eLife assessment

This manuscript utilizes a Drosophila explant system and modeling to provide important insights into the mechanism of microtubule aster positioning. Although the intellectual framework of aster positioning has been worked out by the same authors in their previous work, this study provides additional solid evidence to solidify their model.

Reviewer #1 (Public Review):

Summary:<br /> This paper addresses the mechanisms positioning microtubule asters in Drosophila explants. Taking advantage of a genetic mutant, blocking the cell cycle in early embryos, the authors generate embryos with centrosomes detached from nuclei and then study the positioning mechanisms of such asters in explants. They conclude that asters interact via pushing forces. While this is an artificial system, understanding the mechanics of asters positioning, in particular, whether forces are pushing or pulling is an important one.

Strengths:<br /> The major strength of this paper is the series of laser cutting experiments supporting that asters position via pushing forces acting both on the boundary (see below for a relevant comment) and between asters. The combination of imaging, data analysis and mathematical modeling is also powerful.

Weaknesses:<br /> This paper has overlap in the conclusions with a previous paper from the same authors, so its impact is reduced. In Figure 2, the tracking of fluid flows is hard to see and better quantifications/analyses would lead to stronger conclusions. In Figure 4, it is not clear that the acceleration is significant and no statistical test is provided or described, as far as I can tell.

Reviewer #2 (Public Review):

Summary:<br /> The aster, consisting of microtubules, plays important roles in spindle positioning and the determination of the cleavage site in animals. The mechanics of aster movement and positioning have been extensively studied in several cell types. However, there is no unified biophysical model, as different mechanisms appear to predominate in different model systems. In the present manuscript, the authors studied aster positioning mechanics in the Drosophila syncytial embryo, in which short-ranged aster repulsion generates a separation force. Taking advantage of the ex vivo system developed by the group and the fly gnu mutant, in which the nuclear number can be minimized, the authors performed time-lapse observations of single asters and multiple asters in the explant. The observed aster dynamics were interpreted by building a mathematical model dealing with forces. They found that aster dissociation from the boundary depends on the microtubule pushing force. Additionally, laser ablation targeting two separating asters showed that aster-aster separation is also mediated by the microtubule pushing force. Furthermore, they built a simulation model based on the experimental results, which reproduced aster movement in the explant under various conditions. Notably, the actual aster dynamics were best reproduced in the model by including a short-ranged inhibitory term when asters are close to the boundary or each other.

Strengths:<br /> This study reveals a unique aster positioning mechanics in the syncytial embryo explant, which leads to an understanding of the mechanism underlying the positioning of multiple asters associated with nuclei in the embryo. The use of explants enabled accurate measurement of aster motility and, therefore, the construction of a quantitative model. This is a notable achievement.

Weaknesses:<br /> The main conclusion that aster repulsion predominates in this system has already been drawn by the same authors in their recent study (de-Carvalho et al., Development, 2022). Therefore, the conceptual advance in the current study is marginal. The molecular mechanisms underlying aster repulsion remain unexplored since the authors were unable to identify specific factor(s) responsible for aster repulsion in the explant.

Specific suggestions on the original manuscript:<br /> Microtubules should be visualized more clearly (either in live or fixed samples). This is particularly important in Figure 4E and Video 4 (laser ablation experiment to create asymmetric asters).

Comments on the revised manuscript:<br /> Despite my suggestion, the authors did not provide evidence confirming the actual ablation of microtubules in the specified target region. The authors argue, "Given our controls and previous experience, we are confident we are ablating the microtubules." Then, at the very least, the authors should describe (in Materials and Methods) the "controls" they employed and provide a citation to the previous study where proper ablation was validated using the same laser settings. Otherwise, readers might not be convinced of the authors' claim.

eLife assessment

This important report studies the recovery of genome-reduced bacterial cells in laboratory evolution experiments to understand how they regain their fitness. Through the analysis of gene expression and a series of tests, the authors discover distinct molecular changes in the evolved bacterial strains and propose that various mechanisms are employed to offset the effects of a reduced genome. While the findings have intriguing implications for understanding genome evolution, it is crucial to note that the evidence supporting these claims is incomplete due to insufficient experimental tests and statistical analysis.

Reviewer #1 (Public Review):

In this study, the authors explored how the reduced growth fitness, resulting from genome reduction, can be compensated through evolution. They conducted an evolution experiment with a strain of Escherichia coli that carried a reduced genome, over approximately 1,000 generations. The authors carried out sequencing and found no clear genetic signatures of evolution across replicate populations. They carry out transcriptomics and a series of analyses that lead them to conclude that there are divergent mechanisms at play in individual evolutionary lineages. The authors used gene network reconstruction to identify three gene modules functionally differentiated, correlating with changes in growth fitness, genome mutation, and gene expression, respectively, due to evolutionary changes in the reduced genome.

I think that this study addresses an interesting question. Many microbial evolution experiments evolve by loss of function mutations, but presumably, a cell that has already lost so much of its genome needs to find other mechanisms to adapt. Experiments like this have the potential to study "constructive" rather than "destructive" evolution.

At the top of the results, the authors should say what species they're working with and give some background about the nature of the reduced genome. It is important to know what the changes were and especially how much of the genome was deleted. Some insights into the genes that were deleted would also be useful context for understanding the evolution experiment. This could be included in the introduction or results.

Reviewer #2 (Public Review):

This manuscript describes an adaptive laboratory evolution (ALE) study with a previously constructed genome-reduced E. coli. The growth performance of the end-point lineages evolved in M63 medium was comparable to the full-length wild-type level at lower cell densities. Subsequent mutation profiling and RNA-Seq analysis revealed many changes in the genome and transcriptomes of the evolved lineages. The authors did a great deal of analyzing the patterns of evolutionary changes between independent lineages, such as the chromosomal periodicity of transcriptomes, pathway enrichment analysis, weight gene co-expression analysis, and so on. They observed a striking diversity in the molecular characteristics amongst the evolved lineages, which, as they suggest, reflect divergent evolutionary strategies adopted by the genome-reduced organism.

As for the overall quality of the manuscript, I am rather torn. The manuscript leans towards elaborating observed findings, rather than explaining their biological significance. For this reason, readers are left with more questions than answers. For example, fitness assay on reconstituted (single and combinatorial) mutants was not performed, nor was any supporting evidence on the proposed contributions of each mutant provided. This leaves the nature of mutations - be they beneficial, neutral, or deleterious, the presence of epistatic interactions, and the magnitude of fitness contribution, largely elusive. Also, it is difficult to tell whether the RNA-Seq analysis in this study managed to draw biologically meaningful conclusions or instill insight into the nature of genome-reduced bacteria. The analysis primarily highlighted the differences in transcriptome profiles among each lineage based on metrics such as 'DEG counts' and the 'GO enrichment'. However, I could not see any specific implications regarding the biology of the evolved minimal genome drawn. In their concluding remark, 'Multiple evolutionary paths for the reduced genome to improve growth fitness were likely all roads leading to Rome,' the authors observed the first half of the sentence, but the distinctive characteristics of 'all roads' or 'evolutionary paths', which I think should have been the key aspect in this investigation, remains elusive.

Reviewer #3 (Public Review):

Summary:<br /> Studying evolutionary trajectories provides important insight into the genetic architecture of adaptation and provides a potential contribution to evaluating the predictability (or unpredictability) of biological processes involving adaptation. While many papers in the field address adaptation to environmental challenges, the number of studies on how genomic contexts, such as large-scale variation, can impact evolutionary outcomes adaptation is relatively low. This research experimentally evolved a genome-reduced strain for ~1000 generations with 9 replicates and dissected their evolutionary changes. Using the fitness assay of OD measurement, the authors claimed that there is a general trend of increasing growth rate and decreasing carrying capacity, despite a positive correlation among all replicates. The authors also performed genomic and transcriptomic research at the end of experimental evolution, claiming the dissimilarity in the evolution at the molecular level.

Strengths:<br /> The experimental evolution approach with a high number of replicates provides a good way to reveal the generality/diversity of the evolutionary routes.

The assay of fitness, genome, and transcriptome all together allows a more thorough understanding of the evolutionary scenarios and genetic mechanisms.

Weaknesses:<br /> My major concern is the current form of statistical analysis leads to the conclusion that the dissimilarity is not very strong. Adding some more statistical analysis should substantially improve the strength of the manuscript. As mentioned in the Discussion, I understand that there are more available methods to test for generality in experimental evolution but less for diversity. When it is improper to use a canonical statistical test, a test with some simulation and resampling can be useful. For example, I particularly appreciate the analysis done in Figure 2B. An analysis like that should be done more throughout the entire manuscript.

eLife assessment

This important work by Hann et al. advances our understanding of the role of the survival motor neuron (SMN) protein in coordinating pathogenesis of the spinal muscular atrophy (SMA). The authors addressed many concerns raised by the reviewers, providing convincing evidence in terms of skeletal analyses not being able to satisfactorily elucidate SMN regulation of bone development.

Reviewer #1 (Public Review):

Summary:

The manuscript by Hann et. al examines the role of survival motor neuron protein (SMN) in lateral plate mesoderm-derived cells using the Prrx1Cre to elucidate how changing cell-specific SMN levels coordinate aspects of the spinal muscular atrophy (SMA) pathology. SMN has generally been studied in neuronal cells, and this is one of the first insights into non-neuronal cells that may contribute to SMA disease. The authors generated 3 mouse lines: a Prrx1;Smnf/f conditional null mouse, as well as, single and double copy Prrx1;Smnf/f;SMN2 mice carrying either one or two copies of a human SMN2 transgene. First, the bone development and growth of all three were assessed; the conditional null Smn mutation was lethal shortly after birth, while the SMN2 2-copy mutant did not exhibit bone growth phenotypes. Meanwhile, single-copy SMN2 mutant mice showed reduced size and shorter limbs with shorter proliferative and hypertrophic chondrocyte zones. The authors suggested that this was cell autonomous by assessing the expression of extrinsic factors known to modulate proliferation/differentiation of growth plate chondrocytes. After assessing bone phenotypes, the authors transitioned to the assessments of neuromuscular junction (NMJ) phenotypes, since there are documented neuromuscular impairments in SMA and the Prrx1Cre transgene is expressed in muscle-associated fibro-adipogenic progenitors (FAPs). Neonatal NMJ development was unchanged in mutant mice with two copies of SMN2 , but adult single-copy SMN2 mutant mice had abnormal NMJ morphology, altered presynaptic neurotransmission, and problematic nerve terminal structure. Finally, the authors sought to assess the ability to rescue NMJ phenotypes via FAP cell transplantation and showed wild-type FAPs were able to reduce pre/postsynaptic fragmentation and neurofilament varicosities.

Strengths:

The conditional genetic approaches are novel and interestingly demonstrate the potential for chondrocyte and fibro-adipogenic progenitor-specific contributions to the SMA pathology.

The characterizations of the neuromuscular and NMJ phenotypes are relatively strong.

The data strongly suggest a non-neuronal contribution to SMA, which indicates a need for further mechanistic (cellular and molecular) studies to better understand SMA.

Weaknesses:

The skeletal analyses are not rigorous and likely do not get to the core of how SMN regulates bone development.

The overall work is descriptive and lacks convincing mechanisms.

Additional experimentation is likely needed to fully justify the conclusions.

Reviewer #2 (Public Review):

Summary:

Sang-Hyeon et al. laid out a compelling rationale to explore the role of the SMN protein in mesenchymal cells, to determine whether SMN deficiency there could be a pathologic mechanism of SMA. They crossed Smnf7/f7 mice with Prrx1Cre mice to produce SmnΔMPC mice where exon 7 was specifically deleted and thus SMN protein was eliminated in limb mesenchymal progenitor cells (MPCs). To demonstrate gene dosage-dependence of phenotypes, SmnΔMPC mice were crossed with transgenic mice expressing human SMN2 to produce SmnΔMPC mice with different copies of SMN2 (0, 1, or 2). The paper provides genetic evidence that SMN in mesenchymal cells regulates the development of bones and neuromuscular junctions. Genetic data were convincing and revealed novel functions of SMN.

Strengths:

Overall, the paper provided genetic evidence that SMN deficiency in mesenchymal cells caused abnormalities in bones and NMJs, revealing novel functions of SMN and leading to future experiments. As far as genetics is concerned, the data were convincing (except for the rescue experiment, see below); the conclusions are important.

Weaknesses:

The paper seemed to be descriptive in nature and could be improved with more experiments to investigate underlying mechanisms. In addition, some data appeared to be contradicting or difficult to explain. The rescue data were not convincing.

Reviewer #3 (Public Review):

Summary:

SMN expression in non-neuronal cells, particularly in limb mesenchymal progenitors is essential for the proper growth of chondrocytes and the formation of adult NMJ junctions.

Strengths:

The authors show copy numbers of smndelta7 in MPC influence NMJ structure.

Weaknesses:

Functional recovery by FAP transplantation is not complete. Mesenchymal progenitors are heterogeneous, and how heterogeneity influences this study is not clear. Part of the main findings to show the importance of SMN expression in non-neuronal cells is partly published by the same group (Kim et al., JCI Insight 2022). In the study, the authors used Dpp4(+) cells. The difference between the current study and the previous study is not so clear.

Reviewer #1 (Public Review):

People can perform a wide variety of different tasks, and a long-standing question in cognitive neuroscience is how the properties of different tasks are represented in the brain. The authors develop an interesting task that mixes two different sources of difficulty, and find that the brain appears to represent this mixture on a continuum, in the prefrontal areas involved in resolving task difficulty. While these results are interesting and in several ways compelling, they overlap with previous findings and rely on novel statistical analyses that may require further validation.

Strengths<br /> 1. The authors present an interesting and novel task for combining the contributions of stimulus-stimulus and stimulus-response conflict. While this mixture has been measured in the multi-source interference task (MSIT), this task provides a more graded mixture between these two sources of difficulty.

2. The authors do a good job triangulating regions that encoding conflict similarity, looking for the conjunction across several different measures of conflict encoding. These conflict measures use several best-practice approaches towards estimating representational similarity.

3. The authors quantify several salient alternative hypothesis, and systematically distinguish their core results from these alternatives.

4. The question that the authors tackle is important to cognitive control, and they make a solid contribution.

Concerns<br /> 1. The framing of 'infinite possible types of conflict' feels like a strawman. While they might be true across stimuli (which may motivate a feature-based account of control), the authors explore the interpolation between two stimuli. Instead, this work provides confirmatory evidence that task difficulty is represented parametrically (e.g., consistent with literatures like n-back, multiple object tracking, and random dot motion). This parametric encoding is standard in feature-based attention, and it's not clear what the cognitive map framing is contributing.

2. The representations within DLPFC appear to treat 100% Stoop and (to a lesser extent) 100% Simon differently than mixed trials. Within mixed trials, the RDM within this region don't strongly match the predictions of the conflict similarity model. It appears that there may be a more complex relationship encoded in this region.

3. To orthogonalized their variables, the authors need to employ a complex linear mixed effects analysis, with a potential influence of implementation details (e.g., high-level interactions and inflated degrees of freedom).

Reviewer #2 (Public Review):

Summary<br /> This study examines the construct of "cognitive spaces" as they relate to neural coding schemes present in response conflict tasks. The authors use a novel experimental design in which different types of response conflict (spatial Stroop, Simon) are parametrically manipulated. These conflict types are hypothesized to be encoded jointly, within an abstract "cognitive space", in which distances between task conditions depend only on the similarity of conflict types (i.e., where conditions with similar relative proportions of spatial-Stroop versus Simon conflicts are represented with similar activity patterns). Authors contrast such a representational scheme for conflict with several other conceptually distinct schemes, including a domain-general, domain-specific, and two task-specific schemes. The authors conduct a behavioral and fMRI study to test whether prefrontal cortex activity is correlated to one of these coding schemes. Replicating the authors' prior work, this study demonstrates that sequential behavioral adjustments (the congruency sequence effect) are modulated as a function of the similarity between conflict types. In fMRI data, univariate analyses identified activation in left prefrontal and dorsomedial frontal cortex that was modulated by the amount of Stroop or Simon conflict present, and representational similarity analyses that identified coding of conflict similarity, as predicted under the cognitive space model, in right lateral prefrontal cortex.

Strengths

This study addresses an important question regarding how conflict or difficulty might be encoded in the brain within a computationally efficient representational format. Relative to the other models reported in the paper, the evidence in support of the cognitive space model is solid. The ideas postulated by the authors are interesting and valuable ones, worthy of follow-up work that provides additional necessary scrutiny of the cognitive-space account.

Weaknesses

Future, within-subject experiments will be necessary to disentangle the cognitive space model from confounded task variables. A between-subjects manipulation of stimulus orientation/location renders the results difficult to interpret, as the source and spatial scale of the conflict encoding on cortex may differ from more rigorous (and more typical) within-subject manipulations.

Results are also difficult to interpret because Stroop and Simon conflict are confounded with each other. For interpretability, these two sources of conflict need to be manipulated orthogonally, so that each source of conflict (as well as their interaction) could be separately estimated and compared in terms of neural encoding. For example, it is therefore not clear whether the RSA results are due to encoding of only one type of conflict (Stroop or Simon), to a combination of both, and/or to interactive effects.

Finally, the motivation for the use of the term "cognitive space" to describe results is unclear. Evidence for the mere presence of a graded/parametric neural encoding (i.e., the reported conflict RSA effects) would not seem to be sufficient. Indeed, it is discussed in the manuscript that cognitive spaces/maps allow for flexibility through inference and generalization. Future work should therefore focus on linking neural conflict encoding to inference and generalization more directly.

Author Response

Reviewer #1 (Public Review):

[...] Genes expressed in the same direction in lowland individuals facing hypoxia (the plastic state) as what is found in the colonised state are defined as adaptative, while genes with the opposite expression pattern were labelled as maladaptive, using the assumption that the colonised state must represent the result of natural selection. Furthermore, genes could be classified as representing reversion plasticity when the expression pattern differed between the plasticity and colonised states and as reinforcement when they were in the same direction (for example more expressed in the plastic state and the colonised state than in the ancestral state). They found that more genes had a plastic expression pattern that was labelled as maladaptive than adaptive. Therefore, some of the genes have an expression pattern in accordance with what would be predicted based on the plasticity-first hypothesis, while others do not.

