the interaction domain can be determined, structural modeling of the complex would be feasible using AlphaFold2 or other programs. Then, it is possible to test point mutations that may disrupt the interaction and if so, the functional effect.

The revision added AlphaFold models for the protein interaction. However, the models were not analyzed and potential mutations that would disrupt the interact were not predicted, made and tested. The revision did not addressed the request for the protease inhibitor.

Author response:

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

Public Reviews:

Reviewer #1 (Public review):

Summary:

In this manuscript, the authors have tried to dissect the functions of Proteasome activator 28γ (PA28γ) which is known to activate proteasomal function in an ATP-independent manner. Although there are multiple works that have highlighted the role of this protein in tumours, this study specifically tried to develop a correlation with Complement C1q binding protein (C1QBp) that is associated with immune response and energy homeostasis.

Strengths:

The observations of the authors hint that beyond PA28y's association with the proteasome, it might also stabilize certain proteins such as C1QBP which influences energy metabolism.

Weaknesses:

The strength of the work also becomes its main drawback. That is, how PA28y stabilizes C1QBP or how C1QBP elicits its pro-tumourigenic role under PA28y OE.<br /> In most of the experiments, the authors have been dependent on the parallel changes in the expression of both the proteins to justify their stabilizing interaction. However, this approach is indirect at best and does not confirm the direct stabilizing effect of this interaction. IP experiments do not indicate direct interaction and have some quality issues. The upregulation of C1QBP might be indirect at best. It is quite possible that PA28y might be degrading some secondary protein/complex that is responsible for C1QBP expression. Since the core idea of the work is PA28y direct interaction with C1QBP stabilizing it, the same should be demonstrated in a more convincing manner.

Thank you very much for the important comments. Using AlphaFold 3, we found that interaction between PA28γ and C1QBP may depend on amino acids 1-167 and 1-213 (Revised Appendix Figure 1D-H), which was confirmed by our immunoprecipitation (Revised Figure 1I). In the future, we will use nuclear magnetic resonance spectroscopy to analyze protein-protein interaction between PA28γ and C1QBP and demonstrate it by GST pull down in vitro experiments.

In all of the assays, C1QBP has been detected as doublet. However, the expression pattern of the two bands varies depending on the experiment. In some cases, the upper band is intensely stained and in some the lower bands. Do C1QBP isoforms exist and are they differentially regulated depending on experiment conditions/tissue types?

Thank you very much for the important comments. We have rechecked the experimental results with two bands, which may have been caused by using polyclonal antibody of C1QBP (Abcam: ab101267). Therefore, we conducted the experiment with monoclonal antibody of C1QBP (Cell Signaling Technology: #6502) and replaced the corresponding images in revised figure (Revised Figure 1E and Revised Appendix Figure 3D).

Problems with the background of the work: Line 76. This statement is far-fetched. There are presently a number of works of literature that have dealt with the metabolic programming of OSCC including identification of specific metabolites. Moreover, beyond the estimation of OCR, the authors have not conducted any experiments related to metabolism. In the Introduction, the significance of this study and how it will extend our understanding of OSCC needs to be elaborated.

Thank you very much for the important comments. Based on your suggestion, we have revised the content and updated the references (“Introduction”, Paragraph 2, Line 13-17 and Paragraph 4, Line 5-8). In addition, we plan to conduct experiments to investigate the regulation of metabolism by PA28γ and C1QBP and update our data in the future.

The modified content is as follows:

“Current research on metabolic reprogramming in OSCC primarily focused on mechanism of glycolytic metabolism and metabolic shift from glycolysis to oxidative phosphorylation (OXPHOS) of oral squamous cell carcinoma, which lays the groundwork for novel therapeutic interventions to counteract OSCC (Chen et al., 2024; Zhang et al., 2020).”

“It is the first study to describe the undiscovered role of PA28γ in promoting the malignant progression of OSCC by elevating mitochondrial function, providing new clinical insights for the treatment of OSCC.”

Reviewer #2 (Public review):

Summary:

The authors tried to determine how PA28g functions in oral squamous cell carcinoma (OSCC) cells. They hypothesized it may act through metabolic reprogramming in the mitochondria.

Strengths:

They found that the genes of PA28g and C1QBP are in an overlapping interaction network after an analysis of a genome database. They also found that the two proteins interact in coimmunoprecipitation and pull-down assays using the lysate from OSCC cells with or without expression of the exogenous genes. They used truncated C1QBP proteins to map the interaction site to the N-terminal 167 residues of C1QBP protein. They observed the levels of the two proteins are positively correlated in the cells. They provided evidence for the colocalization of the two proteins in the mitochondria, the effect on mitochondrial form and function in vitro and in vivo OSCC models, and the correlation of the protein expression with the prognosis of cancer patients.

Weaknesses:

Many data sets are shown in figures that cannot be understood without more descriptions, either in the text or the legend, e.g., Figure 1A. Similarly, many abbreviations are not defined.

Thank you very much for the important comments. We have revised the descriptions in the legend to make it easier to understand.

Some of the pull-down and coimmunoprecipitation data do not support the conclusion about the PA28g-C1QBP interaction. For example, in Appendix Figure 1B the Flag-C1QBP was detected in the Myc beads pull-down when the protein was expressed in the 293T cells without the Myc-PA28g, suggesting that the pull-down was not due to the interaction of the C1QBP and PA28g proteins. In Appendix Figure 1C, assume the SFB stands for a biotin tag, then the SFB-PA28g should be detected in the cells expressing this protein after pull-down by streptavidin; however, it was not. The Western blot data in Figure 1E and many other figures must be quantified before any conclusions about the levels of proteins can be drawn.

Thank you very much for the meticulous review. We have rechecked the experimental results, and we made a mistake in the labeling of the image. Therefore, we have corrected it in the revised figure (Revised Appendix Figure 1B, C). In addition, we have conducted a quantitative analysis of gray values to confirm the results of western blot data are accurate by Image J software.

The immunoprecipitation method is flawed as it is described. The antigen (PA28g or C1QBP) should bind to the respective antibody that in turn should binds to Protein G beads. The resulting immunocomplex should end up in the pellet fraction after centrifugation and be analyzed further by Western blot for coprecipitates. However, the method in the Appendix states that the supernatant was used for the Western blot.

Thank you very much for the careful review. We have corrected it in the revised appendix file (“Supplemental Materials and Methods”, Part“Immunoprecipitation assay”, Line 4-6).

The modified content is as follows:

The sample was shaken on a horizontal shaker for 4 h, after which the deposit was collected for western blotting.

To conclude that PA28g stabilizes C1QBP through their physical interaction in the cells, one must show whether a protease inhibitor can substitute PA28q and prevent C1QBP degradation, and show whether a mutation that disrupts the PA28g-C1QBP interaction can reduce the stability of C1QBP. In Figure 1F, all cells expressed Myc-PA28g. Therefore, the conclusion that PA28g prevented C1QBP degradation cannot be reached. Instead, since more Myc-PA28g was detected in the cells expressing Flag-C1QBP compared to the cells not expressing this protein, a conclusion would be that the C1QBP stabilized the PA28g. Figure 1G is a quantification of Western blot data that should be shown.

Thank you very much for the meticulous review. We have rechecked the experimental results, and we made a mistake in the labeling of the image. Therefore, we have corrected it in the revised figure. Compared with the control group, the presence of Myc-PA28γ significantly increased the expression level of Flag-C1QBP (Revised Figure 1F). Gray value analysis showed that in cells transfected with Myc-PA28γ, the decay rate of Flag-C1QBP was significantly slower than that of the control group (Revised Figure 1G), suggesting that PA28γ can delay the protein degradation of C1QBP and stabilize its protein level. This indicates that an increase in the level of PA28γ protein can significantly enhance the expression level of C1QBP protein, while PA28γ can slow down the degradation rate of C1QBP and improve its stability. In addition, we plan to conduct experiments to investigate the effects of protease inhibitors and PA28γ mutants on the stability of C1QBP and update our data in the future.

The binding site for PA28g in C1QBP was mapped to the N-terminal 167 residues using truncated proteins. One caveat would be that some truncated proteins did not fold correctly in the absence of the sequence that was removed. Thus, the C-terminal region of the C1QBP with residues 168-283 may still bind to the PA29g in the context of full-length protein. In Figure 1I, more Flag-C1QBP 1-167 was pulled down by Myc-PA28g than the full-length protein or the Flag-C1QBP 1-213. Why?

Thank you very much for the important comments. Immunoprecipitation is a qualitative experiment. Using AlphaFold 3, we found that interaction between PA28γ and C1QBP may depend on amino acids 1-167 and 1-213 (Revised Appendix Figure 1D-H), which was confirmed by our immunoprecipitation (Revised Figure 1I).

The interaction site in PA28g for C1QBP was not mapped, which prevents further analysis of the interaction. Also, if the interaction domain can be determined, structural modeling of the complex would be feasible using AlphaFold2 or other programs. Then, it is possible to test point mutations that may disrupt the interaction and if so, the functional effect.

Thank you very much for the important comments. Based on your suggestion, we have added relevant content to the revised appendix figure. (Revised Appendix Figure 1D-H).

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

(1) There are a lot of typos in the figure and manuscript that need to be addressed.

Thank you very much for the important comments. We have corrected the typos in the revised figure and manuscript.

(2) Figure 1A: The amount of protein that has been immunoprecipitated is more than the actual amount present in the lysate. The authors should calculate the efficiency of the precipitation to support their results.

Thank you very much for the important comments. Immunoprecipitation is a qualitative experiment. Moreover, it can enrich specific proteins and their binding partners, increase their concentration in the sample, and thus improve the sensitivity of detection.

(3) Figure 1D: The relative expression levels of C1QBP look similar in almost all cell lines except for HN12. It seems that the relation of PA28y with C1QBP is more of a cell type-specific effect. It would be better if the blots were quantified, and the differences were statistically determined.

Thank you very much for the important comments. We have conducted a quantitative analysis of gray values to confirm the results of western blot data are accurate by Image J software.

(4) Figure 1E: How do the authors quantify the expression of the protein in absolute terms? From the methods, it is understood that the flag-tagged construct is stably expressed. Under such conditions, how the authors observed the variable expression of the protein should be elaborated.

Thank you very much for the important comments. We transfected Flag-PA28γ plasmids at 0ug, 0.5ug, 1ug, and 2ug in 293T cells. After collecting the protein for Western Blot, we found that the protein expression of Flag-PA28γ gradually increased. Moreover, the increased protein expression of C1QBP is consistent with the expression of Flag-PA28γ, which indicated a dose-dependent relationship between the two proteins.

(5) Figures 1F, G: The data does not correlate with the arguments presented in the text. The authors propose that interaction with PA28y increases the stability of C1QBP. However, the experiment lacks appropriate controls. Ideally, the expression of C1QBP should be tested in the presence and absence of PA28y. Moreover, the observed difference in expression between lanes 1-4 and 5-8 for myc-PA28y needs to be explained. Are the samples from different sources with variable PA28y expression? Figure 1G quantification for C1QBP does not correlate with the figure presented in F since the expression of the protein in the first four lanes is undetectable.

Thank you very much for the meticulous review. We have rechecked the experimental results, and we made a mistake in the labeling of the image. Therefore, we have corrected it in the revised figure. Compared with the control group, the presence of Myc-PA28γ significantly increased the expression level of Flag-C1QBP (Revised Figure 1F). Gray value analysis showed that in cells transfected with Myc-PA28γ, the decay rate of Flag-C1QBP was significantly slower than that of the control group (Revised Figure 1G), suggesting that PA28γ can delay the protein degradation of C1QBP and stabilize its protein level. This indicates that an increase in the level of PA28γ protein can significantly enhance the expression level of C1QBP protein, while PA28γ can slow down the degradation rate of C1QBP and improve its stability. In addition, we plan to conduct experiments to investigate the effects of protease inhibitors and PA28γ mutants on the stability of C1QBP and update our data in the future.

(6) Appendix Figure 1B: Lane 1 does not express Myc-tagged protein but pull-down has been performed using Myc beads. Then how come flag-C1qbp is getting pulled down in lane 1 if there is no PA28y? This indicates a non-specific interaction of C1qbp with the substrata under the experimental conditions used. Similarly, in Figure 1C SFB-PA28y is expressed in both lanes but is reflected only in lane 2 and not in lane 1 even when pull-down is being performed using SFB beads, again reflecting the non-specificity of the interactions shown through immunoprecipitated.

Thank you very much for the meticulous review. We have rechecked the experimental results, and we made a mistake in the labeling of the image. Therefore, we have corrected it in the revised figure (Revised Appendix Figure 1B, C).

(7) Figure 2A: Figure 2A the co-localization of P28y with C1QBP in mitochondria is not very convincing. The authors are urged to provide high-resolution images for the same along with quantification of co-localization coefficients.

Thank you very much for the important comments. We plan to obtain high-resolution images of co-localization of PA28γ with C1QBP in mitochondria and add the quantification analysis. We will update our data in the future.

(8) Figure 2C: Mitochondria dynamics is an interplay of multiple factors. From the images, it seems that PA28y OE elevates mitochondria biogenesis in general which is having an umbrella effect on mitochondria fusion/fission and OCR. Images also do not convincingly indicate changes in mitochondrial length. The role of PA28y on mitochondria dynamics requires further justification. However, the presented data does not underline whether the changes in mitochondria behaviour are a consequence of PA28y and C1QBP interaction. Correlating higher mitochondria respiration with ROS generation is a far-fetched conclusion since, at present, there are multiple reports that suggest otherwise.

Thank you very much for the important comments. We plan to knock out the interaction regions between PA28γ and C1QBP (like amino acids 1-167 and 1-213) to confirm whether PA28γ affects mitochondrial function through C1QBP and update our data in the future.

(9) Line 157: The presented data does not substantiate the claims made that Pa28y regulates mitochondrial function through C1QBP.

Thank you very much for the important comments. Based on your suggestion, we have made some modifications to make it more accurate (“Results”, Part “PA28γ and C1QBP colocalize in mitochondria and affect mitochondrial functions”, Paragraph 3, Line 1-2).

The modified content is as follows:

“Collectively, these data suggest that PA28γ, which co-localizes with C1QBP in mitochondria, may involve in regulating mitochondrial morphology and function.”

(10) Line 159: From the past data it is not very clear how PA28y upregulates C1QBP, hence the statement is not well supported. The presented data indicates the presence of a functional association between the two proteins.

Thank you very much for the important comments. We detected the expression of C1QBP in two PA28γ-overexpressing OSCC cells (UM1 and 4MOSC2) and found an increase in C1QBP expression (Revised Figure 4B). Based on the results of the protein levels of the mitochondrial respiratory chain complex and other mitochondrial functional proteins, we believe that PA28γ regulates mitochondrial function by upregulating C1QBP.

(11) Figure 4A, B: Given the mitochondrial role of C1QBP, the lesser levels of mitochondrial proteins upon C1QBP silencing are expected. Does it get phenocopied upon PA28y silencing? Similarly, all the subsequent mitochondrial phenotypes in D should be seen in a PA28y-depleted background.

Thank you very much for the important comments. We plan to detect the mitochondrial protein expressions and OCRs of PA28γ-silenced OSCC cells. We will update our data in the future.

(12) Line 198: The presented data do indicate a functional association between these two proteins but it does not provide a solid evidence for the same.

Thank you very much for the important comments. Based on your suggestion, we have made some modifications to make it more accurate (“Discussion”, Paragraph 1, Line 9-10).

The modified content is as follows:

“Excitingly, we found the evidence that PA28γ interacts with and stabilizes C1QBP.”

(13) Line 218-220: In this work, the authors highlight the non-degradome role of PA28y and hence, this fact should be treated appropriately in discussion in line with the presented data.

Thank you very much for the important comments. Based on your suggestion, we have added relevant content to the revised manuscript (“Discussion”, Paragraph 2, Line 16-19).

The modified content is as follows:

“In addition, PA28γ can also play as a non-degradome role on tumor angiogenesis. For example, PA28γ can regulate the activation of NF-κB to promote the secretion of IL-6 and CCL2 in OSCC cells, thus promoting the angiogenesis of endothelial cells ( S. Liu et al., 2018).”

(14) Line 236-240: Although the authors' statement on organ heterogeneity being the cause for getting the contrasting result is justifiable but here there is no direct evidence of PA28y involvement in regulation of OXPHOS and its impact on cellular metabolism (glycolysis, metabolic signalling, etc).

Thank you very much for the important comments. Based on your suggestion, we have made some modifications to make it more accurate (“Discussion”, Paragraph 3, Line 7-9).

The modified content is as follows:

“Therefore, PA28γ's regulation of OXPHOS may impact cellular energy metabolism.”

(15) Line 249: No conclusive data supporting this statement.

Thank you very much for the important comments. Based on your suggestion, we have made some modifications to make it more accurate (“Discussion”, Paragraph 5, Line 1-3).

The modified content is as follows:

“Furthermore, our study reveals that PA28γ can regulate C1QBP and influence mitochondrial morphology and function by enhancing the expression of OPA1, MFN1, MFN2 and the mitochondrial respiratory complex.”

Reviewer #2 (Recommendations for the authors):

(1) The images shown in Figure 2A need to be quantified before the conclusion about the mitochondrial colocalization of the two proteins can be drawn. In Figure 2B and Appendix Figure 2A, the mitochondrial vacuoles and ridge should be indicated for general readers, and quantification should be performed before the conclusion is drawn.

Thank you very much for the important comments. We will update our data in the future.

(2) The OCR data from two cell lines are shown in Figure 2E and F. Which is which? The sentence, "The results indicated ... compared to control cells" in lines 130-132, was confusing; perhaps, it would be clear if "were significantly greater" could be deleted.

Thank you very much for the important comments. We have re-labeled the Figure 2E and F to make it clearly (Revised Figure 2E, F). Based on your suggestion, we have deleted the words in revised manuscript. (“Results”, Part “PA28γ and C1QBP colocalize in mitochondria and affect mitochondrial functions”, Paragraph 1, Line 9-11).

The modified content is as follows:

“The results indicated significantly higher basal respiration, maximal OCRs and ATP production in PA28γ-overexpressing cells compared to control cells (Fig. 2G-I and Appendix Fig. 2B-D).”

(3) Figures 4E-H show the migration, invasive, and proliferation capabilities of the cells. Which for which?

Thank you very much for the important comments. We have re-labeled the Figure 4F-H to make it clearly (Revised Figure 4F-H).

(4) In the Discussion, lines 198-201, it states that "C1QBP enhances ... function of OPA1, MNF1, MFN2..." What is the evidence? In lines 222-224, it says that "the binding sites ... may mask the specific ... modification sites". Please justify. In lines 253-254, "fuse" and fuses" are misleading, Did the authors mean "localize" and "localizes"?

Thank you very much for the important comments. Based on your suggestion, we have made some modifications to make it more accurate (“Discussion”, Paragraph 1, Line 9-13, Paragraph 2, Line 20-23, and Paragraph 5, Line 3-6).

The modified content is as follows:

“Excitingly, we found the evidence that PA28γ interacts with and stabilizes C1QBP. We speculate that aberrantly accumulated C1QBP enhances the function of mitochondrial OXPHOS and leads to the production of additional ATP and ROS by activating the expression and function of OPA1, MNF1, MFN2 and mitochondrial respiratory chain complex proteins.”

“Our study reveals that PA28γ interacts with C1QBP and stabilizes C1QBP at the protein level. Therefore, we speculate that the binding sites of PA28γ and C1QBP may mask the specific post-translational modification sites of C1QBP and inhibit its degradation.”

“Mitochondrial fusion, crucial for oxidative metabolism and cell proliferation, is regulated by MFN1, MFN2, and OPA1. The first two fuse with the outer mitochondrial membrane, while the last fuses with the inner mitochondrial membrane (Westermann, 2010).”

(5) Figure 6 was not referred to in the text. In this figure, PA28g and C1QBP are located in the inner membrane and matrix. Has this been determined? What is the blue ovals that are intermediaries of PA28g/C1QBP and OPA1/MFN1/MFN2?

Thank you very much for the important comments. According to our immunofluorescence assay (Figure 2A), PA28γ is in both the nucleus and cytoplasm. A recent study has demonstrated that PA28γ can shuttle between the nucleus and cytoplasm, participating in various cellular processes. Furthermore, GeneCard information indicates that the subcellular localization of PA28γ includes the nucleus, cytoplasm and mitochondria (Author response image 1). In this article, we mainly focus on the functions of PA28γ and C1QBP located in the cytoplasm. Therefore, figure 6 mainly displays PA28γ and C1QBP in the cytoplasm. Based on your suggestion, we have made some modifications to make it more accurate in revised figure (Revised Figure 6).

Author response image 1.

eLife Assessment

This study provides valuable insights into our understanding of the development of the enteric nervous system. The authors use genetically engineered mice to study the behavior of stem cells in organizing the enteric nervous system and the secreted signals that regulate these cells. The study rests on a degree of incomplete evidence since the characterization of some of the mouse resources is not complete in the current version.

Reviewer #1 (Public Review):

Summary:

The manuscript by Poltavski and colleagues describes the discovery of previously unreported enteric neural crest-derived cells (ENCDC) which are marked by Pax2 and originating from the Placodes. By creating multiple conditional mouse mutants, the authors demonstrate these cells are a distinct population from the previously reported ENCDCs which originate from the Vagal neural crest cells and express Wnt1.

These Pax2-positive ENCDCs are affected due to the loss of both Ret and Ednrb highlighting that these cells are also ultimately part of the canonical processes governing ENCDC and enteric nervous system (ENS) development. The authors also make explant cultures from the mouse GI tract to detect how Ednrb signaling is important for Ret signaling pathways in these cells and rediscovers the interactions between these 2 pathways. One important observation the authors make is that CGRP-positive neurons in the adult distal colon seem to be primarily derived from these Pax2-positive ENCDCs, which are significantly reduced in the Ednrb mutants, thus highlighting the role of Ednrb in maintaining this neuronal type.

Comments on latest version:

Author response: We disagree that the datasets from previous studies provide additional insights that are relevant to the current study. It must be appreciated that Wnt1Cre and Pax2Cre are genetic lineage tracers and that migratory ENS progenitor cells labeled with these reagents do not maintain expression of Wnt1 and Pax2 mRNA or protein. The Wnt1 and Pax2 genes are only transiently expressed within their distinct regions of the ectoderm, and their expression turns off as cells delaminate and begin migration. Thus, Pax2Cre-labeled ENS progenitor cells are not Pax2-positive thereafter. The single cell RNA-Seq studies suggested by the reviewer were collected from older embryos and postnatal mice, and do not represent the E10.5-E11.5 period that accounts for genesis of Ret-mediated and Ednrb-mediated Hirschsprung disease pathology. Even with the most recent work by Zhou et al (Dev Cell, 2024) that included E10.5 cells, this analysis only evaluated neural crest-derived Sox10Cre lineage cells, which does not include the placode-derived Pax2Cre lineage (as we show explicitly in Fig. 2-figure supplement 2). Consequently, it would not be possible to find the "Pax2-positive cells" in these datasets. Performing a new transcriptomic analysis by isolating Pax2Cre-lineage and Wnt1Cre-lineage cells at the appropriate developmental time points could be the basis of future studies, but we think these are beyond the scope of the present paper.

Reviewer comment: Since these cells are a completely new discovery, additional validation would be beneficial. Whole early GI tract datasets are available, such as human 6-week fetal gut data (PMID: 29802404) and whole mouse embryo studies spanning development that include ENS (PMID: 38355799). If the authors believe that none of these existing datasets can detect these cells in their developmental state and that targeted cell studies with specific Cre drivers would be required, they should make this explicitly clear.

A key advantage of discovering a new cell type, particularly in the relatively understudied field of ENS, is the opportunity for the broader community to leverage this finding to inform their own research. If these cells are absent from current datasets, even those covering the whole GI tract, this should be clearly communicated.

I aim to support the authors here. New discoveries in science require robust validation to enhance their impact. The authors have generated an important reagent with great potential for broader use, and addressing these straightforward requests would strengthen the study and make it more valuable to the scientific community.

Author response: The observation that human mutations in RET and EDNRB both cause Hirschsprung disease is decades old, and of course numerous studies in human, mouse, and cells have addressed the relation between the two signaling pathways. We did not mean to imply that we were the first to discover that Ret and Ednrb signaling pathways interact. The reviewer cites a number of papers all from the Chakravarti lab that address this phenomenon; while these are a valuable contribution to the field, there is still more to be learned. The model elaborated in PMID: 31313802, in which Ret and Ednrb are both enmeshed in a common gene regulatory network, does not readily explain why each has a different phenotypic manifestation and doesn't take into account the importance of the placodal lineage. The main new contributions of our paper are the existence of a new cell lineage that contributes to the ENS, and that the placodal and neural crest lineages utilize Ret and Ednrb signaling differently. The clarification of how these elements are differentially used by the two lineages explains long-segment and short-segment Hirschsprung disease (Ret and Ednrb mutants, respectively) far better than in past studies. The reviewer unfortunately dismisses these insights and seems to feel that a biochemical exploration of one specific component of the signaling interaction (Y1015 phosphorylation) would be more relevant. This should be the basis of future studies and are beyond the scope of the new findings reported in the present paper

Reviewer comment: The authors completely miss the point. There is no association between phenotypic severity (L-HSCR, S-HSCR, or TCA) and mutations in a given gene in HSCR. EDNRB, for example, has a syndromic association with Waardenburg-Shah syndrome (WS4-A), which includes pigmentation anomalies due to EDNRB expression in neural crest cells that give rise to pigment cells.

The authors' discovery reinforces the current paradigm that nearly all HSCR is mediated by mutations in genes within the GRN, accounting for 72% of the population attributable risk. This is valuable; reinforcing established paradigms with new data is crucial, and the authors should appreciate the significance of this contribution.

The discovery of the signaling interaction is particularly important, as it offers a potential explanation for disease severity and provides a basis for classifying patients in future sequencing studies. It is surprising that the authors seem reluctant to highlight this novel finding, as it could greatly benefit future research, including the development of specific mouse mutants and advancing human genetics studies.

Author response: The reviewer overlooked that one of the review articles that we cited (Chen, Hsu, & Hung, 2020) has a dedicated paragraph for RET (section 3.14), which summarizes the work by Barheri-Yarmand et al (PMID: 25795775) which is the very paper noted by the reviewer in the comment above. The reviewer also somewhat misstated the results of the Barheri-Yarmand et al study. By immunostaining, this paper showed nuclear localization of endogenous Ret, albeit a version of Ret with a disease-associated mutation that makes it constitutively active by constitutive autophosphorylation. Nonetheless, this was endogenous Ret. The paper also used overexpression of GFP-tagged RET in HEK293 cells to show that wildtype RET can behave in a similar manner, at least under these circumstances. Our point is simply that Ret (and other receptor tyrosine kinases) can be found in the nucleus in certain biological contexts, and our observations are consistent with this precedent. The reviewer also suggests a biochemical follow-up analysis related to this observation, which we agree would be of interest. Such an investigation however is beyond the scope of the present study.

Reviewer comment: As the authors themselves now highlight from the cited paper that any evidence of RET entering the nucleus is of a mutant RET protein, How does this align with their discovery for wildtype protein?

This finding of nuclear localization of RET is both intriguing and unprecedented. Despite extensive biochemical studies on RET, given its role as an oncogene, this feature has not been identified before. If validated, this discovery could significantly advance the field and improve interpretation of future studies. I reiterate my previous point: a novel finding that challenges the current paradigm requires additional evidence.

Reviewer #2 (Public review):

Summary:

This manuscript by Poltavski and colleagues explores the relative contributions of Pax2- and Wnt1- lineage derived cells in the enteric nervous system (ENS) and how they are each affected by disruptions in Ret and Endrb signaling. The current understanding of ENS development in mice is that vagal neural crest progenitors derived from a Wnt1+ lineage migrate into and colonize the developing gut. The sacral neural crest was thought to make a small contribution to the hindgut in addition but recent work has questioned that contribution and shown that the ENS is entirely populated by vagal crest (PMID: 38452824). GDNF-Ret and Endothelin3-Ednrb signaling are both known to be essential for normal ENS development and loss of function mutations are associated with a congenital disorder called Hirschsprung's disease. The transcription factor Pax2 has been studied in CNS and cranial placode development but has not been previously implicated in ENS development. In this work, the authors begin with the unexpected observation that conditional knockout of Ednrb in Pax2-expressing cells causes a similar aganglionosis, growth retardation, and obstructed defecation as conditional knockout of Ednrb in Wnt1-expressing cells. The investigators then use the Pax2 and Wnt1 Cre transgenic lines to lineage-trace ENS derivatives and assess the effects of loss of Ret or Ednrb during embryonic development in these lineages. Finally, they use explants from the corresponding embryos to examine the effects of GDNF on progenitor outgrowth and differentiation.

Strengths:

- The manuscript is overall very well illustrated with high resolution images and figures. Extensive data are presented.

- The identification of Pax2 expression as a lineage marker that distinguishes a subset of cells in the ENS that may be distinct from cells derived from Wnt1+ progenitors is an interesting new observation that challenges current understanding of ENS development

- Pax2 has not been previously implicated in ENS development - this manuscript does not directly test that role but hints at the possibility

- Interrogation of two distinct signaling pathways involved in ENS development and their relative effects on the two purported lineages

Weaknesses:

- The major challenge with interpreting this work is the use of two transgenic lines, Wnt1-Cre and Pax2-Cre, which are not well characterized in terms of fidelity to native gene expression and recombination efficiency in the ENS. If 100% of cells that express Wnt1 do not express Cre or if the Pax2 transgene is expressed in cells that do not normally express Pax2, then these observations would have very different interpretations and would not support the conclusions made. The two lineages are never compared in the same embryo, which also makes it difficult to assess relative contributions and renders the evidence more circumstantial than definitive.

- Visualization of the Pax2-Cre and Wnt-1Cre induced recombination in cross-sections at postnatal ages would help with data interpretation. If there is recombination evident in the mesenchyme, this would particularly alter interpretation of Ednrb mutant experiments, since that pathway has been shown to alter gut mesenchyme and ECM, which could indirectly alter ENS colonization.

- The data on distinct lineages in Fig 3 is somewhat weak and the description in the Results section tends to over-interpretation. For example, "A minimum number (approx. 3%) of CGRP+ neurons were labeled by Wnt1Cre ... which indicates that Wnt1Cre-derived cells have little or no commitment to a mechanosensory fate in the distal colon." The data panel in Fig 3f shows that most of the CGRP-IR cells in Wnt1-Cre-Tomato mice are tdTomato+ though their tdTomato fluorescence is less intense than in neighboring smaller, likely glial cells. This suggests that CGRP+/Tomato+ neurons were likely undercounted. IHC for tdTomato to ensure detection of low levels of Tomato expression and quantification of observations would strengthen the authors' claim. CGRP+ enteric neurons have been visualized and functionally described by several investigators in the field using Wnt1-Cre-GCaMP mice, which also challenges the authors' conclusions. Finally, quantification of CGRP+ enteric neurons by measuring CGRP mucosal fiber immunoreactivity is not accurate because it would reflect both ENS CGRP-expressing neurons and visceral afferents from DRG. Moreover, it is not known if all CGRP+ enteric neurons project to the mucosa or if all mucosal-projecting neurons are mechanosensory. Finally, most of the signal seems to be non-specific background staining in the mucosa and quantification of mucosal signal in this context does not seem meaningful.

- No consideration of glia - are these derived from both lineages?

- No discussion of how these observations may fit in with recent work that suggests a mesenchymal contribution of enteric neurons (PMID: 38108810)

- Phospho-RET staining in Figure 7 is difficult to discern and interpret with high background. Positive and negative controls would strengthen these data.

Comments on revised version:

The authors have responded to the weaknesses identified above. Based on my own assessment of the revised manuscript, my assessment is unchanged because the manuscript is largely unchanged.

Author response:

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

Reviewer #1 (Public Review):

The manuscript by Poltavski and colleagues describes the discovery of previously unreported enteric neural crestderived cells (ENCDC) which are marked by Pax2 and originating from the Placodes. By creating multiple conditional mouse mutants, the authors demonstrate these cells are a distinct population from the previously reported ENCDCs which originate from the Vagal neural crest cells and express Wnt1.

These Pax2-positive ENCDCs are affected due to the loss of both Ret and Ednrb highlighting that these cells are also ultimately part of the canonical processes governing ENCDC and enteric nervous system (ENS) development. The authors also make explant cultures from the mouse GI tract to detect how Ednrb signaling is important for Ret signaling pathways in these cells and rediscovers the interactions between these 2 pathways. One important observation the authors make is that CGRP-positive neurons in the adult distal colon seem to be primarily derived from these Pax2-positive ENCDCs, which are significantly reduced in the Ednrb mutants, thus highlighting the role of Ednrb in maintaining this neuronal type.

I appreciate the amount of work the authors have put into generating the mouse models to detect these cells, but there isn't any new insight on either the nature of ENCDC development or the role of Ret and Ednrb. Also, there are sophisticated single-cell genomics methods to detect rare cell type/states these days and the authors should either employ some of those themselves in these mouse models or look at extensively publicly available single-cell datasets of the developing wildtype and mutant mouse and human ENS to map out the global transcriptional profile of these cells. A more detailed analysis of these Pax2-positive cells would be really helpful to both the ENS community as well as researchers studying gut motility disorders.

We would like to point out that the reviewer’s comments in both Public Review and in some cases reiterated in Recommendations for the Authors are rooted in several misunderstandings. The reviewer writes “Pax2-positive ENCDCs”, as if the Pax2 lineage (properly, the Pax2Cre-labeled lineage) of the ENS is a subset of neural crest, and states that “there isn’t any new insight” from our study on ENS development. Our conclusion is quite different, that the Pax2Cre lineage (placode-derived) is distinct from the neural crest-derived cell lineage. The reviewer may not have appreciated that our study establishes a fundamental reinterpretation of the very long-standing dogma that the ENS is derived solely from neural crest. We believe that finding and characterizing the unique contribution of an independent cell lineage to the ENS provides critical new perspectives into ENS development and the etiology of Hirschsprung disease. One feature of the Pax2Cre (placodal) lineage is as the source of CGRP-positive mechanosensory neurons in the colon (as the reviewer mentioned), but this is one feature of the larger conceptual discovery of the existence of a separate lineage contribution to the ENS, not the most important observation in and of itself.

The reviewer continues by saying that we “rediscovered” the interaction between Ednrb and Ret in ENS development. In our study we show that the two lineages (placode-derived and neural crest-derived) employ Ednrb and Ret signaling in distinct ways. This isn’t simply rediscovery, this is new insight. To the extent that both lineages utilize both signaling axes (albeit with mechanistic differences) is a primary reason why the unique placodal lineage contribution to the ENS remained unsuspected until now. We have revised the text to make these points more clear in our revised manuscript.

The reviewer also suggests single cell genomic methods, which is addressed below in our response to the reviewer’s first recommendation.

Reviewer #2 (Public Review):

This manuscript by Poltavski and colleagues explores the relative contributions of Pax2- and Wnt1- lineagederived cells in the enteric nervous system (ENS) and how they are each affected by disruptions in Ret and Endrb signaling. The current understanding of ENS development in mice is that vagal neural crest progenitors derived from a Wnt1+ lineage migrate into and colonize the developing gut. The sacral neural crest was thought to make a small contribution to the hindgut in addition but recent work has questioned that contribution and shown that the ENS is entirely populated by the vagal crest (PMID: 38452824). GDNF-Ret and Endothelin3-Ednrb signaling are both known to be essential for normal ENS development and loss of function mutations are associated with a congenital disorder called Hirschsprung's disease. The transcription factor Pax2 has been studied in CNS and cranial placode development but has not been previously implicated in ENS development. In this work, the authors begin with the unexpected observation that conditional knockout of Ednrb in Pax2-expressing cells causes a similar aganglionosis, growth retardation, and obstructed defecation as conditional knockout of Ednrb in Wnt1-expressing cells. The investigators then use the Pax2 and Wnt1 Cre transgenic lines to lineage-trace ENS derivatives and assess the effects of loss of Ret or Ednrb during embryonic development in these lineages. Finally, they use explants from the corresponding embryos to examine the effects of GDNF on progenitor outgrowth and differentiation.

Strengths:

-  The manuscript is overall very well illustrated with high-resolution images and figures. Extensive data are presented.

-  The identification of Pax2 expression as a lineage marker that distinguishes a subset of cells in the ENS that may be distinct from cells derived from Wnt1+ progenitors is an interesting new observation that challenges the current understanding of ENS development.

-  Pax2 has not been previously implicated in ENS development - this manuscript does not directly test that role but hints at the possibility.

-  Interrogation of two distinct signaling pathways involved in ENS development and their relative effects on the two purported lineages.

The reviewer provided a succinct and accurate summary of our analysis. We correct just the one statement that the ENS is entirely populated by vagal crest. The paper cited by the reviewer (PMID: 38452824) used Wnt1DreERT2 to lineage label the NC population, so of course only looked at neural crest (comparing vagal vs. sacral NC). The advance in our study is to newly document the independent contribution of the placodal lineage.

Weaknesses:

-  The major challenge with interpreting this work is the use of two transgenic lines, rather than knock-ins, Wnt1Cre and Pax2-Cre, which are not well characterized in terms of fidelity to native gene expression and recombination efficiency in the ENS. If 100% of cells that express Wnt1 do not express this transgene or if the Pax2 transgene is expressed in cells that do not normally express Pax2, then these observations would have very different interpretations and not support the conclusions made. The two lineages are never compared in the same embryo, which also makes it difficult to assess relative contributions and renders the evidence more circumstantial than definitive.

We do not agree that the Cre lines being transgenics rather than knock-ins changes the utility of these reagents or the interpretation of the results; there are also potential problems with knock-in alleles. Wnt1Cre has been in use for 25 years as a pan-neural crest lineage cell marker with exceptional efficiency and specificity (including numerous studies of the ENS), so we disagree that it is not well characterized. Pax2Cre of course has not previously been studied in the ENS, but it has been broadly used in other contexts (e.g., craniofacial, kidney). That said, and as noted in our original manuscript, we are aware that an issue of this study is the uniqueness of the recombination domains of the two Cre lines.  As we wrote, Wnt1Cre and Pax2Cre cannot be combined into the same embryo because they are both Cre lines, and we do not have a suitable nonCre recombinase line to substitute for either. Instead, we demonstrate that the two lines recombine in distinct territories of the early embryonic ectoderm, and that the two lineages thus labeled are distinct in marker expression at the initial onset of their delamination, utilize Edn3-Ednrb and GDNF-Ret in distinct ways during their migration to the hindgut, and contribute to different terminal cell fates in the colon. We think this evidence of the distinct nature of the two lineages from start to finish is compelling rather than merely circumstantial.

-  Visualization of the Pax2-Cre and Wnt-1Cre induced recombination in cross-sections at postnatal ages would help with data interpretation. If there is recombination induced in the mesenchyme, this would particularly alter the interpretation of Ednrb mutant experiments, since that pathway has been shown to alter gut mesenchyme and ECM, which could indirectly alter ENS colonization.

We have several thoughts about this comment. First, we are uncertain why postnatal analysis would be informative, as ENS colonization occurs (or fails to occur in mutants) during embryogenesis. The reviewer might be thinking of a juvenile stage additional contribution to the ENS, which is addressed below (responses to Recommendations for the Authors) but as we discuss there is not relevant to our analysis. Second, we did examine recombination in the distal hindgut at E12.5 during ENS colonization (Fig. 1f and 1h) and did not see overlap between either Cre recombination domain and Edn3 mRNA expression (which is expressed by the nonENS mesenchyme). Furthermore, Ednrb is not expressed in the gut mesenchyme during ENS colonization (Fig. 7figure supplement 1), thus ectopic mesenchymal Cre expression, if any, by either line would have no impact in Cre/Ednrb mutants. Lastly, the reviewer’s idea could have been a plausible hypothesis at the onset of the project, but here we show positive evidence for a different explanation. We do not rigorously exclude the reviewer’s hypothesis, nor other theoretically possible models, but we think we have provided a strong case to support the direct involvement of Ret and Ednrb in ENS progenitors rather than in surrounding non-neural mesenchyme.

-  No consideration of glia - are these derived from both lineages?

To properly address this question would require new reagents and analyses that we have not yet initiated. While an interesting question from a developmental biology standpoint, we don’t think that this investigation would change any of the interpretations that we make in the manuscript.

-  No discussion of how these observations may fit in with recent work that suggests a mesenchymal contribution of enteric neurons (PMID: 38108810).

The recent paper cited by the reviewer is very explicit in describing this mesenchymal contribution to the ENS as occurring after postnatal day P11. Other than the terminal Hirschsprung phenotype, all of our analysis of cell lineage migration and fate and colonic aganglionosis was conducted at embryonic or early (P9) postnatal stages. We therefore do not see a relation of our work to this study. In light of this paper, however, we do agree that it would be worthwhile in a future study to explore Wnt1Cre and Pax2Cre lineage dynamics in the ENS of older mice.

Reviewer #1 (Recommendations For The Authors):

(1) The authors should reanalyze multiple single-cell RNA-seq datasets available now, to see if these cells are detected in those studies and then look at the global transcriptional profile of these Pax2-positive cells compared to the other vagal neural crest-derived ENCDCs. Some of these datasets can be found here - PMIDs: 33288908, 37585461, and https://www.gutcellatlas.org/.

We disagree that the datasets from previous studies provide additional insights that are relevant to the current study. It must be appreciated that Wnt1Cre and Pax2Cre are genetic lineage tracers and that migratory ENS progenitor cells labeled with these reagents do not maintain expression of Wnt1 and Pax2 mRNA or protein. The Wnt1 and Pax2 genes are only transiently expressed within their distinct regions of the ectoderm, and their expression turns off as cells delaminate and begin migration. Thus, Pax2Cre-labeled ENS progenitor cells are not Pax2-positive thereafter. The single cell RNA-Seq studies suggested by the reviewer were collected from older embryos and postnatal mice, and do not represent the E10.5-E11.5 period that accounts for genesis of Ret-mediated and Ednrb-mediated Hirschsprung disease pathology. Even with the most recent work by Zhou et al (Dev Cell, 2024) that included E10.5 cells, this analysis only evaluated neural crest-derived Sox10Cre lineage cells, which does not include the placode-derived Pax2Cre lineage (as we show explicitly in Fig. 2-figure supplement 2).  Consequently, it would not be possible to find the “Pax2-positive cells” in these datasets. Performing a new transcriptomic analysis by isolating Pax2Cre-lineage and Wnt1Cre-lineage cells at the appropriate developmental time points could be the basis of future studies, but we think these are beyond the scope of the present paper. 

(2) Even in their current quantification method of using immunofluorescent cells in a microscopic field, the authors count very few cells. The quantification in Figures 2v-2z is only from 4 embryos and is in the hundreds. This leads to misrepresentation of cell numbers and is best reflected in Figure 2x, where Wnt1Cre/Ret GI tracts have 0 Ret +ve cells, which we now know is not true even in ubiquitous Ret null embryos, where Ret null cells are detected as late as E14.5 (PMID 37585461)

Because of the reviewer’s comment, we recognize that the specific detail about cell numbers wasn’t properly written. We didn’t count a few hundred cells total, it was a few hundred cells per embryo. Exact numbers are provided in the revised figure legend where “cells/embryo” is now explicitly stated. Multiplied by the number of embryos, this means that we evaluated approx. 1000 total cells per genotype and time point in cases where Ret+ and/or GFP+ (lineage+) cells were found. The total absence of such cells in Wnt1Cre/Ret mutants is a rigorous conclusion. Our results do not misrepresent nor contradict the study by Vincent et al (PMID 37585461). Our analyses were performed on gut tissue isolated at E10.5 and E11.5 stages, which is long before Schwann cell precursors (SCPs, the primary focus of the Vincent et al study) colonize the gut (E14.5; Uesaka et al, 2015. PMID: 26156989). Indeed, as the reviewer notes, SCPs migrate into the gut in a Retindependent manner. For being at a much earlier time point, our focus is on the cranial ectoderm sources of ENS progenitors. We have adjusted the text associated with Fig. 2 to make this more clear.

(3) There are multiple sections in the manuscript that rehash already known facts, like the whole section about Wnt1 conditional Ret null mice which show failure of migration of ENCDCs. This has been shown multiple times and doesn't add anything to the author's story.

We think this comment stems from the reviewer’s perception that the Pax2Cre lineage is a subset of neural crest. The Wnt1Cre data (including Ret-deficient and Ednrb-deficient embryos) presented in the manuscript are not intended to rehash what is already known but to establish important similarities and differences between the newly identified placode-derived and the well-established neural crest-derived ENS progenitor cells. In light of the reviewer’s suggestion #8 below, to move the Wnt1Cre lineage analysis to a supplement, this information remains in the main text to provide proper comparison to the Pax2Cre-lineage profile. We think we were fair in the text to the legacy of work on neural crest and ENS development and were explicit in using our Wnt1Cre analysis to compare to the Pax2Cre lineage. Finally, we point out that our analysis was conducted on a different genetic background (outbred ICR) compared to previous studies, and there are strain-specific differences in Hirschsprung-associated lethality between our background and previous studies, so it was not impossible that the behavior of the neural crest cell lineage in the ICR background could be different from past observations on different backgrounds. Although we did not identify any major differences, it is important that the information on NC behavior in this background be presented. 

(4) Also, the conclusion drawn for Figure 5C "this indicates that the Wnt1Cre-derived cells do not harbor a cellautonomous response to GDNF" seems to suggest the authors are not very well versed with the ENS literature. GDNF as well as EDN3 are expressed from surrounding mesenchyme and are cell non-autonomous.

The reviewer seems to have misread or misunderstood the specific statement as well as the more important broader conclusion of the experiment. First, of course the source of GDNF ligand in vivo is the mesenchyme. The explant assay was designed to eliminate this and then to substitute GDNF as provided experimentally. The focus of the experiment was to address the response to GDNF, not the source of GDNF. But more importantly, the experiment revealed a surprising outcome that the reviewer did not appreciate. In Pax2Cre/Ret mutants, the Wnt1Cre lineage still expresses Ret, yet does not grow out from the gut explant when provided with GDNF. This shows that the neural crest lineage requires Ret function in placode-derived cells in order to respond to GDNF. In other words, despite expressing Ret, the NC lineage does not harbor a cellautonomous response to GDNF, as we wrote. Because this might be confusing to some readers, we have revised the description of this analysis to hopefully be more clear.

(5) The fact that Ret and Ednrb signaling pathways interact is not a novel finding and has been reported multiple times in Ret and Ednrb mutant mice and cell lines (PMID: 12355085, 12574515 , 27693352, 31818953), potentially through shared transcription factors (PMID:31313802).It would have been more relevant if the authors could show how the specific tyrosine residue (Y 1015) in Ret is phosphorylated in the presence of Ednrb.

The observation that human mutations in RET and EDNRB both cause Hirschsprung disease is decades old, and of course numerous studies in human, mouse, and cells have addressed the relation between the two signaling pathways. We did not mean to imply that we were the first to discover that Ret and Ednrb signaling pathways interact. The reviewer cites a number of papers all from the Chakravarti lab that address this phenomenon; while these are a valuable contribution to the field, there is still more to be learned. The model elaborated in PMID: 31313802, in which Ret and Ednrb are both enmeshed in a common gene regulatory network, does not readily explain why each has a different phenotypic manifestation and doesn’t take into account the importance of the placodal lineage. The main new contributions of our paper are the existence of a new cell lineage that contributes to the ENS, and that the placodal and neural crest lineages utilize Ret and Ednrb signaling differently. The clarification of how these elements are differentially used by the two lineages explains long-segment and short-segment Hirschsprung disease (Ret and Ednrb mutants, respectively) far better than in past studies. The reviewer unfortunately dismisses these insights and seems to feel that a biochemical exploration of one specific component of the signaling interaction (Y1015 phosphorylation) would be more relevant. This should be the basis of future studies and are beyond the scope of the new findings reported in the present paper. 

(6) What is the mechanism of the presence of Y1015 phosphorylation in 33% of Ednrb deficient Pax2Cre cells? It appears to me what the authors report as absent phosphorylation in the 67% of cells could be just weak staining or cells missing in prep.

The reviewer, referring to Fig. 7q, presumably meant to say Wnt1Cre rather than Pax2Cre. The reviewer overlooked that we provided an explanation for this observation in our original manuscript. This sentence reads “Because Ednrb is expressed only in a subset of Wnt1Cre-derived enteric progenitor cells (Figure 7 – figure supplement 1), the residual Y1015 phosphorylation observed in Wnt1Cre/Ednrb mutant cells is likely to occur in the Ednrb-negative Wnt1Cre-derived cell population”. The sentence is retained unchanged in the revised manuscript. The explanation is not because of weak staining or problems with tissue preparation.

(7) The references the authors cite regarding the previous discovery of Ret expression in the nucleus are incorrect. The review articles the authors cite do not mention anything about Ret expression in the nucleus. The evidence of nuclear localization of Ret previously comes from overexpression studies in HEK293 cells (PMID: 25795775). Such overexpression studies are fraught with generating noisy data for well-documented reasons. But if this observation is correct, the authors miss a great opportunity to identify what the Ret protein is doing in the nucleus. Is it in direct contact with its known transcription factors like Sox10 and Rarb? This would shed a lot of light on the possible mechanism of Ret LoF observed in Ret mutant mice

The reviewer overlooked that the one of the review articles that we cited (Chen, Hsu, & Hung, 2020) has a dedicated paragraph for RET (section 3.14), which summarizes the work by Barheri-Yarmand et al (PMID: 25795775) which is the very paper noted by the reviewer in the comment above. The reviewer also somewhat misstated the results of the Barheri-Yarmand et al study. By immunostaining, this paper showed nuclear localization of endogenous Ret, albeit a version of Ret with a disease-associated mutation that makes it constitutively active by constitutive autophosphorylation. Nonetheless, this was endogenous Ret. The paper also used overexpression of GFP-tagged RET in HEK293 cells to show that wildtype RET can behave in a similar manner, at least under these circumstances. Our point is simply that Ret (and other receptor tyrosine kinases) can be found in the nucleus in certain biological contexts, and our observations are consistent with this precedent.

The reviewer also suggests a biochemical follow-up analysis related to this observation, which we agree would be of interest. Such an investigation however is beyond the scope of the present study.

(8) The manuscript could benefit from a major rewrite by reorganizing sections to make it easy for the readers to follow the narrative.

Many sections about the role of Ret and Ednrb in Wnt1cre-derived ENCDCs can be moved to a supplement. These facts are well-documented and have been proven before.

This was addressed in our response to comment #3 of this reviewer. The figures have been kept as main figures in the revised manuscript to allow side-by-side comparison to parallel analysis of the Pax2Cre lineage.

- The observation that only a handful of Pax2Cre cells at E10.5 express Ret and the observation that conditional Ret null abrogates these cells at E11.5, are not presented together and makes connecting these two facts difficult.

Ret expression at E10.5 and E11.5 are both shown in the same figure (Fig. 2). In the presentation of these results, we first describe in normal development that Ret is expressed differently in E10.5 ENS progenitors between the Pax2Cre and Wnt1Cre lineages. This is additional support for the argument that the two lineages are molecularly distinct. Then comes evaluation of postnatal fates with different markers before we return to embryonic Ret expression. We acknowledge that this can make it difficult to connect these observations. We decided to retain the original organization in order to not lose this important conclusion. However, we have revised the text to hopefully make this connection between the sections more congruent.