Thank you for a precise summary of our work. We appreciate the very encouraging comments recognizing the value of our work. We have addressed concerns from the reviewer in greater detail below.

Q1. As pointed out by the authors themselves, the fact that temperature was not included as a variable, which would make the experimental design much more complex, misses the opportunity to more accurately reflect the environmental conditions that the colonizer individuals face at high altitude. Also pointed out by the authors, the acclimation experiment in hypoxia lasted 4 weeks. It is possible that longer term effects would be identifiable in gene expression in the lowland individuals facing hypoxia on a longer time scale. Furthermore, a sample size of 3 or 4 individuals per group depending on the tissue for wild individuals may miss some of the natural variation present in these populations. Stating that they have a n=7 for the plastic stage and n= 14 for the ancestral and colonized stages refers to the total number of tissue samples and not the number of individuals, according to supplementary table 1.

We shared the same concerns as the reviewer. This is partly because it is quite challenging to bring wild birds into captivity to conduct the hypoxia acclimation experiments. We had to work hard to perform acclimation experiments by taking lowland sparrows in a hypoxic condition for a month. We indeed have recognized the similar set of limitations as the review pointed out and have discussed the limitations in the study, i.e., considering hypoxic condition alone, short time acclimation period, etc. Regarding sample sizes, we have collected cardiac muscle from nine individuals (three individuals for each stage) and flight muscle from 12 individuals (four individuals for each stage). We have clarified this in Supplementary Table 1.

Q2. Finally, I could not find a statement indicating that the lowland individuals placed in hypoxia (plastic stage) were from the same population as the lowland individuals for which transcriptomic data was already available, used as the "ancestral state" group (which themselves seem to come from 3 populations Qinghuangdao, Beijing, and Tianjin, according to supplementary table 2) nor if they were sampled in the same time of year (pre reproduction, during breeding, after, or if they were juveniles, proportion of males or females, etc). These two aspects could affect both gene expression (through neutral or adaptive genetic variation among lowland populations that can affect gene expression, or environmental effects other than hypoxia that differ in these populations' environments or because of their sexes or age). This could potentially also affect the FST analysis done by the authors, which they use to claim that strong selective pressure acted on the expression level of some of the genes in the colonised group.

The reviewer asked how individual tree sparrows used in the transcriptomic analyses were collected. The individuals used for the hypoxia acclimation experiment and represented the ancestral lowland population were collected from the same locality (Beijing) and at the same season (i.e., pre-breeding) of the year. They are all adults and weight approximately 18g. We have clarified this in the Supplementary Table S1 and Methods. We did not distinguish males from females (both sexes look similar) under the assumption that both sexes respond similarly to hypoxia acclimation in their cardiac and flight muscle gene expression.

The Supplementary Table 2 lists the individuals that were used for sequence analyses. These individuals were only used for sequence comparisons but not for the transcriptomic analyses. The population genetic structure analyzed in a previously published study showed that there is no clear genetic divergence within the lowland population (i.e., individuals collected from Beijing, Tianjing and Qinhuangdao) or the highland population (i.e., Gangcha and Qinghai Lake). In addition, there was no clear genetic divergence between the highland and lowland populations (Qu et al. 2020).

Author response image 1.

Figure 1. Population genetic structure of the Eurasian Tree Sparrow (Passer montanus). The genetic structure generated using FRAPPE. The colors in each column represent the contribution from each subcluster (Qu et al. 2020). Yellow, highland population; blue, lowland population.

Q4. Impact of the work There has been work showing that populations adapted to high altitude environments show changes in their hypoxia response that differs from the short-term acclimation response of lowland population of the same species. For example, in humans, see Erzurum et al. 2007 and Peng et al. 2017, where they show that the hypoxia response cascade, which starts with the gene HIF (Hypoxia-Inducible Factor) and includes the EPO gene, which codes for erythropoietin, which in turns activates the production of red blood cell, is LESS activated in high altitude individuals compared to the activation level in lowland individuals (which gives it its name). The present work adds to this body of knowledge showing that the short-term response to hypoxia and the long term one can affect different pathways and that acclimation/plasticity does not always predict what physiological traits will evolve in populations that colonize these environments over many generations and additional selection pressure (UV exposure, temperature, nutrient availability). Altogether, this work provides new information on the evolution of reaction norms of genes associated with the physiological response to one of the main environmental variables that affects almost all animals, oxygen availability. It also provides an interesting model system to study this type of question further in a natural population of homeotherms.

Erzurum, S. C., S. Ghosh, A. J. Janocha, W. Xu, S. Bauer, N. S. Bryan, J. Tejero et al. "Higher blood flow and circulating NO products offset high-altitude hypoxia among Tibetans." Proceedings of the National Academy of Sciences 104, no. 45 (2007): 17593-17598. Peng, Y., C. Cui, Y. He, Ouzhuluobu, H. Zhang, D. Yang, Q. Zhang, Bianbazhuoma, L. Yang, Y. He, et al. 2017. Down-regulation of EPAS1 transcription and genetic adaptation of Tibetans to high-altitude hypoxia. Molecular biology and evolution 34:818-830.

Thank you for highlighting the potential novelty of our work in light of the big field. We found it very interesting to discuss our results (from a bird species) together with similar findings from humans. In the revised version of manuscript, we have discussed short-term acclimation response and long-term adaptive evolution to a high-elevation environment, as well as how our work provides understanding of the relative roles of short-term plasticity and long-term adaptation. We appreciate the two important work pointed out by the reviewer and we have also cited them in the revised version of manuscript.

Reviewer #2 (Public Review):

This is a well-written paper using gene expression in tree sparrow as model traits to distinguish between genetic effects that either reinforce or reverse initial plastic response to environmental changes. Tree sparrow tissues (cardiac and flight muscle) collected in lowland populations subject to hypoxia treatment were profiled for gene expression and compared with previously collected data in 1) highland birds; 2) lowland birds under normal condition to test for differences in directions of changes between initial plastic response and subsequent colonized response. The question is an important and interesting one but I have several major concerns on experimental design and interpretations.

Thank you for a precise summary of our work and constructive comments to improve this study. We have addressed your concerns in greater detail below.

Q1. The datasets consist of two sources of data. The hypoxia treated birds collected from the current study and highland and lowland birds in their respective native environment from a previous study. This creates a complete confounding between the hypoxia treatment and experimental batches that it is impossible to draw any conclusions. The sample size is relatively small. Basically correlation among tens of thousands of genes was computed based on merely 12 or 9 samples.

We appreciate the critical comments from the reviewer. The reviewer raised the concerns about the batch effect from birds collected from the previous study and this study. There is an important detail we didn’t describe in the previous version. All tissues from hypoxia acclimated birds and highland and lowland birds have been collected at the same time (i.e., Qu et al. 2020). RNA library construction and sequencing of these samples were also conducted at the same time, although only the transcriptomic data of lowland and highland tree sparrows were included in Qu et al. (2020). The data from acclimated birds have not been published before.

In the revised version of manuscript, we also compared log-transformed transcript per million (TPM) across all genes and determined the most conserved genes (i.e., coefficient of variance ≤  0.3 and average TPM ≥ 1 for each sample) for the flight and cardiac muscles, respectively (Hao et al. 2023). We compared the median expression levels of these conserved genes and found no difference among the lowland, hypoxia-exposed lowland, and highland tree sparrows (Wilcoxon signed-rank test, P<0.05). As these results suggested little batch effect on the transcriptomic data, we used TPM values to calculate gene expression level and intensity. This methodological detail has been further clarified in the Methods and we also provided a new supplementary Figure (Figure S5) to show the comparative results.

Author response image 2.

The median expression levels of the conserved genes (i.e., coefficient of variance ≤ 0.3 and average TPM ≥ 1 for each sample) did not differ among the lowland, hypoxia-exposed lowland, and highland tree sparrows (Wilcoxon signed-rank test, P<0.05).

The reviewer also raised the issue of sample size. We certainly would have liked to have more individuals in the study, but this was not possible due to the logistical problem of keeping wild bird in a common garden experiment for a long time. We have acknowledged this in the manuscript. In order to mitigate this we have tested the hypothesis of plasticity following by genetic change using two different tissues (cardiac and flight muscles) and two different datasets (co-expressed gene-set and muscle-associated gene-set). As all these analyses show similar results, they indicate that the main conclusion drawn from this study is robust.

Q2. Genes are classified into two classes (reversion and reinforcement) based on arbitrarily chosen thresholds. More "reversion" genes are found and this was taken as evidence reversal is more prominent. However, a trivial explanation is that genes must be expressed within a certain range and those plastic changes simply have more space to reverse direction rather than having any biological reason to do so.

Thank you for the critical comments. There are two questions raised we should like to address them separately. The first concern centered on the issue of arbitrarily chosen thresholds. In our manuscript, we used a range of thresholds, i.e., 50%, 100%, 150% and 200% of change in the gene expression levels of the ancestral lowland tree sparrow to detect genes with reinforcement and reversion plasticity. By this design we wanted to explore the magnitudes of gene expression plasticity (i.e., Ho & Zhang 2018), and whether strength of selection (i.e., genetic variation) changes with the magnitude of gene expression plasticity (i.e., Campbell-Staton et al. 2021).

As the reviewer pointed out, we have now realized that this threshold selection is arbitrarily. We have thus implemented two other categorization schemes to test the robustness of the observation of unequal proportions of genes with reinforcement and reversion plasticity. Specifically, we used a parametric bootstrap procedure as described in Ho & Zhang (2019), which aimed to identify genes resulting from genuine differences rather than random sampling errors. Bootstrap results suggested that genes exhibiting reversing plasticity significantly outnumber those exhibiting reinforcing plasticity, suggesting that our inference of an excess of genes with reversion plasticity is robust to random sampling errors. We have added these analyses to the revised version of manuscript, and provided results in the Figure 2d and Figure 3d.

Author response image 3.

Figure 2a (left) and Figure 2b (right). Frequencies of genes with reinforcement and reversion plasticity (>50%) and their subsets that acquire strong support in the parametric bootstrap analyses (≥ 950/1000).

In addition, we adapted a bin scheme (i.e., 20%, 40% and 60% bin settings along the spectrum of the reinforcement/reversion plasticity). These analyses based on different categorization schemes revealed similar results, and suggested that our inference of an excess of genes with reversion plasticity is robust. We have provided these results in the Supplementary Figure S2 and S4.

Author response image 4.

(A) and Figure S4 (B). Frequencies of genes with reinforcement and reversion plasticity in the flight and cardiac muscle. (A) For genes identified by WGCNA, all comparisons show that there are more genes showing reversion plasticity than those showing reinforcement plasticity for both the flight and cardiac msucles. (B) For genes that associated with muscle phentoypes, all comparisons show that there are more genes showing reversion plasticity than those showing reinforcement plasticity for the flight muscle, while more than 50% of comparisons support an excess of genes with reversion plasticity for the cardiac muscle. Two-tailed binomial test, NS, non-significant; , P < 0.05; , P < 0.01; **, P < 0.001.

The second issue that the reviewer raised is that the plastic changes simply have more space to reverse direction rather than having any biological reason to do so. While a causal reason why there are more genes with expression levels being reversed than those with expression levels being reinforced at the late stages is still contentious, increasingly many studies show that genes expression plasticity at the early stage may be functionally maladapted to novel environment that the species have recently colonized (i.e., lizard, Campbell-Staton et al. 2021; Escherichia coli, yeast, guppies, chickens and babblers, Ho and Zhang 2018; Ho et al. 2020; Kuo et al. 2023). Our comparisons based on the two genesets that are associated with muscle phenotypes corroborated with these previous studies and showed that initial gene expression plasticity may be nonadaptive to the novel environments (i.e., Ghalambor et al. 2015; Ho & Zhang 2018; Ho et al. 2020; Kuo et al. 2023; Campbell-Staton et al. 2021).

Q3. The correlation between plastic change and evolved divergence is an artifact due to the definitions of adaptive versus maladaptive changes. For example, the definition of adaptive changes requires that plastic change and evolved divergence are in the same direction (Figure 3a), so the positive correlation was a result of this selection (Figure 3d).

The reviewer raised an issue that the correlation between plastic change and evolved divergence is an artifact because of the definition of adaptive versus maladaptive changes, for example, Figure 3d. We agree with the reviewer that the correlation analysis is circular because the definition of adaptive and maladaptive plasticity depends on the direction of plastic change matched or opposed that of the colonized tree sparrows. We have thus removed previous Figure 3d-e and related texts from the revised version of manuscript. Meanwhile, we have changed Figure 3a to further clarify the schematic framework.

Author Response:

Reviewer #1 (Public Review):

Despite numerous studies on quinidine therapies for epilepsies associated with GOF mutant variants of Slack, there is no consensus on its utility due to contradictory results. In this study Yuan et al. investigated the role of different sodium selective ion channels on the sensitization of Slack to quinidine block. The study employed electrophysiological approaches, FRET studies, genetically modified proteins and biochemistry to demonstrate that Nav1.6 N- and C-tail interacts with Slack's C-terminus and significantly increases Slack sensitivity to quinidine blockade in vitro and in vivo. This finding inspired the authors to investigate whether they could rescue Slack GOF mutant variants by simply disrupting the interaction between Slack and Nav1.6. They find that the isolated C-terminus of Slack can reduce the current amplitude of Slack GOF mutant variants co-expressed with Nav1.6 in HEK cells and prevent Slack induced seizures in mouse models of epilepsy. This study adds to the growing list of channels that are modulated by protein-protein interactions, and is of great value for future therapeutic strategies.

I have a few comments with regard to how Nav1.6 sensitize Slack to block by quinidine.

(1) It is not clear to me if the Slack induced current amplitude varies depending on the specific Nav subtype. To this end, it would be valuable to test if Slack open probability is affected by the presence of specific Nav subtypes. Nav induced differences in Slack current amplitude and open probability could explain why individual Nav subtypes show varied ability to sensitize Slack to quinidine blockade.

We appreciate the reviewer for raising this point. In order to address whether the whole-cell current amplitudes of Slack varies depending on the specific NaV subtype, we examined Slack current amplitudes upon co-expression of Slack with specific NaV subtypes in HEK293 cells. The results have shown that there are no significant differences in Slack current amplitudes upon co-expression of Slack with different NaV channel subtypes (Author response image 1), suggesting whole-cell Slack current amplitudes cannot explain the varied ability of NaV subtypes to sensitize Slack to quinidine blockade. To investigate the effect of different NaV channel subtypes on Slack open probability, we will perform the single-channel recordings in the future studies.

Author response image 1.

The amplitudes of Slack currents upon co-expression of Slack with specific NaV subtypes in HEK293 cells. ns, p > 0.05, one-way ANOVA followed by Bonferroni’s post hoc test.

(2) It has previously been shown that INaP (persistent sodium current) is important for inducing Slack currents. Here the authors show that INaT (transient sodium current) of Nav1.6 is necessary for the sensitization of Slack to quinidine block whereas INaP surprisingly has no effect. The authors then show that the N-tail together with C-tail of Nav1.6 can induce same effect on Slack as full-length Nav1.6 in presence of high intracellular concentrations of sodium. However, it is not clear to me how the isolated N- and C-tail of Nav1.6 can induce sensitization of Slack to quinidine by interacting with C-terminus of Slack, while sensitization also is dependant on INaT. The authors speculate on different slack open conformation, but one could speculate if there is a missing link, such as an un-identified additional interacting protein that causes the coupling.

We fully agree the importance of investigating the detailed mechanism underlying the sensitization of Slack to quinidine blockade mediated by the N- and C-termini of NaV1.6. Regarding the possibility of additional interacting proteins (“missing link”) that mediate the coupling between Slack and NaV1.6, our GST-pull down assays involving Slack and the N- and C-termini of NaV1.6 (Fig. S7) suggest a direct interaction between Slack and NaV1.6 channels. This finding leads us to consider the possibility of additional interacting proteins might be excluded. In order to further address these questions, we plan to employ structural biological methods, such as cryo-electron microscopy (cryo-EM).

Reviewer #2 (Public Review):

This is a very interesting paper about the coupling of Slack and Nav1.6 and the insight this brings to the effects of quinidine to treat some epilepsy syndromes.

Slack is a sodium-activated potassium channel that is important to hyperpolarization of neurons after an action potential. Slack is encoded by KNCT1 which has mutations in some epilepsy syndromes. These types of epilepsy are treated with quinidine but this is an atypical antiseizure drug, not used for other types of epilepsy. For sufficient sodium to activate Slack, Slack needs to be close to a channel that allows robust sodium entry, like Na channels or AMPA receptors. but more mechanistic information is not available. Of particular interest to the authors is what allows quinidine to be effective in reducing Slack.

In the manuscript, the authors show that Nav, not AMPA receptors are responsible for Slack activation, at least in cultured neurons (HeK293, primary cortical neurons). Most of the paper focuses on the evidence that Nav1.6 promotes Slack sensitivity to quinidine.

(1) The paper is very well written although there are reservations about the use of non-neuronal cells or cultured primary neurons rather than a more intact system.

We appreciate the reviewer's positive evaluation of our work. We acknowledge that utilizing a more intact system would provide valuable insights into the inhibitory effect of quinidine on Slack-NaV1.6. However, there are certain challenges associated with studying Slack currents in their entirety.

First, in our experiments, isolating Slack currents from Na+-activated K+ currents in an intact system is challenging as selective inhibitors for Slick are currently unavailable. To address this, we propose using Slick gene knockout mice to specifically measure Slack currents under physiological conditions in the future investigations. Second, we have observed that the interaction between Slack and NaV1.6 primarily occurs at the axon initial segment of neurons. This poses a difficulty when using brain slices for measurements, as employing the whole-cell voltage-clamp technique to assess Slack at the axon initial segment may introduce systemic errors.

We believe that testing the pharmacological effects of quinidine on Slack-NaV1.6 in primary neurons remains the optimal approach. Although non-neuronal cells or cultured primary neurons may not fully replicate the complexity of an intact system, they still provide valuable insights into the interactions between Slack and NaV1.6, and the effects of quinidine.

(2) I also have questions about the figures.

We will make the necessary modifications and clarifications based on the reviewer's comments:

(3) Finally, riluzole is not a selective drug, so the limitations of this drug should be discussed.

We thank the reviewer for raising this point. We will discuss the limitations of riluzole in our revised version of the manuscript.

(4) On a minor point, the authors use the term in vivo but there are no in vivo experiments.

We thanks the reviewer for raising this point. In our experiments, although we did not conduct experiments directly in living organisms, our results demonstrated the co-immunoprecipitation of NaV1.6 with Slack in homogenates from mouse cortical and hippocampal tissues (Fig. 3C). This result may support that the interaction between Slack and NaV1.6 occurs in vivo.

Reviewer #3 (Public Review):

Yuan et al., set out to examine the role of functional and structural interaction between Slack and NaVs on the Slack sensitivity to quinidine. Through pharmacological and genetic means they identify NaV1.6 as the privileged NaV isoform in sensitizing Slack to quinidine. Through biochemical assays, they then determine that the C-terminus of Slack physically interacts with the N- and C-termini of NaV1.6. Using the information gleaned from the in vitro experiments the authors then show that virally-mediated transduction of Slack's C-terminus lessens the extent of SlackG269S-induced seizures. These data uncover a previously unrecognized interaction between a sodium and a potassium channel, which contributes to the latter's sensitivity to quinidine.

The conclusions of this paper are mostly well supported by data, but some aspects of functional and structural studies in vivo as well as physically interaction need to be clarified and extended.

(1) Immunolabeling of the hippocampus CA1 suggests sodium channels as well as Slack colocalization with AnkG (Fig 3A). Proximity ligation assay for NaV1.6 and Slack or a super-resolution microscopy approach would be needed to increase confidence in the presented colocalization results. Furthermore, coimmunoprecipitation studies on the membrane fraction would bolster the functional relevance of NaV1.6-Slac interaction on the cell surface.

We thank the reviewer for good suggestions. We acknowledge that employing proximity ligation assay and high-resolution techniques would significantly enhance our understanding of the localization of the Slack-NaV1.6 coupling.

At present, the technical capabilities available in our laboratory and institution do not support high-resolution testing. However, we are enthusiastic about exploring potential collaborations to address these questions in the future. Furthermore, we fully recognize the importance of conducting co-immunoprecipitation (Co-IP) assays from membrane fractions. While we have already completed Co-IP assays for total protein and quantified the FRET efficiency values between Slack and NaV1.6 in the membrane region, the Co-IP assays on membrane fractions will be conducted in our future investigations.

(2) Although hippocampal slices from Scn8a+/- were used for studies in Fig. S8, it is not clear whether Scn8a-/- or Scn8a+/- tissue was used in other studies (Fig 1J & 1K). It will be important to clarify whether genetic manipulation of NaV1.6 expression (Fig. 1K) has an impact on sodium-activated potassium current, level of surface Slack expression, or that of NaV1.6 near Slack.

We thank the reviewer for pointing this out. In Fig. 1G,J,K, primary cortical neurons from homozygous NaV1.6 knockout (Scn8a-/-) mice were used. We will clarify this information in the revised manuscript. In terms of the effects of genetic manipulation of NaV1.6 expression on IKNa and surface Slack expression, we compared the amplitudes of IKNa measured from homozygous NaV1.6 knockout (NaV1.6-KO) neurons and wild-type (WT) neurons. The results showed that homozygous knockout of NaV1.6 does not alter the amplitudes of IKNa (Author response image 2). The level of surface Slack expression will be tested further.

Author response image 2.

The amplitudes of IKNa in WT and NaV1.6-KO neurons (data from manuscript Fig. 1K). ns, p > 0.05, unpaired two-tailed Student’s t test.

(3) Did the epilepsy-related Slack mutations have an impact on NaV1.6-mediated sodium current?

We thank the reviewer’s question. We examined the amplitudes of NaV1.6 sodium current upon expression alone or co-expression of NaV1.6 with epilepsy-related Slack mutations (K629N, R950Q, K985N). The results showed that the tested epilepsy-related Slack mutations do not alter the amplitudes of NaV1.6 sodium current (Author response image 3).

Author response image 3.

The amplitudes of NaV1.6 sodium currents upon co-expression of NaV1.6 with epilepsy-related Slack mutant variants (SlackK629N, SlackR950Q, and SlackK985N). ns, p>0.05, one-way ANOVA followed by Bonferroni’s post hoc test.

4) Showing the impact of quinidine on persistent sodium current in neurons and on NaV1.6-expressing cells would further increase confidence in the role of persistent sodium current on sensitivity of Slack to quinidine.

We appreciate the reviewer’s question. Previous studies have shown that quinidine can inhibit persistent sodium currents at low concentrations1. In our experiments, blocking persistent sodium currents by application of riluzole in the bath solution showed no significant effects on the sensitivity of Slack to quinidine blockade upon co-expression of Slack with NaV1.6 (Fig. 2F,H). This result suggested that persistent sodium currents were not involved in the sensitization of Slack to quinidine blockade.

1. Ju YK, Saint DA, Gage PW. Effects of lignocaine and quinidine on the persistent sodium current in rat ventricular myocytes. Br J Pharmacol. Oct 1992; 107(2):311-6. doi:10.1111/j.1476-5381.1992.tb12743.x

Author Response:

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

Reviewer #1 (Public Review):

Chan et al. tried identifying the binding sites or pockets for the KCNQ1-KCNE1 activator mefenamic acid. Because the KCNQ1-KCNE1 channel is responsible for cardiac repolarization, genetic impairment of either the KCNQ1 or KCNE1 gene can cause cardiac arrhythmia. Therefore, the development of activators without side effects is highly demanded. Because the binding of mefenamic acid requires both KCNQ1 and KCNE1 subunits, the authors performed drug docking simulation by using KCNQ1-KCNE3 structural model (because this is the only available KCNQ1-KCNE structure) with substitution of the extracellular five amino acids (R53-Y58) into D39-A44 of KCNE1. That could be a limitation of the work because the binding mode of KCNE1 might differ from that of KCNE3. Still, they successfully identified some critical amino acid residues, including W323 of KCNQ1 and K41 and A44 of KCNE1. They subsequently tested these identified amino acid residues by analyzing the point mutants and confirmed that they attenuated the effects of the activator. They also examined another activator, yet structurally different DIDS, and reported that DIDS and mefenamic acid share the binding pocket, and they concluded that the extracellular region composed of S1, S6, and KCNE1 is a generic binding pocket for the IKS activators.

The data are solid and well support their conclusions, although there are a few concerns regarding the choice of mutants for analysis and data presentation.

Other comments:

1. One of the limitations of this work is that they used psKCNE1 (mostly KCNE3), not real KCNE1, as written above. It is also noted that KCNQ1-KCNE3 is in the open state. Unbinding may be facilitated in the closed state, although evaluating that in the current work is difficult.

We agree that it is difficult to evaluate the role of unbinding from our model. Our data showing that longer interpulse intervals have a normalizing effect on the GV curve (Figure 3-figure supplement 2) could be interpreted to suggest that unbinding occurs in the closed state. Alternatively, the slowing of deactivation caused by S1-S6 interactions and facilitated by the activators may effectively be exceeded at the longer interpulse intervals.

1. According to Figure 2-figure supplement 2, some amino acid residues (S298 and A300) of the turret might be involved in the binding of mefenamic acid. On the other hand, Q147 showing a comparable delta G value to S298 and A300 was picked for mutant analysis. What are the criteria for the following electrophysiological study?

EP experiments interrogated selected residues with significant contributions to mefenamic acid and DIDs coordination as revealed by the MM/GBSA and MM/PBSA methods. A300 was identified as potentially important. We did attempt A300C but were never able to get adequate expression for analysis.

1. It is an interesting speculation that K41C and W323A stabilize the extracellular region of KCNE1 and might increase the binding efficacy of mefenamic acid. Is it also the case for DIDS? K41 may not be critical for DIDS, however.

Yes, we found K41 was not critical to the binding/action of DIDS compared to MEF. In electrophysiological experiments with the K41C mutation, DIDS induced a leftward GV shift (~ -25 mV) whereas the normalized response was statistically non-significant. In MD simulation studies, we observed detachment of DIDS from K41C-Iks only in 3 runs out of 8 simulations. This is in contrast to Mef, where the drug left the binding site of K41C-Iks complex in all simulations.

1. Same to #2, why was the pore turret (S298-A300) not examined in Figure 7?

Again, we attempted A300C but could not get high enough expression.

Reviewer #3 (Public Review):

Weaknesses:

1. The computational aspect of the work is rather under-sampled - Figure 2 and Figure 4. The lack of quantitative analysis on the molecular dynamic simulation studies is striking, as only a video of a single representative replica is being shown per mutant/drug. Given that the simulations shown in the video are extremely short; some video only lasts up to 80 ns. Could the author provide longer simulations in each simulation condition (at least to 500 ns or until a stable binding pose is obtained in case the ligand does not leave the binding site), at least with three replicates per each condition? If not able to extend the length of the simulations due to resources issue, then further quantitative analysis should be conducted to prove that all simulations are converged and are sufficient. Please see the rest of the quantitative analysis in other comments.

We provide more quantitative analysis for the existing MD simulations and ran five additional simulations with 500 ns duration by embedding the channel in a POPC lipid membrane. For the new MD simulations, we used a different force field in order to minimize ambiguity related to force fields as well. Analysis of these data has led to new data and supplemental figures regarding RMSD of ligands during the simulations (Figure 4-figure supplement 1 and Figure 6-figure supplement 3), clustering of MD trajectories based on Mef conformation (Figure 2-figure supplement 3 and Figure 6 -figure supplement 2), H-bond formation over the simulations (Figure 2-figure supplement 4 and Figure 6-figure supplement 1). We have edited the manuscript to include this new information where appropriate.

1. Given that the protein is a tetramer, at least 12 datasets could have been curated to improve the statistic. It was also unclear how frequently the frames from the simulations were taken in order to calculate the PBSA/GBSA.

By using one ligand for each ps-IKs channel complex we tried to keep the molecular system and corresponding analysis as simple as was possible. Our initial results have shown that 4D docking and subsequent MD simulations with only one ligand bound to ps-IKs was complicated enough. Our attempts to dock 4 ligands simultaneously and analyze the properties of such a system were ineffective due to difficulties in: i) obtaining stable complexes during conformational sampling and 4D docking procedures, since the ligand interaction covers a region including three protein chains with dynamic properties, ii) possible changes of receptor conformation properties at three other subunits when one ligand is already occupying its site, iii) marked diversity of the binding poses of the ligand as cluster analysis of ligand-channels complex shows (Figure 2-figure supplement 3).

We have added a line in the methods to clarify the use of only one ligand per channel complex in simulations.

In order to calculate MMPBSA/MMGBSA we used a frame every 0.3 ns throughout the 300 ns simulation (1000 frames/simulation) or during the time the ligand remained bound. We have clarified this in the Methods.

1. The lack of labels on several structures is rather unhelpful (Figure 2B, 2C, 4B). The lack of clarity of the interaction map in Figures 2D and 6A.

We updated figures considering the reviewer's comments and added labels. For 2D interaction maps, we provided additional information in figure legends to improve clarity.

1. The RMSF analysis is rather unclear and unlabelled thoroughly. In fact, I still don't quite understand why n = 3, given that the protein is a tetramer. If only one out of four were docked and studied, this rationale needs to be explained and accounted for in the manuscript.

The rationale of conducting MD simulations with one ligand bound to IKs is explained in response to point 2 of the reviewer’s comments.

RMSF analysis in Figure 4C-E was calculated using the chain to which Mef was docked but after Mef had left the binding site. Details were added to the methods.

1. For the condition that the ligands suppose to leave the site (K42C for Mef and Y46A for DIDS), can you please provide simulations at a sufficient length of time to show that ligand left the site over three replicates? Given that the protein is a tetramer, I would be expecting three replicates of data to have four data points from each subunit. I would be expecting distance calculation or RMSD of the ligand position in the binding site to be calculated either as a time series or as a distribution plot to show the difference between each mutant in the ligand stability within the binding pocket. I would expect all the videos to be translatable to certain quantitative measures.

We have shown in the manuscript that the MEF molecule detaches from the K41C/IKs channel complex in all three simulations (at 25 ns, 70 ns and 20 ns, Table. 4). Similarly, the ligand left the site in all five new 500 ns duration simulations. We did not provide simualtions for Y46A, but Y46C left the binding site in 4 of 5 500 ns simulations and changed binding pose in the other.

Difficulties encountered upon extending the docking and MD simulations for 4 receptor sites of the channel complex is discussed in our response to point # 2 of the reviewer.

1. Given that K41 (Mef) and Y46 are very important in the coordination, could you calculate the frequency at which such residues form hydrogen bonds with the drug in the binding site? Can you also calculate the occupancy or the frequency of contact that the residues are making to the ligand (close 4-angstrom proximity etc.) and show whether those agree with the ligand interaction map obtained from ICM pro in Figure 2D?

We thank the reviewer for the suggestion to analyze the H-bond contribution to ligand dynamics in the binding site. In the plots shown in Figure 2-figure supplement 4 and Figure 6-figure supplement 1, we now provide detailed information about the dynamics of the H-bond formation between the ligand and the channel-complex throughout simulations. In addition, we have quantified this and have added these numbers to a table (Table 2) and in the text of the results.

1. Given that the author claims that both molecules share the same binding site and the mode of ligand binding seems to be very dynamic, I would expect the authors to show the distribution of the position of ligand, or space, or volume occupied by the ligand throughout multiple repeats of simulations, over sufficient sampling time that both ligand samples the same conformational space in the binding pocket. This will prove the point in the discussion - Line 463-464. "We can imagine a dynamic complex... bind/unbind from Its at a high frequency".

To support our statement regarding a dynamic complex we analyzed longer MD simulations and clustered trajectories, from this an average conformation from each cluster was extracted and provided as supplementary information which shows the different binding modes for Mef (Figure 2-figure supplement 3). DIDS was more stable in MD simulations and though there were also several clusters, they were similar enough that when using the same cut-off distance as for mefenamic acid, they could be grouped into one cluster. (Note the scale differences on dendrogram between Figure 2-figure supplement 3 and Figure 6-figure supplement 2).

1. I would expect the authors to explain the significance and the importance of the PBSA/GBSA analysis as they are not reporting the same energy in several cases, especially K41 in Figure 2 - figure supplement 2. It was also questionable that Y46, which seems to have high binding energy, show no difference in the EPhys works in figure 3. These need to be commented on.

Several studies indicate that G values calculated using MM/PBSA and MM/GBSA methods may vary. Some studies report marked differences and the reasons for such a discrepancy is thoroughly discussed in a review by Genheden and Ryde (PMID: 25835573). Therefore, we used both methods to be sure that key residues contributing to ligand binding identified with one method appear in the list of residues for which the calculations are done with the other method.

Y46C which showed only a slightly less favorable binding energy and did not unbind during 300 ns simulations, unbound, or changed pose in 4 out of 5 of the longer simulations in the presence of a lipid membrane (Figure 4-figure supplement 1). The discrepancy between electrophysiological and MD data is commented in the manuscript (pages 12-13).

1. Can the author prove that the PBSA/GBSA analysis yielded the same average free energy throughout the MD simulation? This should be the case when the simulations are converged. The author may takes the snapshots from the first ten ns, conduct the analysis and take the average, then 50, then 100, then 250 and 500 ns. The author then hopefully expects that as the simulations get longer, the system has reached equilibrium, and the free energy obtained per residue corresponds to the ensemble average.

As we mention in the manuscript, MEF- channel interactions are quite dynamic and vary even from simulation to simulation. The frequent change of the binding pose of the ligands observed during simulations (represented in Figure 2 - figure supplement 3 as clusters) is a clear reflection of such a dynamic process. Therefore, we do not expect the same average energy throughout the simulation but we do expect that G values stands above the background for key residues, which was generally the case (Figure 2 - figure supplement 2 and Figure 6.)

1. The phrase "Lowest interaction free energy for residues in ps-KCNE1 and selected KCNQ1 domains are shown as enlarged panels (n=3 for each point)" needs further explanation. Is this from different frames? I would rather see this PBSA and GBSA calculated on every frame of the simulations, maybe at the one ns increment across 500 ns simulations, in 4 binding sites, in 3 replicas, and these are being plotted as the distribution instead of plotting the smallest number. Can you show each data point corresponding to n = 3?

The MMPBSA/MMGBSA was calculated for 1000 frames across 3x300 ns simulations with 0.3 ns sampling interval, together 3000 frames, shown in Figure 2-figure supplement 2 and includes error bars to show the differences across runs. We have updated the legend for greater clarity.

1. I cannot wrap my head around what you are trying to show in Figure 2B. This could be genuinely improved with better labelling. Can you explain whether this predicted binding pose for Mef in the figure is taken from the docking or from the last frame of the simulation? Given that the binding mode seems to be quite dynamic, a single snapshot might not be very helpful. I suggest a figure describing different modes of binding. Figure 2B should be combined with figure 2C as both are not very informative.

We have updated Figure 2B with better labelling and added a new figure showing the different modes of binding (Figure 2-figure supplement 3).

1. Similar to the comment above, but for Figure 4B. I do not understand the argument. If the author is trying to say that the pocket is closed after Mef is removed - then can you show, using MD simulation, that the pocket is openable in an apo to the state where Mef can bind? I am aware that the open pocket is generated through batches of structures through conformational sampling - but as the region is supposed to be disordered, can you show that there is a possibility of the allosteric or cryptic pocket being opened in the simulations? If not, can you show that the structure with the open pocket, when the ligand is removed, is capable of collapsing down to the structure similar to the cryo-EM structure? If none of the above work, the author might consider using PocketMiner tools to find an allosteric pocket (https://doi.org/10.1038/s41467-023-36699-3) and see a possibility that the pocket exists.