Reviewer #2 (Recommendations For The Authors):

- The labeling of some as "figure supplements" is really hard to follow in the text and confusing to interpret when a main figure or supplemental figure is being referenced, and which one.

We understand this comment, but this is journal style and outside of our control. We have kept the journal format in the revised manuscript.

- The data in Figures 3b-c is well established in the field and somewhat misinterpreted. NOS1 neurons in the mouse ENS and their projections have been well described (Sang and Young, 1996, and other studies). CGRP immunoreactivity would reflect both ENS CGRP-expressing neurons and visceral afferents from DRG.

There of course is a history of analysis of NOS1, CGRP, and other markers in the ENS. The focus of the analysis in Fig. 3 is to demonstrate how the cells that express these markers are impacted by gene manipulation in the Wnt1Cre and Pax2Cre lineages. For the giant migrating contractions that are associated with defecation, ample past electrophysiological studies have established that mechanosensory CGRP+ neurons trigger NOS+ inhibitory neurons (and ACh+ excitatory neurons) of the myenteric plexus to propel colonic contents. Thus, these are the relevant markers to explain the lack of colonic peristalsis in Ednrb-deficient mice. To our awareness, our results with NOS1 do not contradict any past study, including the Sang and Young 1996 description. Regarding CGRP, indeed the reviewer is correct that this marker is expressed by both neuronal subtypes. Two arguments support the specific derivation of ENS mechanosensory neurons from the Pax2 lineage. First, the ENS and DRG neurons can be distinguished by the location of their cell bodies and their axon extensions in the gut wall; only the ENS neurons are deficient in Pax2Cre/Ednrb mutants (as documented in Fig. 3). Second, the DRG population is derived from neural crest and is not labeled by Pax2Cre. If this population of CGRP+ neurons had functional relevance to colonic peristalsis, this would not be altered in Pax2Cre/Ednrb mutants. Indeed, the CGRP+ afferent nerve endings of DRG origin in the distal colon are mechanical distension sensors but do not modulate either ENS or autonomic nervous system activity (PMID: 37541195). We believe that our interpretation is correct.

- The evidence in Figure 3 supporting the claim that NOS1 and CGRP-expressing enteric neurons come from distinct lineages is weak. IHC for CGRP is notoriously poor at labeling soma in the ENS. IHC for tdTomato to ensure the detection of low levels of Tomato expression and quantification of observations would strengthen this claim.

CGRP is a vesicular peptide which is stored and transported in vesicles, therefore the antibody against CGRP labels vesicular particles of soma and synaptic vesicles along the axons of those CGRP-producing neurons.

It is not expected to label the entire cytoplasm (or the range of subcellular organelles) as NOS antibody does. We did included quantification of data in Figure 3-figure supplement 1 in the manuscript to support the claim of lineage derivation. As described in the Methods section of the manuscript, we used binary threshold selection for Tomato+ cell count using Fiji-Image J, which detects both TomatoHigh and TomatoLow cells as Tomato+; we feel this is equal to or even superior to IHC for this analysis. 

- IHC panels in Figures 3h-o are largely uninterpretable. Most of the signal seems to be non-specific background staining in the mucosa and quantification of mucosal signal in this context does not seem meaningful.  

We disagree with the reviewer’s comment. As described in the response above, CGRP+ mechanosensory neurons send their peripheral axon projections to innervate mucosa (sensory epithelial cells), and NOS+ inhibitory motor axons innervate the circular muscle. Thus, panels h-o of Fig. 3 focus on the axonal profile and are not intended to visualize soma, which is why sagittal views are presented instead of flatmount views. All of the controls were performed side-by-side to confirm that the signal is real and interpretable.

Note also that the colon does not have villi so this annotation should be revised.

We appreciate that the reviewer brought this misstatement to our attention. We corrected this error in the revised manuscript.

- Phospho-RET staining in Figure 7 is difficult to discern and interpret with high background. Positive and negative controls would strengthen these data.

Fig. 7 shows phospho Ret-Y1015 staining in lineage-labeled Wnt1Cre/Ednrb/R26nTnG mutants. The strength of the signal to noise in the figure is a matter of Ret expression level and the quality of the anti-pY1015 antibody. We are not aware of a meaningful positive control that has been validated in the literature that we could use for comparison. The ideal negative control would be to perform the same analysis in Wnt1Cre/Ret/R26nTnG mutants, but because this manipulation eliminates the entire NC cell lineage from the colon, there would be no NC cells in which to visualize background staining in this lineage with this antibody when Ret protein is not present. We note that anti-pY1096 did not show a difference in staining between control and mutant, which supports the interpretation of a specific impact on pY1015. We also point out here, as in the text, that we do not yet have any validation that phosphorylation of Y1015 is functionally important in NC migration to the distal colon. Clearly, more work to address this role and to demonstrate the mechanism of phosphorylation of this specific residue in response to Edn3-Ednrb signaling will be needed.

eLife Assessment

This fundamental study addresses the regulation of the MICAL-family of actin regulators by Rab GTPases, which play a key role in directing membrane trafficking within cells. The compelling evidence explains how Rab8 family members bind at two sites to allosterically regulate MICAL1, and relieve an auto-inhibited state unable to bind actin. This study lays the basis for further progress in understanding membrane trafficking and cytoskeleton dynamics in eukaryotic cells.

Reviewer #1 (Public review):

The manuscript describes comprehensive structure-function studies combining structural studies, Alphafold2-based modelling, and extensive structural validation by mutagenesis and biochemical experiments. Consequently, a sophisticated activation mechanism of Mical1 as a representative of the MICAL family is elucidated at the molecular level. Since MICAL proteins are important regulators of membrane trafficking and cytoskeleton dynamics, the study is of high relevance for many groups. Structural data are of high quality, the modelling data appear to be sound, and the subsequent biochemical analyses are carried out in great detail, yielding a complete story. I have little to criticize on this beautiful work.

Reviewer #2 (Public review):

Summary:

Rai and coworkers have studied the regulation of the MICAL-family of actin regulators by Rab 8 family GTPases. Their work uses a combination of structural biology, biochemistry, and modelling approaches to identify the regions and specific residues interacting with Rabs and understand the consequences of MICAL1 regulation. The study extends previous work on individual domains by incorporating analysis of the full-length MICAL1 protein and provides compelling evidence for allosteric regulation by Rab binding to two low and high-affinity regulatory sites.

Strengths:

Excellent biochemical and structural analysis.

Weaknesses:

Additional data to test the model for Rab regulation of MICAL1 in the actin-pelleting assay would enhance the study.

eLife Assessment

This study offers valuable insights into the molecular mechanisms by which Osx influences osteocyte function, particularly through its regulation of Cx43. However, the evidence supporting the authors' claims is incomplete, necessitating additional experimental data and further investigation to fully substantiate these findings. While this study presents a new perspective on the complex role of Osx in bone biology, it also raises significant questions about the intricacies of its regulatory network.

Reviewer #1 (Public review):

The manuscript "Osterix Facilitates Osteocytic Communication by Targeting Connexin43" investigates the role of Osterix (Osx) in osteocytes using a Col1α1-CreER;Osxfl/fl mouse model and cultured cells. The study reveals that Osx is expressed in osteocytes, and its deletion in vitro leads to a significant reduction in osteocyte dendrite formation, highlighting its critical role in maintaining cellular communication. Through ChIP-seq analysis, the authors identified Connexin43 (Cx43) as a direct downstream target of Osx. Moreover, treatment with all-trans retinoic acid (ATRA), a known agonist of Cx43, was able to rescue the dendritic network in osteocytes, restoring their communication capabilities in vitro.

This research provides valuable insights into the molecular mechanisms by which Osx influences osteocyte function, particularly through its regulation of Cx43. However, despite these findings, the study does not fully elucidate all the mechanisms involved in Osx-mediated osteocytic communication. Several conclusions, particularly those related to the broader signaling pathways, require additional experimental evidence and further investigation to be fully substantiated. This study provides a new aspect in understanding the complex role of Osx in bone biology but leaves open questions regarding the intricacies of its regulatory network.

Major Comments:

(1) In the Col1a1-CreER;tdTomato mice, the number of tdTomato+ cells in the cortical bone appears lower compared to Osx+ cells. The overlap between tdTomato+ and Osx+ cells in Figure 1 is limited. Could this affect the knockout efficiency? Can the authors provide data on Osx knockout efficiency in vivo? While immunostaining of Osx is shown in both control and mutant mice in Figure 2A, the Osx expression pattern differs from Figure 1A. Osx expression is relatively low in the bone marrow in Figure 1A, but much stronger in Figure 2A.

Additionally, Osx+ cells in Figure 1A seem confined to the bone surface, whereas Figure 2A shows a broader distribution. What developmental stage of mice was used in Figure 1? Could the authors also provide co-staining with other osteocyte markers alongside Osx?

(2) The authors mentioned using both siRNA and Lenti-Osx to modulate Osx expression. What was the specific purpose of these experiments? If the authors aim to demonstrate that Osx plays a critical role in osteocytes, they should provide data on downstream targets or markers relevant to osteocyte function. Additionally, did these treatments affect processes like differentiation or cell viability in osteocytes? The current results only demonstrate that siRNA and Lenti-Osx can successfully modulate Osx expression in vitro, but further evidence is needed to support broader functional conclusions.

(3) Osx knockout mice exhibited a decreased osteocyte dendritic network both in vivo and in vitro. To better understand how this affects overall bone health, could the authors provide additional parameters, such as bone thickness, bone strength, and other relevant metrics? Furthermore, to determine whether these phenotypes are primarily due to defects in the osteocyte dendritic network or a reduction in osteocyte numbers, the authors should also assess the number of osteocytes in the knockout mice Figure 2.

(4) Regarding the Lucifer Yellow Dye Transfer Assay in Figure 3, the authors should provide data on cell density and cell viability for both control and mutant groups. Additionally, although less dye is observed in the mutant group, the migration distance appears comparable to the control group. Could the authors explain this result? Furthermore, how was transmission speed between the groups evaluated in Figure 3D? More details on the method used to assess transmission speed would be helpful.

(5) For a more comprehensive and unbiased analysis of Osx function in osteocytes, the authors should present a full analysis of differentially expressed genes, rather than focusing solely on integrins. Additionally, it would be beneficial to include an analysis of the knockdown group alongside the other groups, considering the animal model used in this study involves knockout mice.

(6) In the immunofluorescence staining of integrin αvβ1 in the si-Osx and Lenti-Osx groups, the cellular localization of integrin αvβ1 appears altered. Unlike the control group, where it is mainly localized in the cytoplasm, positive signals are observed in the nucleus of the si-Osx and Lenti-Osx groups. Additionally, since integrin αvβ1 is a membrane protein, shouldn't it primarily be observed on the cell membrane rather than in the cytoplasm? Could the authors clarify this observation?

(7) The results regarding Cx43 expression after Lenti-Osx treatment are questionable. It appears that the images for the Lenti-GFP and Lenti-Osx groups have been misrepresented. The merged images for the Lenti-GFP control group seem to belong to the Lenti-Osx group, and vice versa. If the images were presented in their correct order, the conclusions would contradict the authors' claims. This issue needs to be addressed to ensure an accurate interpretation of the data.

(8) The authors demonstrated that ATRA treatment elevates Cx43 protein levels in the control group, where Osx function is normal. However, can ATRA also restore Cx43 protein levels in the si-Osx treated group, where Osx transcriptional function is impaired? Theoretically, Cx43 protein levels should not be restored in the si-Osx group. Could the observed rescue phenotype be due to effects downstream of Cx43? This possibility should be considered and clarified.

(9) Does the Cx43 mutation of knockout cause similar phenotypes in the animal model? Can restoration of Cx43 rescue the bone phenotype?

Reviewer #2 (Public review):

This study shows that Osx plays a pivotal role in the dendritic network and intercellular communication of Col1α1-positive osteocytes via targeting Connexin43 (Cx43). It provides solid evidence to broaden our understanding of Osx's roles during bone homeostasis. This work will be of interest to investigators studying bone diseases involving osteocytes, such as delayed fracture healing or osteoporosis.

Comments:

(1) In Figure 1, it appears that the Osx- and Col1α1-positive cells may not be exclusively expressed by osteocytes. Possibly periosteum cells and osteoblasts are also included. This could potentially impact the interpretation of results. The authors should provide a clearer analysis to distinguish the cell types precisely.

(2) Jialiang S. Wang et al. (Nat Commun. 2021 Nov 1;12(1):6274.) have previously reported on the direct role of Osx in osteocytes. In light of this prior research, it is essential for the authors to thoroughly discuss how this study differs from previous findings.

(3) In the methods section, it is crucial to provide detailed information about the manufacturer and country of origin of reagents, like ATRA.

(4) The morphology of osteocytes in cortical bone can vary between the metaphysis site and the middle shaft site of long bones. For SEM data of osteocytes in Figure 2, it is necessary to address this issue. The authors should clarify whether morphological difference was observed between these sites and, if so, how these differences might impact the interpretation of the data.

(5) In the bone research field, two different Col1α1 - CreER mice were used. The authors should specify which type of Col1α1 - CreER mice were utilized in this research.

(6) A more detailed description of the statistical method used in Figure 2G - I is required, particularly with regard to quantifying the number of osteocyte dendritic processes.

(7) In Figure 6C and Figure 6D, while the legend indicates N = 3, there are five data points presented in the statistical graph.

Reviewer #3 (Public review):

Summary:

This study investigated the expression of Osterix (Osx) not only in osteoblasts but also significantly in osteocytes. Through Osx knockout, the osteocytic dendritic network was damaged, leading to communication disruption. This study investigated the regulatory role of Osx on osteoblast dendrites through Cx43.

Strengths:

This paper provides a good explanation of the role of Osx in osteocyte synapse and cell communication, enriching the understanding of Osx's functional significance. The results of the experiment support the conclusions of the study. This is an interesting study with a clear logical structure.

Weaknesses:

Some experimental results need to be supplemented, and there are still some details and errors in the text that need to be revised.

eLife Assessment

This work presents an important method for depleting ribosomal RNA from bacterial single-cell RNA sequencing libraries, enabling the study of cellular heterogeneity within microbial biofilms. The approach convincingly identifies a small subpopulation of cells at the biofilm's base with upregulated PdeI expression, offering invaluable insights into the biology of bacterial biofilms and the formation of persister cells. Further integrated analysis of gene interactions within these datasets could deepen our understanding of biofilm dynamics and resilience.

Reviewer #1 (Public review):

Summary:

In this manuscript, Yan and colleagues introduce a modification to the previously published PETRI-seq bacterial single cell protocol to include a ribosomal depletion step based on a DNA probe set that selectively hybridizes with ribosome-derived (rRNA) cDNA fragments. They show that their modification of the PETRI-seq protocol increases the fraction of informative non-rRNA reads from ~4-10% to 54-92%. The authors apply their protocol to investigating heterogeneity in a biofilm model of E. coli, and convincingly show how their technology can detect minority subpopulations within a complex community.

Strengths:

The method the authors propose is a straightforward and inexpensive modification of an established split-pool single cell RNA-seq protocol that greatly increases its utility, and should be of interest to a wide community working in the field of bacterial single cell RNA-seq.

Reviewer #2 (Public review):

Summary:

This work introduces a new method of depleting the ribosomal reads from the single-cell RNA sequencing library prepared with one of the prokaryotic scRNA-seq techniques, PETRI-seq. The advance is very useful since it allows broader access to the technology by lowering the cost of sequencing. It also allows more transcript recovery with fewer sequencing reads. The authors demonstrate the utility and performance of the method for three different model species and find a subpopulation of cells in the E.coli biofilm that express a protein, PdeI, which causes elevated c-di-GMP levels. These cells were shown to be in a state that promotes persister formation in response to ampicillin treatment.

Strengths:

The introduced rRNA depletion method is highly efficient, with the depletion for E.coli resulting in over 90% of reads containing mRNA. The method is ready to use with existing PETRI-seq libraries which is a large advantage, given that no other rRNA depletion methods were published for split-pool bacterial scRNA-seq methods. Therefore, the value of the method for the field is high. There is also evidence that a small number of cells at the bottom of a static biofilm express PdeI which is causing the elevated c-di-GMP levels that are associated with persister formation. This finding highlights the potentially complex role of PdeI in regulation of c-di-GMP levels and persister formation in microbial biofilms.

Weaknesses:

Given many current methods that also introduce different techniques for ribosomal RNA depletion in bacterial single-cell RNA sequencing, it is unclear what is the place and role of RiboD-PETRI. The efficiency of rRNA depletion varies greatly between species for the majority of the available methods, so it is not easy to select the best fitting technique for a specific application.

Despite transcriptome-wide coverage, the authors focused on the role of a single heterogeneously expressed gene, PdeI. A more integrated analysis of multiple genes and\or interactions between them using these data could reveal more insights into the biofilm biology.

The authors should also present the UMIs capture metrics for RiboD-PETRI method for all cells passing initial quality filter (>=15 UMIs/cell) both in the text and in the figures. Selection of the top few cells with higher UMI count may introduce biological biases in the analysis (the top 5% of cells could represent a distinct subpopulation with very high gene expression due to a biological process). For single-cell RNA sequencing, showing the statistics for a 'top' group of cells creates confusion and inflates the perceived resolution, especially when used to compare to other methods (e.g. the parent method PETRI-seq itself).

Author response:

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

eLife Assessment 

The work introduces a valuable new method for depleting the ribosomal RNA from bacterial single-cell RNA sequencing libraries and shows that this method is applicable to studying the heterogeneity in microbial biofilms. The evidence for a small subpopulation of cells at the bottom of the biofilm which upregulates PdeI expression is solid. However, more investigation into the unresolved functional relationship between PdeI and c-di-GMP levels with the help of other genes co-expressed in the same cluster would have made the conclusions more significant. 

Many thanks for eLife’s assessment of our manuscript and the constructive feedback. We are encouraged by the recognition of our bacterial single-cell RNA-seq methodology as valuable and its efficacy in studying bacterial population heterogeneity. We appreciate the suggestion for additional investigation into the functional relationship between PdeI and c-di-GMP levels. We concur that such an exploration could substantially enhance the impact of our conclusions. To address this, we have implemented the following revisions: We have expanded our data analysis to identify and characterize genes co-expressed with PdeI within the same cellular cluster (Fig. 3F, G, Response Fig. 10); We conducted additional experiments to validate the functional relationships between PdeI and c-di-GMP, followed by detailed phenotypic analyses (Response Fig. 9B). Our analysis reveals that while other marker genes in this cluster are co-expressed, they do not significantly impact biofilm formation or directly relate to c-di-GMP or PdeI. We believe these revisions have substantially enhanced the comprehensiveness and context of our manuscript, thereby reinforcing the significance of our discoveries related to microbial biofilms. The expanded investigation provides a more thorough understanding of the PdeI-associated subpopulation and its role in biofilm formation, addressing the concerns raised in the initial assessment.

Public Reviews: 

Reviewer #1 (Public Review): 

Summary: 

In this manuscript, Yan and colleagues introduce a modification to the previously published PETRI-seq bacterial single-cell protocol to include a ribosomal depletion step based on a DNA probe set that selectively hybridizes with ribosome-derived (rRNA) cDNA fragments. They show that their modification of the PETRI-seq protocol increases the fraction of informative non-rRNA reads from ~4-10% to 54-92%. The authors apply their protocol to investigating heterogeneity in a biofilm model of E. coli, and convincingly show how their technology can detect minority subpopulations within a complex community. 

Strengths: 

The method the authors propose is a straightforward and inexpensive modification of an established split-pool single-cell RNA-seq protocol that greatly increases its utility, and should be of interest to a wide community working in the field of bacterial single-cell RNA-seq. 

Weaknesses: 

The manuscript is written in a very compressed style and many technical details of the evaluations conducted are unclear and processed data has not been made available for evaluation, limiting the ability of the reader to independently judge the merits of the method. 

Thank you for your thoughtful and constructive review of our manuscript. We appreciate your recognition of the strengths of our work and the potential impact of our modified PETRI-seq protocol on the field of bacterial single-cell RNA-seq. We are grateful for the opportunity to address your concerns and improve the clarity and accessibility of our manuscript.

We acknowledge your feedback regarding the compressed writing style and lack of technical details, which are constrained by the requirements of the Short Report format in eLife. We have addressed these issues in our revised manuscript as follows:

(1) Expanded methodology section: We have provided a more comprehensive description of our experimental procedures, including detailed protocols for the ribosomal depletion step (lines 435-453) and data analysis pipeline (lines 471-528). This will enable readers to better understand and potentially replicate our methods.

(2) Clarification of technical evaluations: We have elaborated on the specifics of our evaluations, including the criteria used for assessing the efficiency of ribosomal depletion (lines 99-120), and the methods employed for identifying and characterizing subpopulations (lines 155-159, 161-163 and 163-167).

(3) Data availability: We apologize for the oversight in not making our processed data readily available. We have deposited all relevant datasets, including raw and source data, in appropriate public repositories (GEO: GSE260458) and provide clear instructions for accessing this data in the revised manuscript.

(4) Supplementary information: To maintain the concise nature of the main text while providing necessary details, we have included additional supplementary information. This will cover extended methodology (lines 311-318, 321-323, 327-340, 450-453, 533, and 578-589), detailed statistical analyses (lines 492-493, 499-501 and 509-528), and comprehensive data tables to support our findings.

We believe these changes significantly improved the clarity and reproducibility of our work, allowing readers to better evaluate the merits of our method.

Reviewer #2 (Public Review): 

Summary: 

This work introduces a new method of depleting the ribosomal reads from the single-cell RNA sequencing library prepared with one of the prokaryotic scRNA-seq techniques, PETRI-seq. The advance is very useful since it allows broader access to the technology by lowering the cost of sequencing. It also allows more transcript recovery with fewer sequencing reads. The authors demonstrate the utility and performance of the method for three different model species and find a subpopulation of cells in the E.coli biofilm that express a protein, PdeI, which causes elevated c-di-GMP levels. These cells were shown to be in a state that promotes persister formation in response to ampicillin treatment. 

Strengths: 

The introduced rRNA depletion method is highly efficient, with the depletion for E.coli resulting in over 90% of reads containing mRNA. The method is ready to use with existing PETRI-seq libraries which is a large advantage, given that no other rRNA depletion methods were published for split-pool bacterial scRNA-seq methods. Therefore, the value of the method for the field is high. There is also evidence that a small number of cells at the bottom of a static biofilm express PdeI which is causing the elevated c-di-GMP levels that are associated with persister formation. Given that PdeI is a phosphodiesterase, which is supposed to promote hydrolysis of c-di-GMP, this finding is unexpected. 

Weaknesses: 

With the descriptions and writing of the manuscript, it is hard to place the findings about the PdeI into existing context (i.e. it is well known that c-di-GMP is involved in biofilm development and is heterogeneously distributed in several species' biofilms; it is also known that E.coli diesterases regulate this second messenger, i.e. https://journals.asm.org/doi/full/10.1128/jb.00604-15). 

There is also no explanation for the apparently contradictory upregulation of c-di-GMP in cells expressing higher PdeI levels. Perhaps the examination of the rest of the genes in cluster 2 of the biofilm sample could be useful to explain the observed association. 

Thank you for your thoughtful and constructive review of our manuscript. We are pleased that the reviewer recognizes the value and efficiency of our rRNA depletion method for PETRI-seq, as well as its potential impact on the field. We would like to address the points raised by the reviewer and provide additional context and clarification regarding the function of PdeI in c-di-GMP regulation.

We acknowledge that c-di-GMP’s role in biofilm development and its heterogeneous distribution in bacterial biofilms are well studied. We appreciate the reviewer's observation regarding the seemingly contradictory relationship between increased PdeI expression and elevated c-di-GMP levels. This is indeed an intriguing finding that warrants further explanation.

PdeI is predicted to function as a phosphodiesterase involved in c-di-GMP degradation, based on sequence analysis demonstrating the presence of an intact EAL domain, which is known for this function. However, it is important to note that PdeI also harbors a divergent GGDEF domain, typically associated with c-di-GMP synthesis. This dual-domain structure indicates that PdeI may play complex regulatory roles. Previous studies have shown that knocking out the major phosphodiesterase PdeH in E. coli results in the accumulation of c-di-GMP. Moreover, introducing a point mutation (G412S) in PdeI's divergent GGDEF domain within this PdeH knockout background led to decreased c-di-GMP levels2. This finding implies that the wild-type GGDEF domain in PdeI contributes to maintaining or increasing cellular c-di-GMP levels.

Importantly, our single-cell experiments demonstrated a positive correlation between PdeI expression levels and c-di-GMP levels (Figure 4D). In this revision, we also constructed a PdeI(G412S)-BFP mutation strain. Notably, our observations of this strain revealed that c-di-GMP levels remained constant despite an increase in BFP fluorescence, which serves as a proxy for PdeI(G412S) expression levels (Figure 4D). This experimental evidence, coupled with domain analyses, suggests that PdeI may also contribute to c-di-GMP synthesis, rebutting the notion that it acts solely as a phosphodiesterase. HPLC LC-MS/MS analysis further confirmed that the overexpression of PdeI, induced by arabinose, resulted in increased c-di-GMP levels (Fig. 4E) . These findings strongly suggest that PdeI plays a pivotal role in upregulating c-di-GMP levels.

Our further analysis indicated that PdeI contains a CHASE (cyclases/histidine kinase-associated sensory) domain. Combined with our experimental results showing that PdeI is a membrane-associated protein, we hypothesize that PdeI acts as a sensor, integrating environmental signals with c-di-GMP production under complex regulatory mechanisms.

We understand your interest in the other genes present in cluster 2 of the biofilm and their potential relationship to PdeI and c-di-GMP. Upon careful analysis, we have determined that the other marker genes in this cluster do not significantly impact biofilm formation, nor have we identified any direct relationship between these genes, c-di-GMP, or PdeI. Our focus on PdeI within this cluster is justified by its unique and significant role in c-di-GMP regulation and biofilm formation, as demonstrated by our experimental results. While other genes in this cluster may be co-expressed, their functions appear unrelated to the PdeI-c-di-GMP pathway we are investigating. Therefore, we opted not to elaborate on these genes in our main discussion, as they do not contribute directly to our understanding of the PdeI-c-di-GMP association. However, we can include a brief mention of these genes in the manuscript, indicating their lack of relevance to the PdeI-c-di-GMP pathway. This addition will provide a more comprehensive view of the cluster's composition while maintaining our focus on the key findings related to PdeI and c-di-GMP.

We have also included the aforementioned explanations and supporting experimental data within the manuscript to clarify this important point (lines 193-217). Thank you for highlighting this apparent contradiction, allowing us to provide a more detailed explanation of our findings.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors): 

Overall, I found the main text of the manuscript well written and easy to understand, though too compressed in parts to fully understand the details of the work presented, some examples are outlined below. The materials and methods appeared to be less carefully compiled and could use some careful proof-reading for spelling (e.g. repeated use of "minuts" for minutes, "datas" for data) and grammar and sentence fragments (e.g. "For exponential period E. coli data." Line 333). In general, the meaning is still clear enough to be understood. I also was unable to find figure captions for the supplementary figures, making these difficult to understand. 

We appreciate your careful review, which has helped us improve the clarity and quality of our manuscript. We acknowledge that some parts of the main text may have been overly compressed due to Short Report format in eLife. We have thoroughly reviewed the manuscript and expanded on key areas to provide more comprehensive explanations. We have carefully revised the Materials and Methods section to address the following: Corrected all spelling and grammatical error, including "minuts" to "minutes" and "datas" to "data". Corrected grammatical issues and sentence fragments throughout the section. We sincerely apologize for the omission of captions for the supplementary figures. We have now added detailed captions for all supplementary figures to ensure they are easily understandable. We believe these revisions address your concerns and enhance the overall readability and comprehension of our work.

General comments: 

(1) To evaluate the performance of RiboD-PETRI, it would be helpful to have more details in general, particularly to do with the development of the sequencing protocol and the statistics shown. Some examples: How many reads were sequenced in each experiment? Of these, how many are mapped to the bacterial genome? How many reads were recovered per cell? Have the authors performed some kind of subsampling analysis to determine if their sequencing has saturated the detection of expressed genes? The authors show e.g. correlations between classic PETRI-seq and RiboD-PETRI for E. coli in Figure 1, but also have similar data for C. crescentus and S. aureus - do these data behave similarly? These are just a few examples, but I'm sure the authors have asked themselves many similar questions while developing this project; more details, hard numbers, and comparisons would be very much appreciated. 

Thank you for your valuable feedback. To address your concerns, we have added a table in the supplementary material that clarifies the details of sequencing.

The correlation values of PETRI-seq and RiboD-PETRI data in C. crescentus are relatively good. However, the correlation values between PETRI-seq and RiboD-PETRI data in SA data are relatively less high. The reason is that the sequencing depths of RiboD-PETRI and PETRI-seq are different, resulting in much higher gene expression in the RiboD-PETRI sequencing results than in PETRI-seq, and the calculated correlation coefficient is only about 0.47. This indicates that there is some positive correlation between the two sets of data, but it is not particularly strong. This indicates that there is a certain positive correlation between these two sets of data, but it is not particularly strong. However, we have counted the expression of 2763 genes in total, and even though the calculated correlation coefficient is relatively low, it still shows that there is some consistency between the two groups of samples.

Author response image 1.

Assessment of the effect of rRNA depletion on transcriptional profiles of (A) C. crescentus (CC) and (B) S. aureus (SA) . The Pearson correlation coefficient (r) of UMI counts per gene (log2 UMIs) between RiboD-PETRI and PETRI-seq was calculated for 4097 genes (A) and 2763 genes (B). The "ΔΔ" label represents the RiboD-PETRI protocol; The "Ctrl" label represents the classic PETRI-seq protocol we performed. Each point represents a gene.

(2) Additionally, I think it is critical that the authors provide processed read counts per cell and gene in their supplementary information to allow others to investigate the performance of their method without going back to raw FASTQ files, as this can represent a significant hurdle for reanalysis. 

Thank you for your suggestion. However, it's important to clarify that reads and UMIs (Unique Molecular Identifiers) are distinct concepts in single-cell RNA sequencing. Reads can be influenced by PCR amplification during library construction, making their quantity less stable. In contrast, UMIs serve as a more reliable indicator of the number of mRNA molecules detected after PCR amplification. Throughout our study, we primarily utilized UMI counts for quantification. To address your concern about data accessibility, we have included the UMI counts per cell and gene in our supplementary materials provided above (Table S7-15. Some of the files are too large in memory and are therefore stored in GEO: GSE260458). This approach provides a more accurate representation of gene expression levels and allows for robust reanalysis without the need to process raw FASTQ files.

(3) Finally, the authors should also discuss other approaches to ribosomal depletion in bacterial scRNA-seq. One of the figures appears to contain such a comparison, but it is never mentioned in the text that I can find, and one could read this manuscript and come away believing this is the first attempt to deplete rRNA from bacterial scRNA-seq. 

We have addressed this concern by including a comparison of different methods for depleting rRNA from bacterial scRNA-seq in Table S4 and make a short text comparison as follows: “Additionally, we compared our findings with other reported methods (Fig. 1B; Table S4). The original PETRI-seq protocol, which does not include an rRNA depletion step, exhibited an mRNA detection rate of approximately 5%. The MicroSPLiT-seq method, which utilizes Poly A Polymerase for mRNA enrichment, achieved a detection rate of 7%. Similarly, M3-seq and BacDrop-seq, which employ RNase H to digest rRNA post-DNA probe hybridization in cells, reported mRNA detection rates of 65% and 61%, respectively. MATQ-DASH, which utilizes Cas9-mediated targeted rRNA depletion, yielded a detection rate of 30%. Among these, RiboD-PETRI demonstrated superior performance in mRNA detection while requiring the least sequencing depth.” We have added this content in the main text (lines 110-120), specifically in relation to Figure 1B and Table S4. This addition provides context for our method and clarifies its position among existing techniques.

Detailed comments: 

Line 78: the authors describe the multiplet frequency, but it is not clear to me how this was determined, for which experiments, or where in the SI I should look to see this. Often this is done by mixing cultures of two distinct bacteria, but I see no evidence of this key experiment in the manuscript. 

The multiplet frequency we discuss in the manuscript is not determined through experimental mixing of distinct bacterial cultures.The PETRI-seq and mirco-SPLIT articles have also done experiments mixing the two libraries to determine the single-cell rate, and both gave good results. Our technique is derived from these two articles (mainly PETRI-seq), and the biggest difference is the difference in the later RiboD part, so we did not do this experiment separately. So the multiple frequencies here are theoretical predictions based on our sequencing results, calculated using a Poisson distribution. We have made this distinction clearer in our manuscript (lines 93-97). The method is available in Materials and Methods section (lines 520-528). The data is available in Table S2. To elaborate:

To assess the efficiency of single-cell capture in RiboD-PETRI, we calculated the multiplet frequency using a Poisson distribution based on our sequencing results

(1) Definition: In our study, multiplet frequency is defined as the probability of a non-empty barcode corresponding to more than one cell.

(2) Calculation Method: We use a Poisson distribution-based approach to calculate the predicted multiplet frequency. The process involves several steps:

We first calculate the proportion of barcodes corresponding to zero cells: . Then, we calculate the proportion corresponding to one cell: . We derive the proportion for more than zero cells: P(≥1) = 1 - P(0). And for more than one cell: P(≥2) = 1 - P(1) - P(0). Finally, the multiplet frequency is calculated as:

(3) Parameter λ: This is the ratio of the number of cells to the total number of possible barcode combinations. For instance, when detecting 10,000 cells, .

Line 94: the concept of "percentage of gene expression" is never clearly defined. Does this mean the authors detect 99.86% of genes expressed in some cells? How is "expressed" defined - is this just detecting a single UMI? 

The term "percentage gene expression" refers to the proportion of genes in the bacterial strain that were detected as expressed in the sequenced cell population. Specifically, in this context, it means that 99.86% of all genes in the bacterial strain were detected as expressed in at least one cell in our sequencing results. To define "expressed" more clearly: a gene is considered expressed if at least one UMI (Unique Molecular Identifier) detected in a cell in the population. This definition allows for the detection of even low-level gene expression. To enhance clarity in the manuscript, we have rephrased the sentence as “transcriptome-wide gene coverage across the cell population”.

Line 98: The authors discuss the number of recovered UMIs throughout this paragraph, but there is no clear discussion of the number of detected expressed genes per cell. Could the authors include a discussion of this as well, as this is another important measure of sensitivity? 

We appreciate your suggestion to include a discussion on the number of detected expressed genes per cell, as this is indeed another important measure of sensitivity. We would like to clarify that we have actually included statistics on the number of genes detected across all cells in the main text of our paper. This information is presented as percentages. However, we understand that you may be looking for a more detailed representation, similar to the UMI statistics we provided. To address this, we have now added a new analysis showing the number of genes detected per cell (lines 132-133, 138-139, 144-145 and 184-186, Fig. 2B, 3B and S2B). This additional result complements our existing UMI data and provides a more comprehensive view of the sensitivity of our method. We have included this new gene-per-cell statistical graph in the supplementary materials.

Figure 1B: I presume ctrl and delta delta represent the classic PETRI-seq and RiboD protocols, respectively, but this is not specified. This should be clarified in the figure caption, or the names changed. 

We appreciate you bringing this to our attention. We acknowledge that the labeling in the figure could have been clearer. We have now clarified this information in the figure caption. To provide more specificity: The "ΔΔ" label represents the RiboD-PETRI protocol; The "Ctrl" label represents the classic PETRI-seq protocol we performed. We have updated the figure caption to include these details, which should help readers better understand the protocols being compared in the figure.​

Line 104: the authors claim "This performance surpassed other reported bacterial scRNA-seq methods" with a long number of references to other methods. "Performance" is not clearly defined, and it is unclear what the exact claim being made is. The authors should clarify what they're claiming, and further discuss the other methods and comparisons they have made with them in a thorough and fair fashion. 

We appreciate your request for clarification, and we acknowledge that our definition of "performance" should have been more explicit. We would like to clarify that in this context, we define performance primarily in terms of the proportion of mRNA captured. Our improved method demonstrates a significantly higher rate of rRNA removal compared to other bacterial single-cell library construction methods. This results in a higher proportion of mRNA in our sequencing data, which we consider a key performance metric for single-cell RNA sequencing in bacteria. Additionally, when compared to our previous method, PETRI-seq, our improved approach not only enhances rRNA removal but also reduces library construction costs. This dual improvement in both data quality and cost-effectiveness is what we intended to convey with our performance claim.

We recognize that a more thorough and fair discussion of other methods and their comparisons would be beneficial. We have summarized the comparison in Table S4 and make a short text discussion in the main text (lines 106-120). This addition provides context for our method and clarifies its position among existing techniques.

Figure 1D: Do the authors have any explanation for the relatively lower performance of their C. crescentus depletion? 

We appreciate your attention to detail and the opportunity to address this point. The lower efficiency of rRNA removal in C. crescentus compared to other species can be attributed to inherent differences between species. It's important to note that a single method for rRNA depletion may not be universally effective across all bacterial species due to variations in their genetic makeup and rRNA structures. Different bacterial species can have unique rRNA sequences, secondary structures, or associated proteins that may affect the efficiency of our depletion method. This species-specific variation highlights the challenges in developing a one-size-fits-all approach for bacterial rRNA depletion. While our method has shown high efficiency across several species, the results with C. crescentus underscore the need for continued refinement and possibly species-specific optimizations in rRNA depletion techniques. We thank you for bringing attention to this point, as it provides valuable insight into the complexities of bacterial rRNA depletion and areas for future improvement in our method.

Line 118: The authors claim RiboD-PETRI has a "consistent ability to unveil within-population heterogeneity", however the preceding paragraph shows it detects potential heterogeneity, but provides no evidence this inferred heterogeneity reflects the reality of gene expression in individual cells. 

We appreciate your careful reading and the opportunity to clarify this point. We acknowledge that our wording may have been too assertive given the evidence presented. We acknowledge that the subpopulations of cells identified in other species have not undergone experimental verification. Our intention in presenting these results was to demonstrate RiboD-PETRI's capability to detect “potential” heterogeneity consistently across different bacterial species, showcasing the method's sensitivity and potential utility in exploring within-population diversity. However, we agree that without further experimental validation, we cannot definitively claim that these detected differences represent true biological heterogeneity in all cases. We have revised this section to reflect the current state of our findings more accurately, emphasizing that while RiboD-PETRI consistently detects potential heterogeneity across species, further experimental validation would be required to confirm the biological significance of the observations (lines 169-171).

Figure 1 H&I: I'm not entirely sure what I am meant to see in these figures, presumably some evidence for heterogeneity in gene expression. Are there better visualizations that could be used to communicate this? 

We appreciate your suggestion for improving the visualization of gene expression heterogeneity. We have explored alternative visualization methods in the revised manuscript. Specifically, for the expression levels of marker genes shown in Figure 1H (which is Figure 2D now), we have created violin plots (Supplementary Fig. 4). These plots offer a more comprehensive view of the distribution of expression levels across different cell populations, making it easier to discern heterogeneity. However, due to the number of marker genes and the resulting volume of data, these violin plots are quite extensive and would occupy a significant amount of space. Given the space constraints of the main figure, we propose to include these violin plots as a Fig. S4 immediately following Figure 1 H&I (which is Figure 2D&E now). This arrangement will allow readers to access more detailed information about these marker genes while maintaining the concise style of the main figure.

Regarding the pathway enrichment figure (Figure 2E), we have also considered your suggestion for improvement. We attempted to use a dot plot to display the KEGG pathway enrichment of the genes. However, our analysis revealed that the genes were only enriched in a single pathway. As a result, the visual representation using a dot plot still did not produce a particularly aesthetically pleasing or informative figure.

Line 124: The authors state no significant batch effect was observed, but in the methods on line 344 they specify batch effects were removed using Harmony. It's unclear what exactly S2 is showing without a figure caption, but the authors should clarify this discrepancy. 

We apologize for any confusion caused by the lack of a clear figure caption for Figure S2 (which is Figure S3D now). To address your concern, in addition to adding figure captions for supplementary figure, we would also like to provide more context about the batch effect analysis. In Supplementary Fig. S3, Panel C represents the results without using Harmony for batch effect removal, while Panel D shows the results after applying Harmony. In both panels A and B, the distribution of samples one and two do not show substantial differences. Based on this observation, we concluded that there was no significant batch effect between the two samples. However, we acknowledge that even subtle batch effects could potentially influence downstream analyses. Therefore, out of an abundance of caution and to ensure the highest quality of our results, we decided to apply Harmony to remove any potential minor batch effects. This approach aligns with best practices in single-cell analysis, where even small technical variations are often accounted for to enhance the robustness of the results.

To improve clarity, we have revised our manuscript to better explain this nuanced approach: 1. We have updated the statement to reflect that while no major batch effect was observed, we applied batch correction as a precautionary measure (lines 181-182). 2. We have added a detailed caption to Figure S3, explaining the comparison between non-corrected and batch-corrected data. 3. We have modified the methods section to clarify that Harmony was applied as a precautionary step, despite the absence of obvious batch effects (lines 492-493).

Figure 2D: I found this panel fairly uninformative, is there a better way to communicate this finding? 

Thank you for your feedback regarding Figure 2D. We have explored alternative ways to present this information, using a dot plot to display the enrichment pathways, as this is often an effective method for visualizing such data. Meanwhile, we also provided a more detailed textual description of the enrichment results in the main text, highlighting the most significant findings.

Figure 2I: the figure itself and caption say GFP, but in the text and elsewhere the authors say this is a BFP fusion. 

We appreciate your careful review of our manuscript and figures. We apologize for any confusion this may have caused. To clarify: Both GFP (Green Fluorescent Protein) and BFP (Blue Fluorescent Protein) were indeed used in our experiments, but for different purposes: 1. GFP was used for imaging to observe location of PdeI in bacteria and persister cell growth, which is shown in Figure 4C and 4K. 2. BFP was used for cell sorting, imaging of location in biofilm, and detecting the proportion of persister cells which shown in Figure 4D, 4F-J. To address this inconsistency and improve clarity, we will make the following corrections: 1. We have reviewed the main text to ensure that references to GFP and BFP are accurate and consistent with their respective uses in our experiments. 2. We have added a note in the figure caption for Figure 4C to explicitly state that this particular image shows GFP fluorescence for location of PdeI. 3. In the methods section, we have provided a clear explanation of how both fluorescent proteins were used in different aspects of our study (lines 326-340).

Line 156: The authors compare prices between RiboD and PETRI-seq. It would be helpful to provide a full cost breakdown, e.g. in supplementary information, as it is unclear exactly how the authors came to these numbers or where the major savings are (presumably in sequencing depth?) 

We appreciate your suggestion to provide a more detailed cost breakdown, and we agree that this would enhance the transparency and reproducibility of our cost analysis. In response to your feedback, we have prepared a comprehensive cost breakdown that includes all materials and reagents used in the library preparation process. Additionally, we've factored in the sequencing depth (50G) and the unit price for sequencing (25¥/G). These calculations allow us to determine the cost per cell after sequencing. As you correctly surmised, a significant portion of the cost reduction is indeed related to sequencing depth. However, there are also savings in the library preparation steps that contribute to the overall cost-effectiveness of our method. We propose to include this detailed cost breakdown as a supplementary table (Table S6) in our paper. This table will provide a clear, itemized list of all expenses involved, including: 1. Reagents and materials for library preparation 2. Sequencing costs (depth and price per G) 3. Calculated cost per cell.

Line 291: The design and production of the depletion probes are not clearly explained. How did the authors design them? How were they synthesized? Also, it appears the authors have separate probe sets for E. coli, C. crescentus, and S. aureus - this should be clarified, possibly in the main text.

Thank you for your important questions regarding the design and production of our depletion probes. We included the detailed probe information in Supplementary Table S1, however, we didn’t clarify the information in the main text due to the constrains of the requirements of the Short Report format in eLife. We appreciate the opportunity to provide clarifications. ​

The core principle behind our probe design is that the probe sequences are reverse complementary to the r-cDNA sequences. This design allows for specific recognition of r-cDNA. The probes are then bound to magnetic beads, allowing the r-cDNA-probe-bead complexes to be separated from the rest of the library. To address your specific questions: 1. Probe Design: We designed separate probe sets for E. coli, C. crescentus, and S. aureus. Each set was specifically constructed to be reverse complementary to the r-cDNA sequences of its respective bacterial species. This species-specific approach ensures high efficiency and specificity in rRNA depletion for each organism. The hybrid DNA complex wasthen removed by Streptavidin magnetic beads. 2. Probe Synthesis: The probes were synthesized based on these design principles. 3. Species-Specific Probe Sets: You are correct in noting that we used separate probe sets for each bacterial species. We have clarified this important point in the main text to ensure readers understand the specificity of our approach. To further illustrate this process, we have created a schematic diagram showing the principle of rRNA removal and clarified the design principle in figure legend, which we have included in the figure legend of Fig. 1A.

Line 362: I didn't see a description of the construction of the PdeI-BFP strain, I assume this would be important for anyone interested in the specific work on PdeI. 

Thank you for your astute observation regarding the construction of the PdeI-BFP strain. We appreciate the opportunity to provide this important information. The PdeI-BFP strain was constructed as follows: 1. We cloned the pdeI gene along with its native promoter region (250bp) into a pBAD vector. 2. The original promoter region of the pBAD vector was removed to avoid any potential interference. 3. This construction enables the expression of the PdeI-BFP fusion protein to be regulated by the native promoter of pdeI, thus maintaining its physiological control mechanisms. 4. The BFP coding sequence was fused to the pdeI gene to create the PdeI-BFP fusion construct. We have added a detailed description of the PdeI-BFP strain construction to our methods section (lines 327-334).

Reviewer #2 (Recommendations For The Authors): 

(1) General remarks: 

Reconsider using 'advanced' in the title. It is highly generic and misleading. Perhaps 'cost-efficient' would be a more precise substitute. 

Thank you for your valuable suggestion. After careful consideration, we have decided to use "improved" in the title. Firstly, our method presents an efficient solution to a persistent challenge in bacterial single-cell RNA sequencing, specifically addressing rRNA abundance. Secondly, it facilitates precise exploration of bacterial population heterogeneity. We believe our method encompasses more than just cost-effectiveness, justifying the use of the term "advanced."

Consider expanding the introduction. The introduction does not explain the setup of the biological question or basic details such as the organism(s) for which the technique has been developed, or which species biofilms were studied. 

Thank you for your valuable feedback regarding our introduction. We acknowledge our compressed writing style due to constrains of the requirements of the Short Report format in eLife. We appreciate opportunity to expand this crucial section of our manuscript, which will undoubtedly improve the clarity and impact of our manuscript's introduction.

We revised our introduction (lines 53-80) according to following principles:

(1) Initial Biological Question: We explained the initial biological question that motivated our research—understanding the heterogeneity in E. coli biofilms—to provide essential context for our technological development.

(2) Limitations of Existing Techniques: We briefly described the limitations of current single-cell sequencing techniques for bacteria, particularly regarding their application in biofilm studies.

(3) Introduction of Improved Technique: We introduced our improved technique, initially developed for E. coli.

(4) Research Evolution: We highlighted how our research has evolved, demonstrating that our technique is applicable not only to E. coli but also to Gram-positive bacteria and other Gram-negative species, showcasing the broad applicability of our method.

(5) Specific Organisms Studied: We provided examples of the specific organisms we studied, encompassing both Gram-positive and Gram-negative bacteria.

(6) Potential Implications: Finally, we outlined the potential implications of our technique for studying bacterial heterogeneity across various species and contexts, extending beyond biofilms.

(2) Writing remarks: 

43-45 Reword: "Thus, we address a persistent challenge in bacterial single-cell RNA-seq regarding rRNA abundance, exemplifying the utility of this method in exploring biofilm heterogeneity.". 

Thank you for highlighting this sentence and requesting a rewording. I appreciate the opportunity to improve the clarity and impact of our statement. We have reworded the sentence as: "Our method effectively tackles a long-standing issue in bacterial single-cell RNA-seq: the overwhelming abundance of rRNA. This advancement significantly enhances our ability to investigate the intricate heterogeneity within biofilms at unprecedented resolution." (lines 47-50)

49 "Biofilms, comprising approximately 80% of chronic and recurrent microbial infections in the human body..." - probably meant 'contribute to'. 

Thank you for catching this imprecision in our statement. We have reworded the sentence as: "​Biofilms contribute to approximately 80% of chronic and recurrent microbial infections in the human body...​"

54-55 Please expand on "this". 

Thank you for your request to expand on the use of "this" in the sentence. You're right that more clarity would be beneficial here. We have revised and expanded this section in lines 54-69.

81-84 Unclear why these species samples were either at exponential or stationary phases. The growth stage can influence the proportion of rRNA and other transcripts in the population. 

Thank you for raising this important point about the growth phases of the bacterial samples used in our study. We appreciate the opportunity to clarify our experimental design. To evaluate the performance of RiboD-PETRI, we designed a comprehensive assessment of rRNA depletion efficiency under diverse physiological conditions, specifically contrasting exponential and stationary phases. This approach allows us to understand how these different growth states impact rRNA depletion efficacy. Additionally, we included a variety of bacterial species, encompassing both gram-negative and gram-positive organisms, to ensure that our findings are broadly applicable across different types of bacteria. By incorporating these variables, we aim to provide insights into the robustness and reliability of the RiboD-PETRI method in various biological contexts. We have included this rationale in our result section (lines 99-106), providing readers with a clear understanding of our experimental design choices.

86 "compared TO PETRI-seq " (typo). 

We have corrected this typo in our manuscript.

94 "gene expression collectively" rephrase. Probably this means coverage of the entire gene set across all cells. Same for downstream usage of the phrase. 

Thank you for pointing out this ambiguity in our phrasing. Your interpretation of our intended meaning is accurate. We have rephrased the sentence as “transcriptome-wide gene coverage across the cell population”.

97 What were the median UMIs for the 30,000 cell library {greater than or equal to}15 UMIs? Same question for the other datasets. This would reflect a more comparable statistic with previous studies than the top 3% of the cells for example, since the distributions of the single-cell UMIs typically have a long tail. 

Thank you for this insightful question and for pointing out the importance of providing more comparable statistics. We agree that median values offer a more robust measure of central tendency, especially for datasets with long-tailed distributions, which are common in single-cell studies. The suggestion to include median Unique Molecular Identifier (UMI) counts would indeed provide a more comparable statistic with previous studies. We have analyzed the median UMIs for our libraries as follows and revised our manuscript according to the analysis (lines 126-130, 133-136, 139-142 and 175-180).

(1) Median UMI count in Exponential Phase E. coli:

Total: 102 UMIs per cell

Top 1,000 cells: 462 UMIs per cell

Top 5,000 cells: 259 UMIs per cell

Top 10,000 cells: 193 UMIs per cell

(2) Median UMI count in Stationary Phase S. aureus:

Total: 142 UMIs per cell

Top 1,000 cells: 378 UMIs per cell

Top 5,000 cells: 207 UMIs per cell

Top 8,000 cells: 167 UMIs per cell

(3) Median UMI count in Exponential Phase C. crescentus:

Total: 182 UMIs per cell

Top 1,000 cells: 2,190 UMIs per cell

Top 5,000 cells: 662 UMIs per cell

Top 10,000 cells: 225 UMIs per cell

(4) Median UMI count in Static E. coli Biofilm:

Total of Replicate 1: 34 UMIs per cell

Total of Replicate 2: 52 UMIs per cell

Top 1,621 cells of Replicate 1: 283 UMIs per cell

Top 3,999 cells of Replicate 2: 239 UMIs per cell

104-105 The performance metric should again be the median UMIs of the majority of the cells passing the filter (15 mRNA UMIs is reasonable). The top 3-5% are always much higher in resolution because of the heavy tail of the single-cell UMI distribution. It is unclear if the performance surpasses the other methods using the comparable metric. Recommend removing this line. 