Please see the attached screenshot which depicts the binding pocket from the longest run we performed (1250 ns) before drug detachment (grey superimposed structures) and after (red superimposed structures). Mefenamic acid is represented as licorice and colored green. Snapshots for superimposition were collected every 10 ns. As can be seen in the figure, when the drug leaves the binding site (after 500 ns, structures colored red), the N-terminal residue of psKCNE1, W323, and other residues that form the pocket shift toward the binding site, overlapping with where Mefenamic acid once resided. The surface structure in Figure 4B shows this collapse.

Author response image 1.

In the manuscript, we propose that drug binding occurs by the mechanism that could be best described by induced fit models, which state that the formation of the firm complexes (channel-Mef complex) is a result of multiple-states conformational adjustments of the bimolecular interaction. These interactions do not necessarily need to have large interfaces at the initial phase. This seems to be the case in Mef with IKS interactions, since we could not identify a pocket of appropriate size either using PocketMiner software suggested by the reviewer or with PocketFinder tool of ICM-pro software.

1. Figure 4C - again, can you show the RMSF analysis of all four subunits leading to 12 data points? If it is too messy to plot, can you plot a mean with a standard deviation? I would say that a 1-1.5 angstroms increase in the RMSF is not a "markedly increased", as stated on line 280. I would also encourage the authors to label whether the RMSF is calculated from the backbone, side-chain or C-alpha atoms and, ideally, compare them to see where the dynamical properties are coming from.

Please see the answer to comment #4. We agree that the changes are not so dramatic and modified the text accordingly. RMSD was calculated for backbone atom to compare residues with different side chains, a note of this is now in the methods and statistical significance of ps-IKs vs K41C, W323A and Y46C is indicated in Figures 4C-4E.

1. In the discussion - Lines 464-467. "Slowed deactivation of the S1/KCNE1/Pore domain/drug complex... By stabilising the activated complex. MD simulation suggests the latter is most likely the case." Can you point out explicitly where this has been proven? If the drug really stabilised the activated complex, can you show which intermolecular interaction within E1/S1/Pore has the drug broken and re-form to strengthen the complex formation? The authors have not disproven the point on steric hindrance either. Can this be disproved by further quantitative analysis of existing unbiased equilibrium simulations?

The stabilization of S1/KCNE1/Pore by drugs does not necessarily have to involve a creation of new contacts between protein parts or breakage of interfaces between them. The stabilization of activated complexes by drugs may occur when the drug simultaneously binds to both moveable parts of the channel, such as voltage sensor(s) or upper KCNE1 region, and static region(s) of the channel, such as the pore domain. We have changed the corresponding text for better clarity.

1. Figure 4D - Can you show this RMSF analysis for all mutants you conducted in this study, such as Y46C? Can you explain the difference in F dynamics in the KCNE3 for both Figure 4C and 4D?

We now show the RMSF for K41C, W323A and Y46C in Figure 4C-E. We speculate that K41 (magenta) and W323 (yellow), given their location at the lipid interface (see Author response image 1), may be important stabilizing residues for the KCNE N-terminus, whereas Y46 (green) which is further down the TMD has less of an impact.

Author response image 2.

1. Line 477: the author suggested that K41 and Mef may stabilise the protein-protein interface at the external region of the channel complex. Can you prove that through the change in protein-protein interaction, contact is made over time on the existing MD trajectories, whether they are broken or formed? The interface from which residues help to form and stabilise the contact? If this is just a hypothesis for future study, then this has to be stated clearly.

It is known that crosslinking of several residues of external E1 with the external pore residues dramatically stabilizes voltage-sensors of KCNQ1/KCNE1 complex in the up-state conformation. This prevents movable protein regions in the voltage-sensors returning to their initial positions upon depolarization, locking the channel in an open state. We suggest that MEF may restrain the backward movement of voltage-sensors in a similar way that stabilizes open conformation of the channel. The stabilization of the voltage sensor domain through MEF occurs due to contacts of the drug with both static (pore domain) and dynamic protein parts (voltage-sensors and external KCNE1 regions). We have changed the corresponding part of the text.

1. The author stated on lines 305-307 that "DIDS is stabilised by its hydrophobic and vdW contacts with KCNQ1 and KCNE1 subunits as well as by two hydrogen bonds formed between the drug and ps-KCNE1 residue L42 and KCNQ1 residue Q147" Can you show, using H-bond analysis that these two hydrogen bonds really exist stably in the simulations? Can you show, using minimum distance analysis, that L42 are in the vdW radii stably and are making close contact throughout the simulations?

We performed a detailed H-bond analysis (Figure 6-supplement figure 1) which shows that DIDS forms multiple H-bond over the simulations, though only some of them (GLU43, TYR46, ILE47, SER298, TYR299, TRP323 ) are stable. Thus, the H-bonds that we observed in DIDS-docking experiments were unstable in MD simulations. As in the case of the IKs-MEF complex, the prevailing H-bonds exhibit marked quantitative variability from simulation to simulation. We have added a table detailing the most frequent H-bonds during MD simulations (Table 2).

1. Discussion - In line 417, the author stated that the "S1 appears to pull away from the pore" and supplemented the claim with the movie. This is insufficient. The author should demonstrate distance calculation between the S1 helix and the pore, in WT and mutants, with and without the drug. This could be shown as a time series or distribution of centre-of-mass distance over time.

We tried to analyze the distance changes between the upper S1 and the pore domain but failed to see a strong correlation We have removed this statement from the discussion.

1. Given that all the work were done in the open state channel with PIP2 bound (PDB entry: 6v01), could the author demonstrate, either using docking, or simulations, or alignment, or space-filling models - that the ligand, both DIDS and Mef, would not be able to fit in the binding site of a closed state channel (PDB entry: 6v00). This would help illustrate the point denoted Lines 464-467. "Slowed deactivation of the S1/KCNE1/Pore domain/drug complex... By stabilising the activated complex. MD simulation suggests the latter is most likely the case."

As of now, a structure representing the closed state of the channel does not exist. 6V00 is the closed inactivated state of the channel pore with voltage-sensors in the activated conformation. In order to create simulation conditions that reliably describe the electrophysiological experiments, at least a good model for closed channels with resting state voltage sensors is necessary.

1. The author stated that the binding pose changed in one run (lines 317 to 318). Can you comment on those changes? If the pose has changed - what has it changed to? Can you run longer simulations to see if it can reverse back to the initial confirmation? Or will it leave the site completely?

Longer simulations and trajectory clustering revealed several binding modes, where one pose dominated in approximately 50% of all simulations in Figure 2-figure supplement 3 encircled with a blue frame.

1. Binding free energy of -32 kcal/mol = -134 kJ/mol. If you try to do dG = -RTlnKd, your lnKd is -52. Your Kd is e^-52, which means it will never unbind if it exists. I am aware that this is the caveat with the methodologies. But maybe these should be highlighted throughout the manuscript.

We thank the reviewer for this comment. G values, and corresponding Kd values, calculated from simulation of Mef-ps-IKs complex do not reflect the apparent Kd values determined in electrophysiological experiments, nor do they reflect Kd values of drug binding that could be determined in biochemical essays. Important measures are the changes observed in simulations of mutant channel complexes relative to wild type. We now briefly mention this issue in the manuscript.

Reviewer #1 (Recommendations For The Authors):

1) It would be nice to have labels of amino acid residues in Figure 2B.

We updated Figure 2B and added some residue labels.

2) Fig. 3A and 7A. In what order the current traces are presented? I don't see the rule.

We have now arranged the current traces in a more orderly manner, listing them first by ascending KCNE1 residue numbers and then by ascending KCNQ1 residue numbers. Now consistent with Fig 3 and 7 (normalized response and delta V1/2).

3) Line 312 "A44 and Y46 were more so." A44 may be more critical, but I can't see Y46 is more, according to Figure 2-figure supplement2 and Figure 6.

Indeed, comparison of the energy decomposition data indicates approximately the same ∆G values for Y46. We have revised this in the text correspondingly.

4) Line 267 "Mefenamic acid..." I would like to see the movie.

We no longer have access to this original movie

5) In supplemental movies 5-7, the side chains of some critical amino acid residues (W323, K41) would be better presented as in movies 1-4.

We have retained the original presentations of these movies as the original files are no longer available.

Reviewer #2 (Recommendations For The Authors):

General comments:

1) To determine the effect of mefenamic acid and DIDS on channel closing kinetics, a protocol in which they step from an activating test pulse to a repolarizing tail pulse to -40 mV for 1 s is used. If I understand it right, the drug response is assessed as the difference in instantaneous tail current amplitude and the amplitude after 1 s (row 599-603). The drug response of each mutant is then normalized to the response of the WT channel. However, for several mutants there is barely any sign of current decay during this relatively brief pulse (1 s) at this specific voltage. To determine drug effects more reliably on channel closing kinetics/the extent of channel closing, I wonder if these protocols could be refined? For instance, to cover a larger set of voltages and consider longer timescales?

To clarify, the drug response of each mutant is not normalized to the response of the WT channel. In fact, our analysis is not meant to compare mutant and WT tail current decay but rather how isochronal tail current decay is changed in response to drug treatment in each channel construct. As acknowledged by the reviewer, the peak to end difference currents were calculated by subtracting the minimum amplitude of the deactivating current from the peak amplitude of the deactivating current. But the difference current in mefenamic acid or DIDS was normalized to the maximum control (in the absence of drug) difference current and subtracted from 1.0 to obtain the normalized response. Thus, the difference in tail current decay in the absence and in the presence of drug is measured within the same time scale and allow a direct comparison between before and after drug treatment. As shown in Fig 3D and 7C, a large drug response such as the one measured in WT channels is reflected by a value close to 1. A smaller drug response is indicated by low values. We recognize that some mutations resulted in an intrinsic inhibition of tail current decay in the absence of drug, which potentially lead to underestimating the normalized response value. Our goal was not to study in detail the effects of the drug on channel closing kinetics, but only to determine the impact of the mutation on drug binding by using tail current decay as a readout. Consequently, we believe that the duration of the deactivating tail current used in this experiment was sufficient to detect drug-induced tail current decay inhibition.

2) The effect of mefenamic acid seems to be highly dependent on the pulse-to-pulse interval in the experiments. For instance, for WT in Figure 3 - Figure supplement 1, a 15 s pulse-to-pulse interval provides a -100 mV shift in V1/2 induced by mefenamic acid, whereas there is no shift induced when using a 30 s pulse-to-pulse interval. Can the authors explain why they generally consider a 15 s pulse-to-pulse interval more suitable (physiologically relevant?) in their experiments to assess drug effects?

In our previous experiments, we have determined that a 15 s inter-pulse interval is generally adequate for the WT IKs channels to fully deactivate before the onset of the next pulse. Consistent with our previous work (Wang et al. 2019), we observed that in wild-type EQ channels, there is no current summation from one pulse to the next one (see Fig 1A, bottom panel). This is important as the IKs channel complex is known to be frequency dependent i.e. current amplitude increases as the inter-pulse interval gets shorter. Such current summation results in a leftward shift of the conductance-voltage (GV) relationship. This is also important with regards to drug effects. As indicated by the reviewer, mefenamic acid effects are prominent with a 15 sec inter-pulse interval but less so with a 30 sec inter-pulse interval when enough time is given for channels to more completely deactivate. Full effects of mefenamic acid would have therefore been concealed with a 30sec inter-pulse interval.

Moreover, our patch-clamp recordings aim to explore the distinct responses of mutant channels to mefenamic acid and DIDS in comparison to the wild-type channel. It is important to note that the inter-pulse interval's physiological relevance is not necessarily crucial in this context.

3) Related to comment 1 and 2, there is a large diversity in the intrinsic properties of tested mutants. For instance, V1/2 ranges from 4 to 70 mV. Also, there is large variability in the slope of the G-V curves. Whether channel closing kinetics, or the impact of pulse-to-pulse interval, vary among mutants is not clear. Could the authors please discuss whether the intrinsic properties of mutants may affect their ability to respond to mefenamic acid and DIDS? Also, please provide representative current families and G-V curves for all assessed mutants in supplementary figures.

The intrinsic properties of some mutants vary from the WT channels and influence their responsiveness to mefenamic acid and DIDS. The impact of the mutations on the IKs channel complex are reflected by changes in V1/2 (Table 1, 4) and tail current decay (Figs. 3, 7). But, it is the examination of the drug effects on these intrinsic properties (i.e. GV curve and tail current decay) that constitutes the primary endpoint of our study. We consider that the degree by which mef and DIDS modify these intrinsic properties reflects their ability to bind or not to the mutated channel. In our analysis, we compared each mutant's response to mefenamic acid and DIDS with its respective control. Consequently, the intrinsic properties of the mutant channels have already been considered in our evaluation. As requested, we have provided representative current families and G-V curves for all assessed mutants in Figure 3-figure supplement 1 and Figure 7-figure supplement 1.

4) The A44C and Y148C mutants give strikingly different currents in the examples shown in Figure 3 and Figure 7. What is the reason for this? In the examples in figure 7, it almost looks like KCNE1 is absent. Although linked constructs are used, is there any indication that KCNE1 is not co-assembled properly with KCNQ1 in those examples?

The size of the current is critical to determining its shape, as during the test pulse there is some endogenous current mixed in which impacts shape. A44C and Y148C currents shown in Figure 7 are smaller with a larger contribution of the endogenous current, mostly at the foot of the current trace. In our experience there is little endogenous current in the tail current at -40 mV and for this reason we focus our measurements there.

Although constructs with tethered KCNQ1 and KCNE1 were used, we cannot rule out the possibility that Q1 and E1 interaction was altered by some of the mutations. Several KCNE1 and KCNQ1 residues have been identified as points of contact between the two subunits. For instance, the KCNE1 loop (position 36-47) has been shown to interact with the KCNQ1 S1-S2 linker (position 140-148) (Wang et al, 2011). Thus, it is conceivable that mutation of one or several of those residues may alter KCNQ1/KCNE1 interaction and modify the activation/deactivation kinetics of the IKs channel complex.

5) I had a hard time following the details of the simulation approaches used. If not already stated (I could not find it), please provide: i) details on whether the whole channel protein was considered for 4D docking or a docking box was specified, ii) information on how simulations with mutant ps-IKs were prepared (for instance with the K41C mutant), especially whether the in silico mutated channel was allowed to relax before evaluation (and for how long). Also, please make sure that information on simulation time and number of repeats are provided in the Methods section.

For 4D docking, only residues within 0.8 nm of psKCNE1 residues D39-A44 were selected. Complexes with mutated residues were relaxed using the same protocol as the WT channel, (equilibration with gradually releasing restraints with a final equilibration for 10 ns where only the backbone was constrained with 50 kcal/mol/nm2). We have updated the methods accordingly.

Specific comments:

In figure legends, please provide information on whether data represents mean +/- SD or SEM. Also, please provide information on which statistical test was used in each figure.

We revised the figure legend to add the nature of the statistical test used.

G-V curves are normalized between 0 and 1. However, for many mutants the G-V relationship does not reach saturation at depolarized voltages. Does this affect the estimated V1/2? I could not really tell as I was not sure how V1/2 was determined for different mutants (could the explanation on row 595-598 be clarified)?

The primary focus here is in the shift between the control response and drug response for each mutant, rather than the absolute V1/2 values. The isochronal G-V curves that are generated for each construct (WT and mutant) utilize an identical voltage protocol. This approach ensures a uniform comparison among all mutants. By observing the shifts in these curves, we can gain insight into the response of mutant channels to the drug. This information ultimately helps elucidate the inherent properties of the mutant channels and contributes to our understanding of the drug's binding mechanism to the channel.

As requested by the reviewer, we also clarified the way V1/2 was generated: When the G-V curve did not reach zero, the V1/2 value was directly read from the plot at the voltage point where the curve crossed the 0.5 value on the y coordinate.

A general comment is that the Discussion is fairly long and some sections are quite redundant to the Results section. The authors could consider focusing the text in the Discussion.

We changed the discussion correspondingly wherever it was appropriate.

I found it a bit hard to follow the authors interpretation on whether their drug molecules remain bound throughout the experiments, or whether there is fast binding/unbinding. Please clarify if possible.

In the 300 ns MD simulations mefenamic acid and DIDS remained stably bound to WT-ps-IKS, binding of drugs to mutant complexes are described in the Table 3 and Table 5. In longer simulations with the channel embedded in a lipid environment, mefenamic acid unbinds in two out of five runs for WT-ps-IKs (Figure 4 – figure supplement 1), and DIDS shows a few events where it briefly unbinds (Figure 6 -figure supplement 3). Based on electrophysiological data we speculate that drugs might bind and unbind to WT-ps-IKs during the gating process. We do not see bind-unbinding in MD simulations, since the model we used in simulations reflects only open conformation of the channel-complex with an activated-state voltage-sensor, whereas a resting-state voltage sensor condition was not considered.

The authors have previously shown that channels with no, one or two KCNE1 subunits are not, or only to a small extent, affected by mefenamic acid (Wang et al., 2020). Could the details of the binding site and proposed mechanisms of action provide clues as to why all binding sites need to be occupied to give prominent drug effects?

In the manuscript, we propose that the binding of drugs induces conformational changes in the pocket region that stabilize S1/KCNE1/Pore complex. In the tetrameric channel with 4:4 alpha to beta stoichiometry the drugs are likely to occupy all four sites with complete stabilization of S1/KCNE1/Pore. When one or more KCNE1 subunits is absent, as in case of EQQ, or EQQQQ constructs, drugs will bind to the site(s) where KCNE1 is available. This will lead to stabilization of the only certain part of the S1/KCNE1/Pore complex. We believe that the corresponding effect of the drug, in this case will be partially effective.

There is a bit of jumping in the order of when some figures are introduced (e.g. row 178 and 239). The authors could consider changing the order to make the figures easier to follow.

We have changed the corresponding section appropriately to improve the reading flow.

Row 237: "Data not shown", please show data.

The G-V curve of the KCNE1 Y46C mutant displays a complex, double Boltzmann relationship which does not allow for the calculation of a meaningful V1/2 nor would it allow for an accurate determination of drug effects. Consequently, we have excluded it from the manuscript.

In the Discussion, the author use the term "KCNE1/3". Does this correspond to the previous mention of "ps-KCNE1"?

Yes, this refers to ps-KCNE1. We have changed it correspondingly.