We appreciate your suggestion regarding the use of median UMIs as a more appropriate performance metric, and we agree that comparing the top 3-5% of cells can be misleading due to the heavy tail of the single-cell UMI distribution. We have removed the line in question (104-105) that compares our method's performance based on the top 3-5% of cells in the revised manuscript. Instead, we focused on presenting the median UMI counts for cells passing the filter (≥15 mRNA UMIs) as the primary performance metric. This will provide a more representative and comparable measure of our method's performance. We have also revised the surrounding text to reflect this change, ensuring that our claims about performance are based on these more robust statistics (lines 126-130, 133-136, 139-142 and 175-180).

106-108 The sequencing saturation of the libraries (in %), and downsampling analysis should be added to illustrate this point. 

Thank you for your valuable suggestion. Your recommendation to add sequencing saturation and downsampling analysis is highly valuable and will help better illustrate our point. Based on your feedback, we have revised our manuscript by adding the following content:

To provide a thorough evaluation of our sequencing depth and library quality, we performed sequencing saturation analysis on our sequencing samples. The findings reveal that our sequencing saturation is 100% (Fig. 8A & B), indicating that our sequencing depth is sufficient to capture the diversity of most transcripts. To further illustrate the impact of our downstream analysis on the datasets, we have demonstrated the data distribution before and after applying our filtering criteria (Fig. S1B & C). These figures effectively visualized the influence of our filtering process on the data quality and distribution. After filtering, we can have a more refined dataset with reduced noise and outliers, which enhances the reliability of our downstream analyses.

We have also ensured that a detailed description of the sequencing saturation method is included in the manuscript to provide readers with a comprehensive understanding of our methodology. We appreciate your feedback and believe these additions significantly improve our work.

122: Please provide more details about the biofilm setup, including the media used. I did not find them in the methods. 

We appreciate your attention to detail, and we agree that this information is crucial for the reproducibility of our experiments. We propose to add the following information to our methods section (lines 311-318):

"For the biofilm setup, bacterial cultures were grown overnight. The next day, we diluted the culture 1:100 in a petri dish. We added 2ml of LB medium to the dish. If the bacteria contain a plasmid, the appropriate antibiotic needs to be added to LB. The petri dish was then incubated statically in a growth chamber for 24 hours. After incubation, we performed imaging directly under the microscope. The petri dishes used were glass-bottom dishes from Biosharp (catalog number BS-20-GJM), allowing for direct microscopic imaging without the need for cover slips or slides. This setup allowed us to grow and image the biofilms in situ, providing a more accurate representation of their natural structure and composition.​"

125: "sequenced 1,563 reads" missing "with" 

Thank you for correcting our grammar. We have revisd the phrase as “sequenced with 1,563 reads”.

126: "283/239 UMIs per cell" unclear. 283 and 239 UMIs per cell per replicate, respectively? 

Thank you for correcting our grammar. We have revised the phrase as “283 and 239 UMIs per cell per replicate, respectively” (lines 184).

Figure 1D: Please indicate where the comparison datasets are from. 

We appreciate your question regarding the source of the comparison datasets in Figure 1D. All data presented in Figure 1D are from our own sequencing experiments. We did not use data from other publications for this comparison. Specifically, we performed sequencing on E. coli cells in the exponential growth phase using three different library preparation methods: RiboD-PETRI, PETRI-seq, and RNA-seq. The data shown in Figure 1D represent a comparison of UMIs and/or reads correlations obtained from these three methods. All sequencing results have been uploaded to the Gene Expression Omnibus (GEO) database. The accession number is GSE260458. We have updated the figure legend for Figure 1D to clearly state that all datasets are from our own experiments, specifying the different methods used.

Figure 1I, 2D: Unable to interpret the color block in the data. 

We apologize for any confusion regarding the interpretation of the color blocks in Figures 1I and 2D (which are Figure 2E, 3E now). The color blocks in these figures represent the p-values of the data points. The color scale ranges from red to blue. Red colors indicate smaller p-values, suggesting higher statistical significance and more reliable results. Blue colors indicate larger p-values, suggesting lower statistical significance and less reliable results. We have updated the figure legends for both Figure 2E and Figure 3E to include this explanation of the color scale. Additionally, we have added a color legend to each figure to make the interpretation more intuitive for readers.

Figure1H and 2C: Gene names should be provided where possible. The locus tags are highly annotation-dependent and hard to interpret. Also, a larger size figure should be helpful. The clusters 2 and 3 in 2C are the most important, yet because they have few cells, very hard to see in this panel. 

We appreciate your suggestions for improving the clarity and interpretability of Figures 1H and 2C (which is Figure 2D, 3D now). We have replaced the locus tags with gene names where possible in both figures. We have increased the size of both figures to improve visibility and readability. We have also made Clusters 2 and 3 in Figure 3D more prominent in the revised figure. Despite their smaller cell count, we recognize their importance and have adjusted the visualization to ensure they are clearly visible. We believe these modifications will significantly enhance the clarity and informativeness of Figures 2D and 3D.​

(3) Questions to consider further expanding on, by more analyses or experiments and in the discussion: 

What are the explanations for the apparently contradictory upregulation of c-di-GMP in cells expressing higher PdeI levels? How could a phosphodiesterase lead to increased c-di-GMP levels? 

We appreciate the reviewer's observation regarding the seemingly contradictory relationship between increased PdeI expression and elevated c-di-GMP levels. This is indeed an intriguing finding that warrants further explanation.

PdeI was predicted to be a phosphodiesterase responsible for c-di-GMP degradation. This prediction is based on sequence analysis where PdeI contains an intact EAL domain known for degrading c-di-GMP. However, it is noteworthy that PdeI also contains a divergent GGDEF domain, which is typically associated with c-di-GMP synthesis (Fig S8). This dual-domain architecture suggests that PdeI may engage in complex regulatory roles. Previous studies have shown that the knockout of the major phosphodiesterase PdeH in E. coli leads to the accumulation of c-di-GMP. Further, a point mutation on PdeI's divergent GGDEF domain (G412S) in this PdeH knockout strain resulted in decreased c-di-GMP levels2, implying that the wild-type GGDEF domain in PdeI contributes to the maintenance or increase of c-di-GMP levels in the cell. Importantly, our single-cell experiments showed a positive correlation between PdeI expression levels and c-di-GMP levels (Response Fig. 9B). In this revision, we also constructed PdeI(G412S)-BFP mutation strain. Notably, our observations of this strain revealed that c-di-GMP levels remained constant despite increasing BFP fluorescence, which serves as a proxy for PdeI(G412S) expression levels (Fig. 4D). This experimental evidence, along with domain analysis, suggests that PdeI could contribute to c-di-GMP synthesis, rebutting the notion that it solely functions as a phosphodiesterase. HPLC LC-MS/MS analysis further confirmed that PdeI overexpression, induced by arabinose, led to an upregulation of c-di-GMP levels (Fig. 4E). These results strongly suggest that PdeI plays a significant role in upregulating c-di-GMP levels. Our further analysis revealed that PdeI contains a CHASE (cyclases/histidine kinase-associated sensory) domain. Combined with our experimental results demonstrating that PdeI is a membrane-associated protein, we hypothesize that PdeI functions as a sensor that integrates environmental signals with c-di-GMP production under complex regulatory mechanisms.

We have also included this explanation (lines 193-217) and the supporting experimental data (Fig. 4D & 4J) in our manuscript to clarify this important point. Thank you for highlighting this apparent contradiction, as it has allowed us to provide a more comprehensive explanation of our findings.

What about the rest of the genes in cluster 2 of the biofilm? They should be used to help interpret the association between PdeI and c-di-GMP. 

We understand your interest in the other genes present in cluster 2 of the biofilm and their potential relationship to PdeI and c-di-GMP. After careful analysis, we have determined that the other marker genes in this cluster do not have a significant impact on biofilm formation. Furthermore, we have not found any direct relationship between these genes and c-di-GMP or PdeI. Our focus on PdeI in this cluster is due to its unique and significant role in c-di-GMP regulation and biofilm formation, as demonstrated by our experimental results. While the other genes in this cluster may be co-expressed, their functions appear to be unrelated to the PdeI and c-di-GMP pathway we are investigating. We chose not to elaborate on these genes in our main discussion as they do not contribute directly to our understanding of the PdeI and c-di-GMP association. Instead, we could include a brief mention of these genes in the manuscript, noting that they were found to be unrelated to the PdeI-c-di-GMP pathway. This would provide a more comprehensive view of the cluster composition while maintaining focus on the key findings related to PdeI and c-di-GMP.

Author response image 2.

Protein-protein interactions of marker genes in cluster 2 of 24-hour static biofilms of E coli data.

A verification is needed that the protein fusion to PdeI functional/membrane localization is not due to protein interactions with fluorescent protein fusion. 

We appreciate your concern regarding the potential impact of the fluorescent protein fusion on the functionality and membrane localization of PdeI. It is crucial to verify that the observed effects are attributable to PdeI itself and not an artifact of its fusion with the fluorescent protein. To address this matter, we have incorporated a control group expressing only the fluorescent protein BFP (without the PdeI fusion) under the same promoter. This experimental design allows us to differentiate between effects caused by PdeI and those potentially arising from the fluorescent protein alone.

Our results revealed the following key observations:

(1) Cellular Localization: The GFP alone exhibited a uniform distribution in the cytoplasm of bacterial cells, whereas the PdeI-GFP fusion protein was specifically localized to the membrane (Fig. 4C).

(2) Localization in the Biofilm Matrix: BFP-positive cells were distributed throughout the entire biofilm community. In contrast, PdeI-BFP positive cells localized at the bottom of the biofilm, where cell-surface adhesion occurs (Fig 4F).

(3) c-di-GMP Levels: Cells with high levels of BFP displayed no increase in c-di-GMP levels. Conversely, cells with high levels of PdeI-BFP exhibited a significant increase in c-di-GMP levels (Fig. 4D).

(4) Persister Cell Ratio: Cells expressing high levels of BFP showed no increase in persister ratios, while cells with elevated levels of PdeI-BFP demonstrated a marked increase in persister ratios (Fig. 4J).

These findings from the control experiments have been included in our manuscript (lines 193-244, Fig. 4C, 4D, 4F, 4G and 4J), providing robust validation of our results concerning the PdeI fusion protein. They confirm that the observed effects are indeed due to PdeI and not merely artifacts of the fluorescent protein fusion.

(!) Vrabioiu, A. M. & Berg, H. C. Signaling events that occur when cells of Escherichia coli encounter a glass surface. Proceedings of the National Academy of Sciences of the United States of America 119, doi:10.1073/pnas.2116830119 (2022). https://doi.org/10.1073/pnas.2116830119

(2)bReinders, A. et al. Expression and Genetic Activation of Cyclic Di-GMP-Specific Phosphodiesterases in Escherichia coli. J Bacteriol 198, 448-462 (2016). https://doi.org:10.1128/JB.00604-15

Author response:

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

eLife Assessment

This study attempts to resolve an apparent paradox of rapid evolutionary rates of multi-copy gene systems by using a theoretical model that integrates two classic population models. While the conceptual framework is intuitive and thus useful, the specific model is perplexing and difficult to penetrate for non-specialists. The data analysis of rRNA genes provides inadequate support for the conclusions due to a lack of consideration of technical challenges, mutation rate variation, and the relationship between molecular processes and model parameters.

Overall Responses:

Since the eLife assessment succinctly captures the key points of the reviews, the reply here can be seen as the overall responses to the summed criticisms. We believe that the overview should be sufficient to address the main concerns, but further details can be found in the point-by-point responses below. The overview covers the same grounds as the provisional responses (see the end of this rebuttal) but is organized more systematically in response to the reviews. The criticisms together fall into four broad areas. 

First, the lack of engagement with the literature, particularly concerning Cannings models and non-diffusive limits. This is the main rebuttal of the companion paper (eLife-RP-RA-2024-99990). The literature in question is all in the WF framework and with modifications, in particular, with the introduction of V(K). Nevertheless, all WF models are based on population sampling. The Haldane model is an entirely different model of genetic drift, based on gene transmission. Most importantly, the WF models and the Haldane model differ in the ability to handle the four paradoxes presented in the two papers. These paradoxes are all incompatible with the WF models.

Second, the poor presentation of the model that makes the analyses and results difficult to interpret. In retrospect, we fully agree and thank all the reviewers for pointing them out. Indeed, we have unnecessarily complicated the model. Even the key concept that defines the paradox, which is the effective copy number of rRNA genes, is difficult to comprehend. We have streamlined the presentation now. Briefly, the complexity arose from the general formulation permitting V(K) ≠ E(K) even for single copy genes. (It would serve the same purpose if we simply let V(K) = E(K) for single copy genes.) The sentences below, copied from the new abstract, should clarify the issue. The full text in the Results section has all the details.

“On average, rDNAs have C ~ 150 - 300 copies per haploid in humans. While a neutral mutation of a single-copy gene would take 4N generations (N being the population size of an ideal population) to become fixed, the time should be 4NC* generations for rRNA genes (C* being the effective copy number). Note that C* >> 1, but C* < (or >) C would depend on the drift strength. Surprisingly, the observed fixation time in mouse and human is < 4N, implying the paradox of C* < 1.”

Third, the confusion about which rRNA gene is being compared with which homology, as there are hundreds of them. We should note that the effective copy number C* indicates that the rRNA gene arrays do not correspond with the “gene locus” concept. This is at the heart of the confusion we failed to remove clearly. We now use the term “pseudo-population” to clarify the nature of rDNA variation and evolution. The relevant passage is reproduced from the main text shown below.

“The pseudo-population of ribosomal DNA copies within each individual

While a human haploid with 200 rRNA genes may appear to have 200 loci, the concept of "gene loci" cannot be applied to the rRNA gene clusters. This is because DNA sequences can spread from one copy to others on the same chromosome via replication slippage. They can also spread among copies on different chromosomes via gene conversion and unequal crossovers (Nagylaki 1983; Ohta and Dover 1983; Stults, et al. 2008; Smirnov, et al. 2021). Replication slippage and unequal crossovers would also alter the copy number of rRNA genes. These mechanisms will be referred to collectively as the homogenization process. Copies of the cluster on the same chromosome are known to be nearly identical in sequences (Hori, et al. 2021; Nurk, et al. 2022). Previous research has also provided extensive evidence for genetic exchanges between chromosomes (Krystal, et al. 1981; Arnheim, et al. 1982; van Sluis, et al. 2019).

In short, rRNA gene copies in an individual can be treated as a pseudo-population of gene copies. Such a pseudo-population is not Mendelian but its genetic drift can be analyzed using the branching process (see below). The pseudo-population corresponds to the "chromosome community" proposed recently (Guarracino, et al. 2023). As seen in Fig. 1C, the five short arms harbor a shared pool of rRNA genes that can be exchanged among them. Fig. 1D presents the possible molecular mechanisms of genetic drift within individuals whereby mutations may spread, segregate or disappear among copies. Hence, rRNA gene diversity or polymorphism refers to the variation across all rRNA copies, as these genes exist as paralogs rather than orthologs. This diversity can be assessed at both individual and population levels according to the multi-copy nature of rRNA genes.”

Fourth, the lack of consideration of many technical challenges. We have responded to the criticisms point-by-point below. One of the main criticisms is about mutation rate differences between single-copy and rRNA genes. We did in fact alluded to the parity in mutation rate between them in the original text but should have presented this property more prominently as is done now. Below is copied from the revised text:

“We now consider the evolution of rRNA genes between species by analyzing the rate of fixation (or near fixation) of mutations. Polymorphic variants are filtered out in the calculation. Note that Eq. (3) shows that the mutation rate, m, determines the long-term evolutionary rate, l. Since we will compare the l values between rRNA and single-copy genes, we have to compare their mutation rates first by analyzing their long-term evolution. As shown in Table S1, l falls in the range of 50-60 (differences per Kb) for single copy genes and 40 – 70 for the non-functional parts of rRNA genes. The data thus suggest that rRNA and single-copy genes are comparable in mutation rate. Differences between their l values will have to be explained by other means.”

While the overview should address the key issues, we now present the point-by-point response below. 

Public Reviews:

Reviewer #1 (Public Review):

The manuscript by Wang et al is, like its companion paper, very unusual in the opinion of this reviewer. It builds off of the companion theory paper's exploration of the "Wright-Fisher Haldane" model but applies it to the specific problem of diversity in ribosomal RNA arrays.

The authors argue that polymorphism and divergence among rRNA arrays are inconsistent with neutral evolution, primarily stating that the amount of polymorphism suggests a high effective size and thus a slow fixation rate, while we, in fact, observe relatively fast fixation between species, even in putatively non-functional regions.

They frame this as a paradox in need of solving, and invoke the WFH model.

The same critiques apply to this paper as to the presentation of the WFH model and the lack of engagement with the literature, particularly concerning Cannings models and non-diffusive limits. However, I have additional concerns about this manuscript, which I found particularly difficult to follow.

Response 1: We would like to emphasize that, despite the many modified WF models, there has not been a model for quantifying genetic drift in multi-copy gene systems, due to the complexity of two levels of genetic drift – within individuals as well as between individuals of the population. We will address this question in the revised manuscript (Ruan, et al. 2024) and have included a mention of it in the text as follows:

“In the WF model, gene frequency is governed by 1/N (or 1/2_N_ in diploids) because K would follow the Poisson distribution whereby V(K) = E(K). As E(K) is generally ~1, V(K) would also be ~ 1. In this backdrop, many "modified WF" models have been developed(Der, et al. 2011), most of them permitting V(K) ≠ E(K) (Karlin and McGregor 1964; Chia and Watterson 1969; Cannings 1974). Nevertheless, paradoxes encountered by the standard WF model apply to these modified WF models as well because all WF models share the key feature of gene sampling (see below and (Ruan, et al. 2024)). ”

My first, and most major, concern is that I can never tell when the authors are referring to diversity in a single copy of an rRNA gene compared to when they are discussing diversity across the entire array of rRNA genes. I admit that I am not at all an expert in studies of rRNA diversity, so perhaps this is a standard understanding in the field, but in order for this manuscript to be read and understood by a larger number of people, these issues must be clarified.

Response 2: We appreciate the reviewer’s feedback and acknowledge that the distinction between the diversity of individual rRNA gene copies and the diversity across the entire array of rRNA genes may not have been clearly defined in the original manuscript. The diversity in our manuscript is referring to the genetic diversity of the population of rRNA genes in the cell. To address this concern, we have revised the relevant paragraph in the text:

“Hence, rRNA gene diversity or polymorphism refer to the variation across all rRNA copies, as these genes exist as paralogs rather than orthologs. This diversity can be assessed at both individual and population levels according to the multi-copy nature of rRNA genes.”

Additionally, we have updated the Methods section to include a detailed description of how diversity is measured as follows:

“All mapping and analysis are performed among individual copies of rRNA genes.

Each individual was considered as a psedo-population of rRNA genes and the diversity of rRNA genes was calculated using this psedo-population of rRNA genes.”

The authors frame the number of rRNA genes as roughly equivalent to expanding the population size, but this seems to be wrong: the way that a mutation can spread among rRNA gene copies is fundamentally different than how mutations spread within a single copy gene. In particular, a mutation in a single copy gene can spread through vertical transmission, but a mutation spreading from one copy to another is fundamentally horizontal: it has to occur because some molecular mechanism, such as slippage, gene conversion, or recombination resulted in its spread to another copy. Moreover, by collapsing diversity across genes in an rRNA array, the authors are massively increasing the mutational target size.   

For example, it's difficult for me to tell if the discussion of heterozygosity at rRNA genes in mice starting on line 277 is collapsed or not. The authors point out that Hs per kb is ~5x larger in rRNA than the rest of the genome, but I can't tell based on the authors' description if this is diversity per single copy locus or after collapsing loci together. If it's the first one, I have concerns about diversity estimation in highly repetitive regions that would need to be addressed, and if it's the second one, an elevated rate of polymorphism is not surprising, because the mutational target size is in fact significantly larger.

Response 3: As addressed in previous Response2, the measurement of diversity or heterozygosity of rRNA genes is consistently done by combining copies, as there is no concept of single gene locus for rDNAs. We agree that by combining the diversity across multiple rRNA gene copies into one measurement, the mutational target size is effectively increased, leading to higher observed levels of diversity than one gene. This is in line with our text:

“If we use the polymorphism data, it is as if rDNA array has a population size 5.2 times larger than single-copy genes. Although the actual copy number on each haploid is ~ 110, these copies do not segregate like single-copy genes and we should not expect N* to be 100 times larger than N. The HS results confirm the prediction that rRNA genes should be more polymorphic than single-copy genes.”

Under this consensus, the reviewer points out that the having a large number of rRNA genes is not equivalent to having a larger population size, because the spreading of mutations among rDNA copies within a species involves two stages: within individual (horizontal transmission) and between individuals (vertical transmission). Let’s examine how the mutation spreading mechanisms influence the population size of rRNA genes.

First, an increase in the copy number of rRNA genes dose increase the actual population size (CN) of rRNA genes. If reviewer is referring to the effective population size of rRNA genes in the context of diversity (N* = CN/V*(K)), then an increase in C would also increase N*. In addition, the linkage among copies would reduce the drift effect, leading to increase diversity. Conversely, homogenization mechanism, like gene conversion and unequal crossing-over would reduce genetic variations between copies and increase V*(K), leading to lower diversity. Therefore, the C* =C/V*(K) in mice is about 5 times larger for rRNA genes than the rest of the genome (which mainly single-copy genes), even though the actual copy number is about 110, indicating a high homogenization rate.

Even if these issues were sorted out, I'm not sure that the authors framing, in terms of variance in reproductive success is a useful way to understand what is going on in rRNA arrays. The authors explicitly highlight homogenizing forces such as gene conversion and replication slippage but then seem to just want to incorporate those as accounting for variance in reproductive success. However, don't we usually want to dissect these things in terms of their underlying mechanism? Why build a model based on variance in reproductive success when you could instead explicitly model these homogenizing processes? That seems more informative about the mechanism, and it would also serve significantly better as a null model, since the parameters would be able to be related to in vitro or in vivo measurements of the rates of slippage, gene conversion, etc.

In the end, I find the paper in its current state somewhat difficult to review in more detail, because I have a hard time understanding some of the more technical aspects of the manuscript while so confused about high-level features of the manuscript. I think that a revision would need to be substantially clarified in the ways I highlighted above.

Response 4: We appreciate your perspective on modeling the homogenizing processes of rRNA gene arrays.

We employ the WFH model to track the drift effect of the multi-copy gene system. In the context of the Haldane model, the term K is often referred to as reproductive success, but it might be more accurate to interpret it as “transmission rate” in this study. As stated in the caption of Figure 1D, two new mutations can have very large differences in individual output (K) when transmitted to the next generation through homogenization process.

Regarding why we did not explicitly model different mechanisms of homogenization, previous elegant models of multigene families have involved mechanisms like unequal crossing over(Smith 1974a; Ohta 1976; Smith 1976) or gene conversion (Nagylaki 1983; Ohta 1985) for concerted evolution, or using conversion to approximate the joint effect of conversion and crossing over (Ohta and Dover 1984). However, even when simplifying the gene conversion mechanism, modeling remains challenging due to controversial assumptions, such as uniform homogenization rate across all gene members (Dover 1982; Ohta and Dover 1984). No models can fully capture the extreme complexity of factors, while these unbiased mechanisms are all genetic drift forces that contribute to changes in mutant transmission. Therefore, we opted for a more simplified and collective approach using V*(K) to see the overall strength of genetic drift.

We have discussed the reason for using V*(K) to collectively represent the homogenization effect in Discussion. As stated in our manuscript:

“There have been many rigorous analyses that confront the homogenizing mechanisms directly. These studies (Smith 1974b; Ohta 1976; Dover 1982; Nagylaki 1983; Ohta and Dover 1983) modeled gene conversion and unequal cross-over head on. Unfortunately, on top of the complexities of such models, the key parameter values are rarely obtainable. In the branching process, all these complexities are wrapped into V*(K) for formulating the evolutionary rate. In such a formulation, the collective strength of these various forces may indeed be measurable, as shown in this study.”

Reviewer #2 (Public Review):

Summary:

Multi-copy gene systems are expected to evolve slower than single-copy gene systems because it takes longer for genetic variants to fix in the large number of gene copies in the entire population. Paradoxically, their evolution is often observed to be surprisingly fast. To explain this paradox, the authors hypothesize that the rapid evolution of multi-copy gene systems arises from stronger genetic drift driven by homogenizing forces within individuals, such as gene conversion, unequal crossover, and replication slippage. They formulate this idea by combining the advantages of two classic population genetic models -- adding the V(k) term (which is the variance in reproductive success) in the Haldane model to the Wright-Fisher model. Using this model, the authors derived the strength of genetic drift (i.e., reciprocal of the effective population size, Ne) for the multi-copy gene system and compared it to that of the single-copy system. The theory was then applied to empirical genetic polymorphism and divergence data in rodents and great apes, relying on comparison between rRNA genes and genome-wide patterns (which mostly are single-copy genes). Based on this analysis, the authors concluded that neutral genetic drift could explain the rRNA diversity and evolution patterns in mice but not in humans and chimpanzees, pointing to a positive selection of rRNA variants in great apes.

Strengths:

Overall, the new WFH model is an interesting idea. It is intuitive, efficient, and versatile in various scenarios, including the multi-copy gene system and other cases discussed in the companion paper by Ruan et al.

Weaknesses:

Despite being intuitive at a high level, the model is a little unclear, as several terms in the main text were not clearly defined and connections between model parameters and biological mechanisms are missing. Most importantly, the data analysis of rRNA genes is extremely over-simplified and does not adequately consider biological and technical factors that are not discussed in the model. Even if these factors are ignored, the authors' interpretation of several observations is unconvincing, as alternative scenarios can lead to similar patterns. Consequently, the conclusions regarding rRNA genes are poorly supported. Overall, I think this paper shines more in the model than the data analysis, and the modeling part would be better presented as a section of the companion theory paper rather than a stand-alone paper. My specific concerns are outlined below.

Response 5: We appreciate the reviewer’s feedback and recognize the need for clearer definitions of key terms. We have made revisions to ensure that each term is properly defined upon its first use.

Regarding the model’s simplicity, as in the Response4, our intention was to create a framework that captures the essence of how mutant copies spread by chance within a population, relying on the variance in transmission rates for each copy (V(K)). By doing so, we aimed to incorporate the various homogenization mechanisms that do not affect single-copy genes, highlighting the substantially stronger genetic drift observed in multi-copy systems compared to single-copy genes. We believe that simplifying the model was necessary to make it more accessible and practical for real-world data analysis and provides a useful approximation that can be applied broadly. It is clearly an underestimate the actual rate as some forces with canceling effects might not have been accounted for.

(1) Unclear definition of terms

Many of the terms in the model or the main text were not clearly defined the first time they occurred, which hindered understanding of the model and observations reported. To name a few:

(i) In Eq(1), although C* is defined as the "effective copy number", it is unclear what it means in an empirical sense. For example, Ne could be interpreted as "an ideal WF population with this size would have the same level of genetic diversity as the population of interest" or "the reciprocal of strength of allele frequency change in a unit of time". A few factors were provided that could affect C*, but specifically, how do these factors impact C*? For example, does increased replication slippage increase or decrease C*? How about gene conversion or unequal cross-over? If we don't even have a qualitative understanding of how these processes influence C*, it is very hard to make interpretations based on inferred C*. How to interpret the claim on lines 240-241 (If the homogenization is powerful enough, rRNA genes would have C*<1)? Please also clarify what C* would be, in a single-copy gene system in diploid species.

Response 6: We apology for the confusion caused by the lack of clear definitions in the initial manuscript. We recognize that this has led to misunderstandings regarding the concept we presented. Our aim was to demonstrate the concerted evolution in multi-copy gene systems, involving two levels of “effective copy number” relative to single-copy genes: first, homogenization within populations then divergence between species. We used C* and Ne* to try to designated the two levels driven by the same homogenization force, which complicated the evolutionary pattern.

To address these issues, we have simplified the model and revised the abstract to prevent any misunderstandings:

“On average, rDNAs have C ~ 150 - 300 copies per haploid in humans. While a neutral mutation of a single-copy gene would take 4_N_ (N being the population size) generations to become fixed, the time should be 4_NC* generations for rRNA genes where 1<< C* (C* being the effective copy number; C* < C or C* > C would depend on the drift strength). However, the observed fixation time in mouse and human is < 4_N, implying the paradox of C* < 1. Genetic drift that encompasses all random neutral evolutionary forces appears as much as 100 times stronger for rRNA genes as for single-copy genes, thus reducing C* to < 1.”

Thus, it should be clear that the fixation time as well as the level of polymorphism represent the empirical measures of C*.We have also revised the relevant paragraph in the text to define C* and V*(K) and removed Eq. 2 for clarity:

“Below, we compare the strength of genetic drift in rRNA genes vs. that of single-copy genes using the Haldane model (Ruan, et al. 2024). We shall use * to designate the equivalent symbols for rRNA genes; for example, E(K) vs. E*(K). Both are set to 1, such that the total number of copies in the long run remains constant.

For simplicity, we let V(K) = 1 for single-copy genes. (If we permit V(K) ≠ 1, the analyses will involve the ratio of V*(K) and V(K) to reach the same conclusion but with unnecessary complexities.) For rRNA genes,  V*(K) ≥ 1 may generally be true because K for rDNA mutations are affected by a host of homogenization factors including replication slippage, unequal cross-over, gene conversion and other related mechanisms not operating on single copy genes. Hence,

where C is the average number of rRNA genes in an individual and V*(K) reflects the homogenization process on rRNA genes (Fig. 1D). Thus,

C* = C/V*(K)

represents the effective copy number of rRNA genes in the population, determining the level of genetic diversity relative to single-copy genes. Since C is in the hundreds and V*(K) is expected to be > 1, the relationship of 1 << C* ≤ C is hypothesized. Fig. 1D is a simple illustration that the homogenizing process may enhance V*(K) substantially over the WF model.

In short, genetic drift of rRNA genes would be equivalent to single copy genes in a population of size NC* (or N*). Since C* >> 1 is hypothesized, genetic drift for rRNA genes is expected to be slower than for single copy genes.”

(ii) In Eq(1), what exactly is V*(K)? Variance in reproductive success across all gene copies in the population? What factors affect V*(K)? For the same population, what is the possible range of V*(K)/V(K)? Is it somewhat bounded because of biological constraints? Are V*(K) and C*(K) independent parameters, or does one affect the other, or are both affected by an overlapping set of factors?

Response 7: - In Eq(1), what exactly is V*(K)?  In Eq(1), V*(K) refers to the variance in the number of progeny to whom the gene copy of interest is transmitted (K) over a specific time interval. When considering evolutionary divergence between species, V*(K) may correspond to the divergence time.

- What factors affect V*(K)? For the same population, what is the possible range of V*(K)/V(K)? Is it somewhat bounded because of biological constraints?  “V*(K) for rRNA genes is likely to be much larger than V(K) for single-copy genes, because K for rRNA mutations may be affected by a host of homogenization factors including replication slippage, unequal cross-over, gene conversion and other related mechanisms not operating on single-copy genes. For simplicity, we let V(K) = 1 (as in a WF population) and V*(K) ≥ 1.” Thus, the V*(K)/V(K) = V*(K) can potentially reach values in the hundreds, and may even exceed C, resulting in C*(= C/V*(K)) values less than 1. Biological constraints that could limit this variance include the minimum copy number within individuals, sequence constraints in functional regions, and the susceptibility of chromosomes with large arrays to intrachromosomal crossover (which may lead to a reduction in copy number)(Eickbush and Eickbush 2007), potentially reducing the variability of K.

- Are V*(K) and C*(K) independent parameters, or does one affect the other, or are both affected by an overlapping set of factors?  There is no C*(K), the C* is defined as follows in the text:

“C* = C/V*(K) represents the effective copy number of rRNA genes, reflecting the level of genetic diversity relative to single-copy genes. Since C is in the hundreds and V*(K) is expected to be > 1, the relationship of 1 << C* ≤ C is hypothesized.” The factors influencing V*(K) directly affect C* due to this relationship.

(iii) In the multi-copy gene system, how is fixation defined? A variant found at the same position in all copies of the rRNA genes in the entire population?

Response 8: We appreciate the reviewer's suggestion and have now provided a clear definition of fixation in the context of multi-copy genes within the manuscript.

“For rDNA mutations, fixation must occur in two stages – fixation within individuals and among individuals in the population. (Note that a new mutation can be fixed via homogenization, thus making rRNA gene copies in an individual a pseudo-population.)”

The evolutionary dynamics of multi-copy genes differ from those of single-copy (Mendelian) genes, which mutate, segregate and evolve independently in the population. Fixation in multi-copy genes, such as rRNA genes, is influenced by their ability to transfer genetic information among their copies through nonreciprocal exchange mechanisms, like gene conversion and unequal crossover (Ohta and Dover 1984). These processes can cause fluctuations in the number of mutant copies within an individual's lifetime and facilitate the spread of a mutant allele across all copies even in non-homologous chromosomes. Over time, this can result in the mutant allele replacing all preexisting alleles throughout the population, leading to fixation (Ohta 1976) meaning that the same variant will eventually be present at the corresponding position in all copies of the rRNA genes across the entire population. Without such homogenization processes, fixation would be unlikely to be obtained in multi-copy genes.

(iv) Lines 199-201, HI, Hs, and HT are not defined in the context of a multi-copy gene system. What are the empirical estimators?

Response 9: We appreciate the reviewer's comment and would like to clarify the definitions and empirical estimators for within the context of a multi-copy gene system in the text:

“A standard measure of genetic drift is the level of heterozygosity (H). At the mutation-selection equilibrium

where μ is the mutation rate of the entire gene and Ne is the effective population size. In this study, Ne = N for single-copy gene and Ne = C*N for rRNA genes. The empirical measure of nucleotide diversity H is given by

where L is the gene length (for each copy of rRNA gene, L ~ 43kb) and pi is the variant frequency at the i-th site.

We calculate H of rRNA genes at three levels – within-individual, within-species and then, within total samples (HI, HS and HT, respectively). HS and HT are standard population genetic measures (Hartl, et al. 1997; Crow and Kimura 2009). In calculating HS, all sequences in the species are used, regardless of the source individuals. A similar procedure is applied to HT. The HI statistic is adopted for multi-copy gene systems for measuring within-individual polymorphism. Note that copies within each individual are treated as a pseudo-population (see Fig. 1 and text above). With multiple individuals, HI is averaged over them.”

(v) Line 392-393, f and g are not clearly defined. What does "the proportion of AT-to-GC conversion" mean? What are the numerator and denominator of the fraction, respectively?

Response 10: We appreciate the reviewer's comment and have revised the relevant text for clarity as well as improved the specific calculation methods for f and g in the Methods section.

“We first designate the proportion of AT-to-GC conversion as f and the reciprocal, GC-to-AT, as g. Specifically, f represents the proportion of fixed mutations where an A or T nucleotide has been converted to a G or C nucleotide (see Methods). Given f ≠ g, this bias is true at the site level.”

Methods:

“Specifically, f represents the proportion of fixed mutations where an A or T nucleotide has been converted to a G or C nucleotide. The numerator for f is the number of fixed mutations from A-to-G, T-to-C, T-to-G, or A-to-C. The denominator is the total number of A or T sites in the rDNA sequence of the specie lineage.

Similarly, g is defined as the proportion of fixed mutations where a G or C nucleotide has been converted to an A or T nucleotide. The numerator for g is the number of fixed mutations from G-to-A, C-to-T, C-to-A, or G-to-T. The denominator is the total number of G or C sites in the rDNA sequence of the specie lineage.

The consensus rDNA sequences for the species lineage were generated by Samtools consensus (Danecek, et al. 2021) from the bam file after alignment. The following command was used:

‘samtools consensus -@ 20 -a -d 10 --show-ins no --show-del yes input_sorted.bam output.fa’.”

(2) Technical concerns with rRNA gene data quality

Given the highly repetitive nature and rapid evolution of rRNA genes, myriads of things could go wrong with read alignment and variant calling, raising great concerns regarding the data quality. The data source and methods used for calling variants were insufficiently described at places, further exacerbating the concern.

(i) What are the accession numbers or sample IDs of the high-coverage WGS data of humans, chimpanzees, and gorillas from NCBI? How many individuals are in each species? These details are necessary to ensure reproducibility and correct interpretation of the results.

Response 11: We apologize for not including the specific details of the sample information in the main text. All accession numbers and sample IDs for the WGS data used in this study, including mice, humans, chimpanzee, and gorilla, are already listed in Supplementary Tables S4-S5. We have revised the table captions and referenced them at the appropriate points in the Methods to ensure clarity.

“The genome sequences of human (n = 8), chimpanzee (n = 1) and gorilla (n = 1) were sourced from National Center for Biotechnology Information (NCBI) (Supplementary Table 4). … Genomic sequences of mice (n = 13) were sourced from the Wellcome Sanger Institute’s Mouse Genome Project (MGP) (Keane, et al. 2011).

The concern regarding the number of individuals needed to support the results will be addressed in Response 13.

(ii) Sequencing reads from great apes and mice were mapped against the human and mouse rDNA reference sequences, respectively (lines 485-486). Given the rapid evolution of rRNA genes, even individuals within the same species differ in copy number and sequences of these genes. Alignment to a single reference genome would likely lead to incorrect and even failed alignment for some reads, resulting in genotyping errors. Differences in rDNA sequence, copy number, and structure are even greater between species, potentially leading to higher error rates in the called variants. Yet the authors provided no justification for the practice of aligning reads from multiple species to a single reference genome nor evidence that misalignment and incorrect variant calling are not major concerns for the downstream analysis.

Response 12: While the copy number of rDNA varies in each individuals, the sequence identity among copies is typically very high (median identity of 98.7% (Nurk, et al. 2022)). Therefore, all rRNA genes were aligned against to the species-specific reference sequences, where the consensus nucleotide nearly accounts for >90% of the gene copies in the population. In minimize genotyping errors, our analysis focused exclusively on single nucleotide variants (SNVs) with only two alleles, discarding other mutation types.

Regarding sequence divergence between species, which may have greater sequence variations, we excluded unmapped regions with high-quality reads coverage below 10. In calculation of substitution rate, we accounted for the mapping length (L), as shown in the column 3 in Table 3-5.

We appreciate the reviewer’s comments and have provide details in the Methods.

(vi) It is unclear how variant frequency within an individual was defined conceptually or computed from data (lines 499-501). The population-level variant frequency was calculated by averaging across individuals, but why was the averaging not weighted by the copy number of rRNA genes each individual carries? How many individuals are sampled for each species? Are the sample sizes sufficient to provide an accurate estimate of population frequencies?

Response 13: Each individual was considered as a psedo-population of rRNA genes, varaint frequency within an individual was the proportions of mutant allele in this psedo-population. The calculation of varaint frequency is based on the number of supported reads of each individual.

The reason for calculating population-level variant frequency by averaging across individuals is relevant in the calculation of FIS and FST. In calculating FST, the standard practice is to weigh each population equally. So, when we show FST in humans, we do not consider whether there are more Africans, Caucasians or Asians. There is a reason for not weighing them even though the population sizes could be orders of magnitude different, say, in the comparison between an ethnic minority and the main population. In the case of FIS, the issue is moot. Although copy number may range from 150 to 400 per haploid, most people have 300 – 500 copies with two haploids.

As for the concern regarding the number the individuals needed to support of the results:

Considering the nature of multi-copy genes, where gene members undergo continuous exchanges at a much slower rate compared to the rapid rate of random distribution of chromosomes at each generation of sexual reproduction, even a few variant copies that arise during an individual's lifetime would disperse into the gene pool in the next generation (Ohta and Dover 1984). Thus, there is minimal difference between individuals. Our analysis is also aligns with this theory, particularly in human population (FIS = 0.059), where each individual carries the majority of the population's genetic diversity. Therefore, even a single chimpanzee or gorilla individual caries sufficient diversity with its hundreds of gene copies to calculate divergence with humans.

(vii) Fixed variants are operationally defined as those with a frequency>0.8 in one species. What is the justification for this choice of threshold? Without knowing the exact sample size of the various species, it's difficult to assess whether this threshold is appropriate.

Response 14: First, the mutation frequency distribution is strongly bimodal (see Figure below) with a peak at zero and the other at 1. This high frequency peak starts to rise slowly at 0.8, similar to FST distribution in Figure 4C. That is why we use it as the cutoff although we would get similar results at the cutoff of 0.90 (see Table below). Second, the sample size for the calculation of mutant frequency is based on the number of reads which is usually in the tens of thousands. Third, it does not matter if the mutation frequency calculation is based on one individuals or multiple individuals because 95% of the genetic diversity of the population is captured by the gene pool within each individual.

Author response image 1.

Author response table 1.

The A/T to G/C and G/C to A/T changes in apes and mouse.

New mutants with a frequency >0.9 within an individual are considered as (nearly) fixed, except for humans, where the frequency was averaged over 8 individuals in the Table 2.

The X-squared values for each species are as follows: 58.303 for human, 7.9292 for chimpanzee, and 0.85385 for M. m. domesticus.

(viii) It is not explained exactly how FIS, FST, and divergence levels of rRNA genes were calculated from variant frequency at individual and species levels. Formulae need to be provided to explain the computation.

Response 15: After we clearly defined the HI, HS, and HT in Response9, understanding FIS and F_ST_ becomes straightforward.

“Given the three levels of heterozygosity, there are two levels of differentiation. First, FIS is the differentiation among individuals within the species, defined by

FIS = [HS - HI]/HS  

FIS is hence the proportion of genetic diversity in the species that is found only between individuals. We will later show FIS ~ 0.05 in human rDNA (Table 2), meaning 95% of rDNA diversity is found within individuals.

Second, FST is the differentiation between species within the total species complex, defined as

FST = [HT – HS]/HT 

FST is the proportion of genetic diversity in the total data that is found only between species.”

(3) Complete ignorance of the difference in mutation rate difference between rRNA genes and genome-wide average

Nearly all data analysis in this paper relied on comparison between rRNA genes with the rest (presumably single-copy part) of the genome. However, mutation rate, a key parameter determining the diversity and divergence levels, was completely ignored in the comparison. It is well known that mutation rate differs tremendously along the genome, with both fine and large-scale variation. If the mutation rate of rRNA genes differs substantially from the genome average, it would invalidate almost all of the analysis results. Yet no discussion or justification was provided.

Response 16: We appreciate the reviewer's observation regarding the potential impact of varying mutation rates across the genome. To address this concern, we compared the long-term substitution rates on rDNA and single-copy genes between human and rhesus macaque, which diverged approximately 25 million years ago. Our analysis (see Table S1 below) indicates that the substitution rate in rDNA is actually slower than the genome-wide average. This finding suggests that rRNA genes do not experience a higher mutation rate compared to single-copy genes, as stated in the text:

“Note that Eq. (3) shows that the mutation rate, m, determines the long-term evolutionary rate, l. Since we will compare the l values between rRNA and single-copy genes, we have to compare their mutation rates first by analyzing their long-term evolution. As shown in Table S1, l falls in the range of 50-60 (differences per Kb) for single copy genes and 40 – 70 for the non-functional parts of rRNA genes. The data thus suggest that rRNA and single-copy genes are comparable in mutation rate. Differences between their l values will have to be explained by other means.”

However, given the divergence time (Td) being equal to or smaller than Tf, even if the mutation rate per nucleotide is substantially higher in rRNA genes, these variants would not become fixed after the divergence of humans and chimpanzees without the help of strong homogenization forces. Thus, the presence of divergence sites (Table 5) still supports the conclusion that rRNA genes undergo much stronger genetic drift compared to single-copy genes.

Related to mutation rate: given the hypermutability of CpG sites, it is surprising that the evolution/fixation rate of rRNA estimated with or without CpG sites is so close (2.24% vs 2.27%). Given the 10 - 20-fold higher mutation rate at CpG sites in the human genome, and 2% CpG density (which is probably an under-estimate for rDNA), we expect the former to be at least 20% higher than the latter.

Response 17: While it is true that CpG sites exhibit a 10-20-fold higher mutation rate, the close evolution/fixation rates of rDNA with and without CpG sites (2.24% vs 2.27%) may be attributed to the fact that fixation rates during short-term evolutionary processes are less influenced by mutation rates alone. As observed in the Human-Macaque comparison in the table above, the substitution rate of rDNA in non-functional regions with CpG sites is 4.18%, while it is 3.35% without CpG sites, aligning with your expectation of 25% higher rates where CpG sites are involved.

This discrepancy between the expected and observed fixation rates may be due to strong homogenization forces, which can rapidly fix or eliminate variants, thereby reducing the overall impact of higher mutation rates at CpG sites on the observed fixation rate. This suggests that the homogenization mechanisms play a more dominant role in the fixation process over short evolutionary timescales, mitigating the expected increase in fixation rates due to CpG hypermutability.

Among the weaknesses above, concern (1) can be addressed with clarification, but concerns (2) and (3) invalidate almost all findings from the data analysis and cannot be easily alleviated with a complete revamp work.

Recommendations for the authors:

Reviewing Editor Comments:

Both reviewers found the manuscript confusing and raised serious concerns. They pointed out a lack of engagement with previous literature on modeling and the presence of ill-defined terms within the model, which obscure understanding. They also noted a significant disconnection between the modeling approach and the biological processes involved. Additionally, the data analysis was deemed problematic due to the failure to consider essential biological and technical factors. One reviewer suggested that the modeling component would be more suitable as a section of the companion theory paper rather than a standalone paper. Please see their individual reviews for their overall assessment.

Reviewer #2 (Recommendations For The Authors):

Beyond my major concerns, I have numerous questions about the interpretation of various findings:

Lines 62-63: Please explain under what circumstance Ne=N/V(K) is biologically nonsensical and why.

Response 18: “Biologically non-sensical” is the term used in (Chen, et al. 2017). We now used the term “biologically untenable” but the message is the same. How does one get V(K) ≠ E(K) in the WF sampling? It is untenable under the WF structure. Kimura may be the first one to introduce V(K) ≠ E(K) into the WF model and subsequent papers use the same sort of modifications that are mathematically valid but biologically dubious. As explained extensively in the companion paper, the modifications add complexities but do not give the WF models powers to explain the paradoxes.

Lines 231-234: The claim about a lower molecular evolution rate (lambda) is inaccurate - under neutrality, the molecular evolution rate is always the same as the mutation rate. It is true that when the species divergence Td is not much greater than fixation time Tf, the observed number of fixed differences would be substantially smaller than 2*mu*Td, but the lower divergence level does not mean that the molecular evolution is slower. In other words, in calculating the divergence level, it is the time term that needs to be adjusted rather than the molecular evolution rate.

Response 19: Thanks, we agree that the original wording was not accurate. It is indeed the substitution rate rather than the molecular evolution rate that is affected when species divergence time Td is not much greater than the fixation time Tf. We have revised the relevant text in the manuscript to correct this and ensure clarity.

Lines 277-279: Hs for rRNA is 5.2x fold than the genome average. This could be roughly translated as Ne*/Ne=5.2. According to Eq 2: (1/Ne*)/(1/Ne)= Vh/C*, it can be drived that mean Ne*/Ne=C*/Vh. Then why do the authors conclude "C*=N*/N~5.2" in line 278? Wouldn't it mean that C*/Vh is roughly 5.2?

Response 20: We apologize for the confusion. To prevent misunderstandings, we have revised Equation 1 and deleted Equation 2 from the manuscript. Please refer to the Response6 for further details.

Lines 291-292: What does "a major role of stage I evolution" mean? How does it lead to lower FIS?

Response 21: We apologize for the lack of clarity in our original description, and we have revised the relevant content to make them more directly.

“In this study, we focus on multi-copy gene systems, where the evolution takes place in two stages: both within (stage I) and between individuals (stage II).”

“FIS for rDNA among 8 human individuals is 0.059 (Table 2), much smaller than 0.142 in M. m. domesticus mice, indicating minimal genetic differences across human individuals and high level of genetic identity in rDNAs between homologous chromosomes among human population. … Correlation of polymorphic sites in IGS region is shown in Supplementary Fig. 1. The results suggest that the genetic drift due to the sampling of chromosomes during sexual reproduction (e.g., segregation and assortment) is augmented substantially by the effects of homogenization process within individual. Like those in mice, the pattern indicates that intra-species polymorphism is mainly preserved within individuals.”

Line 297-300: why does the concentration at very allele frequency indicate rapid homogenization across copies? Suppose there is no inter-copy homogenization, and each copy evolves independently, wouldn't we still expect the SFS to be strongly skewed towards rare variants? It is completely unclear how homogenization processes are expected to affect the SFS.

Response 22: We appreciate the reviewer’s insightful comments and apologize for any confusion in our original explanation. To clarify:

If there is no inter-copy homogenization and each copy evolves independently, it would effectively result in an equivalent population size that is C times larger than that of single-copy genes. However, given the copies are distributed on five chromosomes, if the copies within a chromosome were fully linked, there would be no fixation at any sites. Considering the data presented in Table 4, where the substitution rate in rDNA is higher than in single-copy genes, this suggests that additional forces must be acting to homogenize the copies, even across non-homologous chromosomes.

Regarding the specific data presented in the Figure 3, the allele frequency spectrum is based on human polymorphism sites and is a folded spectrum, as the ancestral state of the alleles was not determined. High levels of homogenization would typically push variant mutations toward the extremes of the SFS, leading to fewer intermediate-frequency alleles and reduced heterozygosity. The statement that "allele frequency spectrum is highly concentrated at very low frequency within individuals" was intended to emphasize the localized distribution of variants and the high identity at each site. However, we recognize that it does not accurately reflect the role of homogenization and this conclusion cannot be directly inferred from the figure as presented. Therefore, we have removed the sentence in the text.

The evidence of gBGC in rRNA genes in great apes does not help explain the observed accelerated evolution of rDNA relative to the rest of the genome. Evidence of gBGC has been clearly demonstrated in a variety of species, including mice. It affects not only rRNA genes but also most parts of the genome, particularly regions with high recombination rates. In addition, gBGC increases the fixation probability of W>S mutations but suppresses the fixation of S>W mutations, so it is not obvious how gBGC will increase or decrease the molecular evolution rate overall.

Response 23: We have thoroughly rewritten the last section of Results. The earlier writing has misplaced the emphasis, raising many questions (as stated above). To answer them, we would have to present a new set of equations thus adding unnecessary complexities to the paper. Here is the streamlined and more logical flow of the new section.

First, Tables 4 and 5 have shown the accelerated evolution of the rRNA genes. We have now shown that rRNA genes do not have higher mutation rates. Below is copied from the revised text:

“We now consider the evolution of rRNA genes between species by analyzing the rate of fixation (or near fixation) of mutations. Polymorphic variants are filtered out in the calculation. Note that Eq. (3) shows that the mutation rate, m, determines the long-term evolutionary rate, l. Since we will compare the l values between rRNA and single-copy genes, we have to compare their mutation rates first by analyzing their long-term evolution. As shown in Table S1 l falls in the range of 50-60 (differences per Kb) for single copy genes and 40 – 70 for the non-functional parts of rRNA genes. The data thus suggest that rRNA and single-copy genes are comparable in mutation rate. Differences between their l values will have to be explained by other means.”

Second, we have shown that the accelerated evolution in mice is likely due to genetic drift, resulting in faster fixation of neutral variants. We also show that this is unlikely to be true in humans and chimpanzees; hence selection is the only possible explanation. The section below is copied from the revised text. It shows the different patterns of gene conversions between mice and apes, in agreement with the results of Tables 4 and 5. In essence, it shows that the GC ratio in apes is shifting to a new equilibrium, which is equivalent to a new adaptive peak. Selection is driving the rDNA genes to move to the new adaptive peak.

Revision - “Thus, the much accelerated evolution of rRNA genes between humans and chimpanzees cannot be entirely attributed to genetic drift. In the next and last section, we will test if selection is operating on rRNA genes by examining the pattern of gene conversion. 

3) Positive selection for rRNA mutations in apes, but not in mice – Evidence from gene conversion patterns

For gene conversion, we examine the patterns of AT-to-GC vs. GC-to-AT changes. While it has been reported that gene conversion would favor AT-to-GC over GC-to-AT conversion (Jeffreys and Neumann 2002; Meunier and Duret 2004) at the site level, we are interested at the gene level by summing up all conversions across sites. We designate the proportion of AT-to-GC conversion as f and the reciprocal, GC-to-AT, as g. Both f and g represent the proportion of fixed mutations between species (see Methods). So defined, f and g are influenced by the molecular mechanisms as well as natural selection. The latter may favor a higher or lower GC ratio at the genic level between species. As the selective pressure is distributed over the length of the gene, each site may experience rather weak pressure.

Let p be the proportion of AT sites and q be the proportion of GC sites in the gene. The flux of AT-to-GC would be pf and the flux in reverse, GC-to-AT, would be qg. At equilibrium, pf = qg. Given f and g, the ratio of p and q would eventually reach p/q \= g/f. We now determine if the fluxes are in equilibrium (pf =qg). If they are not, the genic GC ratio is likely under selection and is moving to a different equilibrium.

In these genic analyses, we first analyze the human lineage (Brown and Jiricny 1989; Galtier and Duret 2007). Using chimpanzees and gorillas as the outgroups, we identified the derived variants that became nearly fixed in humans with frequency > 0.8 (Table 6). The chi-square test shows that the GC variants had a significantly higher fixation probability compared to AT. In addition, this pattern is also found in chimpanzees (p < 0.001). In M. m. domesticus (Table 6), the chi-square test reveals no difference in the fixation probability between GC and AT (p = 0.957). Further details can be found in Supplementary Figure 2. Overall, a higher fixation probability of the GC variants is found in human and chimpanzee, whereas this bias is not observed in mice.

Tables 6-7 here

Based on Table 6, we could calculate the value of p, q, f and g (see Table 7). Shown in the last row of Table 7, the (pf)/(qg) ratio is much larger than 1 in both the human and chimpanzee lineages. Notably, the ratio in mouse is not significantly different from 1. Combining Tables 4 and 7, we conclude that the slight acceleration of fixation in mice can be accounted for by genetic drift, due to gene conversion among rRNA gene copies. In contrast, the different fluxes corroborate the interpretations of Table 5 that selection is operating in both humans and chimpanzees.”

References

Arnheim N, Treco D, Taylor B, Eicher EM. 1982. Distribution of ribosomal gene length variants among mouse chromosomes. Proc Natl Acad Sci U S A 79:4677-4680.

Brown T, Jiricny J. 1989. Repair of base-base mismatches in simian and human cells. Genome / National Research Council Canada = Génome / Conseil national de recherches Canada 31:578-583.

Cannings C. 1974. The latent roots of certain Markov chains arising in genetics: A new approach, I. Haploid models. Advances in Applied Probability 6:260-290.

Chen Y, Tong D, Wu CI. 2017. A New Formulation of Random Genetic Drift and Its Application to the Evolution of Cell Populations. Mol Biol Evol 34:2057-2064.

Chia AB, Watterson GA. 1969. Demographic effects on the rate of genetic evolution I. constant size populations with two genotypes. Journal of Applied Probability 6:231-248.

Crow JF, Kimura M. 2009. An Introduction to Population Genetics Theory: Blackburn Press.

Danecek P, Bonfield JK, Liddle J, Marshall J, Ohan V, Pollard MO, Whitwham A, Keane T, McCarthy SA, Davies RM, et al. 2021. Twelve years of SAMtools and BCFtools. Gigascience 10.

Datson NA, Morsink MC, Atanasova S, Armstrong VW, Zischler H, Schlumbohm C, Dutilh BE, Huynen MA, Waegele B, Ruepp A, et al. 2007. Development of the first marmoset-specific DNA microarray (EUMAMA): a new genetic tool for large-scale expression profiling in a non-human primate. Bmc Genomics 8:190.

Der R, Epstein CL, Plotkin JB. 2011. Generalized population models and the nature of genetic drift. Theoretical Population Biology 80:80-99.

Dover G. 1982. Molecular drive: a cohesive mode of species evolution. Nature 299:111-117.

Eickbush TH, Eickbush DG. 2007. Finely orchestrated movements: evolution of the ribosomal RNA genes. Genetics 175:477-485.

Galtier N, Duret L. 2007. Adaptation or biased gene conversion? Extending the null hypothesis of molecular evolution. Trends in Genetics 23:273-277.

Gibbs RA, Rogers J, Katze MG, Bumgarner R, Weinstock GM, Mardis ER, Remington KA, Strausberg RL, Venter JC, Wilson RK, et al. 2007. Evolutionary and Biomedical Insights from the Rhesus Macaque Genome. Science 316:222-234.

Guarracino A, Buonaiuto S, de Lima LG, Potapova T, Rhie A, Koren S, Rubinstein B, Fischer C, Abel HJ, Antonacci-Fulton LL, et al. 2023. Recombination between heterologous human acrocentric chromosomes. Nature 617:335-343.

Hartl DL, Clark AG, Clark AG. 1997. Principles of population genetics: Sinauer associates Sunderland.

Hori Y, Shimamoto A, Kobayashi T. 2021. The human ribosomal DNA array is composed of highly homogenized tandem clusters. Genome Res 31:1971-1982.

Jeffreys AJ, Neumann R. 2002. Reciprocal crossover asymmetry and meiotic drive in a human recombination hot spot. Nat Genet 31:267-271.

Karlin S, McGregor J. 1964. Direct Product Branching Processes and Related Markov Chains. Proceedings of the National Academy of Sciences 51:598-602.

Keane TM, Goodstadt L, Danecek P, White MA, Wong K, Yalcin B, Heger A, Agam A, Slater G, Goodson M, et al. 2011. Mouse genomic variation and its effect on phenotypes and gene regulation. Nature 477:289-294.

Krystal M, D'Eustachio P, Ruddle FH, Arnheim N. 1981. Human nucleolus organizers on nonhomologous chromosomes can share the same ribosomal gene variants. Proceedings of the National Academy of Sciences of the United States of America 78:5744-5748.

Meunier J, Duret L. 2004. Recombination drives the evolution of GC-content in the human genome. Molecular Biology and Evolution 21:984-990.

Nagylaki T. 1983. Evolution of a large population under gene conversion. Proc Natl Acad Sci U S A 80:5941-5945.

Nurk S, Koren S, Rhie A, Rautiainen M, Bzikadze AV, Mikheenko A, Vollger MR, Altemose N, Uralsky L, Gershman A, et al. 2022. The complete sequence of a human genome. Science 376:44-53.

Ohta T. 1985. A model of duplicative transposition and gene conversion for repetitive DNA families. Genetics 110:513-524.

Ohta T. 1976. Simple model for treating evolution of multigene families. Nature 263:74-76.

Ohta T, Dover GA. 1984. The Cohesive Population Genetics of Molecular Drive. Genetics 108:501-521.

Ohta T, Dover GA. 1983. Population genetics of multigene families that are dispersed into two or more chromosomes. Proc Natl Acad Sci U S A 80:4079-4083.

Ruan Y, Wang X, Hou M, Diao W, Xu S, Wen H, Wu C-I. 2024. Resolving Paradoxes in Molecular Evolution: The Integrated WF-Haldane (WFH) Model of Genetic Drift. bioRxiv:2024.2002.2019.581083.

Smirnov E, Chmúrčiaková N, Liška F, Bažantová P, Cmarko D. 2021. Variability of Human rDNA. Cells 10.

Smith GP. 1976. Evolution of Repeated DNA Sequences by Unequal Crossover. Science 191:528-535.

Smith GP. 1974a. Unequal crossover and the evolution of multigene families. Cold Spring Harbor symposia on quantitative biology 38:507-513.

Smith GP. 1974b. Unequal Crossover and the Evolution of Multigene Families.  38:507-513.

Stults DM, Killen MW, Pierce HH, Pierce AJ. 2008. Genomic architecture and inheritance of human ribosomal RNA gene clusters. Genome Res 18:13-18.

van Sluis M, Gailín M, McCarter JGW, Mangan H, Grob A, McStay B. 2019. Human NORs, comprising rDNA arrays and functionally conserved distal elements, are located within dynamic chromosomal regions. Genes Dev 33:1688-1701.

Wall JD, Frisse LA, Hudson RR, Di Rienzo A. 2003. Comparative linkage-disequilibrium analysis of the beta-globin hotspot in primates. Am J Hum Genet 73:1330-1340.

eLife Assessment

This study presents a useful theoretical model of molecular evolution and attempts to use it to resolve the paradox of rapid evolution of ribosomal RNA genes. While intuitive, the model's underlying issue is grouping many factors under "variance in reproductive success" without explicitly modeling the molecular processes. This limitation, along with insufficient consideration of technical challenges in alignment and variants calling, provides incomplete support for the authors' claim that the observed paradoxical patterns in rRNA genes can largely be explained by homogenizing processes, such as gene conversion, unequal crossover and replication slippage.

Reviewer #1 (Public review):

The revision by Wang et al is a much more clear and readable manuscript than the original version, which I think was a bit too terse and hard to parse. In this version, I think I basically understand all the analyses that the authors undertake and how they argue that those analyses support their conclusions.

The fundamental claim of the manuscript is that rRNA genes experience substitutions much too quickly, given that they are a multi-copy gene system. As clarified by the authors in their response, and as I think is relatively clear in the manuscript, they are collapsing all copies of the rRNA array down. They first quantify polymorphism (in this expanded definition, where polymorphism means variable at a given site across any copy). The authors find elevated levels of heterozygosity in rRNA genes compared to single copy genes, which isn't surprising, given that there is a substantially higher target size; that being said, the increase in polymorphism is smaller than the increase in target size. They then look at substitutions between mouse species and also between human and chimp, and argue that the substitution rate is too fast compared to single copy genes in many cases.

I think that this is an interesting problem and one that obviously occupies some space in the literature. As the authors point out, one possibility for explaining the elevated fixation rate is that there is some kind of positive selection in these putatively non-functional regions. The authors, instead, argue that the elevated rate of evolution is due to neutral homogenizing processes. I'm sympathetic to this argument, I'm a neutralist myself :)

That being said, I find the whole analysis and the connection with the WFH model very strange. As I stated in my previous review, it feels very odd to chalk everything up to variance in reproductive success, rather than explicitly modeling the molecular processes that may lead to the homogenization. For example, the authors bring up gene conversion, and even do a small test of gene conversion. But a force like biased gene conversion is perhaps better modeled as a deterministic force, rather than a stochastic force. Indeed, I think that explicit modeling of mutation dynamics has been very helpful in understanding the role of replicative vs damage-related mutation in humans, as seen in Gao et al (2016) and Spisak et al (2024). I realize, as the authors say in their cover letter, that this is hard! But a major concern with this manuscript is that it's about whether drift can plausibly explain the pattern, but then it's basically impossible to know if it really can, because we have no way to compare the estimated parameters with biophysical or biochemical measurements of the rates of homogenizing forces, because the homogenizing forces are just wrapped up under "variance in reproductive success". I think a much more interesting manuscript would have a more explicit model of homogenizing forces.

I also have some concerns about the data analysis, echoing some concerns of the other reviewer. The biggest issue is that traditional read mapping and SNP calling pipelines for highly duplicated loci don't really make sense. I don't fully understand the variant calling pipeline. The authors state that "All mapping and analysis are performed among individual copies of rRNA genes." which makes it sound like the reads mapping to different copies were somehow deconvolved, which is what you'd need to do to use "normal" variant calling approaches that call look for homozygotes and heterozygotes. But I don't know enough about this literature to understand how they did that and if it makes any sense. If, instead, they called variants against collapsed rRNA copies, then using a standard variant calling approach does not make sense. If you have a variant in 2 out of 100 copies, a standard variant calling algorithm would very likely call that a homozygous ancestral site. Conditional on the variant calls being reasonable, however, I'm basically okay with their use of read counts to estimate "allele frequencies" within individuals.

I have some more minor comments:

(1) In the paragraph starting line 61, the authors say that WF models are unable to handle things like viral epidemics and transposons. I don't think that's really fair: the issue here isn't WF dynamics or not, it's that there is fundamentally evolution on two levels (which is also the case in the rRNA case considered in this manuscript). I certainly agree with the authors that you can't just naively apply standard pop gen theory in these systems, but I think the arrow at the WF model is misaimed, as the real issue is drift and selection on multiple levels.

(2) Line 268-269: The authors argue that the long term rate of evolution in rRNA genes is roughly similar to single copy genes, suggesting not a big influence of increased mutation rate. I'm not sure I understand where this number comes from, as opposed to the divergence numbers they look at in Table 3. These seem to be two different conclusions from roughly the same measurement? Surely I am misunderstanding something.

References:

Gao, Z., Wyman, M. J., Sella, G., & Przeworski, M. (2016). Interpreting the dependence of mutation rates on age and time. PLoS biology, 14(1), e1002355.

Spisak, N., de Manuel, M., Milligan, W., Sella, G., & Przeworski, M. (2024). The clock-like accumulation of germline and somatic mutations can arise from the interplay of DNA damage and repair. PLoS biology, 22(6), e3002678.

Reviewer #2 (Public review):

I appreciate the authors' efforts in addressing previous feedback by correcting typos, clarifying terms, and expanding the methodological descriptions. The revisions have notably improved the manuscript's clarity and readability. However, despite these positive changes, I still have several significant concerns, both conceptual and technical, that need to be addressed to strengthen the conclusions of the paper.

The key idea of this paper is the treatment of rDNA copies in an individual as a pseudo-population and model their sequence evolution with the WFH framework by introducing the parameter V*(K). With this modeling framework, the authors claim that the molecular evolution rate of rDNA relative to that of single-copy genes can be expressed as a simple function V*(K) and C (the copy number per individual). Moreover, when V*(K) is sufficiently large, the neutral molecular evolution of rDNA can be faster than expected under a naïve model without considering horizontal, homogenizing processes and thus be potentially compatible with empirical data. However, several issues persist in the definition, assumptions, and derivation of the model:

(1) Several terms in the model remain undefined. While Ne is clearly defined in the standard single-copy gene model as the reciprocal of genetic drift (i.e., the decay in heterozygosity), its meaning for multiple-copy genes is unclear. Based on the context, it appears that the authors define Ne as the parameter that fits the population polymorphism level (Hs) using the equation in line 165. This definition is reasonable, but it should be explicitly clarified in the text."<br /> (2) Another key parameter V*(K) was still not defined within the paper. In response 9, the authors explained that V*(K) refers to "the number of progeny to whom the gene copy of interest is transmitted (K) over a specific time interval". However, the meaning of "progeny" remains unclear. Are the authors referring to the descendent copies of a gene copy, or the offspring individuals (i.e., the living organisms)? For example, if a variant spreads horizontally through homogenizing processes and transmits vertically to multiple offspring individuals, the number of descent gene copies could differ substantially from the number of descendent individuals to whom a gene copy is transmitted to. This distinction needs to be clarified and clearly stated in the paper.<br /> (3) The authors state that V*(K)>=1 for rDNA genes because of the homogenizing processes (lines 139-141) without providing justification. It is unclear, at least to me, whether homogenizing processes are expected increase or decrease the variance in "reproductive success" across gene copies. Moreover, the authors claim that V*(K) "can potentially reach values in the hundreds and may even exceed C, resulting in C*=C/V*(K)<1" (Response 7). This claim is unlikely to be true, as the minimum value of K is bounded by zero and E(K) is assumed to be 1. Even in the extreme case that 1% gene copies leave large numbers of descends while the others leave none, V*(K) would still be less than 100. Such extreme case seems highly improbable, given realistic rates of the homogenizing processes.<br /> (4) Regardless of how the authors define V*(K), it is not immediately clear why Equation 1 (N*=NC/V*(K)) holds. Both sides of the equation have their independent meanings, so the authors need to provide a step-by-step derivation demonstrating that they are equal. Only by doing this will the implicit underlying assumptions become clearer. I also strongly recommend that the authors conduct forward-in-time simulations with fixed N, C, V*(K) (however they define it) and μ to confirm that the right side of Equation 1 actually predicts the N* as calculated from the polymorphism level using the equation in line 165.<br /> (5) Without providing justification, the authors assumed that a certain number N* exists for rRNA such that it fits both the polymorphism level (line 156) in recent timescales and divergence level in longer timescales (i.e., in the comparison between Tf and Td). However, if N, C or any other relevant parameters have varied substantially throughout evolution, N* is expected to vary with time, and the same value may not fit both polymorphism and divergence data simultaneously.

The authors also provided more detailed description of their data analysis methods, but some of my major concerns remain:<br /> (1) A significant issue with aligning reads to a single reference genome is reference bias, referring to the phenomenon that reads carrying the reference alleles tend to align more easily than those with one or more non-reference alleles, thus creating a bias in genotype calling or variant allele frequency quantification. As a result, there may be an underrepresentation of non-reference alleles in called variants or an underestimate of non-reference allele frequency, particularly in regions with high genetic diversity. Simply focusing on bi-allelic SNVs is insufficient to minimize reference bias. Given the fourfold increase in diversity within rDNA, the authors must either provide evidence that reference bias is not a significant concern or adopt graph-based reference genomes or more sophisticated alignment algorithms to address this issue.<br /> (2) The potential for reference bias also renders the analysis of divergence sites unreliable, as aligning reads from one species (e.g. chimpanzee) to the reference of another species (e.g., human) is likely to introduce biases in variant calling between the two. One commonly adopted approach to address this imbalance is to align reads from both species to a third reference genome that is expected to be equidistantly related to both.<br /> (3) Although it is somewhat reassuring that the estimated divergence rate of rDNA between human and macaque is comparable to that of the rest of the genome, there still remains concern of a under-estimation of divergence in rDNA regions due to reference bias issue. Note that while the "third genome" approach reduces imbalance between two genomes in comparison, it may still under-estimate overall divergence level due to under-calling of non-reference variants.<br /> (4) In response to my question about the similarity in rDNA substitution rates estimated with or without CpG sites, the authors suggest that this "may be due to strong homogenizing forces, which can rapidly fix or eliminate variants" (response17). However, this explanation is insufficient, because the observed substitution rate depends on the mutation rate multiplied by the fixation probability, and accelerated fixation or loss does not alter either. Unless the authors can provide more convincing explanation, technical errors in calling of fixed sites still remain a concern.

Minor points<br /> Line 157: The statement "where μ is the mutation rate of the entire gene" must be wrong, as the heterozygosity calculated with such μ would correspond to the chance of seeing two different haplotypes at gene level, which is incompatible with the empirical calculation specified in Equation 2. Instead, μ must represent the mutation rate per site averaged over the entire gene.

In response 22, the authors explained that the allele frequency spectrum shown in Fig 3 is folded, because the ancestral allele was not determined. However, this is inconsistent with x-axis Fig 3 ranging between 0 and 1. I suspect the x-axis represents the frequency of the alternative (i.e., non-reference) allele. If so, the reported correlation is inflated, as the reference allele is somewhat random, and a variant at joint ALT allele frequencies of (0.9, 0.9) is no different from a variant at (0.1, 0.1). The proper way of calculate this correlation is to first determine the minor allele frequency across individuals and then calculate the correlation between minor allele frequencies.

Similarly, in response 14, it is unclear what the x-axis represents. Is it the ALT allele frequency or derived allele frequency? If the former, why are only variants with AF>0.8 defined as fixed variants, while those with AF<0.2 excluded? If it is the latter, please describe how ancestral state is determined.

eLife Assessment

This is a valuable study on the diffusion rates of drug molecules in human-derived cells, presenting convincing data indicating that their diffusion behavior depends on their charged state. It proposes that blocking drug protonation enhances diffusion and fractional recovery, suggesting improved intracellular availability of weakly basic drugs. The findings are significant for drug design and understanding the biophysical behavior of small molecules in cells.

Reviewer #1 (Public review):

Summary:

The authors set out to measure the diffusion of small drug molecules inside live cells. To do this, they selected a range of fluorescent drugs, as well as some commonly used dyes, and used FRAP to quantify their diffusion. The authors find that drugs diffuse and localize within the cell in a way that is weakly correlated with their charge, with positively charged molecules displaying dramatically slower diffusion and a high degree of subcellular localization.

The study is important because it points to an important issue related to the way drugs behave inside cells beyond the simple "IC50" metric (a decidedly mesoscopic/systemic value). The authors conclude, and I agree, that their results point to nuanced effects that are governed by drug chemistry that could be optimized to make them more effective.

Strengths:

(1) The work examines an understudied aspect of drug delivery.

(2) The work uses well-established methodologies to measure diffusion in cells

(3) The work provides an extensive dataset, covering a range of chemistries that are common in small molecule drug design

(4) The authors consider several explanations as to the origin of changes in cellular diffusion

Comments on revised version:

In general, my comments were addressed, new discussions were added, and the paper has been improved significantly, which is great.

However, despite providing very clear instructions, a lot of my comments re statistical treatment were disregarded. Bar charts still do not show the repeats as individual points. Errors bars still represent SEM, which gives a wrong idea about the spread of the data. FRAP lines are still averages, and still do not show the spread of the data.

Significance assignments are done based on average and SEMs, as opposed to the full dataset. There is nothing technically wrong with this, but it generally creates an impression that things are more reproducible/rigorous/significant than they would be if the data was shown completely.

Reviewer #2 (Public Review):

Summary:

Blocking a weak base compound's protonation increased intracellular diffusion and fractional recovery in the cytoplasm, which may improve the intracellular availability and distribution of weakly basic, small molecule drugs and be impactful in future drug development.

Strengths:

(1) The intracellular distribution of drugs and the chemical properties that drive their distribution are much needed in the literature. Thus, the idea behind this paper is of relevance.

(2) The study used common compounds that were relevant to others.

(3) Altering a compound's pKa value and measuring cytosolic diffusion rates certainly is inciteful on how weak base drugs and their relatively high pKa values affect distribution and pharmacokinetics. This particular experiment demonstrated relevance to drug targeting and drug development.

(4) The manuscript was fairly well written.

Comments on revised version:

After reviewing the authors' responses to my questions and concerns, they have adequately corrected the errors, added new information and data based off the reviewers suggestions that improved the manuscript. The manuscript in its current form would add quality information to a part of the literature that is lacking much needed information.

Author response:

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

Public Reviews: 

Reviewer #1 (Public Review):

Summary:

The authors set out to measure the diffusion of small drug molecules inside live cells. To do this, they selected a range of flourescent drugs, as well as some commonly used dyes, and used FRAP to quantify their diffusion. The authors find that drugs diffuse and localize within the cell in a way that is weakly correalted with their charge, with positively charged molecules displaying dramatically slower diffusion and a high degree of subcellular localization. <br /> The study is important because it points at an important issue related to the way drugs behave inside cells beyond the simple "IC50" metric (a decidedly mesoscopic/systemic value). The authors conclude, and I agree, that their results point to nuanced effects that are governed by drug chemistry that could be optimized to make them more effective. 

We are grateful to the reviewer for summarizing the work and appreciate him/her pointing out that it is high time to consider the drug aggregation and high degree of subcellular localization while optimizing to make them more effective beyond the mesoscopic value like "IC50".

Strengths: 

The work examines an understudied aspect of drug delivery. 

The work uses well-established methodologies to measure diffusion in cells 

The work provides an extensive dataset, covering a range of chemistries that are common in small molecule drug design 

The authors consider several explanations as to the origin of changes in cellular diffusion

We are grateful to the reviewer for pointing out the strengths of the manuscript.

Weaknesses: 

The results are described qualitatively, despite quantitative data that can be used to infer the strength of the proposed correlations. 

The statistical treatment of the data is not rigorous and not visualized according to best practices, making it difficult for readers to assess the significance of the findings. 

Some important aspects of drug behavior are not discussed quantitatively, such as the cell-to-cell or subcellular variability in concentration. 

It is unclear if the observed behavior of each drug in the cell actually relates to its efficacy - though this is clearly beyond the scope of this specific work.

We have addressed the weaknesses found by the reviewer (see bellow in Reviewer #1 Recommendations For The Authors). Concerning the last point, It would have been indeed very valuable to find a relation between drug's observable behavior and their efficacy, but as the reviewer indicates, it is beyond the scope of this work.

Reviewer #2 (Public Review): 

Summary:

Blocking a weak base compound's protonation increased intracellular diffusion and fractional recovery in the cytoplasm, which may improve the intracellular availability and distribution of weakly basic, small molecule drugs and be impactful in future drug development. 

We are thankful to the reviewer for summarizing our work and acknowledging that the points raised above can be impactful in future drug development.

Strengths: 

(1) The intracellular distribution of drugs and the chemical properties that drive their distribution are much needed in the literature. Thus, the idea behind this paper is of relevance. 

(2) The study used common compounds that were relevant to others. 

(3) Altering a compound's pKa value and measuring cytosolic diffusion rates certainly is inciteful on how weak base drugs and their relatively high pKa values affect distribution and pharmacokinetics. This particular experiment demonstrated relevance to drug targeting and drug development. 

(4) The manuscript was fairly well written. 

We are thankful to the reviewer for pointing out the strengths of the manuscript like the intracellular distribution of drugs and properties that drive it, which are missing in the literature.

Weaknesses: 

(1) Small sample sizes. 2 acids and 1 neutral compound vs 6 weak bases (Figure 1). 

We fully agree with the reviewer on this point. However, the major limitation we have faced here is the small number of drug/drug-like molecules that fluorescent with sufficient high quantum yields. For this study, we initially screened 1600 drugs for their fluorescence in the visible spectrum, and penetration into cells, resulting in 16 drugs. Of those, a small number was suitable for FRAP due to low quantum yield. For some of the molecules (Mitoxantrone, Priaquine), recovery was minimal, making them challenging to study. We added this information in the materials and method section under “Selection of drugs used in this study” (p.10).

(2) A comparison between the percentage of neutral and weak base drug accumulation in lysosomes would have helped indicate weak base ion trapping. Such a comparison would have strengthened this study. 

For weakly basic compounds, the ionic form and the non-ionic form of the molecules always remain in equilibrium. The direction of the equilibrium depends on the pH of the medium, which determines the major form of the drug molecules in the solution. Our examples of GSK3 inhibitor (neutral compound, pka~7.0, as predicted by Chemaxon), shows behaviour very similar to the other basic drugs (pka>8) inside the cells. As lysosome pH is about 5.0, the neutral drug also gets protonated inside the lysosomes, as the colocalization study reveals (Figure 4). We added Fig S16 C-D, where we show co-localization of three drugs within the lysosomes showing that all the three weak base drugs colocalize to acidic lysosomes from moderately to extensively. See also in p. 11 under “Confocal microscopy and FRAP Analysis section”.

(3) When cytosolic diffusion rates of compounds were measured, were the lysosomes extracted from the image using Imaris to determine a realistic cytosolic value? In real-time, lysosomes move through the cytosol at different rates. Because weak base drugs get trapped, it is likely the movement of a weak base in the lysosome being measured rather than the movement of a weak base itself throughout the cytosol. This was unclear in the methods. Please explain.

We want to thank the reviewer for pointing this out. To clarify the point, we added to the material and method section in p. 13 the following text: “When the areas of bleach were selected in the drug-treated cell cytoplasm, we avoided the lysosomes as much as possible, within the resolution limits of the confocal microscope. Lysosomes themselves were measured to move within the cytoplasm with an diffusion coefficient of 0.03-0.071 µm2 s−1  (Bandyopadhyay et al., 2014), which is much slower than the diffusion measured for even the slowest compounds using fast Line FRAP, further validating that we did not measure lysosome diffusion.” In addition, we show that in cells after Bafilomycin A1 or Na-Azide treatments the number of lysosomes was reduced drastically (Figures S8& S9, and Figure 7), while the rates of diffusion remain very slow, similar to those measured without lysosomal inhibitors.   

(4) Because weak base drugs can be protonated in the cytoplasm, the authors need to elaborate on why they thought that inhibiting lysosome accumulation of weak bases would increase cytosolic diffusion rates. Ion trapping is different than "micrometers per second" in the cytosol. Moreover, treating cells with sodium azide de-acidifies lysosomes and acidifies the cytosol; thus, more protons in the cytosol means more protonation of weak base drugs. The diffusion rates were slowed down in the presence of lysosome inhibition (Figure 7), which is more fitting of the story about blocking protonation increases diffusion rates, but in this case, increasing cytosolic protonation via lysosome de-acidification agents decreases diffusion rates. Please elaborate.

We thank the reviewer for the comment. We added to the results in p. 7 (top) the following “While we selected bleach spots to be small and located outside of lysosomes, this does not assure that some of the bleached area does not include smaller lysosomes. Therefore we investigated whether inhibiting lysosomal trapping will eliminate slow diffusion of cationic drugs.” In addition, we added to the results in p. 7-8 the following: “Comparative FRAP profiles and diffusion coefficients (Figure 7B-D and 7F-H) were slow, but conversely to Bafilomycin, sodium azide treatment did cause a further reduction is rates from Dconfocal 2.4±0.1 µm2s-1  to 1.8±0.1µm2s-1 for quinacrine and from 0.6 to  0.45 µm2s-1 for the GSK3 inhibitor (Figure 7C and G). Both Bafilomycin and sodium azide treatments resulted in elimination of drug confinement in the lysosome, and the small difference in diffusion rates may be a result of the de-acidification of the lysosomes by sodium azide, which may increase the protons in the cytosol upon treatment.”

Reviewer : A discussion of the likely impact: 

The manuscript certainly adds another dimension to the field of intracellular drug distribution, but the manuscript needs to be strengthened in its current form. Additional experiments need to be included, and there are clarifications in the manuscript that need to be addressed. Once these issues are resolved, then the manuscript, if the conclusions are further strengthened, is much needed and would be inciteful to drug development.

Reviewer #1 (Recommendations For The Authors):

Major issues: 

The paper suffers from poor statistical treatment of the data. FRAP recovery curves should be shown for each repeat, overlaid by an average with SDs as errorbars or shaded regions shown. In bar plots, SEMs should be eliminated in favor of StdDevs. All datapoints should be shown for each bar in Figs. 3-8. To show differences in D_confocal appropriate statistical tests should be conducted. In addition it is unclear what an "independent repeat" is. Does this mean 30 separate imaging sessions/drug treatments/etc? Is it 30 cells on the same coverslip? Is it a combination of both? All reported errors, SD or SEM, should have a single significant digit. Guidelines and best practices for representing quantitative imaging data are all described and visualized in detail in Lord et al. JBS 2020. 

We improved the statistics and added the individual progression curves and did the statistics on them as requested. See Figure S2 for individual FRAP curves of fluorescein, GSK3 inhibitor and and quinacrine. Statistical analysis of the individual FRAP curves is in Figure 3B, 4B, 5B, 7C and G. For details see figures legends and material and methods p. 13 in “Determination of Dconfocal from FRAP results”. Line FRAP was done from the cells taken from different plates, treated independently (see text p. 13).   

The extensive (and commendable!) dataset the authors have collected can be put to better use than what is currently done. The main text figures in the current form of the preprint are mostly descriptive and their discussion is qualitative, to the point where the author's conclusions are supported only anecdotally. Instead, I would much rather see panels that collate the entire dataset (both protein and drugs) numerically, comparing diffusion values in buffer/cytoplasm/nucleus for all drugs (Like Fig. S6, which is in my opinion the most important in the paper but for some reason relegated to the SI). In addition I would like to see correlations within the dataset, such as D_confocal vs. pKa, vs. concentration (as measured by overall fluorescence signal, see my comment below), vs. mw, or vs. specific chemical moieties (number of charges, aromatic rings, etc). Such correlations should be discussed in terms of a correlation coefficient if conclusions were to be drawn from them, and include errors if available. 

We want to thank the reviewer for these suggestions. We now made new Figures 9, and S16 to compare multiple parameters. Figure 9C shows a clear relation between pKa and Dconfocal, but no relation was found between logP, MW or number of aromatic rings and Dconfocal. Fig. S3 also shows the relation between drug concentration and Dconfocal values. These data are now discussed in the discussion section in p. 9 (bottom). 

The drug sequestration hypothesis and other conclusions brought forth by the authors could be further tested by looking at the concentration dependence of the drugs inside eachcell and/or its partitioning between different subcellular compartments. The concentration dependence of these drugs is discussed in a very anecdotal fashion using two concentrations - and despite some cases showing an effect no further studies were done. Drug concentrations in this experiment can vary between cells between repeats or even within a single repeat as a result of drug chemistry and delivery methods (microinjection/passive permeability). This is especially important since it is unclear what clinically-relevant concentrations are for each drug (or at least an IC50 for the cell types tested here). I would like to see a quantitative measure of concentrations as another metric to compare diffusion behavior (see my comment above as well). 

And maybe one thing to consider in addition would be some discussion in the paper about what sub-cellular distributions might actually mean in the context of drug efficacy (asking for myself as well!) - a paragraph describing recent works on the topic with some references could be instructive. 

We want to thank the reviewer for the suggestion. We added now Figure S3, showing the relation between fluorescence intensity in each cell (which is directly related to the concentration of the compound) and FRAP rates and percent recovery for fluorescein, GSK inhibitor and Quinacrine. The results show now relation between drug concentration and FRAP rates, and some relation towards percent recovery. These data are now discussed in the main text (p. 4 bottor and p.6) and in the discussion (p. 9, bottom).

Minor issues: 

Readers could benefit from a schematic showing the line FRAP method. It is difficult to understand from the text.

We show now in Figure 2 the line-FRAP method, and discuss it in the introduction (p. 3 top).

Have the authors considered enrichment in the cell membrane? Summed intensity projections or co-labeling with membrane dyes could prove useful to identify if the membrane is enriched in fluorescence.

The microscopy slides, including the super-resolution image in Figure S15 do not show enrichment of membranes.

Cell extracts obtained by chemical lysis are problematic because they contain surfactants. This comparison might not be meaningful. 

The reviewer is correct about surfactants; However, this is only for illustration to show the crowd density of the cell extracts compared to live cells.

Unclear why "Bleach size" plots are shown. They are not discussed in the main text. 

We show now a bleach size plot in Figure 2, where we explain the method. We removed them from the other figures.

Some figure panels have a strange aspect ratio, causing text to look distorted. 

We corrected the figure distortion in the revised manuscript.

How are the values of D_confocal in buffer compared with past literature? Should these not all be diffusion limited? BCECF - larger than many of the drugs used here - shows ~ 100 μm^2/s in buffer (Verkman TiBS 2002).

We discussed this in our previous work (Ref. 13, iscience 2022, Dey et al.) Dconfocal is a relative diffusion rate and should not be confused with single-molecule diffusion coefficients. FRAP cannot measure the diffusion of more than 100 μm^2/s in the buffer. However, when comparing apparent FRAP rates between different fluorophores, it is not quantitative due to the major implication of the bleach radius towards diffusion rates. The rate constant normalized by bleach radius^2 is the proper way to compare i.e., our Dconfocal. (Ref. JMB 2021, iScience 2022 by Dey et al.).

Reviewer #2 (Recommendations For The Authors): 

Recommendations: 

(1) Page 3 at the bottom of the Introduction states, "...sodium azide (Hiruma et al., 2007) inhibited accumulation in lysosomes, cellular diffusion...increased only slightly." However, Figure 7C, F shows a sodium azide-induced decrease in the Dconfocal cellular diffusion. Please clarify.

Thank you for pointing this out; we corrected it in the revised version, including adding statistics.

(2) Page 6 states, "Quinacrine accumulation in the lysosome was observed also immediately after micro-injection, with aggregation increasing over time. Dconfocal of 4.2{plus minus}0.2 µm2 s-1 was calculated from line-FRAP immediately after micro-injection, slowing to 2.2{plus minus}0.1 µm2 s-1 following 2 hours incubations, with fractional recoveries of 0.63 and 0.57 respectively." If lysosome sequestration does not have an effect on cytosolic diffusion rates as the manuscript concludes, why do the authors think the diffusion rate decreased here within 2 hours? A solid conclusion would strengthen the conclusions of this manuscript rather than passing over it.

Thank you for pointing this out. We added the following text to page 7: “It is notable that the Dconfocal for Quinacrine remained consistent regardless of Bafilomycin treatment, 2 hours after incubation (Fig. S9D, 2.4±0.1 µm2s-1). However, when measured immediately after injection, the diffusion coefficient was higher at 4.2 µm2s-1 (Fig. S5D). This result does not support the notion that the faster diffusion measured immediately after cellular injection relates to lysosomal aggregation, and would better support self-aggregation, or aggregation with other molecules in the cell, which increases over time. This notion is further supported by the almost complete lack in FRAP observed 24 hours after injection (Fig. S5C).”

(3) In the Results section, the subheading states, "Inhibition of lysosomal sequestration is only slightly increasing diffusion in cells", but the conclusion for bafilomycin was...Dconfocal values were not altered by Bafilomycin A1", and the conclusion for sodium azide was diffusion coefficients (Figure 7B-C and 7E-F) were not much changed for the two drugs and stayed low... similarly to what was observed with Bafilomycin." The clear question is what is the result, "slightly increased diffusion, decreased diffusion, or had no significant effect at all"? Please clarify the wording in the manuscript to accurately describe the results. 

Indeed, a small difference is obsevered between the two treatments. We added now statistical significance to Fig. 7D and H and to Fig. S8 and S9. In addition, we clarified this point in the text in p.7-8: “Comparative FRAP profiles and diffusion coefficients (Figure 7B-D and 7F-H) were slow, but conversely to Bafilomycin, sodium azide treatment did cause a further reduction is rates from Dconfocal 2.4±0.1 µm2s-1  to 1.8±0.1µm2s-1 for quinacrine and from 0.6 to  0.45 µm2s-1 for the GSK3 inhibitor (Figure 7C and G). Both Bafilomycin and sodium azide treatments resulted in elimination of drug confinement in the lysosome, and the small difference in diffusion rates may be a result of the de-acidification of the lysosomes by sodium azide, which may increase the protons in the cytosol upon treatment.”

(4) In Figure 8B, why was the Dconfocal for AM-fluorescein with or without sodium azide not included here? Besides consistency, the results might demonstrate significance. Please elaborate on the occlusion of this data. 

Fraction recovery after FRAP of AM-fluorescein was very low. Calculating Dconfocal rates with such low fraction recovery is meaningless, as in the time of measurement only a small fraction recovered. Therefore, we calculated Dconfocal only when fraction recovery was at least 0.5.

(5) Throughout the Results section, the ideas and experiments are of relevance, but the suggestions/conclusions at the end of each paragraph of this section seem lightly thought out. For example, as stated on Page 8, "...however, this did not contribute new information to the puzzle." For a chemistry paper, a chemical suggestion strengthens the manuscript. 

We want to thank the reviewer for these suggestions. We now made new Figures 9, and S16 to compare multiple parameters. Figure 9C shows a clear relation between pKa and Dconfocal, but no relation was found between logP, MW or number of aromatic rings and Dconfocal. Fig. S16 also shows the relation between drug concentration and Dconfocal values. We revised the discussion section to giver more weith to these quantitative assessments. These data are now discussed in p. 9.

In conclusion, the manuscript's ideas are needed, but the conclusions drawn from the experiments need to be strengthened, more explanatory, and consistent with the main conclusion of the manuscript.

See answer to point 5.

eLife Assessment

This is a useful revised manuscript that shows a set of data including the first cryo-EM structures of human PIEZO1 as well as structures of disease-related mutants in complex with the regulatory subunit MDFIC, which generate different inactivation phenotypes. The molecular basis of PIEZO channel inactivation is of great interest due to its association with several pathologies. This manuscript provides some structural insights that may help to ultimately build a molecular picture of PIEZO channel inactivation. While the structures are of use and clear conformational differences can be seen in the presence of the auxiliary subunit MDFIC, the strength of the evidence supporting the conclusions of the paper, especially the proposed role for pore lipids in inactivation, is incomplete.

Reviewer #1 (Public review):

Summary:

This manuscript by Shan, Guo, Zhang, Chen et al., shows a raft of interesting data including the first cryo-EM structures of human PIEZO1. Clearly the molecular basis of PIEZO channel inactivation is of great interest and as such this manuscript provides some valuable extra information that may help to ultimately build a molecular picture of PIEZO channel inactivation. However, the current manuscript though does not provide any compelling evidence for a detailed mechanism of PIEZO inactivation.

Strengths:

This manuscript documents the first cryo-EM structures of human PIEZO1 and gain of function mutants associated with hereditary anaemia. It is also the first evidence showing that PIEZO1 gain of function mutants are also regulated by the auxiliary subunit MDFIC.

Weaknesses:

While the structures are interesting and clear differences can be seen in the presence of the auxiliary subunit MDFIC the major conclusions and central tenets of the paper, especially a role for pore lipids in inactivation, lack data to support them. The post translational modification of PIEZOs auxiliary subunit MDFIC is not modelled as a covalent interaction.

Comments on revisions:

The revisions do absolutely nothing to allay any of the major concerns documented in my initial review of this manuscript.

(1) Mouse vs Human inactivation<br /> Not only is a quantification not provided the literature on this point is still not at all referenced or discussed.<br /> (2) MDFIC -lipidation<br /> Even if they are not assigned in the PDB for illustration they can at least be modelled correctly as covalently bound acyl chains.<br /> (3) Pore lipids and inactivation<br /> None of the explanations are consistent with the data shown.<br /> (4) Cytosolic plug<br /> There is not even any extra discussion provided on this point.<br /> (5) Reduced sensitivity of PIEZO1 in the presence of MDFIC and its regulatory mechanism<br /> No quantification is provided.<br /> (6) Both referencing of the PIEZO1 literature and prose could be improved.<br /> There is little to no attempt to improve the referencing.

Reviewer #2 (Public review):

Notably, the authors provide the first structure of human PIEZO1 (hPIEZO1), which will facilitate future studies in the field. They reveal that hPIEZO1 has a more flattened shape than mouse PIEZO1 (mPIEZO1) and has lipids that insert into the hydrophobic pore region. To understand how PIEZO1 GOF mutations might affect this structure and the underlying mechanistic changes, they solve structures of hPIEZO1 as well as two HX causing mild GOF mutations (A1988V and E756del) and a severe GOF mutation (R2456H). Unable to glean too much information due to poor resolution of the mutant channels, the authors also attempt to resolve MCFIC-bound structures of the mutants. These structures show that MDFIC inserts into the pore region of hPIEZO1, similar to its interaction with mPIEZO1, and results in a more curved and contracted state than hPIEZO1 on its own. The authors use these structures to hypothesize that differences in curvature and pore lipid position underlie the differences in inactivation kinetics between wild-type hPIEZO1, hPIEZO1 GOF mutations, and hPIEZO1 in complex with MDFIC.

Strengths:

This is the first human PIEZO1 structure. Thus, these studies become the steppingstone for future investigations to better understand how disease-causing mutations affect channel gating kinetics.

Comments on revisions:

The revised version of the manuscript is stronger and the authors have addressed most of our concerns. The only clarification that remains is data related to the electrophysiology experiments, Figure S2. In the response, the authors mention that they were referring to previously reported mPIEZO1 mutants. However, it is still missing quantification from the human mutant + MDFIC data. This data should be available to the authors and will be more informative than just the representative traces. In the text line 151-152 "Indeed, electrophysiological studies showed that co-expression of these channelopathy mutants with MDFIC resulted in significantly reduced mechanosensitivity and inactivation rate (Fig. S2)." However the updated version does not have any number or the statistics that were performed to indicate significance. I acknowledge that in the response they describe threshold but very descriptively.

Reviewer #3 (Public review):

Summary:

In this manuscript, the authors used structural biology approaches to determine the molecular mechanism underlying the inactivation of the PIEZO1 ion channel. To this end, the authors presented structures of human PIEZO1 and its slow-inactivating mutants. The authors also determined the structures of these PIEZO1 constructs in complexes with the auxiliary subunit MDFIC, which substantially slows down PIEZO1 inactivation. From these structures, the authors observed a unique feature of human PIEZO1 in which the lipid molecules plugged the channel pore in fast-inactivating constructs. The authors proposed that these lipid molecules prevent ion permeation and underlie the molecular mechanism of human PIEZO1 inactivation.

Strengths:

Notedly, this manuscript reported the first structures of a human PIEZO1 channel, its channelopathy mutants, and their complexes with MDFIC. The proposed role of pore lipids in modulating PIEZO1 ion permeation is interesting.

Weaknesses:

The authors' conclusion regarding the role of pore lipids in PIEZO inactivation is based on the assumption that all structures of human PIEZO1 resolved in this work represent comparable functional states relevant to channel inactivation. The authors should at least acknowledge that this is a critical assumption that is difficult to validate. The fitting of the lipid molecule to cryo-EM density could be improved.

Comments on revisions:

Upon revision, the authors substantially weakened the statement regarding the correlation between curvature and inactivation. The authors also toned down the statement regarding the role of pore lipids in channel inactivation. However, I have a few additional comments.

(1) As I have stated above, the assumption here is that all structures presented in this work represent comparable functional states relevant to channel inactivation. However, this assumption could be invalid. For example, the WT channel could be in the closed conformation, whereas the mutant could be stabilized in a different functional state. I understand that this is very difficult to test structurally and functionally. Therefore, I think the authors should at least acknowledge this limitation/assumption.<br /> (2) This time, I reviewed the coordinates and the map of the PIEZO1 structures. For example, in the WT channel, the fitting of the lipid to the cryo-EM density is questionable and I personally wouldn't model this lipid in this pose.

Author response:

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

eLife Assessment

This useful manuscript shows a set of interesting data including the first cryo-EM structures of human PIEZO1 as well as structures of disease-related mutants in complex with the regulatory subunit MDFIC, which generate different inactivation phenotypes. The molecular basis of PIEZO channel inactivation is of great interest due to its association with several pathologies. This manuscript provides some structural insights that may help to ultimately build a molecular picture of PIEZO channel inactivation. While the structures are of use and clear conformational differences can be seen in the presence of the auxiliary subunit MDFIC, the strength of the evidence supporting the conclusions of the paper, especially the proposed role for pore lipids in inactivation, is incomplete and there is a lack of data to support them.

We thank the editors and reviewers for taking the time and effort to review our manuscript.  The evidence supporting the key role of pore lipids in hPIEZO1 activation is as follows. i. Compared with wild-type hPIEZO1, the hydrophobic acyl chain tails of the pore lipids retracted from the hydrophobic pore region in slower inactivating mutant hPIEZO1-A1988V (Fig. 7a-b). ii. Previous electrophysiological functional studies revealed that substituting this hydrophobic pore formed by I2447, V2450, and F2454 with a hydrophilic pore prolongs the inactivation time for both PIEZO1 and PIEZO2 channels (PMID: 30628892). iii. In the structure of the HX channelopathy mutant R2456H, the interaction between the hydrophilic phosphate group head of pore lipids and R2456 is disrupted, remodeling the blade and pore module and resulting in a significantly slow-inactivating rate. iv. The interaction between pore lipids and lipidated-MDFIC stabilizes the pore lipids to reseal the pore upon activation of the hPIEZO1-MDFIC complex.

According to previously proposed models for the role of pore lipids in mechanosensitive ion channels, such as MscS (PMID: 33568813), MS K2P (PMID: 25500157) and OSCA channels (PMID: 37402734), the pore lipids seal the channel pores in closed state and could be removed in open state by mechanical force induced membrane deformation, which obeys the force-from-lipids principle. Therefore, in our putative model, the pore lipids seal the hydrophobic pore of hPIEZO1 in the closed state. Upon activation of hPIEZO1, the pore lipids retract from the hydrophobic pore and interact with multi-lipidated MDFIC, stabilizing in the inactivation state. The mild channelopathy mutants make the pore lipids retract from the hydrophobic pore and harder to close upon activation. For the severe channelopathy mutant, the interaction between the pore lipids and R2456 is disrupted, resulting in the missing of pore lipids and significantly slow-inactivating. We fully understand the concern of the role of pore lipids in our proposed model. Therefore, we have toned down our putative model.

Public Reviews:  

Reviewer #1 (Public review):  

Summary:  

This manuscript by Shan, Guo, Zhang, Chen et al., shows a raft of interesting data including the first cryo-EM structures of human PIEZO1. Clearly, the molecular basis of PIEZO channel inactivation is of great interest and as such this manuscript provides some valuable extra information that may help to ultimately build a molecular picture of PIEZO channel inactivation. However, the current manuscript though does not provide any compelling evidence for a detailed mechanism of PIEZO inactivation.

Strengths:

This manuscript documents the first cryo-EM structures of human PIEZO1 and the gain of function mutants associated with hereditary anaemia. It is also the first evidence showing that PIEZO1 gain of function mutants are also regulated by the auxiliary subunit MDFIC.

We thank reviewer #1 for the encouragement.

Weaknesses:

While the structures are interesting and clear differences can be seen in the presence of the auxiliary subunit MDFIC the major conclusions and central tenets of the paper, especially a role for pore lipids in inactivation, lack data to support them. The post-translational modification of PIEZOser# auxiliary subunit MDFIC is not modelled as a covalent interaction.

We fully understand the concern of the role of pore lipids in our proposed model. Therefore, we have toned down our putative model.

The lipids densities of the post-transcriptional modification of PIEZO1 auxiliary subunit MDFIC are shown below. As the lipids densities are not confident, we only use the single-chain lipids to represent them. And the lipidated MDFIC is proven by the MDFIC identification paper.

Author response image 1.

Reviewer #2 (Public review):

Summary:

Mechanically activated ion channels PIEZOs have been widely studied for their role in mechanosensory processes like touch sensation and red blood cell volume regulation. PIEZO in vivo roles are further exemplified by the presence of gain-of-function (GOF) or loss-of-function (LOF) mutations in humans that lead to disease pathologies. Hereditary xerocytosis (HX) is one such disease caused due to GOF mutation in Human PIEZO1, which are characterized by their slow inactivation kinetics, the ability of a channel to close in the presence of stimulus. But how these mutations alter PIEZO1 inactivation or even the underlying mechanisms of channel inactivation remains unknown. Recently, MDFIC (myoblast determination family inhibitor proteins) was shown to directly interact with mouse PIEZO1 as an auxiliary subunit to prolong inactivation and alter gating kinetics. Furthermore, while lipids are known to play a role in the inactivation and gating of other mechanosensitive channels, whether this mechanism is conserved in PIEZO1 is unknown. Thus, the structural basis for PIEZO1 inactivation mechanism, and whether lipids play a role in these mechanisms represent important outstanding questions in the field and have strong implications for human health and disease.