Row 576: When was HMR 1556 used?

While HMR 1556 was used in preliminary experiments to confirm that the recorded current was indeed IKs, it does not provide substantial value to the data presented in our study or our experiments. As a result, we have excluded HMR 1556 experiments from the final results and have revised the Methods section accordingly.

Reviewer #3 (Recommendations For The Authors):

1) Figures 2D and 6A are very unclear. Can the authors provide labels as text rather than coloured circles, whether the residue is on Q1 or E1? There is also a distance label in the figure in the small font with the faintest shade of grey, which I believe is supposed to be hydrogen bonds. Can this be improved for clarity?

We feel that additional labels on the ligand diagrams to be more confusing, instead, we updated the description in the legend and added labels to Figure 2B and Figure 6B to improve the clarity of residue positions. In addition, we have added 2 new figures with more detailed information about H-bonds (Figure 2-figure supplement 4, Figure 6- figure supplement 1).

2) Figure 2B - all side chains need labelling in different binding modes. The green ligand on blue protein is very difficult to see. Suddenly, the ligand turns light blue in panel 2C. Can this be consistent throughout the manuscript?

Figure 2B is updated according to this comment.

3) Figure 2 - figure supplement 2, and figure 6B. Can the author show the residue number on the x-axis instead of just the one-letter abbreviation? This requires the reader to count and is not helpful when we try to figure out where the residue is at a glance. I would suggest a structure label adjacent to the plot to show whether they are located with respect to the drug molecule.

Since the numbers for residues on either end of the cluster are indicated at the bottom of each boxed section, we feel that adding residue numbers would just further clutter the figure.

4) Figure 2 - figure supplement 2, and Figure 6B. Can you explain what is being shown in the error bar? I assume standard deviation?

Error bars on Figure 2-figure supplement 2 represent SEM. We added corresponding text in the figure legend.

5) Figure 2 - figure supplement 2, and figure 6B. Can you explain how many frames are being accounted for in this PBSA calculation?

For Figure 2- figure supplement 2 and Figure 6B a frame was made every 0.3 ns over 3x300 ns simulation, 1000 frames for each simulation, 3000 frames overall.

6) Figure 3D/E and 7C/D, it would be helpful to show which mutant show agreeable results with the simulations, PBSA/GBSA and contact analyses as suggested above.

The inconsistencies and discrepancies between the results of MD simulations and electrophysiological experiments are discussed throughout the manuscript.

7) Figure legend, figure 3E - I assume that there is a type that is different mutants with respect to those without the drug. Otherwise, how could WT, with respect to WT, has -105 mV dV1/2?

The reviewer is correct in that the bars indicate the difference in V1/2 between control and drug treatment. Thus, the difference in V1/2 (∆V1/2) between the V1/2 calculated for WT control and the V1/2 for mefenamic acid is indeed -105 mV. We have now revised Figure 3E's legend to accurately reflect this and ensure a clear understanding of the data presented.

8) Figure 3 - figure supplement 1B is very messy, and I could not extract the key point from it. Can this be plotted on a separate trace? At least 1 WT trace and one mutant trace, 1 with WT+drug and one mut+drug as four separate plots for clarity?

The key message of this figure is to illustrate the similarities of EQ WT + Mef and EQ L142C data. Thus, after thorough consideration, we have concluded that maintaining the current figure, which displays the progressive G-V curve shift in EQ WT and L142C in a superimposed manner, best illustrates the gradual shift in the G-V curves. This presentation allows for a clearer and more immediate comparison of the curve shifts, which may be more challenging to discern if the G-V curves were separated into individual figures. We believe that the existing format effectively communicates the relevant information in a comprehensive and accessible manner.

9) Figure 4B - the label Voltage is blended into the orange helix. Can the label be placed more neatly?

We altered the labels for this figure and added that information in the figure description.

10) Can you show the numerical label of the residue, at least only to the KCNE1 portion in Figures 4C and 4D?

We updated these figures and added residue numbering for clarity.

11) Can you hide all non-polar hydrogen atoms in figure 8 and colour each subunit so that it agrees with the rest of the manuscripts? Can you adjust the position of the side chain so that it is interpretable? Can you summarise this as a cartoon? For example, Q147 and Y148 are in grey and are very far hidden away. So as S298. Can you colour-code your label? The methionine (I assume M45) next to T327 is shown as the stick and is unlabelled. Maybe set the orthoscopic view, increase the lighting and rotate the figures in a more interpretable fashion?

We agree that Fig.8 is rather small as originally presented. We have tried to emphasize those residues we feel most critical to the study and inevitably that leads to de-emphasis of other, less important residues. As long as the figure is reproduced at sufficient size we feel that it has sufficient clarity for the purposes of the Discussion.

12) Line 538-539. Can you provide more detail on how the extracellular residues of KCNE3 are substituted? Did you use Modeller, SwissModel, or AlphaFold to substitute this region of the KCNEs?

We used ICM-pro to substitute extracellular residues of KCNE3 and create mutant variants of the Iks channel. This information is provided in the methods section now.

13) Line 551: The PIP2 density was solved using cryo-EM, not X-ray crystallography.

We corrected this.

14) Line 555: The system was equilibrated for ten ns. In which ensemble? Was there any restraint applied during the equilibration run? If yes, at what force constant?

The system was equilibrated in NVT and NPT ensembles with restraints. These details are added to methods. In the new simulations, we did equilibrations gradually releasing spatial from the backbone, sidechains, lipids, and ligands. A final 30 ns equilibration in the NPT ensemble was performed with restraint only for backbone atoms with a force constant of 50 kJ/mol/nm2. Methods were edited accordingly.

15) Line 557: Kelvin is a unit without a degree.

Corrected

16) Line 559: PME is an electrostatic algorithm, not a method.

Corrected

17) Line 566: Collecting 1000 snapshots at which intervals. Given your run are not equal in length, how can you ensure that these are representative snapshots?

Please see comment #5.

18) Table 3 - Why SD for computational data and SEM for experimental data?

There was no particular reason for using SD in some graphs. We used appropriate statistical tests to compare the groups where the difference was not obvious.

Author Response

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

Public Review:

1. Evidence for a disulfide bridge contained in membrane-associated FGF2 dimers

This aspect was brought up in detail by both Reviewer #1 and Reviewer #3. It has been addressed in the revised manuscript by (i) new experimental and computational analyses, (ii) a more detailed discussion of previous work from our lab in which experiments were done the reviewers were asking for and (iii) a more general discussion of known examples of disulfide formation in protein complexes with a particular focus on membrane surfaces facing the cytoplasm, the inner plasma membrane leaflet being a prominent example. Please find our detailed comments in our direct response to Reviewers #1 and #3, see below.

1. Affinity towards PI(4,5)P2 comparing FGF2 dimers versus monomers

This is an aspect that has been raised by Reviewer 3 along with additional comments on the interaction of FGF2 with PI(4,5)P2. Please find our detailed response below. With regard to PI(4,5)P2 affinity aspects of FGF2 dimers versus FGF2 monomers, we think that the increased avidity of FGF2 dimers with two high affinity binding pockets for PI(4,5)P2 are a good explanation for the different values of free energies of binding that were calculated from the atomistic molecular dynamics simulations shown in Fig. 9. This phenomenon is well known for many biomolecular interactions and is also consistent with the cryoEM data contained in our manuscript, showing a FGF2 dimer with two PI(4,5)P2 binding sites facing the membrane surface.

1. C95-C95 FGF2 dimers as signaling units

We have put forward this hypothesis since in structural studies analyzing the FGF ternary signaling complex consisting of FGF2, FGF receptor and heparin, FGF2 mutants were used that lack C95. Nevertheless, two FGF2 molecules are contained in FGF signaling complexes. In addition to the papers on the structure of the FGF signaling complex, we have cited work that showed that C95-C95 crosslinked FGF2 dimers are efficient FGF signaling modules (Decker et al, 2016; Nawrocka et al, 2020). Therefore, being based on an assembly/disassembly mechanism with the transient formation of poreforming FGF2 oligomers, we think it is an interesting idea that the FGF2 secretion pathway produces C95-C95 disulfide-linked FGF2 dimers at the outer plasma membrane leaflet that can engage in FGF2 ternary signaling complexes. While this is a possibility we put forward to stimulate the field, it of course remains a hypothesis which has been clearly indicated as such in the revised manuscript.

Reviewer #1:

1. Evidence for disulfide-bridged FGF2 dimers and higher oligomers on non-reducing versus reducing SDS gels

The experiment suggested by Reviewer #1 is an important one that has been published by our group in previous work. In these studies, we found FGF2 oligomers analyzed on non-reducing SDS gels to be sensitive to DTT, turning the vast majority of oligomeric FGF2 species into monomers [(Müller et al, 2015); Fig. 3, compare panel D with panel H]. This phenomenon could be observed most clearly after short periods of incubations (0.5 hours) of FGF2 with PI(4,5)P2-containing liposomes. These findings constituted the original evidence for PI(4,5)P2-induced FGF2 oligomerization to depend on the formation of intermolecular disulfide bridges.

In the current manuscript, we established the structural principles underlying this process and identified C95 to be the only cysteine residue involved in disulfide formation. Based on biochemical cross-linking experiments in cells, cryo-electron tomography, predictions from AlphaFold-2 Multimer and molecular dynamics simulations, we demonstrated a strong FGF2 dimerization interface in which C95 residues are brought into close proximity when FGF2 is bound to membranes in a PI(4,5)P2-dependent manner. These findings provide the structural basis by which disulfide bridges can be formed from the thiols contained in the side chains of two C95 residues directly facing each other in the dimerization interface. In the revised manuscript, we included additional data that further strengthen this analysis. In the experiments shown in the new Fig. 10, we combined chemical cross-linking with mass spectrometry, further validating the reported FGF2 dimerization interface. In addition, illustrated in the new Fig. 8, we employed a new computational analysis combining 360 individual atomistic molecular dynamics simulations, each spanning 0.5 microseconds, with advanced machine learning techniques. This new data set corroborates our findings, demonstrating that the C95-C95 interface self-assembles independently of C95-C95 disulfide formation, based on electrostatic interactions. Intriguingly, it is consistent with our experimental findings based on cross-linking mass spectrometry (new Fig. 10) where cross-linked peptides could also be observed with the C77/95A variant form of FGF2, suggesting a protein-protein interface whose formation does not depend on disulfide formation. Therefore, we propose that disulfide formation occurs in a subsequent step, representing the committed step of FGF2 membrane translocation with the formation of disulfide-bridged FGF2 dimers being the building blocks for pore-forming FGF2 oligomers.

As a more general remark on the mechanistic principles of disulfide formation in different cellular environments, we would like to emphasize that it is a common misconception that the reducing environment of the cytoplasm generally makes the formation of disulfide bridges unlikely or even impossible. From a biochemical point of view, the formation of disulfide bridges is not limited by a reducing cellular environment but is rather controlled by kinetic parameters when two thiols are brought into proximity. Indeed, it has become well established that disulfide bridges can also be formed in compartments other than the lumen of the ER/Golgi system, including the cytoplasm. For example, viruses maturing in the cytoplasm can form stable structural disulfide bonds in their coat proteins (Locker & Griffiths, 1999; Hakim & Fass, 2010). Moreover, many cytosolic proteins, including phosphatases, kinases and transcriptions factors, are now recognized to be regulated by thiol oxidation and disulfide bond formation, formed as a post-transcriptional modification (Lennicke & Cocheme, 2021). In numerous cases with direct relevance for our studies on FGF2, disulfide bond formation and other forms of thiol oxidation occur in association with membrane surfaces. In fact, many of these processes are linked to the inner plasma membrane leaflet (Nordzieke & Medrano-Fernandez, 2018). Growth factors, hormones and antigen receptors are observed to activate transmembrane NADPH oxidases generating O2·-/H2O2 (Brown & Griendling, 2009). For example, the local and transient oxidative inactivation of membrane-associated phosphatases (e.g., PTEN) serves to enhance receptor associated kinase signaling (Netto & Machado, 2022). It is therefore conceivable that similar processes introduce disulfide bridges into FGF2 while assembling into oligomers at the inner plasma membrane leaflet. In the revised version of our manuscript, we have discussed the above-mentioned aspects in more detail, with the known role of NADPH oxidases in disulfide formation at the inner plasma membrane leaflet being highlighted.

Reviewer #2:

1. Potential effects of a C95A substitution on protein folding and comparison with a C95S substitution with regard to phenotypes observed in FGF2 secretion

A valid point that we indeed addressed at the beginning of this project. Most importantly, we tested whether both FGF2 C95A and FGF2 C95S are characterized by severe phenotypes in FGF2 secretion efficiency. As shown in the revised Fig. 1, cysteine substitutions by serine showed very similar FGF2 secretion phenotypes compared to cysteine to alanine substitutions (Fig. 1C and 1D). In addition, in the pilot phase of this project, we also compared recombinant forms of FGF2 C95A and FGF2 C95S in various in vitro assays. For example, we tested the full set of FGF2 variants in membrane integrity assays as the ones contained in Fig. 4. As shown in Author response image 1, FGF2 variant forms carrying a serine in position 95 behaved in a very similar manner as compared to FGF2 C95A variant forms. Relative to FGF2 wild-type, membrane pore formation was strongly reduced for both types of C95 substitutions. By contrast, both FGF2 C77S and C77A did show activities that were similar to FGF2 wild-type.

Author response image 1.

From these experiments, we conclude that changes in protein structure are not the basis for the phenotypes we report on the C95A substitution in FGF2.

1. Effects of a C77A substitution on FGF2 membrane recruitment in cells

The effect of a C77A substitution in FGF2 recruitment to the inner plasma membrane leaflet is indeed a moderate one. This is likely to be the case because C77 is only one residue of a more complex surface that contacts the α1 subunit of the Na,K-ATPase. Stronger effects can be observed when K54 and K60 are changed, residues that are positioned in close proximity to C77 (Legrand et al, 2020). Nevertheless, as shown in the revised Fig. 1, we consistently observed a reduction in membrane recruitment when comparing FGF2 C77A with FGF2 wild-type. When analyzing the raw data without GFP background subtraction, a significant reduction of FGF2 C77A was observed compared to FGF2 wild-type (Fig. 1A and 1B). We therefore conclude that C77 does not only play a role in FGF2/α1 interactions in biochemical assays using purified components (Fig. 7) but also impairs FGF2/α1 interactions in a cellular context (Fig. 1A and 1B).

1. Identity of the protein band in Fig. 3 labeled with an empty diamond

This is a misunderstanding as we did not assign this band to a FGF2-GFP dimer. When we produced the corresponding cell lines, we used constructs that link FGF2 with GFP via a ‘self-cleaving’ P2A sequence. During translation, even though arranged on one mRNA, this causes the production of FGF2 and GFP as separate proteins in stoichiometric amounts, the latter being used to monitor transfection efficiency. However, a small fraction is always expressed as a complete FGF2-P2A-GFP fusion protein (a monomer). This band can be detected with the FGF2 antibodies used and was labeled in Fig. 3 by an empty diamond.

1. Labeling of subpanels in Fig. 5A

We have revised Fig. 5 according to the suggestion of Reviewer #2.

1. FGF2 membrane binding efficiencies shown in Fig. 5C

It is true that FGF2 variant forms defective in PI(4,5)P2-dependent oligomerization (C95A and C77/95A) bind to membranes with somewhat reduced efficiencies. This is also evident form the intensity profiles shown in Fig. 5A and was observed in biochemical in vitro experiments as well. A plausible explanation for this phenomenon would be the increased avidity when FGF2 oligomerizes, stabilizing membrane interactions (see also Fig. 9B).

1. Residual activities of FGF2 C95A and C77/95A in membrane pore formation?

We do not assign the phenomenon in Fig. 5 Reviewer #2 is referring to as controlled activities of FGF2 C95A and C77/95A in membrane pore formation. Rather, GUVs containing PI(4,5)P2 are relatively labile structures with a certain level of integrity issues upon protein binding and extended incubation times being conceivable. It is basically a technical limitation of this assay with GUVs incubated with proteins for 2 hours. Even after substitution of PI(4,5)P2 with a Ni-NTA membrane lipid, background levels of loss of membrane integrity can be observed (Fig. 6). Therefore, as compared to FGF2 C95A and C77/95A, the critical point here is that FGF2 wt and FGF2 C77A do display significantly higher levels of a loss of membrane integrity in PI(4,5)P2-containing GUVs, a phenomenon that we interpret as controlled membrane pore formation. By contrast, all variant forms of FGF2 show only background levels for loss of membrane integrity in GUVs containing the Ni-NTA lipid.

1. Why does PI(4,5)P2 induce FGF2 dimerization?

This has been studied extensively in previous work (Steringer et al, 2017). As also discussed in the current manuscript, the interaction of FGF2 with membranes through its high affinity PI(4,5)P2 binding pocket orients FGF2 molecules on a 2D surface that increase the likelihood of the formation of the C95containing FGF2 dimerization interface. Moreover, in the presence of cholesterol at levels typical for plasma membranes, PI(4,5)P2 clusters containing up to 4 PI(4,5)P2 molecules (Lolicato et al, 2022), a process that may further facilitate FGF2 dimerization.

1. Is it possible to pinpoint the number of FGF2 subunits in oligomers observed in cryo-electron tomography?

We indeed took advantage of the Halo tags that appear as dark globular structures in cryo-electron tomography. For most FGF2 oligomers with FGF2 subunits on both sides of the membrane, we could observe 4 to 6 Halo tags which is consistent with the functional subunit number that has been analyzed for membrane pore formation (Steringer et al., 2017; Sachl et al, 2020; Singh et al, 2023). However, since the number of higher FGF2 oligomers we observed in cryo-electron tomography was relatively small and the nature of these oligomers appears to be highly dynamic, caution should be taken to avoid overinterpretation of the available data.

Reviewer #3:

1. Conclusive demonstration of disulfide-linked FGF2 dimers

A similar point was raised by Reviewer #1, so that we would like to refer to our response on page 2, see above.