To get at these questions, Shan et al. use cryogenic electron microscopy (Cryo-EM) to investigate the molecular basis underlying differences in inactivation and gating kinetics of PIEZO1 and human disease-causing PIEZO1 mutations. Notably, the authors provide the first structure of human PIEZO1 (hPIEZO1), which will facilitate future studies in the field. They reveal that hPIEZO1 has a more flattened shape than mouse PIEZO1 (mPIEZO1) and has lipids that insert into the hydrophobic pore region. To understand how PIEZO1 GOF mutations might affect this structure and the underlying mechanistic changes, they solve structures of hPIEZO1 as well as two HXcausing mild GOF mutations (A1988V and E756del) and a severe GOF mutation (R2456H). Unable to glean too much information due to poor resolution of the mutant channels, the authors also attempt to resolve MCFIC-bound structures of the mutants. These structures show that MDFIC inserts into the pore region of hPIEZO1, similar to its interaction with mPIEZO1, and results in a more curved and contracted state than hPIEZO1 on its own. The authors use these structures to hypothesize that differences in curvature and pore lipid position underlie the differences in inactivation kinetics between wild-type hPIEZO1, hPIEZO1 GOF mutations, and hPIEZO1 in complex with MDFIC.

Strengths:

This is the first human PIEZO1 structure. Thus, these studies become the stepping stone for future investigations to better understand how disease-causing mutations affect channel gating kinetics.

We thank reviewer #2 for the positive comments.

Weaknesses:

Many of the hypotheses made in this manuscript are not substantiated with data and are extrapolated from mid-resolution structures.

We fully understand the concern of the role of pore lipids in our proposed model. Therefore, we have toned down our putative model.

Reviewer #3 (Public review):

Summary:

In this manuscript, the authors used structural biology approaches to determine the molecular mechanism underlying the inactivation of the PIEZO1 ion channel. To this end, the authors presented structures of human PIEZO1 and its slow-inactivating mutants. The authors also determined the structures of these PIEZO1 constructs in complexes with the auxiliary subunit MDFIC, which substantially slows down PIEZO1 inactivation. From these structures, the authors suggested an anti-correlation between the inactivation kinetics and the resting curvature of PIEZO1 in detergent. The authors also observed a unique feature of human PIEZO1 in which the lipid molecules plugged the channel pore. The authors proposed that these lipid molecules could stabilize human PIEZO1 in a prolonged inactivated state.

We thank reviewer #3 for the summary.

Strengths:

Notedly, this manuscript reported the first structures of a human PIEZO1 channel, its channelopathy mutants, and their complexes with MDFIC. The evidence that lipid molecules could occupy the channel pore of human PIEZO1 is solid. The authors' proposals to correlate PIEZO1 resting curvature and pore-resident lipid molecules with the inactivation kinetics are novel and interesting.

Thanks for the positive comments.

Weaknesses:

However, in my opinion, additional evidence is needed to support the authors' proposals.

(1) The authors determined the apo structure of human PIEZO1, which showed a more flattened architecture than that of the mouse PIEZO1. Functionally, the inactivation kinetics of human PIEZO1 is faster than its mouse counterpart. From this observation (and some subsequent observations such as the complex with MDFIC), the authors proposed the anti-correlation between curvature and inactivation kinetics. However, the comparison between human and mouse PIEZO1 structure might not be justified. For example, the human and mouse structures were determined in different detergent environments, and the choice of detergent could influence the resting curvature of the PIEZO structures.

We apologize for the misleading statement about the anti-correlation between curvature and inactivation kinetics of PIEZOs. We cannot conclude that the observation of curvature variation of mPIEZO1 and hPIEZO1 is related to their inactivation kinetics based on structural studies and electrophysiological assay. The difference in structural basis between mPIEZO1 and hPIEZO1 is what we want to state. To avoid this misleading, we have revised the manuscript. 

For the concern about detergent, we cannot fully exclude its influence on the curvature of PIEZOs. However, previously reported structures of mPiezo1 (PDB: 7WLT, 5Z10, 6B3R) were in the different detergent environments or in lipid bilayer, but the curvature of mPiezo1 is similar as shown below. Considering the high sequence similarity between mPiezo1 and hPiezo1, we hypothesize that the curvature of both hPiezo1 and mPiezo1 may be unaffected by the detergent.

Author response image 2.

Overall structural comparison of curved mPIEZO1 in the lipid bilayer (PDB: 7WLT), mPiezo1 in CHAPS (PDB: 6B3R) and mPiezo1 in Digitonin (PDB: 5Z10).

(2) Related to point 1), the 3.7 Å structure of the A1988V mutant presented by the authors showed a similar curvature as the WT but has a slower inactivating kinetics.

Based on the structural comparison between hPIEZO1 and its A1998V mutant, the retraction of pore lipids from the hydrophobic center pore in hPIEZO1-A1998V is mainly responsible for its slower inactivating kinetics.

(3) Related to point 1), the authors stated that human PIEZO1 might not share the same mechanism as mouse PIEZO1 due to its unique properties. For example, MDFIC only modifies the curvature of human PIEZO1, and lipid molecules were only observed in the pore of the human PIEZO1. Therefore, it may not be justified to draw any conclusions by comparing the structures of PIEZO1 from humans and mice.

Thanks for the constructive suggestion. To avoid this misleading, we have revised the manuscript.

(4) Related to point 1), it is well established that PIEZO1 opening is associated with a flattened structure. If the authors' proposal were true, in which a more flattened structure led to faster inactivation, we would have the following prediction: more opening is associated with faster inactivation. In this case, we would expect a pressure-dependent increase in the inactivation kinetics.

Could the authors provide such evidence, or provide other evidence along this direction?

We appreciate the reviewer’s comment. We are not claiming a relationship between the flattened structure and activation/inactivation. We only present the results of the structure of wild-type/mutant PIEZO1.

(5) In Figure S2, the authors showed representative experiments of the inactivation kinetics of PIEZO1 using whole-cell poking. However, poking experiments have high cell-to-cell variability.

The authors should also show statics of experiments obtained from multiple cells.

We have shown the statics of representative electrophysiology experiments obtained from multiple cells in Figure S2.

(6) In Figure 2 and Figure 5, when the authors show the pore diameter, it could be helpful to also show the side chain densities of the pore lining residues.

We appreciate the reviewer’s suggestion. The side chain of the pore lining restricted residues have been shown in Figure 2 and Figure 5 and the densities of pore domain have been shown in Figure S4 and S14. Interestingly, the pore lining restricted residues in mPIEZO1 and hPIEZO1 is highly conserved.

(7) The authors observed pore-plugging lipids in slow inactivating conditions such as channelopathy mutations or in complex with MDFIC. The authors propose that these lipid molecules stabilize a "deep resting state" of PIEZO1, making it harder to open and harder to inactivate once opened. This will lead to the prediction that the slow-inactivating conditions will lead to a higher activation threshold, such as the mid-point pressure in the activation curve. Is this true?

Yes, it is true. In Figure S2, the MDFIC-induced slow-inactivation conditions in hPIEZO1-MDFIC, hPIEZO1-A1988V-MDFIC, hPIEZO1-E756del-MDFIC and hPIEZO1-R2456H-MDFIC result in larger half-activation thresholds than hPIEZO1, hPIEZO1-A1988V, hPIEZO1-E756del and hPIEZO1-R2456H, respectively.

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

I document the major issues below:

(1) Mouse vs Human inactivation

Line 21- "than the slower inactivating curved mouse PIEZO1 (mPIEZO1)."

Where is the data in this paper or any other paper that human PIEZO1 inactivates faster than mouse PIEZO1? This is central to the way the authors present the paper. In fact, the tau quoted for the hPIEZO1 of ~10 ms is similar to that often measured for mPIEZO1. The reference in the discussion for mouse vs human inactivation times is a review of mechanotransduction. Either the authors need to directly compare the tau of mP1 vs hP1 or quote the relevant primary literature if it exists.

As measured in HEK-PIKO cells transfected with mPiezo1, the inactivation time of mPiezo1 is 13 ± 1 ms (PMID: 29261642) at -80 mV. 

The tau is also voltage-dependent. The tau is beyond 20 ms at -60 mV for mPIEZO1 (PMID:

20813920) and for hPIEZO1 is still around 10 ms.

(2) MDFIC-lipidation

Without seeing the PDB or EMDB I can't guarantee this but from Figure 6d it seems like the Sacylation in the distal C-terminus of MDFIC is not modelled as a covalent interaction, these lipids are covalently added to the Cys residues in S-acylation via zDHHC enzymes. This should be modelled correctly.

Thanks for this suggestion. As the lipid densities of the post-transcriptional modification of PIEZOs auxiliary subunit MDFIC are not confident, we only use the single-chain lipids to represent them.

And the lipidated MDFIC is proven by the MDFIC identification paper (PMID: 37590348).

(3) Pore lipids and inactivation

The lipids close to the pore are interesting and the density for a lipid is also seen in the mouse MDFIC-PIEZO1 complex from Zhou, Ma et al, 2023. However, there is no data provided by the authors that the lipid is functionally relevant to anything. There is not even a correlation with inactivation in Figure 7. P1+MDFIC inactivates slowest yet the lipids are present within the pore. Second, there is no evidence for what these structures are: closed, or inactivated? In fact, the Xiao lab is now interpreting the 7WLU structure as inactivated.

The evidence supporting the key role of pore lipids in hPIEZO1 activation is as follows. i. Compared with wild-type hPIEZO1, the hydrophobic acyl chain tails of the pore lipids retracted from the hydrophobic pore region in slower inactivating mutant hPIEZO1-A1988V (Fig. 7a-b). ii. Previous electrophysiological functional studies revealed that substituting this hydrophobic pore formed by I2447, V2450, and F2454 with a hydrophilic pore prolongs the inactivation time for both PIEZO1 and PIEZO2 channels (PMID: 30628892). iii. In the structure of the HX channelopathy mutant R2456H, the interaction between the hydrophilic phosphate group head of pore lipids and R2456 is disrupted, remodeling the blade and pore module and resulting in a significantly slow-inactivating rate. iv. The interaction between pore lipids and lipidated-MDFIC stabilizes the pore lipids to reseal the pore upon activation of the hPIEZO1-MDFIC complex. Overall, the pore lipid is involved in inactivation, and we have toned down the statement.

(4) Cytosolic plug

There is additional cytosolic density for the human PIEZO1 that the authors intimate could be from a different binding partner. IS it possible to refine this density? Is it from the PIEZO1-tag? At the very least a little more information about this density should be given if it is going to be mentioned like this.

Our purification result shows that the protein is tag-free. We are also curious about the extra cytosolic density, but we do not know what it is.

(5) Reduced sensitivity of PIEZO1 in the presence of MDFIC and its regulatory mechanism

This was reported in the first article however no data is presented by the authors to support MDFIC increasing the mechanical energy required to open PIEZO1. The sentence in the discussion; "MDFIC enables hPIEZO1 to respond to different forces by modifying the pore module through lipid interactions." is not supported by any functional data and seems to be an over-interpretation of the structures.

We appreciate this suggestion. The half-activation threshold of hPEIZO1 and hPEIZO1-MDFIC is measured to be 7 μm and 9 μm, respectively (Fig.S2). In addition, the mechanical currents amplitude of hPIEZO1-MDFIC is extremely small compared to that of WT reaching the nA level (Fig.S2). Therefore, the less mechanosensitive hPIEZO1-MDFIC may require more mechanical energy to open than PIEZO1 WT.

6) Both referencing of the PIEZO1 literature and prose could be improved.

Thanks for the suggestion. We have improved the referencing and prose.

Reviewer #2 (Recommendations for the authors):

(1) The authors speculate that the difference in curvature between human and mouse PIEZO1 results in its fast inactivation but do not provide experimental evidence to support this idea. This claim would have been bolstered by showing that the GOF human mutations have a more curved structure, but these proved too structurally unstable to be solved at high resolution. However, the authors state that the 3.7 angstrom map solved for hPIEZO1-A1988V does have an overall similar architecture as wild-type hPIEZO1; thus, contradicting their hypothesis.

We apologize for the misleading statement. In our revised manuscript, we do not claim a relationship between the flattened structure and activation/inactivation. We only present the results of the structure of wild-type/mutant PIEZO1.

The structure comparison between the A1988V mutant and WT shows a similar architecture but a different occupancy pattern of pore lipids. Therefore, we suggested that the A1988V mutant has slightly slower inactivation kinetics, mainly due to the exit of pore lipids from the pore.

(2) The authors show that interaction with MDFIC alters hPIEZO1 structure to be more curved and use this to support their idea that changing the curvature of the protein underlies the prolonged inactivation kinetics. It has been previously shown that MDFIC does not change the structure of mPIEZO1 but does alter its inactivation and gating kinetics. How does this discrepancy fit into the inactivation model proposed by the authors? Similarly, their claim that MDFIC slows hPIEZO1 inactivation and weakens mechanosensitivity just by affecting the pore module and changing blade curvature is made based on observation and no experimental data to test it.

We have revised the manuscript to avoid misleading the relationship between the curvature and the inaction kinetics of hPIEZO1. The evidence reported previously that substitution of the hydrophobic pore, formed by I2447, V2450, and F2454, with a hydrophilic pore prolongs the inactivation time for both PIEZO1 and PIEZO2 channels (PMID: 30628892). In addition, the severe HX channelopathy mutant R2456H, wherein the interaction between the hydrophilic phosphate group head and R2456 is disrupted, leads to remodeling of the blade and pore module. Indeed, our observation is limited and further experiments will be performed to support our model.

(3) How does their model fit in cell types that have PIEZO1 (or GOF mutant PIEZO1) but not MDFIC?

In cell types that have PIEZO1 or GOF mutant PIEZO1 but not MDFIC, PIEZO1 or GOF mutant PIEZO1 may have a faster inactivation rate than those that bind to MDFIC. It can be proved that overexpressed PIEZOs exhibit faster inactivation kinetics than those in some native cell types with MDFIC expression (PMID: 20813920, 30132757).

(4) Figure S2 is missing quantification of the electrophysiology data. The authors should show summary data in addition to their representative traces including the Imax for all conditions, tau for data shown in b, and sample size for all conditions, and related statistics. The text claims that MDFIC decreases mechanosensitivity (line 156) but there is no data to support this.

For the electrophysiological assay in Figure S2, we referred to previously reported mPIEZO1 mutants (PMID: 23487776, 28716860). We confirmed that the slower inactivation phenotypes of these mutations of hPIEZO1 are similar to those of mPIEZO1.

The half-activation threshold of hPEIZO1 and hPEIZO1-MDFIC is measured to be 7 μm and 9 μm, respectively. This tendency of increased half-activation threshold of hPIEZO1 upon binding with MDFIC is also shown in the electrophysiological result of hPIEZO1 channelopathy mutants.

(5) In line 144, the authors mention that they were able to validate the MDFIC density with multilipidated cysteines on the C-terminal amphipathic helix, but they do not show the density with fitted lipids. While individual densities for some of the lipids are shown in extended Figure 12, it would be helpful to include a figure where they show the map for MDFIC with fitted lipids in it.

Thanks for the valuable suggestion. As the lipid densities of the post-transcriptional modification of PIEZOs auxiliary subunit MDFIC are not confident, we only use the single-chain lipids to represent them. And the lipidated MDFIC is proven by the MDFIC identification paper.

(6) The authors show that R2456 interacts with a lipid at the pore module and hypothesize that this underlies the fast inactivation of hPIEZO1. While they did not obtain a high-resolution structure of this mutant, this hypothesis could be tested by substituting R for side chains with different charges and performing electrophysiology to determine the effects on inactivation.

Thanks for the constructive suggestion. We will perform the electrophysiology assay for R2456 mutants with different side chains.

7) Figure 4 shows overall structure of hPIEZO1 GOF mutations A1988V and E756del in complex with MDFIC. Other than showing an overall similar structure to wildtype hPIEZO1, the authors do not show how the human mutations A1988V alter the structure of the protein at the site of change. Understanding how these mutations affect the local architecture of the protein has important relevance for human physiology.

As the GOF channelopathy mutant hPIEZO1-A1988V is structurally unstable, the density at the site of A1988V is too weak to figure out the related interaction in the structure of the hPIEZO1-A1988V mutant. 

Minor comment:

In general, the manuscript will benefit from heavy copy editing. For example, the word cartoon is misspelled in many of the figure legends.

We apologize for the mistake. The manuscript has been checked and revised.

Reviewer #3 (Recommendations for the authors):

Some portions of this manuscript were not well written. For example, at the end of the 3rd paragraph in the introduction, the authors talked about HX mutations and their correlation with malaria infection and plasma iron. This is irrelevant information and will only distract the readers. It would be ideal if the authors could go through the entire manuscript and improve its clarity.

Thanks for the suggestion. We have revised the sentences about HX mutations as suggested and improved the entire manuscript.

eLife Assessment

In this manuscript, Kaplan et al. study mesenchymal Meis2 in whisker formation and the links between whisker formation and sensory innervation. These useful findings analyze the impact of the conditional deletion of Meis2 using the Wnt1 driver on whisker development and their interaction with trigeminal nerves. The evidence supporting the conclusions is incomplete lacking mechanistic links to the phenotype described and detailed information on the methods used to analyze single-cell RNA sequencing data. The work will be of interest to developmental and skin biologists.

Reviewer #1 (Public review):

Summary:

Mehmet Mahsum Kaplan et al. demonstrate that Meis2 expression in neural crest-derived mesenchymal cells is crucial for whisker follicle (WF) development, as WF fails to develop in wnt1-Cre;Meis2 cKO mice. Advanced imaging techniques effectively support the idea that Meis2 is essential for proper WF development and that nerves, while affected in Meis2 cKO, are dispensable for WF development and not the primary cause of WF developmental failure. The study also reveals that although Meis2 significantly downregulates Foxd1 in the mesenchyme, this is not the main reason for WF development failure. The paper presents valuable data on the role of mesenchymal Meis2 in WF development. However, further quantification and analysis of the WF developmental phenotype would be beneficial in strengthening the claim that Meis2 controls early WF development rather than causing a delay or arrest in development. A deeper sequencing data analysis could also help link Meis2 to its downstream targets that directly impact the epithelial compartment.

Strengths:

(1) The authors describe a novel molecular mechanism involving Mesenchymal Meis2 expression, which plays a crucial role in early WF development.

(2) They employ multiple advanced imaging techniques to illustrate their findings beautifully.

(3) The study clearly shows that nerves are not essential for WF development.

Weaknesses:

(1) The authors claim that Meis2 acts very early during development, as evidenced by a significant reduction in EDAR expression, one of the earliest markers of placode development. While EDAR is indeed absent from the lower panel in Figure 3C of the Meis2 cKO, multiple placodes still express EDAR in the upper two panels of the Meis2 cKO. The authors also present subsequent analysis at E13.3, showing one escaped follicle positive for SHH and Sox9 in Figures 1 and 3. Does this suggest that follicles are specified but fail to develop? Alternatively, could there be a delay in follicle formation? The increase in Foxd1 expression between E12.5 and E13.5 might also indicate delayed follicle development, or as the authors suggest, follicles that have escaped the phenotype. The paper would significantly benefit from robust quantification to accompany their visual data, specifically quantifying EDAR, Sox9, and Foxd1 at different developmental stages. Additionally, analyzing later developmental stages could help distinguish between a delay or arrest in WF development and a complete failure to specify placodes.

(2) The authors show that single-cell sequencing reveals a reduction in the pre-DC population, reduced proliferation, and changes in cell adhesion and ECM. However, these changes appear to affect most mesenchymal cells, not just pre-DCs. Moreover, since E12.5 already contains WFs at different stages of development, as well as pre-DCs and DCs, it becomes challenging to connect these mesenchymal changes directly to WF development. Did the authors attempt to re-cluster only Cluster 2 to determine if a specific subpopulation is missing in Meis2 cKO? Alternatively, focusing on additional secreted molecules whose expression is disrupted across different clusters in Meis2 cKO could provide insights, especially since mesenchymal-epithelial communication is often mediated through secreted molecules. Did the authors include epithelial cells in the single-cell sequencing, can they look for changes in mesenchyme-epithelial cell interactions (Cell Chat) to indicate a possible mechanism?

(3) The authors aim to link Meis2 expression in the mesenchyme with epithelial Wnt signaling by analyzing Lef1, bat-gal, Axin1, and Wnt10b expression. However, the changes described in the figures are unclear, and the phenotype appears highly variable, making it difficult to establish a connection between Meis2 and Wnt signaling. For instance, some follicles and pre-condensates are Lef1 positive in Meis2 cKO. Including quantification or providing a clearer explanation could help clarify the relationship between mesenchymal Meis2 and Wnt signaling in both epidermal and mesenchymal cells. Did the authors include epithelial cells in the sequencing? Could they use single-cell analysis to demonstrate changes in Wnt signaling?

(4) Existing literature, including studies on Neurog KO and NGF KO, as well as the references cited by the authors, suggest that nerves are unlikely to mediate WF development. While the authors conduct a thorough analysis of WF development in Neurog KO, further supporting this notion, this point may not be central to the current work. Additionally, the claim that Meis2 influences trigeminal nerve patterning requires further analysis and quantification for validation.

(5) Meis2 expression seems reduced but has not entirely disappeared from the mesenchyme. Can the authors provide quantification?

Reviewer #2 (Public review):

Summary:

In this manuscript, Kaplan et al. study mesenchymal Meis2 in whisker formation and the links between whisker formation and sensory innervation. To this end, they used conditional deletion of Meis2 using the Wnt1 driver. Whisker development was arrested at the placode induction stage in Meis2 conditional knockouts leading to the absence of expression of placodal genes such as Edar, Lef1, and Shh. The authors also show that branching of trigeminal nerves innervating whisker follicles was severely affected but that whiskers did form in the complete absence of trigeminal nerves.

Strengths:

The analysis of Meis2 conditional knockouts convincingly shows a lack of whisker formation and all epithelial whisker/hair placode markers were analyzed. Using Neurog1 knockout mice, the authors show equally convincingly that whiskers and teeth develop in the complete absence of trigeminal nerves.

Weaknesses:

The manuscript does not provide much mechanistic insight as to why mesenchymal Meis2 leads to the absence of whisker placodes. Using a previously generated scRNA-seq dataset they show that two early markers of dermal condensates, Foxd1 and Sox2, are downregulated in Meis2 mutants. However, given that placodes and dermal condensates do not form in the mutants, this is not surprising and their absence in the mutants does not provide any direct link between Meis2 and Foxd1 or Sox2. (The absence of a structure evidently leads to the absence of its markers.)

Author response:

Reviewer #1 (Public review):

Summary:

Mehmet Mahsum Kaplan et al. demonstrate that Meis2 expression in neural crest-derived mesenchymal cells is crucial for whisker follicle (WF) development, as WF fails to develop in wnt1-Cre;Meis2 cKO mice. Advanced imaging techniques effectively support the idea that Meis2 is essential for proper WF development and that nerves, while affected in Meis2 cKO, are dispensable for WF development and not the primary cause of WF developmental failure. The study also reveals that although Meis2 significantly downregulates Foxd1 in the mesenchyme, this is not the main reason for WF development failure. The paper presents valuable data on the role of mesenchymal Meis2 in WF development. However, further quantification and analysis of the WF developmental phenotype would be beneficial in strengthening the claim that Meis2 controls early WF development rather than causing a delay or arrest in development. A deeper sequencing data analysis could also help link Meis2 to its downstream targets that directly impact the epithelial compartment.

Strengths:

(1) The authors describe a novel molecular mechanism involving Mesenchymal Meis2 expression, which plays a crucial role in early WF development.

(2) They employ multiple advanced imaging techniques to illustrate their findings beautifully.

(3) The study clearly shows that nerves are not essential for WF development.

We thank the reviewer for valuable comments that will help improve our study.

Weaknesses:

(1) The authors claim that Meis2 acts very early during development, as evidenced by a significant reduction in EDAR expression, one of the earliest markers of placode development. While EDAR is indeed absent from the lower panel in Figure 3C of the Meis2 cKO, multiple placodes still express EDAR in the upper two panels of the Meis2 cKO. The authors also present subsequent analysis at E13.3, showing one escaped follicle positive for SHH and Sox9 in Figures 1 and 3. Does this suggest that follicles are specified but fail to develop? Alternatively, could there be a delay in follicle formation? The increase in Foxd1 expression between E12.5 and E13.5 might also indicate delayed follicle development, or as the authors suggest, follicles that have escaped the phenotype. The paper would significantly benefit from robust quantification to accompany their visual data, specifically quantifying EDAR, Sox9, and Foxd1 at different developmental stages. Additionally, analyzing later developmental stages could help distinguish between a delay or arrest in WF development and a complete failure to specify placodes.

The earliest DC (Foxd1) and placodal (EDAR, Lef1) markers tested in this study were observed only in the escaped WFs whereas these markers were missing in expected WF sites in mutants. This was also reflected in the loss of typical placodal morphology in the mutant’s epithelium. On the other hand, escaped WFs developed normally as shown by the analysis in Supp Fig 1A-B showing their normal size. These data suggest that development of escaped WFs is not delayed because they would appear smaller in size. To strengthen this conclusion, we will analyze whiskers at E18.5 in Meis2 cKO mice by staining Edar, Foxd1, Sox9 and/or Lef1 in revision and results will be added in the revised manuscript. Two-week time for this provisional response is too short to gather all these data. As far as quantification is concerned, we have already quantified the number of whiskers in controls and mutants at E12.5 and E13.5 in all whole mount experiments we did, i.e. Shh ISH and Sox9 or EDAR whole mount IFC. We pooled all these numbers together and calculated the whisker number reduction to 5.7+/-2.0% at E12.5 and 17.1+/-5.9 at E13.5 (page 3, row 114). We will also quantify the whisker number at E15.5 and E18.5 in the revised manuscript.

(2) The authors show that single-cell sequencing reveals a reduction in the pre-DC population, reduced proliferation, and changes in cell adhesion and ECM. However, these changes appear to affect most mesenchymal cells, not just pre-DCs. Moreover, since E12.5 already contains WFs at different stages of development, as well as pre-DCs and DCs, it becomes challenging to connect these mesenchymal changes directly to WF development. Did the authors attempt to re-cluster only Cluster 2 to determine if a specific subpopulation is missing in Meis2 cKO? Alternatively, focusing on additional secreted molecules whose expression is disrupted across different clusters in Meis2 cKO could provide insights, especially since mesenchymal-epithelial communication is often mediated through secreted molecules. Did the authors include epithelial cells in the single-cell sequencing, can they look for changes in mesenchyme-epithelial cell interactions (Cell Chat) to indicate a possible mechanism?

We agree with the reviewer that the effect of Meis2 on cell proliferation and expression of cell adhesion and ECM markers are more general because they take place in the whole underlying mesenchyme. Our genetic tools did not allow specific targeting of DC or pre-DCs. Nonetheless, we trust that our data show that mesenchymal Meis2 is required for the initial steps of WF development including Pc formation. As far as bioinformatics data are concerned, this data set was taken from the large dataset GSE262468 covering the whole craniofacial region which led to very limited cell numbers in the cluster 2 (DC): WT_E12_2 --> 28, WT_E13_2 --> 131, MUT_E12_2 --> 19, MUT_E13_2 --> 28. Unfortunately, such small cell numbers did not allow further sub-clustering, efficient normalization, integration and conclusions from their transcriptional profiles. Although a number of interesting differentially expressed genes were identified (see supplementary datasets), none of them convincingly pointed at reasonable secreted molecule candidate.  

We agree with the reviewer that cellchat analysis could provide robust indication of the mesenchymal-epithelial communication, however our datasets included only mesenchymal cell population (Wnt1-Cre2progeny) and epithelial cells were excluded by FACS prior to sc RNA-seq. (Hudacova et al. https://doi.org/10.1016/j.bone.2024.117297)

(3) The authors aim to link Meis2 expression in the mesenchyme with epithelial Wnt signaling by analyzing Lef1, bat-gal, Axin1, and Wnt10b expression. However, the changes described in the figures are unclear, and the phenotype appears highly variable, making it difficult to establish a connection between Meis2 and Wnt signaling. For instance, some follicles and pre-condensates are Lef1 positive in Meis2 cKO. Including quantification or providing a clearer explanation could help clarify the relationship between mesenchymal Meis2 and Wnt signaling in both epidermal and mesenchymal cells. Did the authors include epithelial cells in the sequencing? Could they use single-cell analysis to demonstrate changes in Wnt signaling?

We have now analyzed changes in Lef1 staining intensity in the epithelium and in the upper dermis. According to these quantifications, we observed a considerable decline in the number of Lef1+ placodes in the epithelium which corresponds to the lower number of placodes. On the other hand, Lef1 intensity in the ‘escaped’ placodes were similar between controls and mutants. Lef1 signal in the upper dermis is very strong overall and its quantification did not reveal any changes in the DC and non-DC region of the upper dermis. These data corroborate with our coclusion that Meis2 in the mesenchyme is not crucial for the dermal Wnt signaling but is required for induction of Lef1 expression in the epithelium. However, once ‘escaper’ placodes appear, they display normal wnt signaling in Pc, DC and subsequent development. These quantification data will be added to the revised manuscript.

(4) Existing literature, including studies on Neurog KO and NGF KO, as well as the references cited by the authors, suggest that nerves are unlikely to mediate WF development. While the authors conduct a thorough analysis of WF development in Neurog KO, further supporting this notion, this point may not be central to the current work. Additionally, the claim that Meis2 influences trigeminal nerve patterning requires further analysis and quantification for validation.

We agree with the reviewer that analysis of the Neurogenin knockout mice should not be central to this report. Nonetheless, a thorough analysis of WF development in Neurog1 KO was needed to distinguish between two possible mechanisms: whisker phenotype in Meis2 cKO results from 1. impaired nerve branching 2. Function of Meis2 in the mesenchyme. We will modify the text accordingly to make this clearer to readers. We also agree that nerve branching was not extensively analyzed in the current study but two samples from mutant mice were provided (Fig1 and Supp Videos), reflecting the consistency of the phenotype (see also Machon et al. 2015). This section was not central to this report either but led us to focus fully on the mesenchyme. We think that Meis2 function in cranial nerve development is very interesting and deserves a separate study.

(5) Meis2 expression seems reduced but has not entirely disappeared from the mesenchyme. Can the authors provide quantification?

In the revised manuscript, we will provide wt/mut quantification of Meis2 expression in the dermis.

Reviewer #2 (Public review):

Summary:

In this manuscript, Kaplan et al. study mesenchymal Meis2 in whisker formation and the links between whisker formation and sensory innervation. To this end, they used conditional deletion of Meis2 using the Wnt1 driver. Whisker development was arrested at the placode induction stage in Meis2 conditional knockouts leading to the absence of expression of placodal genes such as Edar, Lef1, and Shh. The authors also show that branching of trigeminal nerves innervating whisker follicles was severely affected but that whiskers did form in the complete absence of trigeminal nerves.

Strengths:

The analysis of Meis2 conditional knockouts convincingly shows a lack of whisker formation and all epithelial whisker/hair placode markers were analyzed. Using Neurog1 knockout mice, the authors show equally convincingly that whiskers and teeth develop in the complete absence of trigeminal nerves.

We thank the reviewer for valuable comments that will help improve our study.

Weaknesses:

The manuscript does not provide much mechanistic insight as to why mesenchymal Meis2 leads to the absence of whisker placodes. Using a previously generated scRNA-seq dataset they show that two early markers of dermal condensates, Foxd1 and Sox2, are downregulated in Meis2 mutants. However, given that placodes and dermal condensates do not form in the mutants, this is not surprising and their absence in the mutants does not provide any direct link between Meis2 and Foxd1 or Sox2. (The absence of a structure evidently leads to the absence of its markers.)

We apologize for unclear explanation of our data. We meant that Meis2 is functionally upstream of Foxd1 because Foxd1 is reduced upon Meis2 deletion. This means that during WF formation, Meis2 operates before Foxd1 induction and does not mean necessarily that Meis2 directly controls expression of Foxd1. Yes, we agree with reviewer’s note that Foxd1 and Sox2, as known DC markers, decline because the number of WF declines. We wanted to convince readers that Meis2 operates very early in the GRN hierarchy during WF development. We also admit that we provide poor mechanistic insights into Meis2 function as a transcription factor. We think that this weak point does not lower the value of the report showing indispensable role of Meis2 in WFs and possibly all HFs.

Reviewer #2 (Public Review):

Summary:

This work introduces a new method of depleting the ribosomal reads from the single-cell RNA sequencing library prepared with one of the prokaryotic scRNA-seq techniques, PETRI-seq. The advance is very useful since it allows broader access to the technology by lowering the cost of sequencing. It also allows more transcript recovery with fewer sequencing reads. The authors demonstrate the utility and performance of the method for three different model species and find a subpopulation of cells in the E.coli biofilm that express a protein, PdeI, which causes elevated c-di-GMP levels. These cells were shown to be in a state that promotes persister formation in response to ampicillin treatment.

Strengths:

The introduced rRNA depletion method is highly efficient, with the depletion for E.coli resulting in over 90% of reads containing mRNA. The method is ready to use with existing PETRI-seq libraries which is a large advantage, given that no other rRNA depletion methods were published for split-pool bacterial scRNA-seq methods. Therefore, the value of the method for the field is high. There is also evidence that a small number of cells at the bottom of a static biofilm express PdeI which is causing the elevated c-di-GMP levels that are associated with persister formation. Given that PdeI is a phosphodiesterase, which is supposed to promote hydrolysis of c-di-GMP, this finding is unexpected.

Weaknesses:

With the descriptions and writing of the manuscript, it is hard to place the findings about the PdeI into existing context (i.e. it is well known that c-di-GMP is involved in biofilm development and is heterogeneously distributed in several species' biofilms; it is also known that E.coli diesterases regulate this second messenger, i.e. https://journals.asm.org/doi/full/10.1128/jb.00604-15).<br /> There is also no explanation for the apparently contradictory upregulation of c-di-GMP in cells expressing higher PdeI levels. Perhaps the examination of the rest of the genes in cluster 2 of the biofilm sample could be useful to explain the observed association.

Reviewer #1 (Public Review):

Summary:

In this manuscript, Yan and colleagues introduce a modification to the previously published PETRI-seq bacterial single-cell protocol to include a ribosomal depletion step based on a DNA probe set that selectively hybridizes with ribosome-derived (rRNA) cDNA fragments. They show that their modification of the PETRI-seq protocol increases the fraction of informative non-rRNA reads from ~4-10% to 54-92%. The authors apply their protocol to investigating heterogeneity in a biofilm model of E. coli, and convincingly show how their technology can detect minority subpopulations within a complex community.

Strengths:

The method the authors propose is a straightforward and inexpensive modification of an established split-pool single-cell RNA-seq protocol that greatly increases its utility, and should be of interest to a wide community working in the field of bacterial single-cell RNA-seq.

Weaknesses:

The manuscript is written in a very compressed style and many technical details of the evaluations conducted are unclear and processed data has not been made available for evaluation, limiting the ability of the reader to independently judge the merits of the method.

eLife assessment

The work introduces a valuable new method for depleting the ribosomal RNA from bacterial single-cell RNA sequencing libraries and shows that this method is applicable to studying the heterogeneity in microbial biofilms. The evidence for a small subpopulation of cells at the bottom of the biofilm which upregulates PdeI expression is solid. However, more investigation into the unresolved functional relationship between PdeI and c-di-GMP levels with the help of other genes co-expressed in the same cluster would have made the conclusions more significant.

Author response:

Public Reviews:

Reviewer #1 (Public Review): 

[...] Strengths: 

The method the authors propose is a straightforward and inexpensive modification of an established split-pool single-cell RNA-seq protocol that greatly increases its utility, and should be of interest to a wide community working in the field of bacterial single-cell RNA-seq. 

Weaknesses: 

The manuscript is written in a very compressed style and many technical details of the evaluations conducted are unclear and processed data has not been made available for evaluation, limiting the ability of the reader to independently judge the merits of the method. 

Thank you for your thoughtful and constructive review of our manuscript. We appreciate your recognition of the strengths of our work and the potential impact of our modified PETRI-seq protocol on the field of bacterial single-cell RNA-seq. We are grateful for the opportunity to address your concerns and improve the clarity and accessibility of our manuscript.

We acknowledge your feedback regarding the compressed writing style and lack of technical details，which are constrained by the requirements of the Short Report format in eLife. We will addresse these issues in our revised manuscript as follows:

(1) Expanded methodology section: We will provide a more comprehensive description of our experimental procedures, including detailed protocols for the ribosomal depletion step and data analysis pipeline. This will enable readers to better understand and potentially replicate our methods.

(2) Clarification of technical evaluations: We will elaborate on the specifics of our evaluations, including the criteria used for assessing the efficiency of ribosomal depletion and the methods employed for identifying and characterizing subpopulations within the E. coli biofilm model.

(3) Data availability: We apologize for the oversight in not making our processed data readily available. We have deposited all relevant datasets, including raw and source data, in appropriate public repositories (GEO number: GSE260458) and provide clear instructions for accessing this data in the revised manuscript.

(4) Supplementary information: To maintain the concise nature of the main text while providing necessary details, we will inculde additional supplementary information. This will cover extended methodology, detailed statistical analyses, and comprehensive data tables to support our findings.

(5) Discussion of limitations: We will include a more thorough discussion of the potential limitations of our modified protocol and areas for future improvement.

​We believe these changes will significantly improve the clarity and reproducibility of our work, allowing readers to better evaluate the merits of our method.

Reviewer #2 (Public Review): 

[...] Strengths: 

The introduced rRNA depletion method is highly efficient, with the depletion for E.coli resulting in over 90% of reads containing mRNA. The method is ready to use with existing PETRI-seq libraries which is a large advantage, given that no other rRNA depletion methods were published for split-pool bacterial scRNA-seq methods. Therefore, the value of the method for the field is high. There is also evidence that a small number of cells at the bottom of a static biofilm express PdeI which is causing the elevated c-di-GMP levels that are associated with persister formation. Given that PdeI is a phosphodiesterase, which is supposed to promote hydrolysis of c-di-GMP, this finding is unexpected. 

Weaknesses: 

With the descriptions and writing of the manuscript, it is hard to place the findings about the PdeI into existing context (i.e. it is well known that c-di-GMP is involved in biofilm development and is heterogeneously distributed in several species' biofilms; it is also known that E.coli diesterases regulate this second messenger, i.e. https://journals.asm.org/doi/full/10.1128/jb.00604-15). <br /> There is also no explanation for the apparently contradictory upregulation of c-di-GMP in cells expressing higher PdeI levels. Perhaps the examination of the rest of the genes in cluster 2 of the biofilm sample could be useful to explain the observed association. 

Thank you for your thoughtful and constructive review of our manuscript. We are pleased that the reviewer recognizes the value and efficiency of our rRNA depletion method for PETRI-seq, as well as its potential impact on the field. We would like to address the points raised by the reviewer and provide additional context and clarification regarding the function of PdeI in c-di-GMP regulation.

We acknowledge that c-di-GMP’s role in biofilm development and its heterogeneous distribution in bacterial biofilms are well studied. We appreciate the reviewer's observation regarding the seemingly contradictory relationship between increased PdeI expression and elevated c-di-GMP levels. This is indeed an intriguing finding that warrants further explanation.

PdeI was predicted to be a phosphodiesterase responsible for c-di-GMP degradation. This prediction is based on sequence analysis where PdeI contains an intact EAL domain known for degrading c-di-GMP. However, it is noteworthy that PdeI also contains a divergent GGDEF domain, which is typically associated with c-di-GMP synthesis. This dual-domain architecture suggests a potential for complex regulatory roles. As reported, the knockout of the major phosphodiesterase PdeH in E. coli leads to the accumulation of c-di-GMP. Further, a point mutation on PdeI's divergent GGDEF domain (G412S) in this PdeH knockout strain resulted in decreased c-di-GMP levels, implying that the wild-type GGDEF domain in PdeI has a role in maintaining or increasing c-di-GMP levels in the cell. Additionally, PdeI contains a CHASE (cyclases/histidine kinase-associated sensory) domain. Combined with our experimental results demonstrating that PdeI is a membrane-associated protein, we predict that PdeI functions as a sensor that integrates environmental signals with c-di-GMP production under complex regulatory mechanisms. The experimental evidence, along with domain analysis, suggests that PdeI could contribute to c-di-GMP synthesis, rebutting the notion that it solely functions as a phosphodiesterase. Furthermore, our single-cell experiments showed a positive correlation between PdeI expression levels and c-di-GMP levels (Fig. 2J). HPLC LC-MS/MS analysis further confirmed that PdeI overexpression (induced by arabinose) upregulated c-di-GMP levels (Fig. 2K). Importantly, in our HPLC LC-MS/MS analysis, we compared the PdeI overexpression strain with the wild-type MG1655 strain, thereby excluding the influence of other genes in cluster 2. In summary, while PdeI is predicted to be a phosphodiesterase based on its sequence and the presence of an EAL domain, the additional presence of a divergent GGDEF domain and experimental evidence suggests that PdeI has a function in upregulating c-di-GMP levels. These findings support the hypothesis that PdeI may have both synthetic and regulatory roles in c-di-GMP metabolism.

eLife Assessment

The paper reports the important discovery that the mouse dorsal inferior colliculus, an auditory midbrain area, encodes sound location. The evidence supporting the claims is solid, being supported by both optical and electrophysiological recordings. The observations described should be of interest to auditory researchers studying the neural mechanisms of sound localization and the role of noise correlations in population coding.

Reviewer #1 (Public review):

Summary:

In this study, the authors address whether the dorsal nucleus of the inferior colliculus (DCIC) in mice encodes sound source location within the front horizontal plane (i.e., azimuth). They do this using volumetric two-photon Ca2+ imaging and high-density silicon probes (Neuropixels) to collect single-unit data. Such recordings are beneficial because they allow large populations of simultaneous neural data to be collected. Their main results and the claims about those results are the following:

(1) DCIC single-unit responses have high trial-to-trial variability (i.e., neural noise);<br /> (2) approximately 32% to 40% of DCIC single units have responses that are sensitive to sound source azimuth;<br /> (3) single-trial population responses (i.e., the joint response across all sampled single units in an animal) encode sound source azimuth "effectively" (as stated in the title) in that localization decoding error matches average mouse discrimination thresholds;<br /> (4) DCIC can encode sound source azimuth in a similar format to that in the central nucleus of the inferior colliculus (as stated in the Abstract);<br /> (5) evidence of noise correlation between pairs of neurons exists;<br /> and 6) noise correlations between responses of neurons help reduce population decoding error.<br /> While simultaneous recordings are not necessary to demonstrate results #1, #2, and #4, they are necessary to demonstrate results #3, #5, and #6.

Strengths:

- Important research question to all researchers interested in sensory coding in the nervous system.<br /> - State-of-the-art data collection: volumetric two-photon Ca2+ imaging and extracellular recording using high-density probes. Large neuronal data sets.<br /> - Confirmation of imaging results (lower temporal resolution) with more traditional microelectrode results (higher temporal resolution).<br /> - Clear and appropriate explanation of surgical and electrophysiological methods. I cannot comment on the appropriateness of the imaging methods.

Strength of evidence for the claims of the study:

(1) DCIC single-unit responses have high trial-to-trial variability -<br /> The authors' data clearly shows this.

(2) Approximately 32% to 40% of DCIC single units have responses that are sensitive to sound source azimuth -<br /> The sensitivity of each neuron's response to sound source azimuth was tested with a Kruskal-Wallis test, which is appropriate since response distributions were not normal. Using this statistical test, only 8% of neurons (median for imaging data) were found to be sensitive to azimuth, and the authors noted this was not significantly different than the false positive rate. The Kruskal-Wallis test was not reported for electrophysiological data. The authors suggested that low numbers of azimuth-sensitive units resulting from the statistical analysis may be due to the combination of high neural noise and a relatively low number of trials, which would reduce the statistical power of the test. This is likely true and highlights a weakness in the experimental design (i.e., a relatively small number of trials). The authors went on to perform a second test of azimuth sensitivity-a chi-squared test-and found 32% (imaging) and 40% (e-phys) of single units to have statistically significant sensitivity. However, the use of a chi-squared test is questionable because it is meant to be used between two categorical variables, and neural response had to be binned before applying the test.

(3) Single-trial population responses encode sound source azimuth "effectively" in that localization decoding error matches average mouse discrimination thresholds -<br /> If only one neuron in a population had responses that were sensitive to azimuth, we would expect that decoding azimuth from observation of that one neuron's response would perform better than chance. By observing the responses of more than one neuron (if more than one were sensitive to azimuth), we would expect performance to increase. The authors found that decoding from the whole population response was no better than chance. They argue (reasonably) that this is because of overfitting of the decoder model-too few trials were used to fit too many parameters-and provide evidence from decoding combined with principal components analysis which suggests that overfitting is occurring. What is troubling is the performance of the decoder when using only a handful of "top-ranked" neurons (in terms of azimuth sensitivity) (Fig. 4F and G). Decoder performance seems to increase when going from one to two neurons, then decreases when going from two to three neurons, and doesn't get much better for more neurons than for one neuron alone. It seems likely there is more information about azimuth in the population response, but decoder performance is not able to capture it because spike count distributions in the decoder model are not being accurately estimated due to too few stimulus trials (14, on average). In other words, it seems likely that decoder performance is underestimating the ability of the DCIC population to encode sound source azimuth.

To get a sense of how effective a neural population is at coding a particular stimulus parameter, it is useful to compare population decoder performance to psychophysical performance. Unfortunately, mouse behavioral localization data do not exist. Instead, the authors compare decoder error to mouse left-right discrimination thresholds published previously by a different lab. However, this comparison is inappropriate because the decoder and the mice were performing different perceptual tasks. The decoder is classifying sound sources to 1 of 13 locations from left to right, whereas the mice were discriminating between left or right sources centered around zero degrees. The errors in these two tasks represent different things. The two data sets may potentially be more accurately compared by extracting information from the confusion matrices of population decoder performance. For example, when the stimulus was at -30 deg, how often did the decoder classify the stimulus to a lefthand azimuth? Likewise, when the stimulus was +30 deg, how often did the decoder classify the stimulus to a righthand azimuth?

(4) DCIC can encode sound source azimuth in a similar format to that in the central nucleus of the inferior colliculus -<br /> It is unclear what exactly the authors mean by this statement in the Abstract. There are major differences in the encoding of azimuth between the two neighboring brain areas: a large majority of neurons in the CNIC are sensitive to azimuth (and strongly so), whereas the present study shows a minority of azimuth-sensitive neurons in the DCIC. Furthermore, CNIC neurons fire reliably to sound stimuli (low neural noise), whereas the present study shows that DCIC neurons fire more erratically (high neural noise).

(5) Evidence of noise correlation between pairs of neurons exists -<br /> The authors' data and analyses seem appropriate and sufficient to justify this claim.

(6) Noise correlations between responses of neurons help reduce population decoding error -<br /> The authors show convincing analysis that performance of their decoder increased when simultaneously measured responses were tested (which include noise correlation) than when scrambled-trial responses were tested (eliminating noise correlation). This makes it seem likely that noise correlation in the responses improved decoder performance. The authors mention that the naïve Bayesian classifier was used as their decoder for computational efficiency, presumably because it assumes no noise correlation and, therefore, assumes responses of individual neurons are independent of each other across trials to the same stimulus. The use of a decoder that assumes independence seems key here in testing the hypothesis that noise correlation contains information about sound source azimuth. The logic of using this decoder could be more clearly spelled out to the reader. For example, if the null hypothesis is that noise correlations do not carry azimuth information, then a decoder that assumes independence should perform the same whether population responses are simultaneous or scrambled. The authors' analysis showing a difference in performance between these two cases provides evidence against this null hypothesis.

Minor weakness:<br /> - Most studies of neural encoding of sound source azimuth are done in a noise-free environment, but the experimental setup in the present study had substantial background noise. This complicates comparison of the azimuth tuning results in this study to those of other studies. One is left wondering if azimuth sensitivity would have been greater in the absence of background noise, particularly for the imaging data where the signal was only about 12 dB above the noise.

Reviewer #2 (Public review):

In the present study, Boffi et al. investigate the manner in which the dorsal cortex of the of the inferior colliculus (DCIC), an auditory midbrain area, encodes sound location azimuth in awake, passively listening mice. By employing volumetric calcium imaging (scanned temporal focusing or s-TeFo), complemented with high-density electrode electrophysiological recordings (neuropixels probes), they show that sound-evoked responses are exquisitely noisy, with only a small portion of neurons (units) exhibiting spatial sensitivity. Nevertheless, a naïve Bayesian classifier was able to predict the presented azimuth based on the responses from small populations of these spatially sensitive units. A portion of the spatial information was provided by correlated trial-to-trial response variability between individual units (noise correlations). The study presents a novel characterization of spatial auditory coding in a non-canonical structure, representing a noteworthy contribution specifically to the auditory field and generally to systems neuroscience, due to its implementation of state-of-the-art techniques in an experimentally challenging brain region. However, nuances in the calcium imaging dataset and the naïve Bayesian classifier warrant caution when interpreting some of the results.

Strengths:

The primary strength of the study lies in its methodological achievements, which allowed the authors to collect a comprehensive and novel dataset. While the DCIC is a dorsal structure, it extends up to a millimetre in depth, making it optically challenging to access in its entirety. It is also more highly myelinated and vascularised compared to e.g., the cerebral cortex, compounding the problem. The authors successfully overcame these challenges and present an impressive volumetric calcium imaging dataset. Furthermore, they corroborated this dataset with electrophysiological recordings, which produced overlapping results. This methodological combination ameliorates the natural concerns that arise from inferring neuronal activity from calcium signals alone, which are in essence an indirect measurement thereof.

Another strength of the study is its interdisciplinary relevance. For the auditory field, it represents a significant contribution to the question of how auditory space is represented in the mammalian brain. "Space" per se is not mapped onto the basilar membrane of the cochlea and must be computed entirely within the brain. For azimuth, this requires the comparison between miniscule differences between the timing and intensity of sounds arriving at each ear. It is now generally thought that azimuth is initially encoded in two, opposing hemispheric channels, but the extent to which this initial arrangement is maintained throughout the auditory system remains an open question. The authors observe only a slight contralateral bias in their data, suggesting that sound source azimuth in the DCIC is encoded in a more nuanced manner compared to earlier processing stages of the auditory hindbrain. This is interesting because it is also known to be an auditory structure to receive more descending inputs from the cortex.

Systems neuroscience continues to strive for the perfection of imaging novel, less accessible brain regions. Volumetric calcium imaging is a promising emerging technique, allowing the simultaneous measurement of large populations of neurons in three dimensions. But this necessitates corroboration with other methods, such as electrophysiological recordings, which the authors achieve. The dataset moreover highlights the distinctive characteristics of neuronal auditory representations in the brain. Its signals can be exceptionally sparse and noisy, which provide an additional layer of complexity in the processing and analysis of such datasets. This will undoubtedly be useful for future studies of other less accessible structures with sparse responsiveness.

Weaknesses:

Although the primary finding that small populations of neurons carry enough spatial information for a naïve Bayesian classifier to reasonably decode the presented stimulus is not called into question, certain idiosyncrasies, in particular the calcium imaging dataset and model, complicate specific interpretations of the model output, and the readership is urged to interpret these aspects of the study's conclusions with caution.

I remain in favour of volumetric calcium imaging as a suitable technique for the study, but the presently constrained spatial resolution is insufficient to unequivocally identify regions of interest as cell bodies (and are instead referred to as "units" akin to those of electrophysiological recordings). It remains possible that the imaging set is inadvertently influenced by non-somatic structures (including neuropil), which could report neuronal activity differently than cell bodies. Due to the lack of a comprehensive ground-truth comparison in this regard (which to my knowledge is impossible to achieve with current technology), it is difficult to imagine how many informative such units might have been missed because their signals were influenced by spurious, non-somatic signals, which could have subsequently misled the models. The authors reference the original Nature Methods article (Prevedel et al., 2016) throughout the manuscript, presumably in order to avoid having to repeat previously published experimental metrics. But the DCIC is neither the cortex nor hippocampus (for which the method was originally developed) and may not have the same light scattering properties (not to mention neuronal noise levels). Although the corroborative electrophysiology data largely alleviates these concerns for this particular study, the readership should be cognisant of such caveats, in particular those who are interested in implementing the technique for their own research.

A related technical limitation of the calcium imaging dataset is the relatively low number of trials (14) given the inherently high level of noise (both neuronal and imaging). Volumetric calcium imaging, while offering a uniquely expansive field of view, requires relatively high average excitation laser power (in this case nearly 200 mW), a level of exposure the authors may have wanted to minimise by maintaining a low number of repetitions, but I yield to them to explain. Calcium imaging is also inherently slow, requiring relatively long inter-stimulus intervals (in this case 5 s). This unfortunately renders any model designed to predict a stimulus (in this case sound azimuth) from particularly noisy population neuronal data like these as highly prone to overfitting, to which the authors correctly admit after a model trained on the entire raw dataset failed to perform significantly above chance level. This prompted them to feed the model only with data from neurons with the highest spatial sensitivity. This ultimately produced reasonable performance (and was implemented throughout the rest of the study), but it remains possible that if the model was fed with more repetitions of imaging data, its performance would have been more stable across the number of units used to train it. (All models trained with imaging data eventually failed to converge.) However, I also see these limitations as an opportunity to improve the technology further, which I reiterate will be generally important for volume imaging of other sparse or noisy calcium signals in the brain.

Transitioning to the naïve Bayesian classifier itself, I first openly ask the authors to justify their choice of this specific model. There are countless types of classifiers for these data, each with their own pros and cons. Did they actually try other models (such as support vector machines), which ultimately failed? If so, these negative results (even if mentioned en passant) would be extremely valuable to the community, in my view. I ask this specifically because different methods assume correspondingly different statistical properties of the input data, and to my knowledge naïve Bayesian classifiers assume that predictors (neuronal responses) are assumed to be independent within a class (azimuth). As the authors show that noise correlations are informative in predicting azimuth, I wonder why they chose a model that doesn't take advantage of these statistical regularities. It could be because of technical considerations (they mention computing efficiency), but I am left generally uncertain about the specific logic that was used to guide the authors through their analytical journey.

In a revised version of the manuscript, the authors indeed justify their choice of the naïve Bayesian classifier as a conservative approach (not taking into account noise correlations), which could only improve with other models (that do). They even tested various other commonly used models, such as support vector machines and k-nearest neighbours, to name a few, but do not report these efforts in the main manuscript. Interestingly, these models, which I supposed would perform better in fact did not overall - a finding that I have no way of interpreting but nevertheless find interesting.

That aside, there remain other peculiarities in model performance that warrant further investigation. For example, what spurious features (or lack of informative features) in these additional units prevented the models of imaging data from converging? In an orthogonal question, did the most spatially sensitive units share any detectable tuning features? A different model trained with electrophysiology data in contrast did not collapse in the range of top-ranked units plotted. Did this model collapse at some point after adding enough units, and how well did that correlate with the model for the imaging data? How well did the form (and diversity) of the spatial tuning functions as recorded with electrophysiology resemble their calcium imaging counterparts? These fundamental questions could be addressed with more basic, but transparent analyses of the data (e.g., the diversity of spatial tuning functions of their recorded units across the population). Even if the model extracts features that are not obvious to the human eye in traditional visualisations, I would still find this interesting.

Although these questions were not specifically addressed in the revised version of the manuscript, I also admit that I did not indent do assert that these should necessarily fall within the scope of the present study. I rather posed them as hypothetical directions one could pursue in future studies. Finally, further concerns I had with statements regarding the physiological meaning of the findings have been ameliorated by nicely modified statements, thus bringing transparency to the readership, which I appreciate.

In summary, the present study represents a significant body of work that contributes substantially to the field of spatial auditory coding and systems neuroscience. However, limitations of the imaging dataset and model as applied in the study muddles concrete conclusions about how the DCIC precisely encodes sound source azimuth and even more so to sound localisation in a behaving animal. Nevertheless, it presents a novel and unique dataset, which, regardless of secondary interpretation, corroborates the general notion that auditory space is encoded in an extraordinarily complex manner in the mammalian brain.

Reviewer #3 (Public review):

Summary:

Boffi and colleagues sought to quantify the single-trial, azimuthal information in the dorsal cortex of the inferior colliculus (DCIC), a relatively understudied subnucleus of the auditory midbrain. They accomplished this by using two complementary recording methods while mice passively listened to sounds at different locations: calcium imaging that recorded large neuronal populations but with poor temporal precision and multi-contact electrode arrays that recorded smaller neuronal populations with exact temporal precision. DCIC neurons respond variably, with inconsistent activity to sound onset and complex azimuthal tuning. Some of this variably was explained by ongoing head movements. The authors used a naïve Bayes decoder to probe the azimuthal information contained in the response of DCIC neurons on single trials. The decoder failed to classify sound location better than chance when using the raw population responses but performed significantly better than chance when using the top principal components of the population. Units with the most azimuthal tuning were distributed throughout the DCIC, possessed contralateral bias, and positively correlated responses. Interestingly, inter-trial shuffling decreased decoding performance, indicating that noise correlations contributed to decoder performance. Overall, Boffi and colleagues, quantified the azimuthal information available in the DCIC while mice passively listened to sounds, a first step in evaluating if and how the DCIC could contribute to sound localization.

Strengths:

The authors should be commended for collection of this dataset. When done in isolation (which is typical), calcium imaging and linear array recordings have intrinsic weaknesses. However, those weaknesses are alleviated when done in conjunction - especially when the data is consistent. This data set is extremely rich and will be of use for those interested in auditory midbrain responses to variable sound locations, correlations with head movements, and neural coding.

The DCIC neural responses are complex with variable responses to sound onset, complex azimuthal tuning and large inter-sound interval responses. Nonetheless, the authors do a decent job in wrangling these complex responses: finding non-canonical ways of determining dependence on azimuth and using interpretable decoders to extract information from the population.

Weaknesses:

The decoding results are a bit strange, likely because the population response is quite noisy on any given trial. Raw population responses failed to provide sufficient information concerning azimuth for significant decoding. Importantly, the decoder performed better than chance when certain principal components or top ranked units contributed but did not saturate with the addition of components or top ranked units. So, although there is azimuthal information in the recorded DCIC populations - azimuthal information appears somewhat difficult to extract.

Although necessary given the challenges associated with sampling many conditions with technically difficult recording methods, the limited number of stimulus repeats precludes interpretable characterization of the heterogeneity across the population. Nevertheless, the dataset is public so those interested can explore the diversity of the responses.

The observations from Boffi and colleagues raises the question: what drives neurons in the DCIC to respond? Sound azimuth appears to be a small aspect of the DCIC response. For example, the first 20 principal components which explain roughly 80% of the response variance are insufficient input for the decoder to predict sound azimuth above chance. Furthermore, snout and ear movements correlate with the population response in the DCIC (the ear movements are particularly peculiar given they seem to predict sound presentation). Other movements may be of particular interest to control for (e.g. eye movements are known to interact with IC responses in the primate). These observations, along with reported variance to sound onsets and inter-sound intervals, question the impact of azimuthal information emerging from DCIC responses. This is certainly out of scope for any one singular study to answer, but, hopefully, future work will elucidate the dominant signals in the DCIC population. It may be intuitive that engagement in a sound localization task may push azimuthal signals to the forefront of DCIC response, but azimuthal information could also easily be overtaken by other signals (e.g. movement, learning).

Boffi and colleagues set out to parse the azimuthal information available in the DCIC on a single trial. They largely accomplish this goal and are able to extract this information when allowing the units that contain more information about sound location to contribute to their decoding (e.g., through PCA or decoding on their activity specifically). Interestingly, they also found that positive noise correlations between units with similar azimuthal preferences facilitate this decoding - which is unusual given that this is typically thought to limit information. The dataset will be of value to those interested in the DCIC and to anyone interested in the role of noise correlations in population coding. Although this work is first step into parsing the information available in the DCIC, it remains difficult to interpret if/how this azimuthal information is used in localization behaviors of engaged mice.

Author response:

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

Recommendations for the authors:

Reviewer #2 (Recommendations for the authors):

I appreciate the efforts the authors made to clarify and justify their statements and methodology, respectively. I additionally appreciate the efforts they made to provide me with detailed information - including figures - to aid my comprehension. However, there are two things I nevertheless recommend the authors to include in the main manuscript.

(1) Statement about animal wellbeing: The authors state that they were constrained in their imaging session duration not because of a commonly reported technical limitation, such as photobleaching (which I honestly assumed), but rather the general wellbeing of the animals, who exhibited signs of distress after longer imaging periods. I find this to be a critical issue and perhaps the best argument against performing longer imaging experiments (which would have increased the number of trials, thus potentially boosting the performance of their model). To say that they put animal welfare above all other scientific and technical considerations speaks to a strong ethical adherence to animal welfare policy, and I believe this should be somehow incorporated into the methods.

We have now included this at the top of page 26:

“Mice fully recovered from the brief isoflurane anesthesia, showing a clear blinking reflex, whisking and sniffing behaviors and normal body posture and movements, immediately after head fixation. In our experimental conditions, mice were imaged in sessions of up to 25 min since beyond this time we started observing some signs of distress or discomfort. Thus, we avoided longer recording times at the expense of collecting larger trial numbers, in strong adherence of animal welfare and ethics policy. A pilot group of mice were habituated to the head fixed condition in daily 20 min sessions for 3 days, however we did not observe a marked contrast in the behavior of habituated versus unhabituated mice beyond our relatively short 25 min imaging sessions. In consequence imaging sessions never surpassed a maximum of 25 min, after which the mouse was returned to its home cage.”

(2) Author response image 2: I sincerely thank the authors for providing us reviewers with this figure, which compares the performance of the naïve Bayesian classifier their ultimately use in the study with other commonly implemented models. Also here I falsely assumed that other models, which take correlated activity into account, did not generally perform better than their ultimate model of choice. Although dwelling on it would be distractive (and outside the primary scope of the study), I would encourage the authors to include it as a figure supplement (and simply mention these controls en passant when they justify their choice of the naïve Bayesian classifier).

This figure was now included in the revised manuscript as supplemental figure 3.

Page 10 now reads:

“We performed cross-validated, multi-class classification of the single-trial population responses (decoding, Fig. 2A) using a naive Bayes classifier to evaluate the prediction errors as the absolute difference between the stimulus azimuth and the predicted azimuth (Fig. 2A). We chose this classification algorithm over others due to its generally good performance with limited available data. We visualized the cross-validated prediction error distribution in cumulative plots where the observed prediction errors were compared to the distribution of errors for random azimuth sampling (Fig. 2B). When decoding all simultaneously recorded units, the observed classifier output was not significantly better (shifted towards smaller prediction errors) than the chance level distribution (Fig. 2B). The classifier also failed to decode complete DCIC population responses recorded with neuropixels probes (Fig. 3A). Other classifiers performed similarly (Suppl. Fig. 3A).”

The bottom paragraph in page 19 now reads:

“To characterize how the observed positive noise correlations could affect the representation of stimulus azimuth by DCIC top ranked unit population responses, we compared the decoding performance obtained by classifying the single-trial response patterns from top ranked units in the modeled decorrelated datasets versus the acquired data (with noise correlations). With the intention to characterize this with a conservative approach that would be less likely to find a contribution of noise correlations as it assumes response independence, we relied on the naive Bayes classifier for decoding throughout the study. Using this classifier, we observed that the modeled decorrelated datasets produced stimulus azimuth prediction error distributions that were significantly shifted towards higher decoding errors (Fig. 6B, C) and, in our imaging datasets, were not significantly different from chance level (Fig. 6B). Altogether, these results suggest that the detected noise correlations in our simultaneously acquired datasets can help reduce the error of the IC population code for sound azimuth. We observed a similar, but not significant tendency with another classifier that does not assume response independence (KNN classifier), though overall producing larger decoding errors than the Bayes classifier (Suppl. Fig. 3B).”

Reviewer #3 (Recommendations for the authors):

I am generally happy with the response to the reviews.

I find the Author response image 3 quite interesting. The neuropixel data looks somewhat like I expected (especially for mouse #3 and maybe mouse #4). I find the distribution of weights across units in the imaging dataset compared to in the pixel dataset intriguing (though it probably is just the dimensionality of the data being so much higher).

I'm not too familiar with facial movements but is it the case that the DCIC would be more modulated by ipsilateral movement compared to contralateral movements? Are face movements in mice conjugate or do both sides of the face move more or less independently? If not it may be interesting in future work to record bilaterally and see if that provides more information about DCIC responses.

We sincerely thank the editors and reviewers for their careful appraisal, commendation of our effort and helpful constructive feedback which greatly improved the presentation of our study. Below in green font is a point by point reply to the comments provided by the reviewers.

Public Reviews:

Reviewer #1 (Public Review):

Summary: In this study, the authors address whether the dorsal nucleus of the inferior colliculus (DCIC) in mice encodes sound source location within the front horizontal plane (i.e., azimuth). They do this using volumetric two-photon Ca2+ imaging and high-density silicon probes (Neuropixels) to collect single-unit data. Such recordings are beneficial because they allow large populations of simultaneous neural data to be collected. Their main results and the claims about those results are the following:

(1) DCIC single-unit responses have high trial-to-trial variability (i.e., neural noise);

(2) approximately 32% to 40% of DCIC single units have responses that are sensitive to sound source azimuth;

(3) single-trial population responses (i.e., the joint response across all sampled single units in an animal) encode sound source azimuth "effectively" (as stated in title) in that localization decoding error matches average mouse discrimination thresholds;

(4) DCIC can encode sound source azimuth in a similar format to that in the central nucleus of the inferior colliculus (as stated in Abstract);

(5) evidence of noise correlation between pairs of neurons exists;

and (6) noise correlations between responses of neurons help reduce population decoding error.

While simultaneous recordings are not necessary to demonstrate results #1, #2, and #4, they are necessary to demonstrate results #3, #5, and #6.

Strengths:

- Important research question to all researchers interested in sensory coding in the nervous system.

- State-of-the-art data collection: volumetric two-photon Ca2+ imaging and extracellular recording using high-density probes. Large neuronal data sets.

- Confirmation of imaging results (lower temporal resolution) with more traditional microelectrode results (higher temporal resolution).

- Clear and appropriate explanation of surgical and electrophysiological methods. I cannot comment on the appropriateness of the imaging methods.

Strength of evidence for claims of the study:

(1) DCIC single-unit responses have high trial-to-trial variability - The authors' data clearly shows this.

(2) Approximately 32% to 40% of DCIC single units have responses that are sensitive to sound source azimuth - The sensitivity of each neuron's response to sound source azimuth was tested with a Kruskal-Wallis test, which is appropriate since response distributions were not normal. Using this statistical test, only 8% of neurons (median for imaging data) were found to be sensitive to azimuth, and the authors noted this was not significantly different than the false positive rate. The Kruskal-Wallis test was not performed on electrophysiological data. The authors suggested that low numbers of azimuth-sensitive units resulting from the statistical analysis may be due to the combination of high neural noise and relatively low number of trials, which would reduce statistical power of the test. This may be true, but if single-unit responses were moderately or strongly sensitive to azimuth, one would expect them to pass the test even with relatively low statistical power. At best, if their statistical test missed some azimuthsensitive units, they were likely only weakly sensitive to azimuth. The authors went on to perform a second test of azimuth sensitivity-a chi-squared test-and found 32% (imaging) and 40% (e-phys) of single units to have statistically significant sensitivity. This feels a bit like fishing for a lower p-value. The Kruskal-Wallis test should have been left as the only analysis. Moreover, the use of a chi-squared test is questionable because it is meant to be used between two categorical variables, and neural response had to be binned before applying the test.

The determination of what is a physiologically relevant “moderate or strong azimuth sensitivity” is not trivial, particularly when comparing tuning across different relays of the auditory pathway like the CNIC, auditory cortex, or in our case DCIC, where physiologically relevant azimuth sensitivities might be different. This is likely the reason why azimuth sensitivity has been defined in diverse ways across the bibliography (see Groh, Kelly & Underhill, 2003 for an early discussion of this issue). These diverse approaches include reaching a certain percentage of maximal response modulation, like used by Day et al. (2012, 2015, 2016) in CNIC, and ANOVA tests, like used by Panniello et al. (2018) and Groh, Kelly & Underhill (2003) in auditory cortex and IC respectively. Moreover, the influence of response variability and biases in response distribution estimation due to limited sampling has not been usually accounted for in the determination of azimuth sensitivity.

As Reviewer #1 points out, in our study we used an appropriate ANOVA test (KruskalWallis) as a starting point to study response sensitivity to stimulus azimuth at DCIC. Please note that the alpha = 0.05 used for this test is not based on experimental evidence about physiologically relevant azimuth sensitivity but instead is an arbitrary p-value threshold. Using this test on the electrophysiological data, we found that ~ 21% of the simultaneously recorded single units reached significance (n = 4 mice). Nevertheless these percentages, in our small sample size (n = 4) were not significantly different from our false positive detection rate (p = 0.0625, Mann-Whitney, See Author response image 1).  In consequence, for both our imaging (Fig. 3C) and electrophysiological data, we could not ascertain if the percentage of neurons reaching significance in these ANOVA tests were indeed meaningfully sensitive to azimuth or this was due to chance.

Author response image 1.

Percentage of the neuropixels recorded DCIC single units across mice that showed significant median response tuning, compared to false positive detection rate (α = 0.05, chance level).

We reasoned that the observed markedly variable responses from DCIC units, which frequently failed to respond in many trials (Fig. 3D, 4A), in combination with the limited number of trial repetitions we could collect, results in under-sampled response distribution estimations. This under-sampling can bias the determination of stochastic dominance across azimuth response samples in Kruskal-Wallis tests. We would like to highlight that we decided not to implement resampling strategies to artificially increase the azimuth response sample sizes with “virtual trials”, in order to avoid “fishing for a smaller p-value”, when our collected samples might not accurately reflect the actual response population variability.

As an alternative to hypothesis testing based on ranking and determining stochastic dominance of one or more azimuth response samples (Kruskal-Wallis test), we evaluated the overall statistical dependency to stimulus azimuth of the collected responses.  To do this we implement the Chi-square test by binning neuronal responses into categories. Binning responses into categories can reduce the influence of response variability to some extent, which constitutes an advantage of the Chi-square approach, but we note the important consideration that these response categories are arbitrary.

Altogether, we acknowledge that our Chi-square approach to define azimuth sensitivity is not free of limitations and despite enabling the interrogation of azimuth sensitivity at DCIC, its interpretability might not extend to other brain regions like CNIC or auditory cortex. Nevertheless we hope the aforementioned arguments justify why the Kruskal-Wallis test simply could not “have been left as the only analysis”.

(3) Single-trial population responses encode sound source azimuth "effectively" in that localization decoding error matches average mouse discrimination thresholds - If only one neuron in a population had responses that were sensitive to azimuth, we would expect that decoding azimuth from observation of that one neuron's response would perform better than chance. By observing the responses of more than one neuron (if more than one were sensitive to azimuth), we would expect performance to increase. The authors found that decoding from the whole population response was no better than chance. They argue (reasonably) that this is because of overfitting of the decoder modeltoo few trials used to fit too many parameters-and provide evidence from decoding combined with principal components analysis which suggests that overfitting is occurring. What is troubling is the performance of the decoder when using only a handful of "topranked" neurons (in terms of azimuth sensitivity) (Fig. 4F and G). Decoder performance seems to increase when going from one to two neurons, then decreases when going from two to three neurons, and doesn't get much better for more neurons than for one neuron alone. It seems likely there is more information about azimuth in the population response, but decoder performance is not able to capture it because spike count distributions in the decoder model are not being accurately estimated due to too few stimulus trials (14, on average). In other words, it seems likely that decoder performance is underestimating the ability of the DCIC population to encode sound source azimuth.

To get a sense of how effective a neural population is at coding a particular stimulus parameter, it is useful to compare population decoder performance to psychophysical performance. Unfortunately, mouse behavioral localization data do not exist. Therefore, the authors compare decoder error to mouse left-right discrimination thresholds published previously by a different lab. However, this comparison is inappropriate because the decoder and the mice were performing different perceptual tasks. The decoder is classifying sound sources to 1 of 13 locations from left to right, whereas the mice were discriminating between left or right sources centered around zero degrees. The errors in these two tasks represent different things. The two data sets may potentially be more accurately compared by extracting information from the confusion matrices of population decoder performance. For example, when the stimulus was at -30 deg, how often did the decoder classify the stimulus to a lefthand azimuth? Likewise, when the stimulus was +30 deg, how often did the decoder classify the stimulus to a righthand azimuth?

The azimuth discrimination error reported by Lauer et al. (2011) comes from engaged and highly trained mice, which is a very different context to our experimental setting with untrained mice passively listening to stimuli from 13 random azimuths. Therefore we did not perform analyses or interpretations of our results based on the behavioral task from Lauer et al. (2011) and only made the qualitative observation that the errors match for discussion.

We believe it is further important to clarify that Lauer et al. (2011) tested the ability of mice to discriminate between a positively conditioned stimulus (reference speaker at 0º center azimuth associated to a liquid reward) and a negatively conditioned stimulus (coming from one of five comparison speakers positioned at 20º, 30º, 50º, 70 and 90º azimuth, associated to an electrified lickport) in a conditioned avoidance task. In this task, mice are not precisely “discriminating between left or right sources centered around zero degrees”, making further analyses to compare the experimental design of Lauer et al (2011) and ours even more challenging for valid interpretation.

(4) DCIC can encode sound source azimuth in a similar format to that in the central nucleus of the inferior colliculus - It is unclear what exactly the authors mean by this statement in the Abstract. There are major differences in the encoding of azimuth between the two neighboring brain areas: a large majority of neurons in the CNIC are sensitive to azimuth (and strongly so), whereas the present study shows a minority of azimuth-sensitive neurons in the DCIC. Furthermore, CNIC neurons fire reliably to sound stimuli (low neural noise), whereas the present study shows that DCIC neurons fire more erratically (high neural noise).

Since sound source azimuth is reported to be encoded by population activity patterns at CNIC (Day and Delgutte, 2013), we refer to a population activity pattern code as the “similar format” in which this information is encoded at DCIC. Please note that this is a qualitative comparison and we do not claim this is the “same format”, due to the differences the reviewer precisely describes in the encoding of azimuth at CNIC where a much larger majority of neurons show stronger azimuth sensitivity and response reliability with respect to our observations at DCIC. By this qualitative similarity of encoding format we specifically mean the similar occurrence of activity patterns from azimuth sensitive subpopulations of neurons in both CNIC and DCIC, which carry sufficient information about the stimulus azimuth for a sufficiently accurate prediction with regard to the behavioral discrimination ability.

(5) Evidence of noise correlation between pairs of neurons exists - The authors' data and analyses seem appropriate and sufficient to justify this claim.

(6) Noise correlations between responses of neurons help reduce population decoding error - The authors show convincing analysis that performance of their decoder increased when simultaneously measured responses were tested (which include noise correlation) than when scrambled-trial responses were tested (eliminating noise correlation). This makes it seem likely that noise correlation in the responses improved decoder performance. The authors mention that the naïve Bayesian classifier was used as their decoder for computational efficiency, presumably because it assumes no noise correlation and, therefore, assumes responses of individual neurons are independent of each other across trials to the same stimulus. The use of decoder that assumes independence seems key here in testing the hypothesis that noise correlation contains information about sound source azimuth. The logic of using this decoder could be more clearly spelled out to the reader. For example, if the null hypothesis is that noise correlations do not carry azimuth information, then a decoder that assumes independence should perform the same whether population responses are simultaneous or scrambled. The authors' analysis showing a difference in performance between these two cases provides evidence against this null hypothesis.

We sincerely thank the reviewer for this careful and detailed consideration of our analysis approach. Following the reviewer’s constructive suggestion, we justified the decoder choice in the results section at the last paragraph of page 18:

“To characterize how the observed positive noise correlations could affect the representation of stimulus azimuth by DCIC top ranked unit population responses, we compared the decoding performance obtained by classifying the single-trial response patterns from top ranked units in the modeled decorrelated datasets versus the acquired data (with noise correlations). With the intention to characterize this with a conservative approach that would be less likely to find a contribution of noise correlations as it assumes response independence, we relied on the naive Bayes classifier for decoding throughout the study.

Using this classifier, we observed that the modeled decorrelated datasets produced stimulus azimuth prediction error distributions that were significantly shifted towards higher decoding errors (Fig. 5B, C) and, in our imaging datasets, were not significantly different from chance level (Fig. 5B). Altogether, these results suggest that the detected noise correlations in our simultaneously acquired datasets can help reduce the error of the IC population code for sound azimuth.”

Minor weakness:

- Most studies of neural encoding of sound source azimuth are done in a noise-free environment, but the experimental setup in the present study had substantial background noise. This complicates comparison of the azimuth tuning results in this study to those of other studies. One is left wondering if azimuth sensitivity would have been greater in the absence of background noise, particularly for the imaging data where the signal was only about 12 dB above the noise. The description of the noise level and signal + noise level in the Methods should be made clearer. Mice hear from about 2.5 - 80 kHz, so it is important to know the noise level within this band as well as specifically within the band overlapping with the signal.

We agree with the reviewer that this information is useful. In our study, the background R.M.S. SPL during imaging across the mouse hearing range (2.5-80kHz) was 44.53 dB and for neuropixels recordings 34.68 dB. We have added this information to the methods section of the revised manuscript.

Reviewer #2 (Public Review):

In the present study, Boffi et al. investigate the manner in which the dorsal cortex of the of the inferior colliculus (DCIC), an auditory midbrain area, encodes sound location azimuth in awake, passively listening mice. By employing volumetric calcium imaging (scanned temporal focusing or s-TeFo), complemented with high-density electrode electrophysiological recordings (neuropixels probes), they show that sound-evoked responses are exquisitely noisy, with only a small portion of neurons (units) exhibiting spatial sensitivity. Nevertheless, a naïve Bayesian classifier was able to predict the presented azimuth based on the responses from small populations of these spatially sensitive units. A portion of the spatial information was provided by correlated trial-to-trial response variability between individual units (noise correlations). The study presents a novel characterization of spatial auditory coding in a non-canonical structure, representing a noteworthy contribution specifically to the auditory field and generally to systems neuroscience, due to its implementation of state-of-the-art techniques in an experimentally challenging brain region. However, nuances in the calcium imaging dataset and the naïve Bayesian classifier warrant caution when interpreting some of the results.

Strengths:

The primary strength of the study lies in its methodological achievements, which allowed the authors to collect a comprehensive and novel dataset. While the DCIC is a dorsal structure, it extends up to a millimetre in depth, making it optically challenging to access in its entirety. It is also more highly myelinated and vascularised compared to e.g., the cerebral cortex, compounding the problem. The authors successfully overcame these challenges and present an impressive volumetric calcium imaging dataset. Furthermore, they corroborated this dataset with electrophysiological recordings, which produced overlapping results. This methodological combination ameliorates the natural concerns that arise from inferring neuronal activity from calcium signals alone, which are in essence an indirect measurement thereof.

Another strength of the study is its interdisciplinary relevance. For the auditory field, it represents a significant contribution to the question of how auditory space is represented in the mammalian brain. "Space" per se is not mapped onto the basilar membrane of the cochlea and must be computed entirely within the brain. For azimuth, this requires the comparison between miniscule differences between the timing and intensity of sounds arriving at each ear. It is now generally thought that azimuth is initially encoded in two, opposing hemispheric channels, but the extent to which this initial arrangement is maintained throughout the auditory system remains an open question. The authors observe only a slight contralateral bias in their data, suggesting that sound source azimuth in the DCIC is encoded in a more nuanced manner compared to earlier processing stages of the auditory hindbrain. This is interesting, because it is also known to be an auditory structure to receive more descending inputs from the cortex.

Systems neuroscience continues to strive for the perfection of imaging novel, less accessible brain regions. Volumetric calcium imaging is a promising emerging technique, allowing the simultaneous measurement of large populations of neurons in three dimensions. But this necessitates corroboration with other methods, such as electrophysiological recordings, which the authors achieve. The dataset moreover highlights the distinctive characteristics of neuronal auditory representations in the brain. Its signals can be exceptionally sparse and noisy, which provide an additional layer of complexity in the processing and analysis of such datasets. This will be undoubtedly useful for future studies of other less accessible structures with sparse responsiveness.

Weaknesses:                                                                                               

Although the primary finding that small populations of neurons carry enough spatial information for a naïve Bayesian classifier to reasonably decode the presented stimulus is not called into question, certain idiosyncrasies, in particular the calcium imaging dataset and model, complicate specific interpretations of the model output, and the readership is urged to interpret these aspects of the study's conclusions with caution.

I remain in favour of volumetric calcium imaging as a suitable technique for the study, but the presently constrained spatial resolution is insufficient to unequivocally identify regions of interest as cell bodies (and are instead referred to as "units" akin to those of electrophysiological recordings). It remains possible that the imaging set is inadvertently influenced by non-somatic structures (including neuropil), which could report neuronal activity differently than cell bodies. Due to the lack of a comprehensive ground-truth comparison in this regard (which to my knowledge is impossible to achieve with current technology), it is difficult to imagine how many informative such units might have been missed because their signals were influenced by spurious, non-somatic signals, which could have subsequently misled the models. The authors reference the original Nature Methods article (Prevedel et al., 2016) throughout the manuscript, presumably in order to avoid having to repeat previously published experimental metrics. But the DCIC is neither the cortex nor hippocampus (for which the method was originally developed) and may not have the same light scattering properties (not to mention neuronal noise levels). Although the corroborative electrophysiology data largely eleviates these concerns for this particular study, the readership should be cognisant of such caveats, in particular those who are interested in implementing the technique for their own research.

A related technical limitation of the calcium imaging dataset is the relatively low number of trials (14) given the inherently high level of noise (both neuronal and imaging). Volumetric calcium imaging, while offering a uniquely expansive field of view, requires relatively high average excitation laser power (in this case nearly 200 mW), a level of exposure the authors may have wanted to minimise by maintaining a low the number of repetitions, but I yield to them to explain.

We assumed that the levels of heating by excitation light measured at the neocortex in Prevedel et al. (2016), were representative for DCIC also. Nevertheless, we recognize this approximation might not be very accurate, due to the differences in tissue architecture and vascularization from these two brain areas, just to name a few factors. The limiting factor preventing us from collecting more trials in our imaging sessions was that we observed signs of discomfort or slight distress in some mice after ~30 min of imaging in our custom setup, which we established as a humane end point to prevent distress. In consequence imaging sessions were kept to 25 min in duration, limiting the number of trials collected. However we cannot rule out that with more extensive habituation prior to experiments the imaging sessions could be prolonged without these signs of discomfort or if indeed influence from our custom setup like potential heating of the brain by illumination light might be the causing factor of the observed distress. Nevertheless, we note that previous work has shown that ~200mW average power is a safe regime for imaging in the cortex by keeping brain heating minimal (Prevedel et al., 2016), without producing the lasting damages observed by immunohistochemisty against apoptosis markers above 250mW (Podgorski and Ranganathan 2016, https://doi.org/10.1152/jn.00275.2016).

Calcium imaging is also inherently slow, requiring relatively long inter-stimulus intervals (in this case 5 s). This unfortunately renders any model designed to predict a stimulus (in this case sound azimuth) from particularly noisy population neuronal data like these as highly prone to overfitting, to which the authors correctly admit after a model trained on the entire raw dataset failed to perform significantly above chance level. This prompted them to feed the model only with data from neurons with the highest spatial sensitivity. This ultimately produced reasonable performance (and was implemented throughout the rest of the study), but it remains possible that if the model was fed with more repetitions of imaging data, its performance would have been more stable across the number of units used to train it. (All models trained with imaging data eventually failed to converge.) However, I also see these limitations as an opportunity to improve the technology further, which I reiterate will be generally important for volume imaging of other sparse or noisy calcium signals in the brain.

Transitioning to the naïve Bayesian classifier itself, I first openly ask the authors to justify their choice of this specific model. There are countless types of classifiers for these data, each with their own pros and cons. Did they actually try other models (such as support vector machines), which ultimately failed? If so, these negative results (even if mentioned en passant) would be extremely valuable to the community, in my view. I ask this specifically because different methods assume correspondingly different statistical properties of the input data, and to my knowledge naïve Bayesian classifiers assume that predictors (neuronal responses) are assumed to be independent within a class (azimuth). As the authors show that noise correlations are informative in predicting azimuth, I wonder why they chose a model that doesn't take advantage of these statistical regularities. It could be because of technical considerations (they mention computing efficiency), but I am left generally uncertain about the specific logic that was used to guide the authors through their analytical journey.

One of the main reasons we chose the naïve Bayesian classifier is indeed because it assumes that the responses of the simultaneously recorded neurons are independent and therefore it does not assume a contribution of noise correlations to the estimation of the posterior probability of each azimuth. This model would represent the null hypothesis that noise correlations do not contribute to the encoding of stimulus azimuth, which would be verified by an equal decoding outcome from correlated or decorrelated datasets. Since we observed that this is not the case, the model supports the alternative hypothesis that noise correlations do indeed influence stimulus azimuth encoding. We wanted to test these hypotheses with the most conservative approach possible that would be least likely to find a contribution of noise correlations. Other relevant reasons that justify our choice of the naive Bayesian classifier are its robustness against the limited numbers of trials we could collect in comparison to other more “data hungry” classifiers like SVM, KNN, or artificial neuronal nets. We did perform preliminary tests with alternative classifiers but the obtained decoding errors were similar when decoding the whole population activity (Supplemental figure 3A). Dimensionality reduction following the approach described in the manuscript showed a tendency towards smaller decoding errors observed with an alternative classifier like KNN, but these errors were still larger than the ones observed with the naive Bayesian classifier (median error 45º). Nevertheless, we also observe a similar tendency for slightly larger decoding errors in the absence of noise correlations (decorrelated, Supplemental figure 3B). Sentences detailing the logic of classifier choice are now included in the results section at page 10 and at the last paragraph of page 18 (see responses to Reviewer 1).

That aside, there remain other peculiarities in model performance that warrant further investigation. For example, what spurious features (or lack of informative features) in these additional units prevented the models of imaging data from converging?

Considering the amount of variability observed throughout the neuronal responses both in imaging and neuropixels datasets, it is easy to suspect that the information about stimulus azimuth carried in different amounts by individual DCIC neurons can be mixed up with information about other factors (Stringer et al., 2019). In an attempt to study the origin of these features that could confound stimulus azimuth decoding we explored their relation to face movement (Supplemental Figure 2), finding a correlation to snout movements, in line with previous work by Stringer et al. (2019).

In an orthogonal question, did the most spatially sensitive units share any detectable tuning features? A different model trained with electrophysiology data in contrast did not collapse in the range of top-ranked units plotted. Did this model collapse at some point after adding enough units, and how well did that correlate with the model for the imaging data?

Our electrophysiology datasets were much smaller in size (number of simultaneously recorded neurons) compared to our volumetric calcium imaging datasets, resulting in a much smaller total number of top ranked units detected per dataset. This precluded the determination of a collapse of decoder performance due to overfitting beyond the range plotted in Fig 4G.

How well did the form (and diversity) of the spatial tuning functions as recorded with electrophysiology resemble their calcium imaging counterparts? These fundamental questions could be addressed with more basic, but transparent analyses of the data (e.g., the diversity of spatial tuning functions of their recorded units across the population). Even if the model extracts features that are not obvious to the human eye in traditional visualisations, I would still find this interesting.

The diversity of the azimuth tuning curves recorded with calcium imaging (Fig. 3B) was qualitatively larger than the ones recorded with electrophysiology (Fig. 4B), potentially due to the larger sampling obtained with volumetric imaging. We did not perform a detailed comparison of the form and a more quantitative comparison of the diversity of these functions because the signals compared are quite different, as calcium indicator signal is subject to non linearities due to Ca2+ binding cooperativity and low pass filtering due to binding kinetics. We feared this could lead to misleading interpretations about the similarities or differences between the azimuth tuning functions in imaged and electrophysiology datasets. Our model uses statistical response dependency to stimulus azimuth, which does not rely on features from a descriptive statistic like mean response tuning. In this context, visualizing the trial-to-trial responses as a function of azimuth shows “features that are not obvious to the human eye in traditional visualizations” (Fig. 3D, left inset).

Finally, the readership is encouraged to interpret certain statements by the authors in the current version conservatively. How the brain ultimately extracts spatial neuronal data for perception is anyone's guess, but it is important to remember that this study only shows that a naïve Bayesian classifier could decode this information, and it remains entirely unclear whether the brain does this as well. For example, the model is able to achieve a prediction error that corresponds to the psychophysical threshold in mice performing a discrimination task (~30 {degree sign}). Although this is an interesting coincidental observation, it does not mean that the two metrics are necessarily related. The authors correctly do not explicitly claim this, but the manner in which the prose flows may lead a non-expert into drawing that conclusion.

To avoid misleading the non-expert readers, we have clarified in the manuscript that the observed correspondence between decoding error and psychophysical threshold is explicitly coincidental.

Page 13, end of middle paragraph:

“If we consider the median of the prediction error distribution as an overall measure of decoding performance, the single-trial response patterns from subsamples of at least the 7 top ranked units produced median decoding errors that coincidentally matched the reported azimuth discrimination ability of mice (Fig 4G, minimum audible angle = 31º) (Lauer et al., 2011).”

Page 14, bottom paragraph:

“Decoding analysis (Fig. 4F) of the population response patterns from azimuth dependent top ranked units simultaneously recorded with neuropixels probes showed that the 4 top ranked units are the smallest subsample necessary to produce a significant decoding performance that coincidentally matches the discrimination ability of mice (31° (Lauer et al., 2011)) (Fig. 5F, G).”

We also added to the Discussion sentences clarifying that a relationship between these two variables remains to be determined and it also remains to be determined if the DCIC indeed performs a bayesian decoding computation for sound localization.

Page 20, bottom:

“… Concretely, we show that sound location coding does indeed occur at DCIC on the single trial basis, and that this follows a comparable mechanism to the characterized population code at CNIC (Day and Delgutte, 2013). However, it remains to be determined if indeed the DCIC network is physiologically capable of Bayesian decoding computations. Interestingly, the small number of DCIC top ranked units necessary to effectively decode stimulus azimuth suggests that sound azimuth information is redundantly distributed across DCIC top ranked units, which points out that mechanisms beyond coding efficiency could be relevant for this population code.

While the decoding error observed from our DCIC datasets obtained in passively listening, untrained mice coincidentally matches the discrimination ability of highly trained, motivated mice (Lauer et al., 2011), a relationship between decoding error and psychophysical performance remains to be determined. Interestingly, a primary sensory representations should theoretically be even more precise than the behavioral performance as reported in the visual system (Stringer et al., 2021).”

Moreover, the concept of redundancy (of spatial information carried by units throughout the DCIC) is difficult for me to disentangle. One interpretation of this formulation could be that there are non-overlapping populations of neurons distributed across the DCIC that each could predict azimuth independently of each other, which is unlikely what the authors meant. If the authors meant generally that multiple neurons in the DCIC carry sufficient spatial information, then a single neuron would have been able to predict sound source azimuth, which was not the case. I have the feeling that they actually mean "complimentary", but I leave it to the authors to clarify my confusion, should they wish.

We observed that the response patterns from relatively small fractions of the azimuth sensitive DCIC units (4-7 top ranked units) are sufficient to generate an effective code for sound azimuth, while 32-40% of all simultaneously recorded DCIC units are azimuth sensitive. In light of this observation, we interpreted that the azimuth information carried by the population should be redundantly distributed across the complete subpopulation of azimuth sensitive DCIC units.

In summary, the present study represents a significant body of work that contributes substantially to the field of spatial auditory coding and systems neuroscience. However, limitations of the imaging dataset and model as applied in the study muddles concrete conclusions about how the DCIC precisely encodes sound source azimuth and even more so to sound localisation in a behaving animal. Nevertheless, it presents a novel and unique dataset, which, regardless of secondary interpretation, corroborates the general notion that auditory space is encoded in an extraordinarily complex manner in the mammalian brain.

Reviewer #3 (Public Review):

Summary: Boffi and colleagues sought to quantify the single-trial, azimuthal information in the dorsal cortex of the inferior colliculus (DCIC), a relatively understudied subnucleus of the auditory midbrain. They used two complementary recording methods while mice passively listened to sounds at different locations: a large volume but slow sampling calcium-imaging method, and a smaller volume but temporally precise electrophysiology method. They found that neurons in the DCIC were variable in their activity, unreliably responding to sound presentation and responding during inter-sound intervals. Boffi and colleagues used a naïve Bayesian decoder to determine if the DCIC population encoded sound location on a single trial. The decoder failed to classify sound location better than chance when using the raw single-trial population response but performed significantly better than chance when using intermediate principal components of the population response. In line with this, when the most azimuth dependent neurons were used to decode azimuthal position, the decoder performed equivalently to the azimuthal localization abilities of mice. The top azimuthal units were not clustered in the DCIC, possessed a contralateral bias in response, and were correlated in their variability (e.g., positive noise correlations). Interestingly, when these noise correlations were perturbed by inter-trial shuffling decoding performance decreased. Although Boffi and colleagues display that azimuthal information can be extracted from DCIC responses, it remains unclear to what degree this information is used and what role noise correlations play in azimuthal encoding.

Strengths: The authors should be commended for collection of this dataset. When done in isolation (which is typical), calcium imaging and linear array recordings have intrinsic weaknesses. However, those weaknesses are alleviated when done in conjunction with one another - especially when the data largely recapitulates the findings of the other recording methodology. In addition to the video of the head during the calcium imaging, this data set is extremely rich and will be of use to those interested in the information available in the DCIC, an understudied but likely important subnucleus in the auditory midbrain.

The DCIC neural responses are complex; the units unreliably respond to sound onset, and at the very least respond to some unknown input or internal state (e.g., large inter-sound interval responses). The authors do a decent job in wrangling these complex responses: using interpretable decoders to extract information available from population responses.

Weaknesses:

The authors observe that neurons with the most azimuthal sensitivity within the DCIC are positively correlated, but they use a Naïve Bayesian decoder which assume independence between units. Although this is a bit strange given their observation that some of the recorded units are correlated, it is unlikely to be a critical flaw. At one point the authors reduce the dimensionality of their data through PCA and use the loadings onto these components in their decoder. PCA incorporates the correlational structure when finding the principal components and constrains these components to be orthogonal and uncorrelated. This should alleviate some of the concern regarding the use of the naïve Bayesian decoder because the projections onto the different components are independent. Nevertheless, the decoding results are a bit strange, likely because there is not much linearly decodable azimuth information in the DCIC responses. Raw population responses failed to provide sufficient information concerning azimuth for the decoder to perform better than chance. Additionally, it only performed better than chance when certain principal components or top ranked units contributed to the decoder but not as more components or units were added. So, although there does appear to be some azimuthal information in the recoded DCIC populations - it is somewhat difficult to extract and likely not an 'effective' encoding of sound localization as their title suggests.

As described in the responses to reviewers 1 and 2, we chose the naïve Bayes classifier as a decoder to determine the influence of noise correlations through the most conservative approach possible, as this classifier would be least likely to find a contribution of correlated noise. Also, we chose this decoder due to its robustness against limited numbers of trials collected, in comparison to “data hungry” non linear classifiers like KNN or artificial neuronal nets. Lastly, we observed that small populations of noisy, unreliable (do not respond in every trial) DCIC neurons can encode stimulus azimuth in passively listening mice matching the discrimination error of trained mice. Therefore, while this encoding is definitely not efficient, it can still be considered effective.

Although this is quite a worthwhile dataset, the authors present relatively little about the characteristics of the units they've recorded. This may be due to the high variance in responses seen in their population. Nevertheless, the authors note that units do not respond on every trial but do not report what percent of trials that fail to evoke a response. Is it that neurons are noisy because they do not respond on every trial or is it also that when they do respond they have variable response distributions? It would be nice to gain some insight into the heterogeneity of the responses.

The limited number of azimuth trial repetitions that we could collect precluded us from making any quantification of the unreliability (failures to respond) and variability in the response distributions from the units we recorded, as we feared they could be misleading. In qualitative terms, “due to the high variance in responses seen” in the recordings and the limited trial sampling, it is hard to make any generalization. In consequence we referred to the observed response variance altogether as neuronal noise. Considering these points, our datasets are publicly available for exploration of the response characteristics.

Additionally, is there any clustering at all in response profiles or is each neuron they recorded in the DCIC unique?

We attempted to qualitatively visualize response clustering using dimensionality reduction, observing different degrees of clustering or lack thereof across the azimuth classes in the datasets collected from different mice. It is likely that the limited number of azimuth trials we could collect and the high response variance contribute to an inconsistent response clustering across datasets.

They also only report the noise correlations for their top ranked units, but it is possible that the noise correlations in the rest of the population are different.

For this study, since our aim was to interrogate the influence of noise correlations on stimulus azimuth encoding by DCIC populations, we focused on the noise correlations from the top ranked unit subpopulation, which likely carry the bulk of the sound location information.  Noise correlations can be defined as correlation in the trial to trial response variation of neurons. In this respect, it is hard to ascertain if the rest of the population, that is not in the top rank unit percentage, are really responding and showing response variation to evaluate this correlation, or are simply not responding at all and show unrelated activity altogether. This makes observations about noise correlations from “the rest of the population” potentially hard to interpret.

It would also be worth digging into the noise correlations more - are units positively correlated because they respond together (e.g., if unit x responds on trial 1 so does unit y) or are they also modulated around their mean rates on similar trials (e.g., unit x and y respond and both are responding more than their mean response rate). A large portion of trial with no response can occlude noise correlations. More transparency around the response properties of these populations would be welcome.

Due to the limited number of azimuth trial repetitions collected, to evaluate noise correlations we used the non parametric Kendall tau correlation coefficient which is a measure of pairwise rank correlation or ordinal association in the responses to each azimuth. Positive rank correlation would represent neurons more likely responding together. Evaluating response modulation “around their mean rates on similar trials” would require assumptions about the response distributions, which we avoided due to the potential biases associated with limited sample sizes.

It is largely unclear what the DCIC is encoding. Although the authors are interested in azimuth, sound location seems to be only a small part of DCIC responses. The authors report responses during inter-sound interval and unreliable sound-evoked responses. Although they have video of the head during recording, we only see a correlation to snout and ear movements (which are peculiar since in the example shown it seems the head movements predict the sound presentation). Additional correlates could be eye movements or pupil size. Eye movement are of particular interest due to their known interaction with IC responses - especially if the DCIC encodes sound location in relation to eye position instead of head position (though much of eye-position-IC work was done in primates and not rodent). Alternatively, much of the population may only encode sound location if an animal is engaged in a localization task. Ideally, the authors could perform more substantive analyses to determine if this population is truly noisy or if the DCIC is integrating un-analyzed signals.

We unsuccessfully attempted eye tracking and pupillometry in our videos. We suspect that the reason behind this is a generally overly dilated pupil due to the low visible light illumination conditions we used which were necessary to protect the PMT of our custom scope.

It is likely that DCIC population activity is integrating un-analyzed signals, like the signal associated with spontaneous behaviors including face movements (Stringer et al., 2019), which we observed at the level of spontaneous snout movements. However investigating if and how these signals are integrated to stimulus azimuth coding requires extensive behavioral testing and experimentation which is out of the scope of this study. For the purpose of our study, we referred to trial-to-trial response variation as neuronal noise. We note that this definition of neuronal noise can, and likely does, include an influence from un-analyzed signals like the ones from spontaneous behaviors.

Although this critique is ubiquitous among decoding papers in the absence of behavioral or causal perturbations, it is unclear what - if any - role the decoded information may play in neuronal computations. The interpretation of the decoder means that there is some extractable information concerning sound azimuth - but not if it is functional. This information may just be epiphenomenal, leaking in from inputs, and not used in computation or relayed to downstream structures. This should be kept in mind when the authors suggest their findings implicate the DCIC functionally in sound localization.

Our study builds upon previous reports by other independent groups relying on “causal and behavioral perturbations” and implicating DCIC in sound location learning induced experience dependent plasticity (Bajo et al., 2019, 2010; Bajo and King, 2012), which altogether argues in favor of DCIC functionality in sound localization.

Nevertheless, we clarified in the discussion of the revised manuscript that a relationship between the observed decoding error and the psychophysical performance, or the ability of the DCIC network to perform Bayesian decoding computations, both remain to be determined (please see responses to Reviewer #2).

It is unclear why positive noise correlations amongst similarly tuned neurons would improve decoding. A toy model exploring how positive noise correlations in conjunction with unreliable units that inconsistently respond may anchor these findings in an interpretable way. It seems plausible that inconsistent responses would benefit from strong noise correlations, simply by units responding together. This would predict that shuffling would impair performance because you would then be sampling from trials in which some units respond, and trials in which some units do not respond - and may predict a bimodal performance distribution in which some trials decode well (when the units respond) and poor performance (when the units do not respond).

In samples with more that 2 dimensions, the relationship between signal and noise correlations is more complex than in two dimensional samples (Montijn et al., 2016) which makes constructing interpretable and simple toy models of this challenging. Montijn et al. (2016) provide a detailed characterization and model describing how the accuracy of a multidimensional population code can improve when including “positive noise correlations amongst similarly tuned neurons”. Unfortunately we could not successfully test their model based on Mahalanobis distances as we could not verify that the recorded DCIC population responses followed a multivariate gaussian distribution, due to the limited azimuth trial repetitions we could sample.

Significance: Boffi and colleagues set out to parse the azimuthal information available in the DCIC on a single trial. They largely accomplish this goal and are able to extract this information when allowing the units that contain more information about sound location to contribute to their decoding (e.g., through PCA or decoding on top unit activity specifically). The dataset will be of value to those interested in the DCIC and also to anyone interested in the role of noise correlations in population coding. Although this work is first step into parsing the information available in the DCIC, it remains difficult to interpret if/how this azimuthal information is used in localization behaviors of engaged mice.

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.

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.

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.

Author response:

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

Public Reviews: 

Reviewer #1 (Public Review):

In this study, Alejandro Rosell et al. uncovers the immunoregulation functions of RAS-p110α pathway in macrophages, including the extravasation of monocytes from the bloodstream and subsequent lysosomal digestion. Disrupting RAS-p110α pathway by mouse genetic tools or by pharmacological intervention, hampers the inflammatory response, leading to delayed resolution and more severe acute inflammatory reactions. The authors proposed that activating p110α using small molecules could be a promising approach for treating chronic inflammation. This study provides insights into the roles and mechanisms of p110α on macrophage function and the inflammatory response, while some conclusions are still questionable because of several issues described below. 