1. Identity of FGF2-P2A-GFP observed in Fig. 3

Again, a similar point has been made, in this case by Reviewer #2 (Point 3). The observed band is not a FGF2-P2A-GFP dimer but rather the complete FGF2-P2A-GFP fusion protein (a monomer) that corresponds to a small population produced during mRNA translation where the P2A sequence did not cause the production of FGF2 and GFP as separate proteins in stoichiometric amounts.

1. Quantification of GFP signals in Fig. 6

Fig. 6 has been revised according to the suggestion of Reviewer #3. A comprehensive comparison of PI(4,5)P2 and the Ni-NTA membrane lipid in FGF2 membrane translocation assays is also contained in previous work that introduced the GUV-based FGF2 membrane translocation assay (Steringer et al., 2017).

1. Experimental evidence for various aspects of FGF2 interactions with PI(4,5)P2

Most of the points raised by Reviewer #3 have been addressed in previous work. For example, FGF2 has been demonstrated to dimerize only on membrane surfaces containing PI(4,5)P2 (Müller et al., 2015). In solution, FGF2 remained a monomer even after hours of incubation as analyzed by native gel electrophoresis and reducing vs. non-reducing SDS gels (see Fig. 3 in Müller et al, 2015). In the same paper, the first evidence for a potential role of C95 in FGF2 oligomerization has been reported, however, at the time, our studies were limited to FGF2 C77/95A. In the current manuscript, the in vitro experiments shown in Figs. 2 to 6 establish the unique role of C95 in PI(4,5)P2-dependent FGF2 oligomerization. As discussed above, FGF2 oligomers have been shown to contain disulfide bridges based on analyses on non-reducing gels in the absence and presence of DTT (Müller et al., 2015).

References

Brown DI, Griendling KK (2009) Nox proteins in signal transduction. Free Radic Biol Med 47: 1239-1253 Decker CG, Wang Y, Paluck SJ, Shen L, Loo JA, Levine AJ, Miller LS, Maynard HD (2016) Fibroblast growth factor 2 dimer with superagonist in vitro activity improves granulation tissue formation during wound healing. Biomaterials 81: 157-168

Hakim M, Fass D (2010) Cytosolic disulfide bond formation in cells infected with large nucleocytoplasmic DNA viruses. Antioxid Redox Signal 13: 1261-1271

Legrand C, Saleppico R, Sticht J, Lolicato F, Muller HM, Wegehingel S, Dimou E, Steringer JP, Ewers H, Vattulainen I et al (2020) The Na,K-ATPase acts upstream of phosphoinositide PI(4,5)P2 facilitating unconventional secretion of Fibroblast Growth Factor 2. Commun Biol 3: 141

Lennicke C, Cocheme HM (2021) Redox metabolism: ROS as specific molecular regulators of cell signaling and function. Mol Cell 81: 3691-3707

Locker JK, Griffiths G (1999) An unconventional role for cytoplasmic disulfide bonds in vaccinia virus proteins. J Cell Biol 144: 267-279

Lolicato F, Saleppico R, Griffo A, Meyer A, Scollo F, Pokrandt B, Muller HM, Ewers H, Hahl H, Fleury JB et al (2022) Cholesterol promotes clustering of PI(4,5)P2 driving unconventional secretion of FGF2. J Cell Biol 221

Müller HM, Steringer JP, Wegehingel S, Bleicken S, Munster M, Dimou E, Unger S, Weidmann G, Andreas H, GarciaSaez AJ et al (2015) Formation of Disulfide Bridges Drives Oligomerization, Membrane Pore Formation and Translocation of Fibroblast Growth Factor 2 to Cell Surfaces. J Biol Chem 290: 8925-8937

Nawrocka D, Krzyscik MA, Opalinski L, Zakrzewska M, Otlewski J (2020) Stable Fibroblast Growth Factor 2 Dimers with High Pro-Survival and Mitogenic Potential. Int J Mol Sci 21

Netto LES, Machado L (2022) Preferential redox regulation of cysteine-based protein tyrosine phosphatases: structural and biochemical diversity. FEBS J 289: 5480-5504

Nordzieke DE, Medrano-Fernandez I (2018) The Plasma Membrane: A Platform for Intra- and Intercellular Redox Signaling. Antioxidants (Basel) 7

Sachl R, Cujova S, Singh V, Riegerova P, Kapusta P, Muller HM, Steringer JP, Hof M, Nickel W (2020) Functional Assay to Correlate Protein Oligomerization States with Membrane Pore Formation. Anal Chem 92: 14861-14866

Singh V, Macharova S, Riegerova P, Steringer JP, Muller HM, Lolicato F, Nickel W, Hof M, Sachl R (2023) Determining the Functional Oligomeric State of Membrane-Associated Protein Oligomers Forming Membrane Pores on Giant Lipid Vesicles. Anal Chem 95: 8807-8815

Steringer JP, Lange S, Cujova S, Sachl R, Poojari C, Lolicato F, Beutel O, Muller HM, Unger S, Coskun U et al (2017) Key steps in unconventional secretion of fibroblast growth factor 2 reconstituted with purified components. eLife 6: e28985

Author Response

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

Reviewer #1 (Public Review):

The paper from Hsu and co-workers describes a new automated method for analyzing the cell wall peptidoglycan composition of bacteria using liquid chromatography and mass spectrometry (LC/MS) combined with newly developed analysis software. The work has great potential for determining the composition of bacterial cell walls from diverse bacteria in high-throughput, allowing new connections between cell wall structure and other important biological functions like cell morphology or host-microbe interactions to be discovered. In general, I find the paper to be well written and the methodology described to be useful for the field. However, there are areas where the details of the workflow could be clarified. I also think the claims connecting cell wall structure and stiffness of the cell surface are relatively weak. The text for this topic would benefit from a more thorough discussion of the weak points of the argument and a toning down of the conclusions drawn to make them more realistic.

Thank you for your thorough and insightful review of our manuscript. We greatly appreciate your positive and constructive feedbacks on our methodology. We have carefully reviewed your comments and have responded to each point as follows:

Specific points:

1) It was unclear to me from reading the paper whether or not prior knowledge of the peptidoglycan structure of an organism is required to build the "DBuilder" database for muropeptides. Based on the text as written, I was left wondering whether bacterial samples of unknown cell wall composition could be analyzed with the methods described, or whether some preliminary characterization of the composition is needed before the high-throughput analysis can be performed. The paper would be significantly improved if this point were explicitly addressed in the main text. We apologize for not making it clearer. The prior knowledge of the peptidoglycan structure of an organism is indeed required to build the “DBuilder” database to accurately identify muropeptides; otherwise, the false discovery rate might increase. While peptidoglycan structures of certain organisms might not have been extensively studied, users still remain the flexibility to adapt the muropeptide compositions based on their study, referencing closely related species for database construction. We have addressed this aspect in the main text to ensure a clearer understanding.

“(Section HAMA platform: a High-throughput Automated Muropeptide Analysis for Identification of PGN Fragments) …(i) DBuilder... Based on their known (or putative) PGN structures, all possible combinations of GlcNAc, MurNAc and peptide were input into DBuilder to generate a comprehensive database that contains monomeric, dimeric, and trimeric muropeptides (Figure 1b)."

2) The potential connection between the structure of different cell walls from bifidobacteria and cell stiffness is pretty weak. The cells analyzed are from different strains such that there are many possible reasons for the change in physical measurements made by AFM. I think this point needs to be explicitly addressed in the main text. Given the many possible explanations for the observed measurement differences (lines 445-448, for example), the authors could remove this portion of the paper entirely. Conclusions relating cell wall composition to stiffness would be best drawn from a single strain of bacteria genetically modified to have an altered content of 3-3 crosslinks.

We understand your concern regarding the weak connection between cell wall structure and cell stiffness. We will make a clear and explicit statement in the main text to acknowledge that the cells analyzed are derived from different strains, introducing the possibility of various factors influencing the observed changes in physical measurements as determined by AFM. Furthermore, we greatly appreciate your suggestion to consider genetically modified strains to investigate the role of cross-bridge length in determining cell envelope stiffness. In this regard, we are in the process of developing a CRISPR/Cas genome editing toolbox for Bifidobacterium longum, and we plan on this avenue of investigation for future work.

Reviewer #2 (Public Review):

The authors introduce "HAMA", a new automated pipeline for architectural analysis of the bacterial cell wall. Using MS/MS fragmentation and a computational pipeline, they validate the approach using well-characterized model organisms and then apply the platform to elucidate the PG architecture of several members of the human gut microbiota. They discover differences in the length of peptide crossbridges between two species of the genus Bifidobacterium and then show that these species also differ in cell envelope stiffness, resulting in the conclusion that crossbridge length determines stiffness.

We appreciate your thoughtful review of our manuscript and your recognition of the potential significance of our work in elucidating the poorly characterized peptidoglycan (PGN) architecture of the human gut microbiota.

The pipeline is solid and revealing the poorly characterized PG architecture of the human gut microbiota is worthwhile and significant. However, it is unclear if or how their pipeline is superior to other existing techniques - PG architecture analysis is routinely done by many other labs; the only difference here seems to be that the authors chose gut microbes to interrogate.

We apologize if this could have been clearer. The HAMA platform stands apart from other pipelines by utilizing automatic analysis of LC-MS/MS data to identify muropeptides. In contrast, most of the routine PGN architecture analyses often use LC-UV/Vis or LC-MS platform, where only the automatic analyzing PGFinder software is supported. To our best knowledge, a comparable pipeline on automatically analyzing LC-MS/MS data was reported by Bern et al., which they used commercial Byonic software with an in-house FASTA database and specific glycan modifications. They achieved accurate and sensitive identification on monomer muropeptides, but struggled with cross-linked muropeptides due to the limitations of the Byonic software. We believe that our pipeline introducing the automatic and comprehensive analysis on muropeptide identification (particularly for Gram-positive bacterial peptidoglycans) would be a valuable addition to the field. To enhance clarity, we have adjusted the context as follows:

(Introduction) … Although they both demonstrated great success in identifying muropeptide monomers, the accurate identification of muropeptide multimers and other various bacterial PGN structures still remains unresolved. This is because deciphering the compositions requires MS/MS fragmentation, but it is still challenging to automatically annotate MS/MS spectra from these complex muropeptide structures."

I do not agree with their conclusions about the correlation between crossbridge length and cell envelope stiffness. These experiments are done on two different species of bacteria and their experimental setup therefore does not allow them to isolate crossbridge length as the only differential property that can influence stiffness. These two species likely also differ in other ways that could modulate stiffness, e.g. turgor pressure, overall PG architecture (not just crossbridge length), membrane properties, teichoic acid composition etc.

Regarding the conclusions drawn about the correlation between cross-bridge length and cell envelope stiffness, we understand your point and appreciate your feedback. We revisit this section of our manuscript and tone down the conclusions drawn from this aspect of the study. We also recognize the importance of considering other potential factors that could influence stiffness, as you mentioned above. In light of this, we mentioned the need for further investigations, potentially involving genetically modified strains, in the main text to isolate and accurately determine the impact of bridge length on cell envelope stiffness.

Reviewer #1 (Recommendations For The Authors):

Minor points:

1) One thing to consider would be testing the robustness of the analysis pipeline with one the well-characterized bacteria studied, but genetically modifying them to change the cell wall composition in predictable ways. Does the analysis pipeline detect the expected changes?

We appreciate the reviewer's suggestion and would like to provide a clear response. Regarding to testing the pipeline with genetically modified strains, our lab previously worked on genetically modified S. maltophilia (KJΔmrdA).1 Inactivation of mrdA turned out the increasing level of N-acetylglucosaminyl-1,6-anhydro-N-acetylmuramyl-L-alanyl-D-glutamyl-meso-diamnopimelic acid-D-alanine (GlcNAc-anhMurNAc tetrapeptide) in muropeptide profiles, which is the critical activator ligands for mutant strain ΔmrdA-mediated β-lactamase expression. In this case, our platform could provide rapid PGN analysis for verifying the expected change of muropeptide profiles (see Author response image 1). Besides, if the predictable changes involve genetically modifications on interpeptide bridges within the PGN structure, for example, the femA/B genes of S. aureus, which are encoded for the synthesis of interpeptide bridges,2 our current HAMA pipeline is capable of detecting these anticipated changes. However, if the genetically modifications involve the introduce of novel components to PGN structures, then it would need to create a dedicated database specific to the genetically modified strain.

Author response image 1.

2) Line 368: products catalyzed > products formed

The sentence has been revised.

“(Section Inferring PGN Cross-linking Types Based on Identified PGN Fragments) …Based on the muropeptide compositional analysis mentioned above, we found high abundances of M3/M3b monomer and D34 dimer in the PGNs of E. faecalis, E. faecium, L. acidophilus, B. breve, B. longum, and A. muciniphila, which may be the PGN products formed by Ldts.”

3) Lines 400-402: Is it possible the effect is related to porosity, not "hardness".

Thank you for the suggestion. The possibility of the slower hydrolysis rate of purified PGN in B. breve being related to porosity is indeed noteworthy. While this could be a potential factor, we would like to acknowledge the limited existing literature that directly addresses the relation between PGN architecture and porosity. It is plausible that current methods available for assessing cell wall porosity may have certain limitations, contributing to the scarcity of relevant studies. In light of this, we would like to propose a speculative explanation for the observed effect. It is plausible that the tighter PGN architecture resulting from shorter interpeptide bridges in B. breve could contribute to its harder texture. This speculation is grounded in the concept that a more compact PGN structure might lead to increased stiffness, aligning with our observations of higher cell stiffness in B. breve.

4) Lines 403-408: See point #2 above.

Thank you for the suggestion. We have explicitly addressed this point in the main text:

“(Section Exploring the Bridge Length-dependent Cell Envelope Stiffness in B. longum and B. breve) … Taken all together, we speculate that a tight peptidoglycan network woven by shorter interpeptide bridges or 3-3 cross-linkages could give bacteria stiffer cell walls. However, it is important to note that cell stiffness is a mechanical property that also depends on PGN thickness, overall architecture, and turgor pressure. These parameters may vary among different bacterial strains. Hence, carefully controlled, genetically engineered strains with similar characteristics will be needed to dissect the role of cross-bridge length in cell envelope stiffness.”

5) Lines 428-429: It is not clear to me how mapping the cell wall architecture provides structural information about the synthetic system. It is also not clear how antibiotic resistance can be inferred. More detail is needed here to flesh out these points.

Thank you for the suggestion. To provide further clarity on these important aspects, the context in the manuscript has been revised.

“(Discussion) …Importantly, our HAMA platform provides a powerful tool for mapping peptidoglycan architecture, giving structural information on the PGN biosynthesis system. This involves the ability to infer possible PGN cross-linkages based on the type of PGN fragments obtained from hydrolysis. For instance, the identification of 3-3 cross-linkage formed by L,D-transpeptidases (Ldts) is of particular significance. Unlike 4-3 cross-linkages, the 3-3 cross-linkage is resistant to inhibition by β-Lactam antibiotics, a class of antibiotics that commonly targets bacterial cell wall synthesis through interference with 4-3 cross-linkages. Therefore, by elucidating the specific cross-linkage types within the peptidoglycan architecture, our approach offers insights into antibiotic resistance mechanisms.”

6) Line 478: "maneuvers are proposed for" > "work is needed to generate". Also, delete "innovative". Also "in silico" > "in silico-based".

The sentence has been revised.

“(Discussion) …To achieve a more comprehensive identification of muropeptides, future work is needed to generate an expanded database, in silico-based fragmentation patterns, and improved MS/MS spectra acquisition.”

7) Line 485: "Its" > "It has potential"

The sentence has been revised.

“(Discussion) …It has potential applications in identifying activation ligands for antimicrobial resistance studies, characterizing key motifs recognized by pattern recognition receptors for host-microbiota immuno-interaction research, and mapping peptidoglycan in cell wall architecture studies.”

8) Figure 1 legend: Define Gb and Pb.

Gb and Pb are the abbreviations of glycosidic bonds and peptide bonds. We have revised the Figure legend 1 as follow:

“(Figure legend 1) …(b) DBuilder constructs a muropeptide database containing monomers, dimers, and trimers with two types of linkage: glycosidic bonds (Gb) and peptide bonds (Pb).”

9) Figure 2: It is hard to see what is going on in panel a and c with all the labels. Consider removing them and showing a zoomed inset with labels in addition to ab unlabeled full chromatogram.

We apologize for not making this clearer. The panel a and c in Figure 2 were directly generated by the Analyzer as a software screenshot of the peak annotations on chromatogram. Our intention was to present a comprehensive PGN mapping (approximately 70% of the peak area was assigned to muropeptide signals) using this platform. We understand the label density might affect clarity, so we have added the output tables of the whole muropeptide identifications as source data (Table 1–Source Data 1&2). Additionally, we have uploaded the Analyzer output files (see Additional Files), which can be better visualized in the Viewer program, and it also allows users zoom in for detailed labeling information.

10) Figure 3: It is worth pointing out what features of the MS/MS fingerprints are helping to discriminate between species.

Thank you for the suggestion. We have revised Figure 3 and the legend as follow:

“(Figure legend 3) …The sequence of each isomer was determined using in silico MS/MS fragmentation matching, with the identified sequence having the highest matching score. The key MS/MS fragments that discriminate between two isomers are labeled in bold brown.”

Author response image 2.

11) Figure 4 and 5 legend: Can you condense the long descriptions of the abbreviations - or at least only refer to them once?

Certainly, to enhance clarity and conciseness in the figure legends, we have revised Figure legend 5 as follow:

“(Figure legend 5) …(b) Heatmap displaying …. Symbols: M, monomer; D, dimer; T, trimer (numbers indicate amino acids in stem peptides). Description of symbol abbreviations as in Figure legend 4, with the addition of "Glycan-T" representing trimers linked by glycosidic bonds.”

Reviewer #2 (Recommendations For The Authors):

1. Please read the manuscript carefully for spelling errors.

We appreciate your careful review of our manuscript. We have thoroughly rechecked the entire manuscript for spelling errors and have made the necessary corrections to ensure the accuracy and quality of the text.

1. Line 46 - "multilayered" is likely only true for Gram-positive bacteria.

We thank reviewer #2 for bringing up this concern. Indeed, Gram-negative bacteria mostly possess single layer of peptidoglycan, but could be up to three layers in some part of the cell surface.3, 4 In order to reduce the confusion, we have rewritten the context as follow: “(Introduction) …PGN is a net-like polymeric structure composed of various muropeptide molecules, with their glycans linearly conjugated and short peptide chains cross-linked through transpeptidation.”

1. Methods section: It seems like pellets from a 10 mL bacterial culture were ultimately suspended in 1.5 L (750 mL water + 750 mL tris) - why such a large volume? And how were PG fragments subsequently washed (centrifugation? There is no information on this in the Methods).

We apologize for the mislabeling on the units. The accurate volume should be “1.5 mL (750 µL water + 750 µL tris)”. We have updated the correct volume in the Methods section (lines 99-100). For the washing process of purified PGN, we added 1 mL water, centrifuged at 10,000 rpm for 5 minutes, and removed supernatant. This information has added to the Methods section (lines 95-98).

1. Line 183 - why were 6 modifications chose as the cutoff? Please make rationale more clear.

We thank reviewer #2 for the comments. We set the maximum modification number of 6 in the assumption of one modification on each sugar of a trimeric muropeptide. A lower cutoff could effectively limit the identification of muropeptides with unlikely numbers of modifications, whereas a higher cutoff could allow for having multiple modifications on a muropeptide. In our hand, muropeptide modifications of E. coli are mostly N-deacetyl-MurNAc and anhydro-MurNAc, and modifications of gut microbes used here are mostly N-deacetyl-GlcNAc, anhydro-MurNAc, O-acetyl-MurNAc, loss of GlcNAc, and amidated iso-Glu. While we recommend starting data analysis with the cutoff of 6 modifications, users are free to adjust this based on their studies.

1. Line 339 - define donor vs. acceptor here (can be added in parentheses after explaining the relevant chemical reactions further above in the text)

Thank you for the suggestion. To provide greater clarity regarding the roles of the donor and acceptor substrates in the transpeptidation process, we have revised the content in the manuscript as follows:

“(Section Inferring PGN Cross-linking Types Based on Identified PGN Fragments) …In general, there are two types of PGN cross-linkage…. Transpeptidation involves two stem peptides which function as acyl donor and acceptor substrates, respectively. As the enzyme names imply, the donor substrates that Ddts and Ldts bind to are terminated as D,D-stereocenters and L,D-stereocenters, which structurally means pentapeptides and tetrapeptides. During D,D-transpeptidation, Ddts recognize D-Ala4-D-Ala5 of the donor stem (pentapeptide) and remove the terminal D-Ala5 residue, forming an intermediate. The intermediate then cross-links the NH2 group in the third position of the neighboring acceptor stem, forming a 4-3 cross-link.”

1. Line 366 following - can you calculate % crosslinks based on these numbers? What does "high abundance" of 3,3 crosslinks mean in this context? Is this the majority of PG?

Thank you for your questions. Calculating the percentage of crosslinks based on the muropeptide compositional numbers is a valid consideration. However, it's important to note that the muropeptides we analyzed were hydrolyzed by mutanolysin, and as such, deriving an accurate % crosslink value from these data might not provide a true representation of the crosslinking percentage within the PGN network. For a more precise determination of % crosslinks, methods such as solid-phase NMR on purified peptidoglycan would be required. Our research provides insights into the characterization of PGN fragments and allows us to infer potential PGN cross-linkage types and the enzymes involved based on the dominant muropeptide fragments. Regarding the phrase "high abundance" in the context, it indicates that the M3b/M4b monomer and D34 dimer muropeptides represent a significant portion of the hydrolysis products. These muropeptides are major constituents within the PGN fragments obtained from the enzymatic hydrolysis.

1. Line 375 - I am not sure PG is a meaningful diffusion barrier for drugs and signaling molecules, give that even larger proteins can apparently diffuse through the pores.

Thank you for raising this point. Peptidoglycan indeed possesses relatively wide pores that allow for the diffusion of larger molecules, including proteins.5 Research has provided a rough estimate of the porosity of the PGN meshwork, suggesting that it allows for the diffusion of proteins with a maximum molecular mass of around 50 kDa.6 Considering this, we acknowledge that PGN may not serve as a significant diffusion barrier for drugs and signaling molecules. The porosity of the PGN scaffold, which is defined by the degree of cross-linking, plays a role in influencing the transport of molecules to the cell membrane. Thus, while PGN may not serve as a strict diffusion barrier, its structural characteristics still impact bacterial cell mechanics and interactions. We have revised the manuscript to reflect this understanding:

“(Section Exploring the Bridge Length-dependent Cell Envelope Stiffness in B. longum and B. breve) …The porosity of the PGN scaffold, defined by the degree of cross-linking, influences the transport of larger molecules such as proteins. Therefore, modifications to PGN structure are anticipated to significantly affect bacterial cell mechanics and interactions.”

1. Line 400 - what does "slower hydrolysis rate" refer to, is this chemical hydrolysis or enzymatic (autolysins?). also, I am not sure hydrolysis rate of either modality allows for solid conclusions about how hard (line 402) the PG is.

Thank you for your comments. The hydrolysis rate here refers to the enzymatic hydrolysis, specifically the mutanolysin cleaving the β-N-acetylmuramyl-(1,4)-N-acetylglucosamine linkage. Indeed, there is no direct correlation between the hydrolysis rate and the hardness of PGN architecture, although the structure rigidity is a key determinant in protein digestion.7 Considering the enzymatic hydrolysis rate depending on the accessibility of the substrate to the enzyme, we proposed that the tighter PGN architecture could also lead to a slower hydrolysis rate. This speculation aligns with our observations of higher cell stiffness or more compact PGN structure of B. breve and its slower hydrolysis rate. We understand this is indirect proof, so the revised sentence now reads:

“(Section Exploring the Bridge Length-dependent Cell Envelope Stiffness in B. longum and B. breve) …Furthermore, B. breve also showed a slower enzymatic hydrolysis rate in purified PGNs, implying that the cell wall structure of B. breve is characterized by a compact PGN architecture.”

1. Line 424 - I am not convinced this pipeline can detect PG architectures that other pipelines cannot; likely, the difference between previous analyses and theirs is due to different growth conditions (3,3 crosslink formation is often modulated by environmental factors/growth stage). In the next sentence, it sounds like mutanolysin treatment is a novelty in PG analysis (which it is not).

We apologize if this could have been clearer and we have revised the paragraph to describe our study more accurately. We agree that different growth conditions could influence PGN architecture and other pipelines could manually identify the PGN architectures or automatically identify them if they are not too complex. Our original intention was to highlight the ability of the HAMA program to automatically identify unreported PGN structure. Here are the revised sentences:

“(Discussion) …We speculate that this finding may be influenced by the comprehensive mass spectrometric approaches we employed or by variations in growth conditions. Moreover, we utilized the well-established enzymatic method involving mutanolysin to cleave the β-N-acetylmuramyl-(1,4)-N-acetylglucosamine linkage, which preserves the original peptide linkage in intact PGN subunits.”

1. Line 440- 442: As outlined in more detail above: I don't think you can conclude something about the relationship between bridge length and envelope stiffness based on these data. Thank you for your valuable feedback. We agree that our data may not definitively support the direct conclusion about the relationship between bridge length and envelope stiffness in Bifidobacterium species. Instead, we will rephrase this section to accurately present the observed correlations without overgeneralizing:

“(Discussion) … Notably, our study suggested a potential correlation between the cell stiffness and the compactness of bacterial cell walls in Bifidobacterium species (Figure 5). B. longum, which predominantly harbors tetrapeptide bridges (Ser-Ala-Thr-Ala), exhibits a trend towards lower stiffness, whereas B. breve, characterized by PGN cross-linked with monopeptide bridges (Gly), demonstrates a trend towards higher stiffness. These findings suggested that it may be correlated between the increased rigidity and the more compact PGN architecture built by shorter cross-linked bridges.”

References: 1. Huang, Y.-W.; Wang, Y.; Lin, Y.; Lin, C.; Lin, Y.-T.; Hsu, C.-C.; Yang, T.-C., Impacts of Penicillin Binding Protein 2 Inactivation on β-Lactamase Expression and Muropeptide Profile in Stenotrophomonas maltophilia. mSystems 2017, 2 (4), 00077-00017.

1. Jarick, M.; Bertsche, U.; Stahl, M.; Schultz, D.; Methling, K.; Lalk, M.; Stigloher, C.; Steger, M.; Schlosser, A.; Ohlsen, K., The serine/threonine kinase Stk and the phosphatase Stp regulate cell wall synthesis in Staphylococcus aureus. Sci. Rep. 2018, 8 (1), 13693.

2. Labischinski, H.; Goodell, E. W.; Goodell, A.; Hochberg, M. L., Direct proof of a "more-than-single-layered" peptidoglycan architecture of Escherichia coli W7: a neutron small-angle scattering study. J. Bacteriol. 1991, 173 (2), 751-756.

3. Rohde, M., The Gram-Positive Bacterial Cell Wall. Microbiol. Spectr. 2019, 7 (3), gpp3-0044-2018.

4. Vollmer, W.; Höltje, J. V., The architecture of the murein (peptidoglycan) in gram-negative bacteria: vertical scaffold or horizontal layer(s)? J. Bacteriol. 2004, 186 (18), 5978-5987.

5. Vollmer, W.; Blanot, D.; De Pedro, M. A., Peptidoglycan structure and architecture. FEMS Microbiol. Rev. 2008, 32 (2), 149-167.

6. Li, Q.; Zhao, D.; Liu, H.; Zhang, M.; Jiang, S.; Xu, X.; Zhou, G.; Li, C., "Rigid" structure is a key determinant for the low digestibility of myoglobin. Food Chem.: X 2020, 7, 100094.

Author Response

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

Response to the Referee Comments We would like to express our appreciation to the editor and the reviewers for their thoughtful comments and constructive suggestions on the manuscript. We agree with most of the comments and have carefully revised the manuscript accordingly. The revisions are highlighted in red font in the revised manuscript. Below are point-by-point responses to the referee’s comments.

Public Reviews:

Reviewer #1 (Public Review):

Microglia are increasingly recognized as playing an important role in shaping the synaptic circuit and regulating neural dynamics in response to changes in their surrounding environment and in brain states. While numerous studies have suggested that microglia contribute to sleep regulation and are modulated by sleep, there has been little direct evidence that the morphological dynamics of microglia are modulated by the sleep/wake cycle. In this work, Gu et al. applied a recently developed miniature two-photon microscope in conjunction with EEG and EMG recording to monitor microglia surveillance in freely-moving mice over extended period of time. They found that microglia surveillance depends on the brain state in the sleep/wake cycle (wake, non-REM, or REM sleep). Furthermore, they subjected the mouse to acute sleep deprivation, and found that microglia gradually assume an active state in response. Finally, they showed that the state-dependent morphological changes depend on norepinephrine (NE), as chemically ablating noradrenergic inputs from locus coeruleus abolished such changes; this is in agreement with previous publications. The authors also showed that the effect of NE is partially mediated by β2-adrenergic receptors, as shown with β2-adrenergic receptor knock-out mice. Overall, this study is a technical tour de force, and its data add valuable direct evidence to the ongoing investigations of microglial morphological dynamics and its relationship with sleep. However, there are a number of details that need to be clarified, and some conclusions need to be corroborated by more control experiments or more rigorous statistical analysis. Specifically:

1. The number of branch points per microglia shown here (e.g., Fig. 2g) is much lower than the values of branch points in the literature, e.g., Liu T et al., Neurobiol. Stress 15: 100342, 2021 (mouse dmPFC, IHC); Liu YU et al., Nat. Neurosci. 22: 1771-81, 2019 (mouse S1, in vivo 2P imaging). The authors need to discuss the possible source of such discrepancy.

Thank you for raising this important point. Two reasons may account for this difference. Firstly, the difference in the definition of branch points in the software. Liu YU et al. used the Sholl analysis of image J software to analyze the number of branch points of microglia. Sholl analysis defines the number of branch points as the number of crossings between branches and concentric circles of increasing radii. We reconstructed microglia morphology using Imaris, a software that defines branching points based on the number of bifurcation points. The number of bifurcations calculated represents the number of microglia branch points. Secondly, this and previous studies found that more branching points present in the state of anesthesia. The morphological characteristics of microglia in head-fixed mice under anesthesia was reported by Liu T et al. and the microglia reconstruction results presented by the authors are indeed more complex than ours. In short, this is an aspect that we have been paying attention to, and the main reasons for this difference may lie in the definition of branch points, analysis methods and related choice of thresholds. True differences in brain states and the heterogeneity of microglia in different brain regions may also contribute to the apparent discrepancy.

1. Microglia process end-point speed (Fig. 2h, o): here the authors show that the speed is highest in the wake state and lowest in NREM, which agrees with the measurement on microglia motility during wakefulness vs NREM in a recent publication (Hristovska I et al., Nat. Commun. 13: 6273, 2022). However, Hristovska et al. also reported lower microglia complexity in NREM vs wake state, which seems to be the opposite of the finding in this paper. The authors need to discuss the possible source of such differences.

This is also an important point. Hristovska et al. reported the morphodynamic characteristics of microglia during wakefulness and NREM sleep. It is worth noting that the sleep state of the mice in their experiments was unnatural due to the head fixation and body limitations, the duration of NREM sleep (sleep stability) being quite different from the NREM sleep analyzed under natural sleep. The limitations of this approach are also discussed by Hristovska et al. “Even though sleep episodes were, as anticipated, shorter than those observed in freely moving animals, changes in neuronal activity characteristic of NREM sleep were monitored by EEG recordings, and changes in morphodynamics were observed during single episodes. Several episodes of REM sleep were detected, but they were too short and rare to be analyzed reliably.” The unnatural sleep state would lead to an increase in the microarousal state, and ultimately a change in the structure of the sleep state, which may be the main reason for the difference in microglia behavior from our natural sleep. We have discussed this in the revised manuscript. Please see line 292298.

1. Fig. 3: the authors used single-plane images to analyze the morphological changes over 3 or 6 hours of SD, which raises the concern that the processes imaged at the baseline may drift out of focus, leading to the dramatic reduction in process lengths, surveillance area, and number of branch points. In fact, a previous study (Bellesi M et al., J. Neurosci. 37(21): 5263-73, 2017) shows that after 8 h SD, the number of microglia process endpoints per cell and the summed process length per cell do not change significantly (although there is a trend to decline). The authors may confirm their findings by either 3D imaging in vivo, or 3D imaging in fixed tissue.

Three lines of evidence indicate that microglia morphology changes in Fig 3 are due to SD, rather than variations in the focal plane. First, our single-plane images were quite stable over 3 or 6 hours of SD, though occasional reversible drifts might happen due to sudden motions. Second, per your suggestion, further experiments and analysis of 3D imaging were performed to monitor microglia dynamics during sleep deprivation. The new result is shown in revised Fig. S3 C-D: the length of microglia branches and the number of branching points were significantly reduced after SD, in agreement with the results of single-plane imaging. Furthermore, we detected no significant difference in microglia branching characteristics during 6h sleep deprivation in 2AR KO mice (Fig.S4), and this indirectly affirmed that singleplane imaging is stable enough for detecting true changes in branching during SD.

1. Fig. 4b: the EEG and EMG signals look significantly different from the example given in Fig. 2a. In particular, the EMG signal appears completely flat except for the first segment of wake state; the EEG power spectrum for REM appears dark; and the wake state corresponds to stronger low frequency components (below ~ 4 Hz) compared to NREM, which is the opposite of Fig. 2a. This raises the concern whether the classification of sleep stage is correct here.

Thank you for insightful comments. We carefully examined the behavioral video of Figure 4b, there were occasionally microarousal events indicated by slow head rotation during NREM sleep, while the companion EMG signals were completely flat, which is atypical during sleep wake cycle. The microarousal events were not excluded from sleep, which makes this set of data unrepresentative and contrary to Fig.4b. In our revised manuscript, we replaced it with more representative data that can clearly and consistently distinguish between different brain states in mice on EMG and EEG. Please see revised Fig.2a, page 34; revised Fig.4b, page 37.

1. Fig. 4 NE dynamics. • How long is a single continuous imaging session for NE? • When monitoring microglia surveillance, the authors were able to identify wake or NREM states longer than 15 min, and REM states longer than 5 min. Here the authors selected wake/NREM states longer than 1 min and REM states longer than 30 s. What makes such a big difference in the time duration selected for analysis? • Also, the definition of F0 is a bit unclear. Is the same F0 used throughout the entire imaging session, or is it defined with a moving window?

A single continuous session of NE imaging usually took about 1 hour. Subsequent analysis was performed on imaging data from each recording that included wake, NREM sleep, and REM sleep. Because of the different time scales of microglia morphological dynamic (relatively slow) and NE signals (fast), we used different time windows in the previous analysis in the previous version of the manuscript.

Per your suggestion, we have now set the same time window selection criteria for both microglia morphological and NE dynamic analysis: for wake and NREM sleep durations longer than 1 minute, and REM sleep durations longer than 30 seconds. We updated the Methods and all statistics in related figures, please see line 151-154, 481-485, 490-492; Fig. 2e-g and 2l-n, page 34. F0 definition is now explained in the Methods section. Please see line 521-522.

1. Fig. 5b: how does the microglia morphology in LC axon ablation mice compare with wild type mice under the wake state? The text mentioned "more contracted" morphology but didn't give any quantification. Also, the morphology of microglia in the wake state (Fig. 5b) appears very different from that shown in Fig. S3C1 (baseline). What is the reason?

The morphology of microglia is indeed heterogeneous and variable, affected by factors including brain state, brain region, microenvironmental changes, along with animal-to-animal difference. We didn’t perform the microglia morphology comparison between the LC axon ablation mice and wild type mice and, in view of this, we removed the description of “more contracted morphology” from the main text. It should also be noted that, as we primarily focused on changes of a microglia in different states over time by selfcomparison, we minimized possible effects of heterogeneity in microglia morphology on our conclusions.