(1) Fig. 1B showed that disruption of RAS-p110α causes the decrease in the activation of NF-κB, which is a crucial transcription factor that regulates the expression of proinflammatory genes. However, the authors observed that disruption of RAS-p110α interaction results in an exacerbated inflammatory state in vivo, in both localized paw inflammation and systemic inflammatory mediator levels. Also, the authors introduced that "this disruption leads to a change in macrophage polarization, favoring a more proinflammatory M1 state" in introduction according to reference 12. The conclusions drew from the signaling and the models seemed contradictory and puzzling. Besides, it is not clear why the protein level of p65 was decreased at 10' and 30'. Was it attributed to the degradation of p65 or experimental variation? 

We thank the reviewer for this insightful comment and apologize for not previously explaining the implications of the observed decrease in NF-κB activation. We found a decrease in NF-κB activation in response to LPS + IFN-γ stimulation in macrophages lacking RAS-PI3K interaction. As the reviewer pointed out, NF-κB is a key transcription factor that regulates the expression of various proinflammatory genes. To better characterize whether the decrease in p-p65 would lead to a reduction in the expression of specific cytokines, we performed a cytokine array using unstimulated and LPS + IFN-γ stimulated macrophages. The results indicated a small number of cytokines with altered expression, validating that RAS-p110α activation of p-p65 regulates the expression of some inflammatory cytokines. These results have been added to the manuscript and to Figure 1 (panels C and D). In brief, the data suggest an impairment in recruitment factors and inflammatory regulators following the disruption of RAS-p110α signaling in macrophages, which aligns with the observed in vivo phenotype. 

Our findings indicate that the disruption of RAS-p110α signaling has a complex and multifaceted role in BMDMs. Specifically, monocytes lacking RAS-PI3K are unable to reach the inflamed area due to an impaired ability to extravasate, caused by altered actin cytoskeleton dynamics. Consequently, inflammation is sustained over time, continuously releasing inflammatory mediators. Moreover, we have shown that macrophages deficient in RAS-p110α interaction fail to mount a full inflammatory response due to decreased activation of p-p65, leading to reduced production of a set of inflammatory regulators. Additionally, these macrophages are unable to effectively process phagocytosed material and activate the resolutive phase of inflammation. As a result of these defects, an exacerbated and sustained inflammatory response occurs. 

Our in vivo data, showing an increase in systemic inflammatory mediators, might be a consequence of the accumulation of monocytes produced by bone marrow progenitors in response to sensed inflammatory stimuli, but unable to extravasate.

Regarding the sentence in the introduction: "this disruption leads to a change in macrophage polarization, favoring a more proinflammatory M1 state" (reference 12), this was observed in an oncogenic context, which might differ from the role of RAS-p110α in a non-oncogenic situation, as analyzed in this work. We introduced these results as an example to establish the role of RAS-p110α in macrophages, demonstrating its participation in macrophage-dependent responses. Together with our study, these findings clearly indicate that p110α signaling is critical when analyzing full immune responses. Previously, little was known about the role of this PI3K isoform in immune responses. Our data, along with those presented by Murillo et al. (ref. 12), demonstrate that p110α plays a significant role in macrophage function in both oncogenic and inflammatory contexts. Additionally, our results suggest that this role is complex and multifaceted, warranting further investigation to fully understand the complexity of p110α signaling in macrophages.

Regarding decreased levels of p65 at 10’ and 30’ in RBD cells we are still uncertain about the possible molecular mechanism leading to the observed decrease. No changes in p65 mRNA levels were observed after 30 minutes of LPS+IFNγ treatment as shown in Author response image 1.

Author response image 1.

Preliminary data not shown here suggest that treating macrophages with BYL exhibits a similar effect, indicating a potential pathway for investigation. Considering that the decrease in protein levels is not due to lower mRNA expression, we may infer that post-translational mechanisms are leading to early protein degradation in RAS-p110α deficient macrophages. This could explain the observed decrease in protein activation. However, the specific molecular mechanism responsible for this degradation remains unclear, and further research is necessary to elucidate it. 

(2) In Fig 3, the authors used bone-marrow derived macrophages (BMDMs) instead of isolated monocytes to evaluate the ability of monocyte transendothelial migration, which is not sufficiently convincing. In Fig. 3B, the authors evaluated the migration in Pik3caWT/- BMDMs, and Pik3caWT/WT BMDMs treated with BYL-719'. Given that the dose effect of gene expression, the best control is Pik3caWT/- BMDMs treated with BYL-719. 

We thank reviewer for this comment. While we agree that using BMDMs might not be the most conventional approach for studying monocyte migration, there were several reasons why we still considered them a valid method. While isolated monocytes are the initial cell type involved in transendothelial migration, bone marrow-derived macrophages (BMDMs) provide a relevant and practical model for studying this process. BMDMs are differentiated from the same bone marrow precursors as monocytes and retain the ability to respond to chemotactic signals, adhere to endothelial cells, and migrate through the endothelium. This makes them a suitable tool for examining the cellular and molecular mechanisms underlying monocyte migration and subsequent macrophage infiltration into tissues. Additionally, BMDMs offer experimental consistency and are easier to manipulate in vitro, enabling more controlled and reproducible studies. 

In response to the comment regarding Fig. 3B, we appreciate the suggestion to use Pik3ca WT/- BMDMs treated with BYL-719 as a control. However, our rationale for using Pik3ca WT/WT BMDMs treated with BYL-719 was based on a conceptual approach rather than a purely experimental control. The BYL-719 treatment in Pik3ca WT/WT cells was intended to simulate the inhibition of p110α in a fully functional, wild-type context. This allows us to directly assess the impact of p110α inhibition under normal physiological conditions, which is more representative of what would occur in an organism where the full dose of Pik3ca is present. Using Pik3ca WT/- BMDMs treated with BYL-719 as a control may not accurately reflect the in vivo scenario, where any therapeutic intervention would likely occur in the context of a fully functional, wild-type background. Our approach aims to provide a clearer understanding of how p110α inhibition affects cell functionality in a wild-type setting, which is relevant for potential therapeutic applications. Therefore, we considered the use of Pik3ca WT/WT BMDMs with BYL-719 treatment to be a more appropriate control for testing the effects of p110α inhibition in normal conditions.

(3) In Fig. 4E-4G, the authors observed that elevated levels of serine 3 phosphorylated Cofilin in Pik3caRBD/- BMDMs both in unstimulated and in proinflammatory conditions, and phosphorylation of Cofilin at Ser3 increase actin stabilization, it is not clear why disruption of RAS-p110α binding caused a decrease in the F-actin pool in unstimulated BMDMs? 

We thank the reviewer for this insightful comment. During the review process, we have carefully quantified all the Western blots conducted. While we did observe an increase in phospho-Cofilin (Ser3) levels in RBD BMDMs, this increase did not reach statistical significance. As a result, we cannot confidently attribute the observed increase in F-actin to this proposed mechanism. We apologize for any confusion this may have caused. Consequently, we have removed these data from Figure 4G and the associated discussion.

Unfortunately, we have not yet identified the underlying mechanism responsible for this phenotype. Future experiments will focus on exploring potential alterations in other actin-nucleating, regulating, and stabilizing proteins that could account for the observed changes in F-actin levels.

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 macrophages 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 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 the function of macrophages in a cell intrinsic manner. The study is well designed, clearly written with overall high-quality figures. 

Weaknesses: 

My main concern is that for many of the readouts, the difference between wild-type and mutant macrophages in vitro or between wild-type and Pik3caRBD mice in vivo is rather modest, even if statistically significant (e.g. Figure 1A, 1C, 2A, 2F, 3B, 4B, 4C). In other cases, such as for the analysis of the H&E images (Figure 1D-E, S1E), the images are not quantified, and it is hard to appreciate what the phenotype in samples from Pik3caRBD mice is or whether this is consistently observed across different animals. Also, the authors claim there is a 'notable decrease' in Akt activation but 'no discernible chance' in ERK activation based on the western blot data presented in Figure 1A. I do not think the data shown supports this conclusion. 

We appreciate the reviewer's careful examination of our data and their observation regarding the modest differences between wild-type and mutant macrophages in vitro, as well as between wild-type and Pik3caRBD mice in vivo. While the differences observed in Figures 1A, 1C, 2A, 2F, 3B, 4B, and 4C are statistically significant but modest, our data demonstrate that they are biologically relevant and should be interpreted within the specific nature of our model. Our study focuses on the disruption of the RASp110α interaction, but it should be noted that alternative pathways for p110α activation, independent of RAS, remain functional in this model. Additionally, the model retains the expression of other p110 isoforms, such as p110β, p110γ, and p110δ, which are known to have significant roles in immune responses. Given the overlapping functions of these p110 isoforms, and the fact that our model involves a subtle modification that specifically affects the RAS-p110α interaction without completely abrogating p110α activity, it is understandable that only modest effects are observed in some readouts. The redundancy and compensation by other p110 isoforms likely mitigate the impact of disrupting RAS-mediated p110α activation.

However, despite these modest in vitro differences, it is crucial to highlight that the in vivo effects on inflammation are both clear and consistent. The persistence of inflammation in our model suggests that the RAS-p110α interaction plays a specific, non-redundant role in resolving inflammation, which cannot be fully compensated by other signaling pathways or p110 isoforms. These findings underscore the importance of RAS-p110α signaling in immune homeostasis and suggest that even subtle disruptions in this pathway can lead to significant physiological consequences over time, particularly in the context of inflammation. The modest differences observed may represent early or subtle alterations that could lead to more pronounced phenotypes under specific stress or stimulation conditions. This could be tested across all the figures mentioned. For instance, in Fig. 1A, the Western blot for AKT has been quantified, demonstrating a significant decrease in AKT levels; in Fig. 1C, although the difference in paw inflammation was only a few millimeters in thickness, considering the size of a mouse paw, those millimeters were very noticeable by eye. Furthermore, pathological examination of the tissue consistently showed an increase in inflammation in RBD mice. Furthermore, the consistency of the observed differences across different readouts and experimental setups reinforces the reliability and robustness of our findings. Even modest changes that are consistently observed across different assays and conditions are indicative of genuine biological effects. The statistical significance of the differences indicates that they are unlikely to be due to random variation. This statistical rigor supports the conclusion that the observed effects, albeit modest, are real and warrant further exploration.

Regarding the analysis of H&E images, we have now quantified the changes with the assistance of the pathologist, Mª Carmen García Macías, who has been added to the author list. We removed the colored arrows from the images and instead quantified fibrin and chromatin remnants as markers of inflammation staging. Loose chromatin, which increases as a consequence of cell death, is higher in the early phases of inflammation and decreases as macrophages phagocytose cell debris to initiate tissue healing. Chromatin content was scored on a scale from 1 to 3, where 1 represents the lowest amount and 3 the highest. The scoring was based on the area within the acute inflammatory abscess where chromatin could be found: 3 for less than 30%, 2 for 30-60%, and 1 for over 60%. Graphs corresponding to this quantification have now been added to Figure 1 and an explanation of the scale has been added to Material and Methods. 

To further substantiate the extent of macrophage function alteration upon disruption of RAS-p110α signaling, the manuscript would benefit from testing macrophage activity in vitro and in vivo across other key macrophage activities such as bacteria phagocytosis, cytokine/chemokine production in response to titrating amounts of different PAMPs, inflammasome function, etc. This would be generally important overall but also useful to determine whether the defects in monocyte motility or macrophage lysosomal function are selectively controlled downstream of RAS-p110α signaling.  

We thank reviewer #2 for this comment. In order to better address the role of RAS-PI3K in macrophage function, we have performed some additional experiments, some of which have been added to the revised version of the manuscript. 

(1) We have performed cytokine microarrays of RAS-p110α deficient macrophages unstimulated and stimulated with LPS+IFN-g. Results have been added to the manuscript and to Supplementary Figure S1E and S1F. In brief, the data obtained suggest an impairment in recruitment factors, as well as in inflammatory regulators after disruption of RAS-p110α signaling in macrophages, which align with the in vivo observed phenotype. 

(2) We also conducted phagocytosis assays to analyze the ability of RAS-p110α deficient macrophages to phagocytose 1 µm Sepharose beads, Borrelia burgdorferi, and apoptotic cells. The data reveal varied behavior of RAS-p110α deficient bone marrow-derived macrophages (BMDMs) depending on the target: 

• Engulfment of Non-biological Particles: RAS-p110α deficient macrophages showed a decreased ability to engulf 1 µm Sepharose beads. This suggests that RAS-p110α signaling is important for the effective phagocytosis of non-biological particles. These findings have now been added to the text and figures have been added to supplementary Fig. S4A

• Response to Bacterial Pathogens: When exposed to Borrelia burgdorferi, RAS-p110α deficient macrophages did not exhibit a change in bacterial uptake. This indicates that RAS-p110α may not play a critical role in the initial phagocytosis of this bacterial pathogen. The observed increase in the phagocytic index, although not statistically significant, might imply a compensatory mechanism or a more complex interaction that warrants further investigation. These findings have now been added to the text and figures have been added to supplementary Fig. S4B. These experiments were performed in collaboration with Dr. Anguita, from CICBioBune (Bilbao, Spain) and, as a consequence, he has been added as an author in the paper. 

• Phagocytosis of Apoptotic Cells: There were no differences in the phagocytosis rate of apoptotic cells between RAS-p110α deficient and control macrophages at early time points. However, the accumulation of engulfed material at later time points suggests a possible delay in the processing and degradation of apoptotic cells in the absence of RAS-p110α signaling.

These findings highlight the complexity of RAS-p110α's involvement in phagocytic processes and suggest that its role may vary with different types of phagocytic targets. 

Furthermore, given the key role of other myeloid cells besides macrophages in inflammation and immunity it remains unclear whether the phenotype observed in vivo can be attributed to impaired macrophage function. Is the function of neutrophils, dendritic cells or other key innate immune cells not affected? 

Thank you for this insightful comment. We understand the key role of other myeloid cells in inflammation and immunity. However, our study specifically focuses on the role of macrophages. Our data show that disruption of RAS-PI3K leads to a clear defect in macrophage extravasation, and our in vitro data demonstrate issues in macrophage cytoskeleton and phagocytosis, aligning with the in vivo phenotype.

Experiments investigating the role of RAS-PI3K in neutrophils, dendritic cells, or other innate immune cells are beyond the scope of this study. Understanding these interactions would indeed require separate, comprehensive studies and the generation of new mouse models to disrupt RAS-PI3K exclusively in specific cell types.

Furthermore, during paw inflammation experiments, polymorphonuclear cells were present from the initial phases of the inflammatory response. What caught our attention was the prolonged presence of these cells. In conversation with our in-house pathologist, she mentioned the lack of macrophages to remove dead polymorphonuclear cells in our RAS-PI3K mutant mice. Specific staining for macrophages confirmed the absence of macrophages in the inflamed node of mutant mice.

We acknowledge that further research is necessary to elucidate the effects on other myeloid cells. However, our current findings provide clear evidence of a decrease in inflammatory monocytes and defective macrophage responses to inflammation, both in vivo and in vitro. We believe these results significantly contribute to understanding the role of RAS-PI3K in macrophage function during inflammation.

Compelling proof of concept data that targeting RAS-p110α signalling constitutes indeed a putative approach for modulation of chronic inflammation is lacking. Addressing this further would increase the conceptual advance of the manuscript and provide extra support to the authors' suggestion that p110α inhibition or activation constitute promising approaches to manage inflammation. 

We thank Reviewer #2 for this insightful comment. In our manuscript, we have demonstrated through multiple experiments that the inhibition of p110α, either by disrupting RAS-p110α signaling or through the use of Alpelisib (BYL-719), has a modulatory effect on inflammatory responses. However, we acknowledge that we have not activated the pathway due to the unavailability of a suitable p110α activator until the concluding phase of our study.

We recognize the importance of this point and are eager about investigating both the inhibition and activation of p110α as potential approaches to managing inflammation in well-established inflammatory disease models. We believe that such comprehensive studies would significantly enhance the conceptual advance and translational relevance of our findings.

However, it is essential to note that the primary aim of our current work was to demonstrate the role of RAS-p110α in the inflammatory responses of macrophages. We have successfully shown that RASp110α influences macrophage behavior and inflammatory signaling. Expanding the scope to include disease models and pathway activation studies would be an extensive project that goes beyond the current objectives of this manuscript. While our present study establishes the foundational role of RASp110α in macrophage-mediated inflammatory responses, we agree that further investigation into both p110α inhibition and activation in disease models is crucial. We are keen to pursue this line of research in future studies, which we believe will provide robust evidence supporting the therapeutic potential of targeting RAS-p110α signaling in chronic inflammation.

Finally, the analysis by FACS should also include information about the total number of cells, not just the percentage, which is affected by the relative change in other populations. On this point, Figure S2B shows a substantial, albeit not significant (with less number of mice analysed), increase in the percentage of CD3+ cells. Is there an increase in the absolute number of T cells or does this apparent relative increase reflect a reduction in myeloid cells? 

We thank the reviewer for this comment, which we have addressed in the revised version of the manuscript. Regarding the total number of cells analyzed, we have added to the Materials and Methods section that in all our studies, a total of 50,000 cells were analyzed (line 749). The percentages of cells are related to these 50,000 events. Additionally, we have increased the number of mice analyzed by including new mice for CD3+ cell analysis. Despite this, the results remain not significant.

Recommendations for the authors:  

Reviewer #1 (Recommendations For The Authors):   

(1) It is recommended to provide a graphical abstract to summarize the multiple functions of RAS-p110α pathway in monocyte/macrophages that the authors proposed 

We thank reviewer for this useful recommendation. A graphical abstract has now been added to the study. 

(2) Western blots in this paper need quantification and a measure of reproducibility 

We have now added a graph with the quantification of the western blots performed in this work as a measure of reproducibility. 

(3) Representative flow data and gating strategy should be included

We have now added the description of the gating strategy followed to material and methods section.

eLife Assessment

This important study investigates the molecular mechanisms underpinning how the tumor necrosis factor alpha-induced protein (TIPE) regulates aerobic glycolysis to promote tumor growth in melanoma. Convincing data using multiple independent approaches provides new insights into the molecular mechanisms underpinning aerobic glycolysis in melanoma cells. The work will be of interest to biomedical researchers working in cancer and metabolism.

Reviewer #1 (Public review):

Summary:

Tian et al. describes how TIPE regulates melanoma progression, stemness, and glycolysis. The authors link high TIPE expression to increased melanoma cell proliferation and tumor growth. TIPE causes dimerization of PKM2, as well as translocation of PKM2 to the nucleus, thereby activating HIF-1alpha. TIPE promotes the phosphorylation of S37 on PKM2 in an ERK-dependent manner. TIPE is shown to increase stem-like phenotype markers. The expression of TIPE is positively correlated with the levels of PKM2 Ser37 phosphorylation in murine and clinical tissue samples. Taken together, the authors demonstrate how TIPE impacts melanoma progression, stemness, and glycolysis through dimeric PKM2 and HIF-1alpha crosstalk.

The authors manipulated TIPE expression using both shRNA and overexpression approaches throughout the manuscript. Using these models, they provide strong evidence of the involvement of TIPE in mediating PKM2 Ser37 phosphorylation and dimerization. The authors also used mutants of PKM2 at S37A to block its interaction with TIPE and HIF-1alpha. In addition, an ERK inhibitor (U0126) was used to block the phosphorylation of Ser37 on PKM2. The authors show how dimerization of PKM2 by TIPE causes nuclear import of PKM2 and activation of HIF-1alpha and target genes. Pyridoxine was used to induce PKM2 dimer formation, while TEPP-46 was used to suppress PKM2 dimer formation. TIPE maintains stem cell phenotypes by increasing expression of stem-like markers. Furthermore, the relationship between TIPE and Ser37 PKM2 was demonstrated in murine and clinical tissue samples.

The evaluation of how TIPE causes metabolic reprogramming can be further assessed using isotope tracing experiments.

Reviewer #2 (Public review):

In this article, Tian et al present a convincing analysis of the molecular mechanisms underpinning TIPE-mediated regulation of glycolysis and tumor growth in melanoma. The authors begin by confirming TIPE expression in melanoma cell lines and identify "high" and "low" expressing models for functional analysis. They show that TIPE depletion slows tumour growth in vivo, and using both knockdown and over expression approaches, show that this is associated with changes in glycolysis in vitro. Compelling data using multiple independent approaches is presented to support an interaction between TIPE and the glycolysis regulator PKM2, and over-expression of TIPE promoted nuclear translocation of PKM2 dimers. Mechanistically, the authors also demonstrate that PKM2 is required for TIPE-mediated activation of HIF1a transcriptional activity, as assessed using an HRE-promoter reporter assay, and that TIPE-mediated PKM2 dimerization is p-ERK dependent. Finally, the dependence of TIPE activity on PKM2 dimerization was demonstrated on tumor growth in vivo and in regulation of glycolysis in vitro, and ectopic expression of HIF1a could rescue inhibition of PKM2 dimerization in TIPE overexpressing cells and reduced induction of general cancer stem cell markers, showing a clear role for HIF1a in this pathway.

The detailed mechanistic analysis of TIPE mediated regulation of PKM2 to control aerobic glycolysis and tumor growth is a major strength of the study and provides new insights into the molecular mechanisms that underpin the Warburg effect in melanoma cells. The main conclusions of this paper are well supported by data, however further investigation of a potential oncogenic effect of TIPE in melanoma patients is warranted to support the tumor promoting role of TIPE identified in the experimental models. Analysis of patient samples showed a significant increase in TIPE protein levels in primary melanoma compared to benign skin tumours, and a further increase upon metastatic progression. Moreover, TIPE levels correlate with proliferation (Ki67) and hypoxia gene sets in the TCGA melanoma patient dataset. However, intriguingly, high TIPE expression associates with better survival outcomes in the TCGA melanoma patient cohort, therefore further investigation of how TIPE-mediated regulation of glycolysis contributes to melanoma progression is warranted to confirm the authors claims of a potential oncogenic function. Regardless, the new insights into the molecular mechanisms underpinning TIPE-mediated aerobic glycolysis in melanoma are convincing and will likely generate interest in the cancer metabolism field.

Author response:

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

Reviewer #1 (Public Review):

Summary:

Tian et al. describe how TIPE regulates melanoma progression, stemness, and glycolysis. The authors link high TIPE expression to increased melanoma cell proliferation and tumor growth. TIPE causes dimerization of PKM2, as well as translocation of PKM2 to the nucleus, thereby activating HIF-1alpha. TIPE promotes the phosphorylation of S37 on PKM2 in an ERK-dependent manner. TIPE is shown to increase stem-like phenotype markers. The expression of TIPE is positively correlated with the levels of PKM2 Ser37 phosphorylation in murine and clinical tissue samples. Taken together, the authors demonstrate how TIPE impacts melanoma progression, stemness, and glycolysis through dimeric PKM2 and HIF-1alpha crosstalk.

Strengths:

The authors manipulated TIPE expression using both shRNA and overexpression approaches throughout the manuscript. Using these models, they provide strong evidence of the involvement of TIPE in mediating PKM2 Ser37 phosphorylation and dimerization. The authors also used mutants of PKM2 at S37A to block its interaction with TIPE and HIF-1alpha. In addition, an ERK inhibitor (U0126) was used to block the phosphorylation of Ser37 on PKM2. The authors show how dimerization of PKM2 by TIPE causes nuclear import of PKM2 and activation of HIF-1alpha and target genes. Pyridoxine was used to induce PKM2 dimer formation, while TEPP-46 was used to suppress PKM2 dimer formation. TIPE maintains stem cell phenotypes by increasing the expression of stem-like markers. Furthermore, the relationship between TIPE and Ser37 PKM2 was demonstrated in murine and clinical tissue samples.

Weaknesses:

The evaluation of how TIPE causes metabolic reprogramming can be better assessed using isotope tracing experiments and improved bioenergetic analysis.

Thank you immensely for your invaluable suggestions. Regrettably, we encountered a significant obstacle in completing the isotope tracing experiments due to an unfortunate shortage of necessary instruments. Furthermore, despite our efforts to consult with several companies, we were unable to secure their assistance, which unfortunately hindered the completion of these experiments. We deeply apologize for this imperfection in our experimental design and have thoroughly discussed this limitation in our manuscript.

Additionally, we acknowledge our oversight in the previous versions of our manuscripts, where only three metabolites were presented. To rectify this and provide a more comprehensive understanding of the metabolic reprogramming induced by TIPE, we have conducted routine untargeted metabolomics analysis. We are pleased to announce that we have incorporated the detailed results of this analysis into our work as a new supplementary figure, designated as Figure S3. This figure specifically highlights the notable decrease in the glycolysis pathway, particularly in pyruvate and lactic acid levels, following TIPE interference.

Reviewer #2 (Public Review):

In this article, Tian et al present a convincing analysis of the molecular mechanisms underpinning TIPE-mediated regulation of glycolysis and tumor growth in melanoma. The authors begin by confirming TIPE expression in melanoma cell lines and identify "high" and "low" expressing models for functional analysis. They show that TIPE depletion slows tumour growth in vivo, and using both knockdown and over-expression approaches, show that this is associated with changes in glycolysis in vitro. Compelling data using multiple independent approaches is presented to support an interaction between TIPE and the glycolysis regulator PKM2, and the over-expression of TIPE-promoted nuclear translocation of PKM2 dimers. Mechanistically, the authors also demonstrate that PKM2 is required for TIPE-mediated activation of HIF1a transcriptional activity, as assessed using an HRE-promoter reporter assay, and that TIPE-mediated PKM2 dimerization is p-ERK dependent. Finally, the dependence of TIPE activity on PKM2 dimerization was demonstrated on tumor growth in vivo and in the regulation of glycolysis in vitro, and ectopic expression of HIF1a could rescue the inhibition of PKM2 dimerization in TIPE overexpressing cells and reduced induction of general cancer stem cell markers, showing a clear role for HIF1a in this pathway. The main conclusions of this paper are well supported by data, but some aspects of the experiments need clarification and some data panels are difficult to read and interpret as currently presented.

The detailed mechanistic analysis of TIPE-mediated regulation of PKM2 to control aerobic glycolysis and tumor growth is a major strength of the study and provides new insights into the molecular mechanisms that underpin the Warburg effect in cancer cells. However, despite these strengths, some weaknesses were noted, which if addressed will further strengthen the study.

(1) The analysis of patient samples should be expanded to more directly measure the relationship between TIPE levels and melanoma patient outcome and progression (primary vs metastasis), to build on the association between TIPE levels and proliferation (Ki67) and hypoxia gene sets that are currently shown.

Thanks for your suggestions. We have expanded the analysis to include the relationship between TIPE levels and melanoma progression, specifically distinguishing between non-lymph node metastasis and lymph node metastasis. In addition, we added the association between TIPE and Ki67 or LDH levels as your advised, as shown in Figure 7.

However, the relationship between TIPE levels and melanoma patient outcome is not presented in this article. One reason is that the tissue microarray lack of the survival data. Interestingly, the TCGA dataset showed that the higher TIPE expression has a favorable prognosis for melanoma. We are also very curious about this. Our following study indicated that TIPE might serve as a positive regulator of PD-L1. Therefore, the higher expression of TIPE presents more sensitive tendency to immunotherapy, resulting in a favorable prognosis in melanoma. The detailed mechanisms will be discussed in our following article, and we hope that it might as a continuous research topic for TIPE in melanoma.

We just only disclose a little information that TIPE shares similar survival and immune signature to PD-L1 and PD-1 in melanoma as following:

Author response image 1.

(2) The duration of the in vivo experiments was not clearly defined in the figures, however, it was clear from the tumor volume measurements that they ended well before standard ethical endpoints in some of the experiments. A rationale for this should be provided because longer-duration experiments might significantly change the interpretation of the data. For example, does TIPE depletion transiently reduce or lead to sustained reductions in tumor growth?

Thanks for your suggestions. Actually, we have performed a pre-experiment before the formal experiments, and all the time points were referred to this. Furthermore, we have added the detailed time points into the figure legends as you suggested.

(3) The analysis of general cancer stem cell markers is solid and interesting, however inclusion of neural crest stem cell markers that are more relevant to melanoma biology would greatly strengthen this aspect of the study.

Thanks for your advices. We have selected two neural crest stem cell markers including Nestin and Sox10 to test their expression after overexpression of TIPE in G361 cells or interference of TIPE in A375 cells.

(4) The authors should take care that all data panels are clearly readable in the figures to facilitate appropriate interpretation by the reader.

Thanks for your suggestions. We have amended the data panels according to you advises to ensure it is clear and professionally presented.

Reviewer #1 (Recommendations for the authors):

It would be suggested to improve the image quality of certain panels (please refer to Fig.1A and Fig.S3B-D).

Thank you for your expert advice. We have optimized the quality of certain panels according to your suggestions.

Reviewer #2 (Recommendations for the authors):

Major comments:

- TCGA survival/patient outcome data relative to TIPE levels should be provided in the supplementary figures, together with TIPE correlation with PKM2.

- Suggest revising how this point is described in the discussion.

We have added the results of TIPE expression and prognosis of melanoma patients from the TCGA database as required by the expert, and discussed it appropriately in the article. In addition, the correlation between TIPE and PKM2 expression has already been described in Supplementary Figure 6.

Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

Summary:

The work by Joseph et al "Impact of the clinically approved BTK inhibitors on the conformation of full-length BTK and analysis of the development of BTK resistance mutations in chronic lymphocytic leukemia" seeks to comparatively analyze the effect of a range of covalent and noncovalent clinical BTK inhibitors upon BTK conformation. The novel aspect of this manuscript is that it seeks to evaluate the differential resistance mutations that arise distinctly from each of the inhibitors.

Strengths:

This is an exciting study that builds upon the fundamental notion of ensemble behavior in solutions for enzymes such as BTK. The HDX-MS and NMR experiments are adequately and comprehensively presented.

We thank the reviewer for this positive feedback.

Weaknesses:

While I commend the novelty of the study, the absence of important controls greatly tempers my enthusiasm for this work. As stated in the abstract, there are no broad takeaways for how resistance mutation bias operated from this study, although the mechanism of action of 2 common resistance mutations is useful. How these 2 resistance mutations connect to ensemble behavior, is not obvious. This is partly because BTK does not populate just binary "open"/"closed" conformations, but there are likely multiple intermediate conformations. Each inhibitor appears to preferentially "select" conformations by the authors' own assessment (line 236) and this carries implications for the emergence of resistance mutations. The most important control that would help is to use ADP or nonhydrolyzable and ATP as a baseline to establish the "inactive" and "active" conformations. All of the HDX-MS and NMR studies use protein that has no nucleotide present. A major question that remains is whether each of the inhibitors preferentially favors/blocks ADP or ATP binding. This then means it is not equivalent to correlate functional kinase assay conditions with either HDX-MS or NMR experiments.

We thank the reviewer for raising this point. The BTK inhibitors studied here are active site inhibitors that completely prevent (block) nucleotide (both ATP and ADP) binding. We believe the other question being asked here is whether the different BTK inhibitors bind preferentially to the ADP or ATP bound kinase (do the conformational states favored by ADP versus ATP bound BTK affect drug binding). We agree this is an interesting question that deserves further study. Here we are focused on the ligand bound state itself rather than on the conformational state selection mechanism of each inhibitor. Thus, HDX-MS and NMR work to compare ligand bound to apo-, ADP, and ATP bound BTK is beyond the scope of this manuscript. That said, previous work (doi: 10.1038/s41598-017-17703-5) has shown that the related TEC kinase, ITK, preferentially binds ADP when the kinase is in the autoinhibited conformation. Since we have previously shown that BTK adopts the autoinhibited conformation in the nucleotide free form (https://doi.org/10.7554/eLife.89489.2), we suggest that the comparison we have carried out here between drug bound and apo-protein is valid. Future work will carefully address the conformational preferences of all three conditions, apo-, ADP- and ATP-bound.

Reviewer #2 (Public Review):

Summary:

Previous NMR and HDX-MS studies on full-length (FL) BTK showed that the covalent BTKi, ibrutinib, causes long-range effects on the conformation of BTK consistent with disruption of the autoinhibited conformation, based on HDX deuterium uptake patterns and NMR chemical shift perturbations. This study extends the analyses to four new covalent BTKi, acalabrutinib, zanubrutinib, tirabrutinib/ONO4059, and a noncovalent ATP competitive BTKi, pirtobrutinib/LOXO405.

The results show distinct conformational changes that occur upon binding each BTKi. The findings show consistent NMR and HDX changes with covalent inhibitors, which move helix aC to an 'out' position and disrupt SH3-kinase interactions, in agreement with X-ray structures of the BTKi complexed with the BTK kinase domain. In contrast, the solution measurements show that pirtobrutinib maintains and even stabilizes the helix aC-in and autoinhibited conformation, even though the BTK:pritobrutinib crystallizes with helix aC-out. This and unexpected variations in NMR and HDX behavior between inhibitors highlight the need for solution measurements to understand drug interactions with the full-length BTK. Overall the findings present good evidence for allosteric effects by each BTKi that induce distal conformational changes which are sensitive to differences in inhibitor structure.

The study goes on to examine BTK mutants T474I and L528W, which are known to confer resistance to pirtobrutinib, zanubritinib, and tirabrutinib. T474I reduces and L528W eliminates BTK autophosphorylation at pY551, while both FL-BTK-WT and FL-BTK-L528W increase HCK autophosphorylation and PLCg phosphorylation. These show that mutants partially or completely inactivate BTK and that inactive FL-BTK can activate HCK, potentially by direct BTK-HCK interactions. But they do not explain drug resistance. However, HDX and NMR show that each mutant alters the effects of BTKi binding compared to WT. In particular, T474I alters the effects of all three inhibitors around W395 and the activation loop, while L528W alters interactions around W395 with tirabrutinib and pirtobrutinib, and does not appear to bind zanubrutinib at all. The study concludes that the mutations might block drug efficacy by reducing affinity or altering binding mode.

Strengths:

The work presents convincing evidence that BTK inhibitors alter the conformation of regions distal to their binding sites, including those involved in the SH3-kinase interface, the activation loop, and a substrate binding surface between helix aF and helix aG. The findings add to the growing understanding of allosteric effects of kinase inhibitors, and their potential regulation of interactions between kinase and binding proteins.

We thank the reviewer for these positive comments.

Weaknesses:

The interpretation of HDX, NMR, and kinase assays is confusing in some places, due to ambiguity in quantifying how much kinase is bound to the inhibitor. It would be helpful to confirm binding occupancy, in order to clarify if mutants lower the amount of BTK complexed with BTKi as implied in certain places, or if they instead alter the binding mode. In addition, the interpretation of the mutant effects might benefit from a more detailed examination of how each inhibitor occupies the ATP pocket and how substitutions of T474 and L528 with Ile and Trp respectively might change the contacts with each inhibitor.

We thank the reviewer for these suggestions. As requested we have now modified the manuscript to clearly state the effects of the mutations on inhibitor binding. Additionally, we have included a new figure to discuss the interaction of the inhibitors within the BTK kinase active site to provide a better explanation for the impact of the resistance mutations.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

Major Comments:

(1) What is the binding affinity of ATP/ADP to BTK? BTK is purified by the authors as an apoenzyme (by the final purification by SEC, all protein should be completely stripped of nucleotide)- but must toggle between ATP and ADP-bound states. Do the inhibitors completely sterically block nucleotide binding? Do they only block one or the other- ADP/ATP binding? Do they weaken ADP/ATP binding? The authors have an opportunity with NMR to establish a clear baseline to compare the inhibitors' effects on BTK. It is not clear if the authors' assumption is that all BTKi share a common mode of action (Line 114).

All BTK inhibitors studied in this work (Ibrutinib, Acalabrutinib, Zanubrutinib, Tirabrutinib and Pirtobrutinib) share a common mode of action. They are active site inhibitors that completely block nucleotide (ATP and ADP) binding. The introduction to the manuscript has been updated to add this information (lines 70-71, pg. 4).

"The covalent BTK inhibitors (Ibrutinib, Acalabrutinib, Zanubrutinib and Tirabrutinib) and the non-covalent BTK inhibitor Pirtobrutinib bind tightly to the BTK active site (Kinact/KI or KD values in the nM range; DOI: 10.1056/NEJMoa2114110). In contrast, previous studies have reported nucleotide affinity for TEC kinases that are lower (KD in the µM range), (doi: 10.1038/s41598-017-17703-5). Additionally, the same work has shown that the conformational state of TEC kinases can impact nucleotide binding. The TEC kinases have a higher affinity for ADP (KD ~ 20 µM), as compared to ATP (KD ~ 15 fold lower than ADP), when the full-length protein adopts the autoinhibited conformation. Disruption of the TEC kinase autoinhibited conformation (by mutation) decreases the affinity for ADP, allowing ATP to bind, enabling kinase activity. Nevertheless, regardless of the conformational state of BTK, all the BTK inhibitors studied here block both ADP and ATP binding to the active site."

(2) Is there an effect of nucleotide binding bias on resistance mutation emergence? Is there a nucleotide binding bias in the resistance mutations characterized in this study? There likely is - BTK L528W is catalytically inactive. It is not clear if this mutant stays bound to ADP or to ATP and cannot transfer the phosphate to its substrate. How does BTK T474I interact with ADP/ATP? This is needed before concluding - in lines 289-291- that mutations cause only minor conformational changes. This needs a qualifier - in the nucleotide-free apo conformation.

The BTK L528W mutation introduces a bulky sidechain into the BTK kinase active site that sterically impedes both ATP and ADP binding. In fact, previous studies (https://doi.org/10.1016/j.jbc.2022.102555) have confirmed the inability of the BTK L528W mutant to bind ATP.

The BTK T474I mutation could alter nucleotide binding. However, The BTK T474I mutation lowers the overall activity of BTK, and is consistent with previous work that have shown the same (https://doi.org/10.1021/acschembio.6b00480). The decrease in overall kinase activity cannot account for the development of resistance (which typically requires increased kinase activity). Hence, a decrease in inhibitor binding is likely driving resistance.

Lines 293 (pg. 14) have been modified to indicate that the conformational changes observed in the BTK mutants are in the absence of nucleotide as requested.

(3) What is the half-life BTK? And does inhibitor binding to BTK change the half-life of the inhibitor?

BTK has a long half-life of 48-72 h (DOI: https://doi.org/10.1124/jpet.113.203489). Unbound covalent inhibitors are rapidly cleared from the body with short half-lives on the order of < 4h. Non-covalent BTK inhibitors typically have a longer half-life on the order of 20h. Once bound to BTK, the irreversible nature of binding by covalent inhibitors make them unavailable to other molecules. CLL patients are treated typically with a once daily or twice daily dose of BTK inhibitor. Hence, inhibitor binding to BTK does not alter the half-life of free inhibitor.

(4) Are there broad differences between covalent and single non-covalent inhibitors upon resistance mutation bias? And nucleotide binding?

The biggest difference observed between BTK covalent and non-covalent inhibitors in the emergence of resistance mutations is the occurrence of the C481S mutation in patients treated with covalent inhibitors. This resistance mutation is absent in patients treated with non-covalent BTK inhibitors. Patients that develop mutations in BTK C481 can no longer be treated with any of the approved covalent BTK inhibitors (as they all use BTK C481 for covalent linkage). To ensure BTK inhibition, patients with mutations in C481 can be treated with non-covalent BTK active site inhibitors. All currently approved BTK inhibitors (covalent and non-covalent) are active site inhibitors that compete with nucleotide binding.

(5) It's unclear why the authors chose to evaluate the impact of inhibitor binding on the linker kinase domain first. This seems unnecessary.

NMR analysis is easier on the smaller BTK linker kinase domain (LKD) fragment compared to the full-length protein. Hence for practical reasons we used the BTK LKD fragment.

(6) Line 508 - there seems to be a gap in understanding protein half-lives, inhibitor half-lives, and the emergence of resistance mutations in this manuscript itself. The manuscript falls short of a mechanistic descriptor of variable inhibitors and resistance mutation bias.

The half-life of the inhibitors assessed in this study are provided in Table 1 of this manuscript. The emergence of resistance mutations such as C481 are likely due to a direct consequence of differences in inhibitor half-life as described in the discussion section of this manuscript (page 23).

(7) HDX-MS reports the conformational average difference across the ensemble but does not distinguish between the number of intermediary conformations. The authors should clarify that this is a limitation of an average readout method such as HDX-MS. This is currently not addressed.

A sentence describing this limitation has been added (lines 219-221, pg. 11) as requested.

Minor  Points:

(1) Some of the qualitative descriptors are unnecessary - line 284 - "Slightly towards....". Line 286 - "Slight stabilizing effect on the conformation..." How slight is slight?

Qualitative descriptors have been removed from the manuscript as requested.

(2) The authors should provide SPR data with Kon and Koff values for Pirtobrutinib binding to BTK ( in the presence of ARP and ADP).

SPR analysis of Pirtobrutinib has previously been reported. Pirtobrutininb binds to BTK wild-type with a KD of 0.9 nM (DOI: 10.1056/NEJMoa2114110). As mentioned earlier in response to comment 1, Pirtobrutinib binds to the BTK kinase active site and is competitive with both nucleotides (ATP and ADP, which bind with lower affinity, KD in the µM range).

(3) In Figure 2, the legend needs to describe the specific time point represented. Same with Figure 5.

The HDX-MS changes that are mapped onto the structure represent the maximal changes observed at any time point. The figure legends have been modified as requested to clarify this.

Reviewer #2 (Recommendations For The Authors):

(1) Figure 7 is an amazing and impressive finding, but it could use two controls: First a blot of pY551 to show more rigorously that FL-BTK-WT and L528W autophosphorylation is unaffected by zanubrutinib binding, just to eliminate the possibility that elevated pY551 accounts for the enhanced HCK phosphorylation.

Both BTK FL enzymes (WT and L528W) in this assay are catalytically inactive and do not contribute to autophosphorylation on BTK Y551 (BTK FL WT is inhibited by Zanubrutinib and BTK FL L528W is catalytically dead). Additionally, BTK FL WT and BTK FL L528W are both able to activate HCK. Hence differences in pY551 levels between these BTK proteins cannot explain how both proteins are able to activate HCK.

Nevertheless, as requested, we probed for pY551 levels on BTK. While BTK cannot autophosphorylate itself on BTK Y551 in this assay, BTK Y551 is able to be phosphorylated by HCK. BTK Y551 phosphorylation levels were higher in BTK FL WT compared to BTK FL L528W likely due to Y551 on the activation loop being less accessible in the BTK L528W mutant (which is more stabilized in the autoinhibited conformation) compared to the WT protein. This data has been added as a new panel in Figure 7a.

Additionally, we tested the ability of the BTK FL L528W/Y551F double mutant to activate HCK. The BTK FL L528W/Y551F double mutant is able to activate HCK similar to BTK FL L528W single mutant, demonstrating that phosphorylation on Y551 is not necessary for HCK activation by BTK FL L528W. This new data has been added as supplemental figure S2a. Taken together, pY551 levels on BTK do not contribute to enhanced HCK phosphorylation. The results section of the manuscript has been modified to include this additional data (Lines 319-335, pg. 15-16).

Second, controls performed in the absence of Zanubrutinib are needed for the time courses with HCK alone, HCK + FL-BTK WT, and HCK + FL-BTK-L528W. This would help show that the ability of BTK to increase the phosphorylation of HCK and PLCg1 is (or isn't) dependent on drug interactions with BTK, HCK, or PLCg.

BTK FL L528W can enhance phosphorylation on PLCg by HCK even in the absence of Zanubrutinib. We have added this data as a new supplemental figure S2b. We have not included BTK FL WT in this analysis as in the absence of Zanubrutinib, we would have two active enzymes (HCK and BTK) in the assay which would complicate the interpretation of the data. The results section of the manuscript has been modified to include this additional data (Lines 333-335, pg. 16).

And please comment: in cells, does zanubrutinib treatment (or any other drug) increase pY phosphorylation of HCK or PLCg?

All clinically approved BTK inhibitors (covalent and non-covalent) inhibit BTK WT activity and decrease PLCg phosphorylation in cells. There have been no reports, to our knowledge, of any clinically approved BTK inhibitor causing an increase in HCK activity.

(2) Sections of the Results discussing Figures 8 and 9 are confusing to read because they variously propose that the mutants (i) reduce inhibitor occupancy, or (ii) alter the inhibitor binding mode. However, some of the results unambiguously show an altered binding mode instead of reduced inhibitor binding.

a) For example, HDX clearly shows protection by tira, zanu, and pirto, therefore reduced inhibitor binding does not seem to be an option. Therefore, I recommend modifying lines 357-363. "The differences in deuterium exchange for drug binding to WT and mutant BTK suggest that the T474I mutation either causes a reduction in inhibitor binding or otherwise alters the mode of drug interaction in the active site. "

While the HDX-MS data of BTK T474I shows protection by Tirabrutinib, Zanubrutinib and Pirtobrutinib, the magnitude of the protection is reduced in the BTK T474I mutant compared to WT BTK (Fig. 8e) suggesting a reduction in inhibitor binding. These results are consistent with previous SPR analysis of the BTK T474I mutant which also showed reduced binding to Zanubrutinib, Acalabrutinib and Pirtobrutinib (DOI: 10.1056/NEJMoa2114110). The manuscript (lines 381-383, pg. 18) has been modified to clearly state that the BTK T474I mutation causes a reduction in inhibitor binding.

b) I recommend modifying lines 370-373.

" In stark contrast to the BTK T474I mutant, the BTK 370 L528W mutant does not show any change in deuterium incorporation in the presence of 371 Zanubrutinib, Tirabrutinib or Pirtobrutinib, providing strong evidence that the BTK L528W 372 mutant does not bind the inhibitors (Fig.8d)."

Lines 432-435: Although the L528W mutation alters binding to both Tirabrutinib 432 and Pirtobrutinib, the NMR data suggests that it retains partial binding unlike the HDX-MS data 433 that suggests complete disruption of binding. The higher inhibitor concentrations used in the NMR 434 experiments compared to the HDX-MS experiments likely explain this discrepancy."

The discordance in the L528W mutant between the lack of any HDX protection by tira and pirto versus the clear chemical shift of W395 by NMR is worrisome. If the HDX experiments were really done under conditions where binding occupancy was too low, then it seems important to redo these experiments at higher drug concentrations.

Alternatively, and perhaps more useful would be to report Kd for binding of these inhibitors to the two mutants. That would allow the authors to interpret these results more definitively.

SPR analysis of inhibitor binding to full-length BTK WT, T474I and L528W has been previously reported (DOI: 10.1056/NEJMoa2114110). The covalent BTK inhibitors (Ibrutinib, Acalabrutinib, and Zanubrutinib) and the non-covalent BTK inhibitor Pirtobrutinib bind tightly to full-length WT BTK (Kinact/KI or KD values in the nM range). The BTK T474I mutation disrupts binding to Zanubrutinib, Acalabrutinib and Pirtobrutinib, but not Ibrutinib and Fenebrutinib. BTK L528W mutation disrupts binding to Zanubrutinib, Acalabrutinib, Ibrutinib and Pirtobrutinib, but not Fenebrutinib. These previously published results are consistent with the HDX-MS and NMR data presented here. The manuscript has been modified to clearly state that the mutations reduce drug binding instead of altered binding.

c) Recommend adding data to confirm statements in lines 419-421:

"Spectral overlays of the BTK L528W mutant with and without Zanubrutinib show no 419 chemical shift changes (Fig. 9a, right panel) suggesting that the mutation completely disrupts 420 inhibitor binding in complete agreement with the HDX-MS data (Fig. 8d).

428-432: The Pirtobrutinib-bound BTK L528W spectrum (Fig. 9c) shows two resonance positions, 428 one of which overlaps with the W395 resonance in the apo protein and the other that corresponds to that of the mutant protein bound to Pirtobrutinib. This data suggests a mixture of inhibitor bound and unbound BTK kinase domain in solution, likely due to a reduction in Pirtobrutinib affinity 431 caused by the L528W mutation."

Likewise, direct measurements of binding affinity to L528W would be helpful. It is not completely convincing that the effects of this mutant are due to the reduced binding of either inhibitor. The effects of pirtobrutinib may instead reflect a slow exchange of W395 instead of 50% occupancy. For example, what happened in the rest of the spectra? Were other chemical shifts apparent in either case, which might address binding stoichiometry? It would be useful to show the full spectra in Supplemental figures, as well as any titrations that may have been done to confirm that the inhibitors are added at saturating concentration.

As requested the full-spectra of Pirtobrutinib bound to BTK L528W has now been added as supplemental figure S1c. In the BTK L528W bound to Pirtobrutinib spectrum, two cross peaks are visible for multiple resonances, one of which overlaps with that of the apo BTK L528W spectrum, suggesting that there is a mixture of apo and inhibitor bound forms of BTK L528W.

The clinically approved inhibitors that we are working with here (Ibrutinib, Acalabrutinib, Zanubrutinib, Tirabrutinib and Pirtobrutinib have reported IC50 values in the nM range (0.5 nM, 3 nM, 0.3 nM, 6.8 nM and 3.68 nM respectively). All the NMR work presented here was carried out at a 1:1.33, protein:inhibitor ratio (absolute concentration of the inhibitor was 200 µM). NMR titrations of BTK WT have been carried out with Ibrutinib (https://doi.org/10.7554/eLife.60470) and Tirabrutinib. Complete binding is observed at a 1:1 molar ratio of protein:inhibitor, consistent with the previously reported binding characteristics. Mass spec analysis also shows one covalent inhibitor bound to each BTK WT protein (Fig. 4a). The BTK T474I and L528W mutants were tested at the same protein:inhibitor ratio as WT BTK for ease of comparison.

(3) The Discussion could use a structural perspective on the likely effects of each mutation on inhibitor binding. Both residues occupy positions in beta7 and the hinge, which are commonly found to form hydrophobic and polar contacts with ATP competitive inhibitors in many kinases. This would be useful to discuss and show as a figure, in order to give the non-kinase expert a better understanding of why the mutations might affect inhibitor binding. The variations in structures of each inhibitor and how they contact these two positions might be useful to inspect, and ask why some inhibitors but not others are affected by mutation, and why some inhibitors but not others induce effects over long distances to W395 and the activation loop.

As requested, we have added a new paragraph in the discussion and a new figure (Fig. 10), to expand on likely effects of the mutations on inhibitor binding. The allosteric effects of some of the BTK inhibitors, on the other hand are currently being investigated and is beyond the scope of the current manuscript.

(4) The authors propose that small differences in Tm and stability of L358W account for its effect on resistance. Does this mutant show elevated expression in patient tumors over those with WT BTK?

Preliminary data indicates that BTK L528W levels are elevated in one of two patients carrying this resistance mutation. However, due to the low number of patients tested, we have chosen to not include the data in this study but will continue to pursue this question in future work.

eLife Assessment

The manuscript reports on an important comparison of a range of approved clinical inhibitors for BTK used for the treatment of chronic lymphocytic leukemia (CLL). The authors provide compelling evidence for their claims, using a combination of HDX-MS and NMR spectroscopy. The novelty is that this study also seeks to evaluate resistance mutation bias. The manuscript will be of high interest to scientists working on protein drug interactions.

Reviewer #1 (Public Review):

Summary:

The work by Joseph et al "Impact of the clinically approved BTK inhibitors on the conformation of full-length BTK and analysis of the development of BTK resistance mutations in chronic lymphocytic leukemia" seeks to comparatively analyze the effect of a range of covalent and noncovalent clinical BTK inhibitors upon BTK conformation. The novel aspect of this manuscript is that it seeks to evaluate the differential resistance mutations that arise distinctly from each of the inhibitors.

Strengths:

This is an exciting study that builds upon the fundamental notion of ensemble behavior in solutions for enzymes such as BTK. The HDX-MS and NMR experiments are adequately and comprehensively presented.

Comments on the revised version:

I am satisfied with the revisions and the clear explanations.

Reviewer #2 (Public Review):

Summary:

Previous NMR and HDX-MS studies on full-length (FL) BTK showed that the covalent BTKi, ibrutinib, causes long-range effects on the conformation of BTK consistent with disruption of the autoinhibited conformation, based on HDX deuterium uptake patterns and NMR chemical shift perturbations. This study extends the analyses to four new covalent BTKi, acalabrutinib, zanubrutinib, tirabrutinib/ONO4059, and a noncovalent ATP competitive BTKi, pirtobrutinib/LOXO405.

The results show distinct conformational changes that occur upon binding each BTKi. The findings show consistent NMR and HDX changes with covalent inhibitors, which move helix aC to an 'out' position and disrupt SH3-kinase interactions, in agreement with X-ray structures of the BTKi complexed with the BTK kinase domain. In contrast, the solution measurements show that pirtobrutinib maintains and even stabilizes the helix aC-in and autoinhibited conformation, even though the BTK:pritobrutinib crystallizes with helix aC-out. This and unexpected variations in NMR and HDX behavior between inhibitors highlight the need for solution measurements to understand drug interactions with the full-length BTK. Overall the findings present good evidence for allosteric effects by each BTKi that induce distal conformational changes which are sensitive to differences in inhibitor structure.

The study goes on to examine BTK mutants T474I and L528W, which are known to confer resistance to pirtobrutinib, zanubritinib, and tirabrutinib. T474I reduces and L528W eliminates BTK autophosphorylation at pY551, while both FL-BTK-WT and FL-BTK-L528W increase HCK autophosphorylation and PLCg phosphorylation. These show that mutants partially or completely inactivate BTK and that inactive FL-BTK can activate HCK, potentially by direct BTK-HCK interactions. But they do not explain drug resistance. However, HDX and NMR show that each mutant alters the effects of BTKi binding compared to WT. In particular, T474I alters the effects of all three inhibitors around W395 and the activation loop, while L528W alters interactions around W395 with tirabrutinib and pirtobrutinib, and does not appear to bind zanubrutinib at all. The study concludes that the mutations might block drug efficacy by reducing affinity or altering binding mode.

Strengths:

The work presents convincing evidence that BTK inhibitors alter the conformation of regions distal to their binding sites, including those involved in the SH3-kinase interface, the activation loop, and a substrate binding surface between helix aF and helix aG. The findings add to the growing understanding of allosteric effects of kinase inhibitors, and their potential regulation of interactions between kinase and binding proteins.

Comments on the revised version:

The authors have satisfactorily addressed my concerns in their revised manuscript.

eLife Assessment

This useful work provides insight into agonist binding to nicotinic acetylcholine receptors, which is the stimulus for channel activation that regulates muscle contraction at the neuromuscular junction. The authors use in silico methods to explore the transient conformational change from a low to high affinity agonist-bound conformation as occurs during channel opening, but for which structural information is lacking owing to its transient nature. The simulations indicating that ligands flip ~180 degrees in the binding site as it transitions from a low to high affinity bound conformation are solid. A limitation is the approximation of binding energies using Poisson-Boltzmann Surface Area and mismatch between calculated and experimental binding energies for two of the four ligands tested. Nonetheless, this work presents an intriguing picture for the nature of a transient conformational change at the agonist binding site correlated with channel opening.

Reviewer #1 (Public Review):

Summary:

The authors want to understand fundamental steps in ligand binding to muscle nicotinic receptors using computational methods. Overall, although the work provides new information and support for existing models of ligand activation of this receptor type, some limitations in the methods and approach mean that the findings are not as conclusive as hoped.

Strengths:

The strengths include the number of ligands tried, and the comparison to the existing mature analysis of receptor function from the senior author's lab.

Weaknesses:

The weakness are the brevity of the simulations, the concomitant lack of scope of the simulations, the lack of depth in the analysis and the incomplete relation to other relevant work. The free energy methods used seem to lack accuracy - they are only correct for 2 out of 4 ligands.

Reviewer #2 (Public Review):

Summary:

The aim of this manuscript is to use molecular dynamics (MD) simulations to describe the conformational changes of the neurotransmitter binding site of a nicotinic receptor. The study uses a simplified model including the alpha-delta subunit interface of the extracellular domain of the channel and describes the binding of four agonists to observe conformational changes during the weak to strong affinity transition.

Strength:

The 200 ns-long simulations of this model suggest that the agonist rotates about its centre in a 'flip' motion, while loop C 'flops' to restructure the site. The changes appear to be reproduced across simulations and different ligands and are thus a strong point of the study.

Weaknesses:

After carrying out all-atom molecular dynamics, the authors revert to a model of binding using continuum Poisson-Boltzmann, surface area and vibrational entropy. The motivations for and limitations associated with this approximate model for the thermodynamics of binding, rather than using modern atomistic MD free energy methods (that would fully incorporate configurational sampling of the protein, ligand and solvent) could be provided. Despite this, the authors report correlation between their free energy estimates and those inferred from the experiment. This did, however, reveal shortcomings for two of the agonists. The authors mention their trouble getting correlation to experiment for Ebt and Ebx and refer to up to 130% errors in free energy. But this is far worse than a simple proportional error, because -24 Vs -10 kcal/mol is a massive overestimation of free energy, as would be evident if it the authors were to instead to express results in terms of KD values (which would have error exceeding a billion fold). The MD analysis could be improved with better measures of convergence, as well as a more careful discussion of free energy maps as function of identified principal components, as described below. Overall, however, the study has provided useful observations and interpretations of agonist binding that will help understand pentameric ligand-gated ion channel activation.

Reviewer #3 (Public Review):

Summary:

The authors use docking and molecular dynamics (MD) simulations to investigate transient conformations that are otherwise difficult to resolve experimentally. The docking and simulations suggest an interesting series of events whereby agonists initially bind to the low affinity site and then flip 180 degrees as the site contracts to its high affinity conformation. This work will be of interest to the ion channel community and to biophysical studies of pentameric ligand-gated channels.

Strengths:

I find the premise for the simulations to be good, starting with an antagonist bound structure as an estimate of the low affinity binding site conformation, then docking agonists into the site and using MD to allow the site to relax to a higher affinity conformation that is similar to structures in complex with agonists. The predictions are interesting and provide a view into what a transient conformation that is difficult to observe experimentally might be like.

Weaknesses:

A weakness is that the relevance of the initial docked low affinity orientations depend solely on in silco results, for which simulated vs experimental binding energies deviate substantially for two of the four ligands tested. This raises some doubt as to the validity of the simulations. I acknowledge that the calculated binding energies for two of the ligands were closer to experiment, and simulated efficiencies were a good representation of experimental measures, which gives some support to the relevance of the in silico observations. Regardless, some of the reviewers comments regarding the simulation methodology were not seriously addressed.

Reviewer #4 (Public Review):

Summary:

In their revised manuscript "Conformational dynamics of a nicotinic receptor neurotransmitter binding site," Singh and colleagues present molecular docking and dynamics simulations to explore the initial conformational changes associated with agonist binding in the muscle nicotinic acetylcholine receptor, in context with the extensive experimental literature on this system. Their central findings are of a consistently preferred pose for agonists upon initial association with a resting channel, followed by a dramatic rotation of the ligand and contraction of a critical loop over the binding site. Principal component analysis also suggests the formation of an intermediate complex, not yet captured in structural studies. Binding free energy estimates are consistent with the evolution of a higher-affinity complex following agonist binding, with a ligand efficiency notably similar to experimental values. Snapshot comparisons provide a structural rationale for these changes on the basis of pocket volume, hydration, and rearrangement of key residues at the subunit interface.

Strengths:

Docking results are clearly presented and remarkably consistent. Simulations are produced in triplicate with each of four different agonists, providing an informative basis for internal validation. They identify an intriguing transition in ligand pose, not well documented in experimental structures, and potentially applicable to mechanistic or even pharmacological modeling of this and related receptor systems. The paper seems a notable example of integrating quantitative structure-function analysis with systematic computational modeling and simulations, likely applicable to the wider journal audience.

Weaknesses:

The response to the initial review is somewhat disappointing, declining in some places to implement suggested clarifications, and propagating apparent errors in at least one table (Fig 2-source data 1). Some legends (e.g. Fig 2-supplement 4, Fig 3, Fig 4) and figure shadings (e.g. Fig 2-supplement 2, Fig 6-supplement 2) remain unclear. Apparent convergence of agonist-docked simulations towards a desensitized state (l 184) is difficult to interpret in absence of comparative values with other states, systems, etc. In more general concerns, aside from the limited timescales (200 ns) that do not capture global rearrangements, it is not obvious that landscapes constructed on two principal components to identify endpoint and intermediate states (Fig 3) are highly robust or reproducible, nor whether they relate consistently to experimental structures.

Author response:

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

The Editors have assessed your revised submission and rather than issuing a further decision letter we are writing to invite you to make a few small amendments to this version of the paper as listed below.

We added a summary paragraph at the end of the introduction for clarity.

(1) RMSD values in Fig 2-source data 1 (and possibly reflected in Fig 2C) appear to be improbably duplicated, specifically ACh runs 1/2, Ebx runs 1/3, and error values for Ebx vs. ACh.

Thanks for bringing this to our attention. The values are now corrected.

(2) Shaded area in Fig 2-supplement 5D is inaccurate for depicting loop C.

The shaded area now reflects residues in loop C, residues 189-198.

(3) In Fig 2-supplement 4 where an abrupt change in ligand RMSD is implied to represent a cis-trans flip, the accompanying figure showing snapshots misleadingly depicts a different simulation of CCh instead of ACh.

The snapshot was from the correct ACh simulation. It was mislabeled as CCh in the legend, which now stands corrected.

(4) Legend to Fig 3 seems misleading regarding colors in the porcupine plots.

The color pattern indicated in the legend represents the FEL plot and not the porcupine plot. Description about the porcupine plot is not associated with any color.

(5) Some shaded regions in Fig 6-supplement 2 do not correspond to intervals reported in Fig 4-source data 1.

Thanks. This is now corrected to match the table.

Given that some of the above points have remained unaddressed from the prior round of review, the authors should double check that they have addressed any other relevant prior comments not explicitly listed here.

Finally, the revised first results section has removed the explanation as to why the authors opted to simulate a dimer (i.e., affinity being affected only by local perturbations). The authors should consider reincorporating this explanation for readers, as well as adding a reference to Wang et al. 1997 (PMID: 9222901) in regard to lines 116-119.

The revised section now includes an added explanation on why dimer was used in simulations. Gupta et. al., J Gen Physiol. 2017 Jan; 149(1): 85–103 was added, as it includes residues from not just the M1 domain that Wang et al covers, but other TMD regions also.

eLife Assessment

Zhang et al. present important findings that reveal a new role for TET2 in controlling glucose production in the liver, showing that both fasting and a high-fat diet increase TET2 levels, while its absence reduces glucose production. TET2 works with HNF4α to activate the FBP1 gene upon glucagon stimulation, while metformin disrupts TET2-HNF4α interaction, lowering FBP1 levels and improving glucose homeostasis. While the results are solid, more details about the mechanisms and methods are needed to strengthen the study's conclusions.

Reviewer #1 (Public review):

Summary:

Zhang et al. describe a delicate relationship between Tet2 and FBP1 in the regulation of hepatic gluconeogenesis.

Strengths:

The studies are very mechanistic, indicating that this interaction occurs via demethylation of HNF4a. Phosphorylation of HNF4a at ser 313 induced by metformin also controls the interaction between Tet2 and FBP1.

Weaknesses:

The results are briefly described, and oftentimes, the necessary information is not provided to interpret the data. Similarly, the methods section is not well developed to inform the reader about how these experiments were performed. While the findings are interesting, the results section needs to be better developed to increase confidence in the interpretation of the results.

Reviewer #2 (Public review):

Summary:

This study reveals a novel role of TET2 in regulating gluconeogenesis. It shows that fasting and a high-fat diet increase TET2 expression in mice, and TET2 knockout reduces glucose production. The findings highlight that TET2 positively regulates FBP1, a key enzyme in gluconeogenesis, by interacting with HNF4α to demethylate the FBP1 promoter in response to glucagon. Additionally, metformin reduces FBP1 expression by preventing TET2-HNF4α interaction. This identifies an HNF4α-TET2-FBP1 axis as a potential target for T2D treatment.

Strengths:

The authors use several methods in vivo (PTT, GTT, and ITT in fasted and HFD mice; and KO mice) and in vitro (in HepG2 and primary hepatocytes) to support the existence of the HNF4alpha-TET-2-FBP-1 axis in the control of gluconeogenesis. These findings uncovered a previously unknown function of TET2 in gluconeogenesis.

Weaknesses:

Although the authors provide evidence of an HNF4α-TET2-FBP1 axis in the control of gluconeogenesis, which contributes to the therapeutic effect of metformin on T2D, its role in the pathogenesis of T2D is less clear. The mechanisms by which TET2 is up-regulated by glucagon should be more explored.

Author response:

eLife Assessment

Zhang et al. present important findings that reveal a new role for TET2 in controlling glucose production in the liver, showing that both fasting and a high-fat diet increase TET2 levels, while its absence reduces glucose production. TET2 works with HNF4α to activate the FBP1 gene upon glucagon stimulation, while metformin disrupts TET2-HNF4α interaction, lowering FBP1 levels and improving glucose homeostasis. While the results are solid, more details about the mechanisms and methods are needed to strengthen the study's conclusions

Thanks for the positive evaluation and constructive comments, which will significantly improve the quality of the manuscript. We will provide more details about the mechanisms and methods in the revised version.

Reviewer #1 (Public review):

Summary:

Zhang et al. describe a delicate relationship between Tet2 and FBP1 in the regulation of hepatic gluconeogenesis.

Strengths:

The studies are very mechanistic, indicating that this interaction occurs via demethylation of HNF4a. Phosphorylation of HNF4a at ser 313 induced by metformin also controls the interaction between Tet2 and FBP1.

Weaknesses:

The results are briefly described, and oftentimes, the necessary information is not provided to interpret the data. Similarly, the methods section is not well developed to inform the reader about how these experiments were performed. While the findings are interesting, the results section needs to be better developed to increase confidence in the interpretation of the results.

We thank the reviewer for the positive evaluation and constructive comments. There is a factual error in the paragraph of “Strengths”. The comment that “The studies are very mechanistic, indicating that this interaction occurs via demethylation of HNF4a. Phosphorylation of HNF4a at ser 313 induced by metformin also controls the interaction between Tet2 and FBP1.” should be revised as follows: “The studies are very mechanistic, indicating that this interaction occurs via demethylation of FBP1. Phosphorylation of HNF4a at ser 313 induced by metformin also controls the interaction between Tet2 and HNF4a.”

Following reviewer’s suggestions, we will provide all the necessary information in methods section to inform the reader about how these experiments were performed, and improve the description of the results in the revised revision.

Reviewer #2 (Public review):

Summary:

This study reveals a novel role of TET2 in regulating gluconeogenesis. It shows that fasting and a high-fat diet increase TET2 expression in mice, and TET2 knockout reduces glucose production. The findings highlight that TET2 positively regulates FBP1, a key enzyme in gluconeogenesis, by interacting with HNF4α to demethylate the FBP1 promoter in response to glucagon. Additionally, metformin reduces FBP1 expression by preventing TET2-HNF4α interaction. This identifies an HNF4α-TET2-FBP1 axis as a potential target for T2D treatment.

Strengths:

The authors use several methods in vivo (PTT, GTT, and ITT in fasted and HFD mice; and KO mice) and in vitro (in HepG2 and primary hepatocytes) to support the existence of the HNF4alpha-TET-2-FBP-1 axis in the control of gluconeogenesis. These findings uncovered a previously unknown function of TET2 in gluconeogenesis.

Weaknesses:

Although the authors provide evidence of an HNF4α-TET2-FBP1 axis in the control of gluconeogenesis, which contributes to the therapeutic effect of metformin on T2D, its role in the pathogenesis of T2D is less clear. The mechanisms by which TET2 is up-regulated by glucagon should be more explored.

We thank the reviewer for the supports and constructive comments, and agree with the reviewer that the current version mainly focused on the function of HNF4α-TET2-FBP1 axis in the control of gluconeogenesis. We will explore the pathogenesis of T2D and the mechanism how TET2 is up-regulated by glucagon in the revised revision.

Both reviewers made positive comments and we will address all the reviewers’ concerns either by new experiments or clarifications. We thank editors and reviewers for the constructive comments, which will significantly improve the quality of the manuscript.

eLife Assessment

Identifying chromatin interactions with high sensitivity and resolution at the genome-wide scale continues to be technically challenging. This study introduces findings based on the improved MNase-based proximity ligation method, MChIP-C, which enables genome-wide measurement of chromatin interactions at single-nucleosome resolution. The evidence presented in this manuscript is convincing, and the technological advancements will be valuable for the study of 3D genome architecture.

Reviewer #1 (Public review):

The authors presented a new MNase-based proximity ligation method called MChIP-C, allowing for the measurement of protein-mediated chromatin interactions at single-nucleosome resolution on a genome-wide scale. With improved resolution and sensitivity, they explored the spatial connectivity of active promoters and identified the potential candidates for establishing/maintaining E-P interactions. Finally, with published CRISPRi screens, they found that most functionally-verified enhancers do physically interact with their cognate promoters, supporting the enhancer-promoter looping model.

While the study's experimental approach and findings are interesting. However, several issues need to be addressed:

(1) The authors described that "the lack of interaction between experimentally-validated enhancers and their cognate promoters in some studies employing C-methods has raised doubts regarding the classical promoter-enhancer looping model", so it's intriguing to see whether the MChIP-C could indeed detect the E-P interactions which were not identified by C-methods as they mentioned (Benabdallah et al., 2019; Gupta et al., 2017). I agree that they identified more E-P interactions using MChIP-C, but specifically, they should show at least 2-3 cases. It's important since this is the main conclusion the authors want to draw.

(2) The authors compared their data to those of Chen et al. (Chen et al., 2022), who used PLAC-seq with anti-H3K4me3 antibodies in K562 cells and standard Micro-C data previously reported for K562, concluding that "MChIP-C achieves superior sensitivity and resolution compared to C-methods based on standard restriction enzymes.". This is not convincing since they only compared their data to one dataset. More datasets from other cell lines should be included.

(3) The reasons to choose Chen's data (Chen et al., 2022) and CRISPRi screens (Fulco et al., 2019; Gasperini et al., 2019) should be provided since there are so many out there.

(4) The authors identify EP300 histone acetyltransferase and the SWI/SNF remodeling complex as potential candidates for establishing and/or maintaining enhancer-promoter interactions, but not RNA polymerase II, mediator complex, YY1 and BRD4. More explanation is needed for this point since they're previously suggested to be associated with E-P interactions.

(5) The limitations of the method should be discussed.

Reviewer #2 (Public review):

Summary:

Golov et al has performed the capture MChIP-C using H3K4me3 antibody. The new method significantly increases the resolution of Micro-C and can detect the clear interactions which is not well described in the previous HiChIP/PLAC-seq method. Overall, the paper represented a significant technological advance which can be valuable to the 3D genomic field in the future.

The authors have addressed all my concerns and comments.

Reviewer #3 (Public review):

Summary:

This manuscript represents a technology development- specifically an micrococcal nuclease chromatin capture approach, termed MChIP-C to identify promoter centered chromatin interactions at single nucleosome resolution via a specific protein, similar to HiChIP, ChIA-PET, etc.. In general the manuscript is technically well done.

Strengths:

Methods appear to hold promise to improve both the sensitivity and resolution of protein-centered chromatin capture approaches.

Weaknesses:

Downsampling analysis gives a better idea of the strengths of the approach, especially related to individual loci. While this method does outperform other approaches, it remains technically sophisticated and for some labs may not be worth the additional effort for the increase in information. Also, until tested and proven by other groups, it is difficult to know how impactful this approach will be.

Author response:

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

We thank the reviewers for their overall positive evaluation of the manuscript and finding MChIP-C to be a valuable technological advance. To address the reviewer’s helpful comments and recommendations, we performed several additional analyses and improved the text and figures.

Briefly, we extended and clarified the main text and methods, added analyses of interactions at consensus and method-specific CTCF/DHS sites (Figure S3), added additional comparison tracks to other methods in specific loci (Figure 4), added examples of MChIP-C E-P interactions at previously-verified loci (Figure S2a) and added extensive MChIP-C downsampling analysis (Figure S6).

Recommendations for authors:

Reviewer #2 (Recommendations For The Authors:

(1) Provide .HiC and .cool files for the community to explore the data.

We thank the reviewer for this suggestion. We have uploaded both the raw and processed data to GEO. We note that .cool and .hic formats may be less useful for this type of data, since it includes only promoter-based interactions and thus the resulting interaction matrix is extremely sparse at the relevant resolutions. In addition, we provide an online genomic browser for our data.

(2) Provide an R or bioconda package for future data processing.

We thank the reviewer for this suggestion. We have organized and streamlined the relevant code for processing MChIP-C data and it is available as a github repository.

(3) The authors should avoid using "mln" for "million".

We thank the reviewer for this suggestion. We have corrected this in the text.

Reviewer #3 (Recommendations For The Authors):

(1) Figure 2- A handful of sites identified by MChIP-C should be verified by 3C or 4C to validate they are true interactions using an orthogonal approach.

We thank the reviewer for this suggestion. As we show in the current manuscript (and supported by several papers using MNase-based C-methods), C-methods based on restriction enzymes are considerably less sensitive than those based on MNase, so using these methods for anecdotal validation may not be adequate. In addition, it is difficult to extract accurate quantitative measurements from 3C and 4C due to challenges in bias normalization. As a large-scale alternative, we analyzed a set of consensus promoter-CTCF and promoter-DHS interactions identified by all 3 methods (PLAC-seq/Micro-C/MChIP-C; Figure S3). We find that MChIP-C shows clearly superior resolution and sensitivity on these consensus sites. In fact, even for sites which were only called by one of the competing methods, we still see better signal in the MChIP-C data (suggesting that our simplistic MChIP-C peak-calling approach could be improved for further gain). However, as this analysis focuses on “easily detectable” consensus sites, we also emphasize the importance of inspecting interactions which are not detected clearly by alternative methods. To this end, we now show in our manuscript interaction profiles for 11 loci (MYC, PTGER3, CITED2, BTG1, ANTXR2, SEMA7A, LMO2, GATA1, HBG2, VEGFA, MYB), each showing high-resolution MChIP-C interactions which coincide with expected genomic features (p300, CTCF, H3K27ac, known enhancers) and are not clearly observable in Micro-C and PLAC-seq. We also note that the extended overlap of detected MChIP-C interactions with functionally validated enhancers (as measured by CRISPRi) provides an additional large-scale orthogonal validation.

(2) A supplemental table indicating read pair depth, etc, similar to S02, should be added for the datasets used for comparison (HiChIP-etc). Given the age differences between some of the reference data used, it may represent simply an improvement by increasing sequencing depth rather than a true technical advantage.

We thank the reviewer for this suggestion. We have added the sequencing depths of the relevant datasets in the methods section. We also performed extensive downsampling analyses as explained in response to the next point.

(3) I would recommend performing a downsampling analysis to determine at what point the MChIP-C data reaches saturation in terms of the number of reads, with a comparison to the HiChIP reference data. This would allow a more objective measure of the sensitivity of the assays with reference to read depth.

We thank the reviewer for this suggestion. First, we note that downsampling does not affect the high sensitivity and resolution results as shown in aggregate plots (e.g. Figure 2 and Figure S3). However, downsampling can affect individual peak calling. We thus downsampled our data to 50%, approximately matching the number of total informative reads of both PLAC-seq and Micro-C (i.e. ~20M). We also further downsampled our data to 25% and 10%. With respect to prediction of K562 functionally validated enhancer-promoter interactions (Figure S6b), even at 25% downsampling MChIP-C achieves both a higher recall and higher precision than the other methods, with a slightly higher false-positive rate. At 10% sampling, recall is slightly worse than Micro-C and PLAC-seq, but both the precision and false-positive rate are better than the alternatives. With respect to saturation, we plotted the number of unique distal cis read pairs versus the total number of reads (Figure S6c), and find that our MChIP-C data does not yet show saturation. We also show that downsampling our data to 50% maintains  ~80% of the called interactions (Figure S6d).

(4) "our results suggest that MChIP-C achieves superior sensitivity and resolution compared to C-methods based on standard restriction enzymes." The sensitivity claims are supported by Figure 2, but not the resolution claims. This is particularly challenging when using histone marks since they can be broad. To directly compare the resolution of MChIP-C to other approaches such as ChIA-PET or HiChIP CTCF or a similar DNA binding protein is required.

We thank the reviewer for this suggestion. We first note that actually both sensitivity and resolution are relevant for the results shown in Figure 2 and for the signal-to-noise calculations. This is because the low resolution of PLAC-seq peaks can result in very broad peaks that cover the entire area of the interrogated window (5kb on each side), which could seem like low sensitivity. However, we believe that the new Figure S3 may show the higher resolution of MChIP-C more clearly, as do the 11 locus interaction profiles tracks shown in Figure 2, Figure 4 and Figure S2.

Public reviews:

Reviewer #1:

The authors presented a new MNase-based proximity ligation method called MChIP-C, allowing for the measurement of protein-mediated chromatin interactions at single-nucleosome resolution on a genome-wide scale. With improved resolution and sensitivity, they explored the spatial connectivity of active promoters and identified the potential candidates for establishing/maintaining E-P interactions. Finally, with published CRISPRi screens, they found that most functionally verified enhancers do physically interact with their cognate promoters, supporting the enhancer-promoter looping model.

The study's experimental approach and findings are interesting. However, several issues need to be addressed.

(1) The authors described that "the lack of interaction between experimentally-validated enhancers and their cognate promoters in some studies employing C-methods has raised doubts regarding the classical promoter-enhancer looping model", so it's intriguing to see whether the MChIP-C could indeed detect the E-P interactions which were not identified by C-methods as they mentioned (Benabdallah et al., 2019; Gupta et al., 2017). I agree that they identified more E-P interactions using MChIP-C, but specifically, they should show at least 2-3 cases. It's important since this is the main conclusion the authors want to draw.

We thank the reviewer for this suggestion. As we show in the current manuscript (and supported by several papers using MNase-based C-methods), C-methods based on restriction enzymes are considerably less sensitive than those based on MNase, so using these methods for anecdotal validation may not be useful. In addition, it is difficult to extract accurate quantitative measurements from 3C and 4C due to challenges in bias normalization. As a large-scale alternative, we analyzed a set of consensus promoter-CTCF and promoter-DHS interactions identified by all 3 methods (PLAC-seq/Micro-C/MChIP-C; new Figure S3). We find that MChIP-C shows clearly superior resolution and sensitivity on these consensus sites. However, as this analysis focuses on “easily detectable” consensus sites, we also emphasize the importance of inspecting interactions which are not detected clearly by alternative methods. To this end, we now show in our manuscript interaction profiles for 11 loci (MYC, PTGER3, CITED2, BTG1, ANTXR2, SEMA7A, LMO2, GATA1, HBG2, VEGFA, MYB), each showing high-resolution MChIP-C interactions which coincide with expected genomic features (p300, CTCF, H3K27ac, known enhancers) and are not clearly observable in Micro-C and PLAC-seq. We also note that the extended overlap of detected MChIP-C interactions with functionally validated enhancers (as measured by CRISPRi) provides an additional large-scale orthogonal validation.

(2) The authors compared their data to those of Chen et al. (Chen et al., 2022), who used PLAC-seq with anti-H3K4me3 antibodies in K562 cells and standard Micro-C data previously reported for K562, concluding that "MChIP-C achieves superior sensitivity and resolution compared to C-methods based on standard restriction enzymes.". This is not convincing since they only compared their data to one dataset. More datasets from other cell lines should be included.

We thank the reviewer for this suggestion. We would like to clarify that all datasets in the paper are K562 datasets, and this cell line is unique in the availability of CRISPRi screens, PLAC-Seq, Micro-C, and hundreds of ChIP-Seq tracks for it. We would expect datasets from other cell types to have changes in their regulatory interactions, so they would be less adequate for direct comparison. In addition, the general resolution and sensitivity limitations (e.g. due to restriction fragment size) are not dependent on cell type and has been shown in other MNase-based method papers.

(3) The reasons for choosing Chen's data (Chen et al., 2022) and CRISPRi screens (Fulco et al., 2019; Gasperini et al., 2019) should be provided since there are so many out there.

We thank the reviewer for this comment. We selected these CRISPRi screen datasets since they match the cell type (K562) which we used for MChIP-C, and we selected the PLAC-seq data as it is the only PLAC-seq/HiChIP dataset which matches both the cell type (K562) and the antibody (H3K4me3).

(4) The authors identify EP300 histone acetyltransferase and the SWI/SNF remodeling complex as potential candidates for establishing and/or maintaining enhancer-promoter interactions, but not RNA polymerase II, mediator complex, YY1, and BRD4. More explanation is needed for this point since they're previously suggested to be associated with E-P interactions.

We thank the reviewer for this comment. We apologize for this point being unclear: as Figure S5 shows, we actually did identify Pol2, mediator YY1 and BRD4 as predictive features, but P300 and SWI/SNF show somewhat higher predictive power. We have now clarified this in the text.

(5) The limitations of the method should be discussed.

We thank the reviewer for this suggestion. We have now added to the text a discussion of what we view as the current main limitation of the method, namely its low fraction of informative reads.

Reviewer #2:

Summary:

Golov et al performed the capture of MChIP-C using the H3K4me3 antibody. The new method significantly increases the resolution of Micro-C and can detect clear interactions which are not well described in the previous HiChIP/PLAC-seq method. Overall, the paper represents a significant technological advance that can be valuable to the 3D genomic field in the future.

Strengths:

(1) The authors established a novel method to profile the promoter center genomic interactions based on the Micro-C method. Such a method could be very useful to dissect the enhancer promoter interaction which has long been an issue for the popular HiC method.

(2) With the MChIP-C method the authors are able to find new genomic interactions with promoter regions enriched in CTCF. The author has significantly increased the detection sensitivity of such methods as PLAC-seq, Micro-C, and HiChIP.

(3) The authors identified a new type of interaction between the CTCF-less promoter and the CTCF binding site. This particular type of interaction could explain the CTCF's function in regulating gene transcription activity as observed in many studies. I personally think the second stripe model of P-CTCF interaction is more likely as this has been proposed for the super-enhancer stripe model before. The author should also discuss this part of the story more.

Weaknesses:

(1) The data presentation should include the contact heat map. The current data presentation makes it hard for the readers to have a comprehensive view of pair-wise interactions between promoters and the PIR. In particular, these maps may directly give answers to the proposed model of promoter-CTCF interactions by the authors in Figure 3a.

We thank the reviewer for this suggestion. We note that since the data mainly includes promoter-based interactions, the resulting interaction matrix is extremely sparse at the relevant resolutions. Specifically with respect to promoter-CTCF interactions, without a good sampling of the entire interaction matrix it is difficult to confidently distinguish between the two models only based on MChIP-C data, as it would require data about interaction between non-promoter regions and CTCF.

(2) In Fig 3D, there seems a very limited increase of power predicting MChIP-C signal for DHS-promoter pairs beyond the addition of CTCF. This figure could be simplified with fewer factors.

We thank the reviewer for this suggestion. We agree that the last factors do not add predictive power, but we do not think this overly complicates the figure and we prefer to leave these for the reader to evaluate.

(3) The current method seems to have a big fraction of unusable reads. How the authors process the data should be included to allow for future reproduction. Ideally, the authors should generate a package on R or Bioconda for this processing.

We thank the reviewer for this suggestion. We agree that the fraction of informative reads is small with respect to some other methods, and expect future versions of MChIP-C to address this limitation. We have organized and streamlined the relevant code for processing MChIP-C data and it is available as a github repository.

Reviewer #3:

Summary:

This manuscript represents a technological development- specifically a micrococcal nuclease chromatin capture approach, termed MChIP-C to identify promoter-centered chromatin interactions at single nucleosome resolution via a specific protein, similar to HiChIP, ChIA-PET, etc.. In general, the manuscript is technically well done. Two major issues raise concerns that need to be addressed. First, it does not appear that novel chromatin interactions identified by MChIP-C which were missed by other approaches such as HiChIP, were validated. This is central to the argument of "improved" sensitivity, which is one of the key factors to assess sensitivity. Second is the question of resolution. Because the authors focus on a histone mark (H3K4me3) it is unclear whether the resolution of the assay truly exceeds other approaches, especially microC. These two issues are not completely supported by the data provided.

Strengths:

The method appears to hold promise to improve both the sensitivity and resolution of protein-centered chromatin capture approaches.

Weaknesses:

(1) Specific validation experiments to demonstrate the identification of previously missed novel interactions are missing.

We thank the reviewer for this suggestion. Given that such interactions are missed by Micro-C and PLAC-seq, it would not make sense to use these methods for validation. We thus propose that MChIP-C interactions can be validated by their overlap with expected genomic features. To this end, we now show in our manuscript interaction profiles for 11 loci (MYC, PTGER3, CITED2, BTG1, ANTXR2, SEMA7A, LMO2, GATA1, HBG2, VEGFA, MYB), each showing high-resolution MChIP-C interactions which coincide with expected genomic features (p300, CTCF, H3K27ac, known enhancers) and are not clearly observable in Micro-C and PLAC-seq. In addition, the higher overlap of MChIP-C interactions with functionally-validated K562 enhancer-promoter interactions (provided by CRISPRi screens) provides further functional validation for novel MChIP-C interactions.

(2) It is unclear if the resolution is really superior based on the data provided.

We thank the reviewer for this comment. We first note that actually both sensitivity and resolution are relevant for the results shown in Figure 2 and for the signal-to-noise calculations. This is because the low resolution of PLAC-seq peaks can result in very broad peaks that cover the entire area of the interrogated window (5kb on each side), which could seem like low sensitivity. However, we believe that the new Figure S3 may show the higher resolution of MChIP-C more clearly, as do the 11 locus interaction profiles tracks shown in Figure 2, Figure 4 and Figure S2.

(3) It is unclear how much advantage the approach has, especially compared to existing approaches such as HiChIP since sequencing depth as a variable is not adequately addressed.

We thank the reviewer for this comment. First, we note that downsampling does not affect the high sensitivity and resolution results as shown in aggregate plots (e.g. Figure 2 and Figure S3). However, downsampling can affect individual peak calling. We thus downsampled our data to 50%, approximately matching the number of total informative reads of both PLAC-seq and Micro-C (i.e. ~20M). We also further downsampled our data to 25% and 10%. With respect to prediction of K562 functionally validated enhancer-promoter interactions (Figure S6b), even at 25% downsampling MChIP-C achieves both a higher recall and higher precision than the other methods, with a slightly higher false-positive rate. At 10% sampling, recall is slightly worse than Micro-C but both the precision and false-positive rate are better than the alternatives.

Reviewer #3 (Public review):

Summary:

Krwawicz et al., present evidence that expression of DNMTs in E. coli results in (1) introduction of alkylation damage that is repaired by AlkB; (2) confers hypersensitivity to alkylating agents such as MMS (and exacerbated by loss of AlkB); (3) confers hypersensitivity to oxidative stress (H2O2 exposure); (4) results in a modest increase in ROS in the absence of exogenous H2O2 exposure; and (5) results in the production of oxidation products of 5mC, namely 5hmC and 5fC, leading to cellular toxicity. The findings reported here have interesting implications for the concept that such genotoxic and potentially mutagenic consequences of DNMT expression (resulting in 5mC) could be selectively disadvantageous for certain organisms. The other aspect of this work which is important for understanding the biological endpoints of genotoxic stress is the notion that DNA damage per se somehow induces elevated levels of ROS.

Strengths:

The manuscript is well-written, and the experiments have been carefully executed providing data that support the authors' proposed model presented in Fig. 7 (Discussion, sources of DNA damage due to DNMT expression).

Weaknesses:

(1) The authors have established an informative system relying on expression of DNMTs to gauge the effects of such expression and subsequent induction of 3mC and 5mC on cell survival and sensitivity to an alkylating agent (MMS) and exogenous oxidative stress (H2O2 exposure). The authors state (p4) that Fig. 2 shows that "Cells expressing either M.SssI or M.MpeI showed increased sensitivity to MMS treatment compared to WT C2523, supporting the conclusion that the expression of DNMTs increased the levels of alkylation damage." This is a confusing statement and requires revision as Fig. 2 does ALL cells shown in Fig. 2 are expressing DNMTs and have been treated with MMS. It is the absence of AlkB and the expression of DNMTs that that causes the MMS sensitivity.

(2) It would be important to know whether the increased sensitivity (toxicity) to DNMT expression and MMS is also accompanied by substantial increases in mutagenicity. The authors should explain in the text why mutation frequencies were not also measured in these experiments.

(3) Materials and Methods. ROS production monitoring. The "Total Reactive Oxygen Species (ROS) Assay Kit" has not been adequately described. Who is the Vendor? What is the nature of the ROS probes employed in this assay? Which specific ROS correspond to "total ROS"?

(4) The demonstration (Fig. 4) that DNMT expression results in elevated ROS and its further synergistic increase when cells are also exposed to H2O2 is the basis for the authors' discussion of DNA damage-induced increases in cellular ROS. S. cerevisiae does not possess DNMTs/5mC, yet exposure to MMS also results in substantial increases in intracellular ROS (Rowe et al, (2008) Free Rad. Biol. Med. 45:1167-1177. PMC2643028). The authors should be aware of previous studies that have linked DNA damage to intracellular increases in ROS in other organisms and should comment on this in the text.

Author response:

Public Reviews:

Reviewer #1 (Public review):

Summary:

The manuscript proposes that 5mC modifications to DNA, despite being ancient and widespread throughout life, represent a vulnerability, making cells more susceptible to both chemical alkylation and, of more general importance, reactive oxygen species. Sarkies et al take the innovative approach of introducing enzymatic genome-wide cytosine methylation system (DNA methyltransferases, DNMTs) into E. coli, which normally lacks such a system. They provide compelling evidence that the introduction of DNMTs increases the sensitivity of E. coli to chemical alkylation damage. Surprisingly they also show DNMTs increase the sensitivity to reactive oxygen species and propose that the DNMT generated 5mC presents a target for the reactive oxygen species that is especially damaging to cells. Evidence is presented that DNMT activity directly or indirectly produces reactive oxygen species in vivo, which is an important discovery if correct, though the mechanism for this remains obscure.

Strengths:

This work is based on an interesting initial premise, it is well-motivated in the introduction and the manuscript is clearly written. The results themselves are compelling.

We thank the reviewer for their positive response to our study.  We also really appreciate the thoughtful comments raised.  Adding the considerations raised below to the manuscript will considerably strengthen our findings.

Weaknesses:

I am not currently convinced by the principal interpretations and think that other explanations based on known phenomena could account for key results. Specific points below.

(1) As noted in the manuscript, AlkB repairs alkylation damage by direct reversal (DNA strands are not cut). In the absence of AlkB, repair of alklylation damage/modification is likely through BER or other processes involving strand excision and resulting in single stranded DNA. It has previously been shown that 3mC modification from MMS exposure is highly specific to single stranded DNA (PMID:20663718) occurring at ~20,000 times the rate as double stranded DNA. Consequently, the introduction of DNMTs is expected to introduce many methylation adducts genome-wide that will generate single stranded DNA tracts when repaired in an AlkB deficient background (but not in an AlkB WT background), which are then hyper-susceptible to attack by MMS. Such ssDNA tracts are also vulnerable to generating double strand breaks, especially when they contain DNA polymerase stalling adducts such as 3mC. The generation of ssDNA during repair is similarly expected follow the H2O2 or TET based conversion of 5mC to 5hmC or 5fC neither of which can be directly repaired and depend on single strand excision for their removal. The potential importance of ssDNA generation in the experiments has not been considered.

We thank the reviewer for this interesting and insightful suggestion.  Our interpretation of our findings is that a subset of MMS-induced DNA damage, specifically 3mC, overlaps with the damage introduced by DNMTs and this accounts for increased sensitivity to MMS when DNMTs are expressed.  However, the idea that the introduction of 3mC by DNMT actually makes the DNA more liable to damage by MMS, potentially through increasing the level of ssDNA, is also a potential explanation, which could operate in addition to the mechanism that we propose.

(2) The authors emphasise the non-additivity of the MMS + DNMT + alkB experiment but the interpretation of the result is essentially an additive one: that both MMS and DNMT are introducing similar/same damage and AlkB acts to remove it. The non-additivity noted would seem to be more consistent with the ssDNA model proposed in #1. More generally non-additivity would also be seen if the survival to DNA methylation rate is non-linear over the range of the experiment, for example if there is a threshold effect where some repair process is overwhelmed. The linearity of MMS (and H2O2) exposure to survival could be directly tested with a dilution series of MMS (H2O2).

We thank the reviewer for this point.  As in the response to point #1, the reviewer’s hypothesis of increased potency of MMS, potentially through increased ssDNA, downstream of 3mC induction by DNMT, is a good one.  The reviewers’ suggestion would produce a highly non-linear response to MMS treatment in the AlkB mutant in the DNMT background, so we agree that investigating non-linearity over a wider range rather than inferring from the non-additivity of a single point would be useful in evaluating the results so we will add a dose-response curve for DNMT-expressing cells to MMS to the revised version of the manuscript.

(3) The substantial transcriptional changes induced by DNMT expression (Supplemental Figure 4) are a cause for concern and highlight that the ectopic introduction of methylation into a complex system is potentially more confounded than it may at first seem. Though the expression analysis shows bulk transcription properties, my concern is that the disruptive influence of methylation in a system not evolved with it adds not just consistent transcriptional changes but transcriptional heterogeneity between cells which could influence net survival in a stressed environment. In practice I don't think this can be controlled for, possibly quantified by single-cell RNA-seq but that is beyond the reasonable scope of this paper.

We fully agree with the reviewer and, indeed, we are very interested in what is driving the transcriptional changes that we observed.  Work is currently underway in the lab to investigate this further but, as the reviewer suggests, is beyond the scope of this paper.  However, we will include a more extensive comment about the transcriptional changes in the discussion of the revised manuscript.

(4) Figure 4 represents a striking result. From its current presentation it could be inferred that DNMTs are actively promoting ROS generation from H2O2 and also to a lesser extent in the absence of exogenous H2O2. That would be very surprising and a major finding with far-reaching implications. It would need to be further validated, for example by in vitro reconstitution of the reaction and monitoring ROS production. Rather, I think the authors are proposing that some currently undefined, indirect consequence of DNMT activity promotes ROS generation, especially when exogenous H2O2 is available. It would help if this were clarified.

We thank the reviewer for picking this up.  In the current version’s discussion, we raised two possible explanations for why DNMT (even without H2O2) increases the ROS levels.  One idea is direct activity of DNMT, and one is through the product of DNMT activity acting as a platform to generate more ROS from endogenous or exogenous sources.  We argued that direct activity is less likely, exactly as the reviewer points out.  It is, however, not impossible and we agree with the reviewer that, if it were to be the case, it would be a striking result.  In the revised version of the manuscript we will include an experiment to test whether DNMTs can generate ROS in vitro, which may provide preliminary evidence to distinguish between the two hypotheses we raised, and we will also edit the text of the discussion to clarify our reasoning. 

Reviewer #2 (Public review):

5-methylcytosine (5mC) is a key epigenetic mark in DNA and plays a crucial role in regulating gene expression in many eukaryotes including humans. The DNA methyltransferases (DNMTs) that establish and maintain 5mC, are conserved in many species across eukaryotes, including animals, plants, and fungi, mainly in a CpG context. Interestingly, 5mC levels and distributions are quite variable across phylogenies with some species even appearing to have no such DNA methylation.

This interesting and well-written paper discusses the continuation of some of the authors' work published several years ago. In that previous paper, the laboratory demonstrated that DNA methylation pathways coevolved with DNA repair mechanisms, specifically with the alkylation repair system. Specifically, they discovered that DNMTs can introduce alkylation damage into DNA, specifically in the form of 3-methylcytosine (3mC). (This appears to be an error in the DNMT enzymatic mechanism where the generation 3mC as opposed to its preferred product 5-methylcytosine (5mC), is caused by the flipped target cytosine binding to the active site pocket of the DNMT in an inverted orientation.) The presence of 3mC is potentially toxic and can cause replication stress, which this paper suggests may explain the loss of DNA methylation in different species. They further showed that the ALKB2 enzyme plays a crucial role in repairing this alkylation damage, further emphasizing the link between DNA methylation and DNA repair.

The co-evolution of DNMTs with DNA repair mechanisms suggests there can be distinct advantages and disadvantages of DNA methylation to different species which might depend on their environmental niche. In environments that expose species to high levels of DNA damage, high levels of 5mC in their genome may be disadvantageous. This present paper sets out to examine the sensitivity of an organism to genotoxic stresses such as alkylation and oxidation agents as the consequence of DNMT activity. Since such a study in eukaryotes would be complicated by DNA methylation controlling gene regulation, these authors cleverly utilize Escherichia coli (E.coli) and incorporate into it the DNMTs from other bacteria that methylate the cytosines of DNA in a CpG context like that observed in eukaryotes; the active sites of these enzymes are very similar to eukaryotic DNMTs and basically utilize the same catalytic mechanism (also this strain of E.coli does not specifically degrade this methylated DNA) .

The experiments in this paper more than adequately show that E. coli expression of these DNMTs (comparing to the same strain without the DNMTS) do indeed show increased sensitivity to alkylating agents and this sensitivity was even greater than expected when a DNA repair mechanism was inactivated. Moreover, they show that this E. coli expressing this DNMT is more sensitive to oxidizing agents such as H2O2 and has exacerbated sensitivity when a DNA repair glycosylase is inactivated. Both propensities suggest that DNMT activity itself may generate additional genotoxic stress. Intrigued that DNMT expression itself might induce sensitivity to oxidative stress, the experimenters used a fluorescent sensor to show that H2O2 induced reactive oxygen species (ROS) are markedly enhanced with DNMT expression. Importantly, they show that DNMT expression alone gave rise to increased ROS amounts and both H2O2 addition and DNMT expression has greater effect that the linear combination of the two separately. They also carefully checked that the increased sensitivity to H2O2 was not potentially caused by some effect on gene expression of detoxification genes by DNMT expression and activity. Finally, by using mass spectroscopy, they show that DNMT expression led to production of the 5mC oxidation derivatives 5-hydroxymethylcytosine (5hmC) and 5-formylcytosine (5fC) in DNA. 5fC is a substrate for base excision repair while 5hmC is not; more 5fC was observed. Introduction of non-bacterial enzymes that produce 5hmC and 5fC into the DNMT expressing bacteria again showed a greater sensitivity than expected. Remarkedly, in their assay with addition of H2O2, bacteria showed no growth with this dual expression of DNMT and these enzymes.

Overall, the authors conduct well thought-out and simple experiments to show that a disadvantageous consequence of DNMT expression leading to 5mC in DNA is increased sensitivity to oxidative stress as well as alkylating agents.

Again, the paper is well-written and organized. The hypotheses are well-examined by simple experiments. The results are interesting and can impact many scientific areas such as our understanding of evolutionary pressures on an organism by environment to impacting our understanding about how environment of a malignant cell in the human body may lead to cancer.

We thank the reviewer for their response to our study, and value the time taken to produce a public review that will aid readers in understanding the key results of our study. 

Reviewer #3 (Public review):

Summary:

Krwawicz et al., present evidence that expression of DNMTs in E. coli results in (1) introduction of alkylation damage that is repaired by AlkB; (2) confers hypersensitivity to alkylating agents such as MMS (and exacerbated by loss of AlkB); (3) confers hypersensitivity to oxidative stress (H2O2 exposure); (4) results in a modest increase in ROS in the absence of exogenous H2O2 exposure; and (5) results in the production of oxidation products of 5mC, namely 5hmC and 5fC, leading to cellular toxicity. The findings reported here have interesting implications for the concept that such genotoxic and potentially mutagenic consequences of DNMT expression (resulting in 5mC) could be selectively disadvantageous for certain organisms. The other aspect of this work which is important for understanding the biological endpoints of genotoxic stress is the notion that DNA damage per se somehow induces elevated levels of ROS.

Strengths:

The manuscript is well-written, and the experiments have been carefully executed providing data that support the authors' proposed model presented in Fig. 7 (Discussion, sources of DNA damage due to DNMT expression).

Weaknesses:

(1) The authors have established an informative system relying on expression of DNMTs to gauge the effects of such expression and subsequent induction of 3mC and 5mC on cell survival and sensitivity to an alkylating agent (MMS) and exogenous oxidative stress (H2O2 exposure). The authors state (p4) that Fig. 2 shows that "Cells expressing either M.SssI or M.MpeI showed increased sensitivity to MMS treatment compared to WT C2523, supporting the conclusion that the expression of DNMTs increased the levels of alkylation damage." This is a confusing statement and requires revision as Fig. 2 does ALL cells shown in Fig. 2 are expressing DNMTs and have been treated with MMS. It is the absence of AlkB and the expression of DNMTs that that causes the MMS sensitivity.

We thank the reviewer for this and agree that this needs to be clarified with regards to the figure presented and will do so in the revised manuscript. 

(2) It would be important to know whether the increased sensitivity (toxicity) to DNMT expression and MMS is also accompanied by substantial increases in mutagenicity. The authors should explain in the text why mutation frequencies were not also measured in these experiments.

This is an important point because it is not immediately obvious that increased sensitivity would be associated with increased mutagenicity (if, for example, 3mC was never a cause of innacurate DNA repair even in the absence of AlkB).  We will carry out this experiment and include these data in the revised version of the manuscript.  Detailed consideration of the types and sources of mutations is beyond the scope of this manuscript, but we are also working on this and hope to produce data on this in the future. 

(3) Materials and Methods. ROS production monitoring. The "Total Reactive Oxygen Species (ROS) Assay Kit" has not been adequately described. Who is the Vendor? What is the nature of the ROS probes employed in this assay? Which specific ROS correspond to "total ROS"?

The ROS measurement was with a kit from ThermoFisher: https://www.thermofisher.com/order/catalog/product/88-5930-74.  The probe is DCFH-DA.  This is a general ROS sensor that is oxidised by a large number of cellular reactive oxygen species hence we cannot attribute the signal to a single species.  Use of a technique with the potential to more precisely identify the species involved is something we plan to do in future, but is beyond what we can do as part of this study.  We will include a comment to this effect in the revised version of the manuscript.

(4) The demonstration (Fig. 4) that DNMT expression results in elevated ROS and its further synergistic increase when cells are also exposed to H2O2 is the basis for the authors' discussion of DNA damage-induced increases in cellular ROS. S. cerevisiae does not possess DNMTs/5mC, yet exposure to MMS also results in substantial increases in intracellular ROS (Rowe et al, (2008) Free Rad. Biol. Med. 45:1167-1177. PMC2643028). The authors should be aware of previous studies that have linked DNA damage to intracellular increases in ROS in other organisms and should comment on this in the text.

We thank the reviewer for this point.  We note that the increased ROS that we observed occur in the presence of DNMTs alone and in the presence of H2O2, not in the presence of MMS; however, the point that DNA damage in general can promote increased ROS in some circumstances is well taken and we will include a comment on this in the discussion of the revised version.