1. The relationship between NE level and microglia dynamics. Fig. 4C shows that the extracellular NE level is the highest in the wake state and the lowest in REM. Previous studies (Liu YU et al., Nat. Neurosci. 22(11):1771-1781, 2019; Stowell RD et al., Nat. Neurosci. 22(11): 1782-1792, 2019) suggest that high NE tone corresponds to reduced microglia complexity and surveillance. Hence, it would be expected that microglia process length, branch point number, and area/volume are higher in REM than in NREM. However, Fig. 2l-n show the opposite. How should we understand this ?

Your point is well-taken. On the one hand, our data clearly showed that NE is critically involved in the brain state-dependent microglia dynamic surveillance, with evidence from the ablation of the LC-NE projection and from the β2AR knockout animal model.

On the other hand, we also understand that NE is not the sole determinant, so the relationship between the NE level and the complexity and surveillance may not be unique.

In this regard, other potential modulators also present dynamic during sleepwake cycle and may partake in the regulation of microglia dynamic surveillance. previous studies (Liu YU et al., 2019; Stowell RD et al., 2019) have shown that microglia can be jointly affected by surrounding neuronal activity and NE level during wake. It has been reported that LC firing stops (Aston-Jones et al., 1981; Rasmussen et al., 1986), while inhibitory neurons, such as PV neurons and VIP neurons, become relatively active during REM sleep (Brécier et al., 2022). ATP level in basal forebrain is shown to be higher in REM than NREM (Peng et al., 2023). In addition, our own preliminary result (Author response image 1) also showed a higher adenosine level in REM than NREM in somatosensory cortex. Last but not the least, we found that β2AR knockout failed to abolish microglial responses to sleep state switch and SD stress altogether.

In brief, microglia are highly sensitive to varied changes in the surrounding environment, and many a modulator may participate in the microglia dynamic during sleep state. This may underlie the microglia complexity difference between REM and NREM. Future investigations are warranted to delineate the signal-integrative role of microglia in physiology and under stress. We have discussed the pertinent points in the revised manuscript. Please see line 343-354.

Author response image 1.

Extracellular adenosine levels in somatosensory cortex in different brain states. AAV2/9-hSyn-GRABAdo1.0 (Peng W. et al., Science. 2020) was injected into the somatosensory cortex (A/P, -1 mm; M/L, +2 mm; D/V, -0.3 mm). Data from the same recording are connected by lines. n = 9 from 3 mice.

Reviewer #2 (Public Review):

The manuscript describes an approach to monitor microglial structural dynamics and correlate it to ongoing changes in brain state during sleep-wake cycles. The main novelty here is the use of miniaturized 2p microscopy, which allows tracking microglia surveillance over long periods of hours, while the mice are allowed to freely behave. Accordingly, this experimental setup would permit to explore long-lasting changes in microglia in a more naturalistic environment, which were previously not possible to identify otherwise. The findings could provide key advances to the research of microglia during natural sleep and wakefulness, as opposed to anesthesia. The main findings of the paper are that microglia increase their process motility and surveillance during REM and NREM sleep as compared to the awake state. The authors further show that sleep deprivation induces opposite changes in microglia dynamics- limiting their surveillance and size. The authors then demonstrate potential causal role for norepinephrine secretion from the locus coeruleus (LC) which is driven by beta 2 adrenergic receptors (b2AR) on microglia. However, there are several methodological and experimental concerns which should be addressed.

The major comments are summarized below:

1. The main technological advantage of the 2p miniaturized microscope is the ability to track single cells over sleep cycles. A main question that is unclear from the analysis and the way the data is presented is: are the structural changes in microglia reversible? Meaning, could the authors provide evidence that the same cell can dynamically change in sleep state and then return to similar size in wakefulness? The same question arises again with the data which is presented for anesthesia, is this change reversible?

As revealed by long-term free behavioral mTPM imaging, the brain-statedependent morphological changes in microglia were reproducible and reversible. Author response image 2 shows that microglia displayed reversible dynamic changes during multiple rounds of sleep-wake transition. Author response image 3 shows that microglia dynamics induced by anesthesia also exhibited reversibility.

Author response image 2.

Long-term tracking of microglia process area in different brain states. Data analysis used 8 cells. Data total of 31 time points were selected from in vivo imaging data and were used to characterize the morphological changes of microglia over a continuous 7-hour period.

Author response image 3.

Reversible changes of microglial process length, area, number of branch points under anesthesia. Wake group: 30 minute-accommodation to new environment; Isoflurane group: 1.5% in air applied at a flow rate of 0.4 L/min for 30 minutes; Recovery group: 30 minutes after recovery from anesthesia. n = 9 cells from 3 mice for each group.

1. The binary comparison between brain states is misleading, shouldn't the changes in structural dynamics compared to the baseline of the state onset? The authors method describes analysis of the last 5 minutes in each sleep/wake state. However, these transitions are directional- for instance, REM usually follows NREM, so the description of a decrease in length during REM sleep could be inaccurate.

As you know, the time scale of microglia morphological dynamic is relatively slow, so we analyzed the microglia morphological dynamic of the last part (30s in the revised manuscript) of each state instead of the state onset, allowing time for stabilization of the microglia response to inter-state transition.

Further, we compared microglia dynamic between two NREM groups transiting to different subsequent states: group1 (NREM to REM) vs group2 (NREM to Wake). This precaution was to exclude the directional effect of state transitions. Our results showed that there was no difference in microglial length, area, number of branching points between the two NREM groups (Author response image 4), indicating that the last 30s of each NREM was not affected by its following state and that it’s reasonable to perform binary comparison.

Author response image 4.

Microglial morphological length, area change, and number of branch points of the last 30s of NREM sleep followed by REM or Wake. n = 9 cells from 3 mice for each group.

1. Sleep deprivation- again, it is unclear whether these structural changes are reversible. This point is straightforward to address using this methodology by measuring sleep following SD. In addition, the authors chose a method to induce sleep deprivation that is rather harsh. It is unclear if the effect shown is the result of stress or perhaps an excess of motor activity.

We adopted the method of forced exercise as it has been commonly used for sleep deprivation (Pandi-Perumal et al., 2007; Nollet M et al., 2020), though it does have the potential limitation of excess of motor activity.

In light of your comments and suggestion, we presented new data demonstrating that sleep duration of the mice, mostly NREM sleep, increased compensatively (ZT9-10) after the 6-hour sleep deprivation (ZT2-8) (revised Fig. S3B). This result shows that sleep deprivation indeed increase sleep pressure in the mice. As the sleep pressure was eased during recovery sleep, morphological changes of microglia were reversed over a timescale of several hours (revised Fig. S3 E-J).

1. The authors perform measurements of norepinephrine with a recently developed GRAB sensor. These experiments are performed to causally link microglia surveillance during sleep to norepinephrine secretion. They perform 2p imaging and collect data points which are single neurons, and it is unclear why the normalization and analysis is performed for bulk fluorescence similar to data obtained with photometry.

We did not perform single-neuron analysis for two reasons. First, our experimental conditions, e.g., the expression of the NE indicator and the control of imaging laser intensity, did not yield sufficient signal-to-noise to clearly discriminate individual neurons with two-photon imaging. Second, NE signal may play a modulatory role, and fluorescence changes appeared to be global, rather than local or cell-specific. Therefore, we analyzed fluorescence changes in different brain states over the whole field-of-view in Fig. 4, rather than at the subregional or single-cell level.

1. The experiments involving b2AR KO mice are difficult to interpret and do not provide substantial mechanistic insight. Since b2AR are expressed throughout numerous cell types in the brain and in the periphery, it is entirely not clear whether the effects on microglia dynamics are direct. The conclusion and the statement regarding the expression of b2AR in microglia is not supported by the references the authors present, which simply demonstrate the existence and function of b2AR in microglia. In addition, these mice show significant changes in sleep pattern and increased REM sleep. This could account for reasons for the changes in microglia structure rather than the interpretation that these are direct effects.

To summarize, the main conclusions of the paper require further support with analysis of existing data and experimental validation.

Previous studies have revealed that norepinephrine (NE) has a modulating effect on microglial dynamics through β2AR pathway (Stowell RD et al., 2019; Liu YU et al., 2019). Stowell et al. and Liu et al. use in vivo two-photon imaging to demonstrate that microglia dynamics differ between awake and anesthetized mice and to highlight the roles of NE and β2AR in these states (Gyoneva S et al., 2013; Stowell RD et al., 2019; Liu YU et al., 2019). To evaluate the direct effect of β2AR on microglial dynamics, Stowell et al. administered the β2AR agonist clenbuterol to anesthetized mice and found that this decreased the motility, arbor complexity, and process coverage of microglia in the parenchyma (Stowell RD et al., 2019). Inhibition of β2AR by antagonist ICI-118,551 in awake mice recapitulated the effects of anesthesia by enhancing microglial arborization and surveillance (Stowell RD et al., 2019). In addition, it has been shown microglia expressed higher numbers of β2ARs than any other cells in the brain (Zhang et al., 2014).

To this end, our current work provided new evidence to support the involvement of the LC-NE-β2AR axis in modulating microglia dynamics both during natural sleep-wake cycle and under SD stress. While we were aware the limitation of using pan-tissue β2AR knockout model that precluded us from pinpointing role of microglial β2AR, it is safe to state that β2-adrenergic receptor signaling plays a significant role in the sleep-state dependent microglia dynamic surveillance, based on the present and previous data.

We have discussed this in the revised manuscript. Please see line 324-354. As you suggested, we added references to support the statement regarding the expression of β2AR in microglia (please see line 333).

Recommendations for the authors: please note that you control which, if any, revisions, to undertake

Reviewer #1 (Recommendations For The Authors):

Some technical details need to be clarified. Also, please double-check for typos.

1. In vivo imaging preparation: how long is the recovery time between window/EEG implantation surgery and imaging/recording?

Imaging data were collected one month after the surgery. We have added descriptions to the methods section of the revised manuscript. Please see line 419.

1. Statistical analysis: the authors used t-test or ANOVA without first checking whether the data pass the normality test. If the data does not follow a normal distribution, nonparametric tests would be more appropriate.

Per your suggestion, we performed the test of statistical significance using parametric (ANOVA) if past the normality test, or the non-parametric (Friedman) tests for non-normal data. Please see line 533-535.

1. Fig. 1b needs a minor change. In the figure, the EMG electrodes appear to be connected to the brain as well.

We have corrected this oversight. Thank you.

1. Fig. 1c: it would be helpful to give examples of raw EEG and EMG traces for REM and NREM separately.

Raw traces are now shown as suggested. Please see Fig. 1c, page 32.

1. Fig. 1h: is each data point one microglia or one end-point?

In Fig. 1h, each data represents the average speed of all branches of one microglia, not one end-point.

1. Sleep deprivation starts at 9 am. What time corresponds to Zeitgeber Time 0 (ZT0, the beginning of the light phase)?

We now clarified that 9 am corresponds to Zeitgeber time 2. Please see line 196.

1. Line 61: the authors referred to Ramon y Cajal's original suggestion that microglia dynamics are coupled to the sleep-wake cycle. However, the cited paper only indicates that Cajal suggested a role of astrocytes in the sleep-wake cycle, not microglia. In addition, there is a typo in the line: there should be a space between "Ramon" and "y" in Cajal's name.

We have updated the statement and reference literature to point out the microglia’s involvement in the sleep-wake cycle. The typo was corrected. Please see line 64-65.

1. Fig. S3B: As each group has only 3 mice, it is unclear how t-test can yield p < 0.01 or even 0.001.

We checked the original data again and it was correct. This small p-values may be due to the small intra-group difference of control group.

1. Line 251-253, "Figure 4h-n" should be "Figure 5h-n"?

We have revised it. Please see line 265-266.

1. Fig. 5h: the receptor should be "adrenergic receptor", not "adrenal receptor".

We changed the term to “adrenergic receptor”. Please see Fig 5h.

1. Fig. 5g, n: the number of data points is apparently less than the sample size given in the figure legend. Perhaps some data points have exactly the same value so they overlap? The authors may consider plotting identical values with a slight shift so that the number of data points shown matches the actual sample size, to avoid confusion.

Yes, we have added small jitters so different data points can be seen to avoid confusion. Please see Fig. 5n.

1. There are some typos (e.g., Line 217, "he" should be "the") and some incomplete references (e.g., [13], [22], [34], [35] lack volume and page number, [15] and [39] lack publisher information). Some references have inconsistent formats (e.g., "Journal of Neuroscience" is sometimes abbreviated and sometimes not). Please correct these.

We have corrected these oversights. Please see references, page 27.

Reviewer #2 (Recommendations For The Authors):

Major issues:

1. Re-analyze the data in a manner that allows to follow and compare the same cells over different state transitions. This is necessary to evaluate the reversibility of microglia structure. In addition, consider analysis of the change from the beginning to the end of each state.

As shown in response figure 2, microglia dynamics were reversible during multiple rounds of sleep-wake transition.

1. It would be nice to see the raw data obtained over time, at least for Figure 1, before offline correction of movement to evaluate the imaging quality and level of drift during imaging.

We agree to your good suggestion. Please see the supporting material video.

1. It would be helpful to add an analysis of the percent time spent in each state for the 10 hour recordings.

Advice has been adopted. Please see revised Fig. S4C.

1. In Figure 2 the results are from 15 cells from several animals. How much do the results vary between mice? It will be helpful to show if this varies between different mice by labeling cells from each mouse differently.

In Author response image 5, in which we have labeled the distribution of data points from seven mice, there was mixed distribution of data from different animals at each brain state, but no clear animal-to-animal difference.

Author response image 5.

Quantitative analysis of microglial length based on multi-plane microglial imaging. n = 17 cells from 7 mice for each group. In right panel, each color codes data from the same animal.

1. SD- please add some quantification for sleep and EEG to show that the manipulation really caused sleep deprivation. To address the confound of forced movement and stress, it might be helpful to add quantification of movement compared to an undisturbed wakefulness.

We have added related data (revised Fig. S3B), as suggested. Please see line 196-197.

1. The DSP4 application should be also performed with NE measurements to verify the specific of the NE signal measured as well as the DSP4 toxin.

Following your suggestion, we have added DSP4 data in revised Fig. S4B.

1. Some suggested refined experiments for the b2AR KO are: a-A conditional b2AR KO in microglia, as cited in the work. b- Local application of a b2 blocker during SD. c- Imaging of NE dynamics in the b2 animals. If NE dynamics during natural sleep cycle are perturbed, then this suggests upstream mechanisms rather than direct microglia effects as suggested by the authors.

We agree that the current study cannot pinpoint a direct effect of microglia harbored β2AR. We have discussed this limitation in the revised manuscript.

Please see line 324-354.

Minor:

1. Typo on page 4 (microcopy instead of microscopy).

It was corrected. Please see line 87.

1. Typo page 11- 'and he largest changes in NE' - supposed to be 'the'.

We have corrected these mistakes. Please see line 228.

1. Fig. 4- there are several units missing in the figure in panel b: the top is Hz, but what does the color bar indicate exactly? 2 what? both for theta/delta and for NE. We have modified this figure and legend for clarity. Please see Fig. 4, page 37.

2. Bottom of page 12- referring to figure 4 but talking about figure 5.

The typo was corrected. Please see line 265-266.

Reference

1. Aston-Jones G, Bloom FE. Activity of norepinephrine-containing locus coeruleus neurons in behaving rats anticipates fluctuations in the sleep-waking cycle. J Neurosci. 1, 876–886 (1981).

2. Bellesi M, de Vivo L, Chini M, Gilli F, Tononi G, Cirelli C. Sleep loss promotes astrocytic phagocytosis and microglial activation in mouse cerebral cortex. J Neurosci. 37, 5263–5273 (2017).

3. Brécier A, Borel M, Urbain N, Gentet LJ. Vigilance and behavioral state-dependent modulation of cortical neuronal activity throughout the sleep/wake cycle. J Neurosci. 42, 4852–66 (2022).

4. Dworak M, McCarley RW, Kim T, Kalinchuk AV, Basheer R. Sleep and brain energy levels: ATP changes during sleep. J Neurosci. 30, 9007-16 (2010).

5. Gyoneva S., Traynelis SF. Norepinephrine modulates the motility of resting and activated microglia via different adrenergic receptors. J Biol Chem. 288, 15291302 (2013).

6. Kjaerby C, Andersen M, Hauglund N, Untiet V, Dall C, Sigurdsson B, Ding F, Feng J, Li Y, Weikop P, Hirase H, Nedergaard M. Memory-enhancing properties of sleep depend on the oscillatory amplitude of norepinephrine. Nat Neurosci. 25, 1059–1070 (2022).

7. Liu T, Lu J, Lukasiewicz K, Pan B, Zuo Y. Stress induces microglia-associated synaptic circuit alterations in the dorsomedial prefrontal cortex. Neurobiology of Stress. 15, 100342 (2021).

8. Liu YU, Ying Y, Li Y, Eyo UB, Chen T, Zheng J, Umpierre AD, Zhu J, Bosco DB, Dong H, Wu LJ. Neuronal network activity controls microglial process surveillance in awake mice via norepinephrine signaling. Nat Neurosci. 22, 1771–1781 (2019).

9. Nollet M, Wisden W, Franks NP. Sleep deprivation and stress: a reciprocal relationship. Interface Focus. 10, 20190092 (2020).

10. Pandi-Perumal SR, Cardinali DP, Chrousos GP. 2007. Neuroimmunology of sleep. New York, NY: Springer.

11. Peng W, Liu X, Ma G, Wu Z, Wang Z, Fei X, Qin M, Wang L, Li Y, Zhang S, Xu M. Adenosine-independent regulation of the sleep-wake cycle by astrocyte activity. Cell Discov. 9, 16 (2023).

12. Peng W, Wu Z, Song K, Zhang S, Li Y, Xu M. Regulation of sleep homeostasis mediator adenosine by basal forebrain glutamatergic neurons. Science. 369, 6508 (2020).

13. Rasmussen K, Morilak DA, Jacobs BL. Single unit activity of locus coeruleus neurons in the freely moving cat: I. During naturalistic behaviors and in response to simple and complex stimuli. Brain Research. 371, 324–334 (1986).

14. Stowell RD, Sipe GO, Dawes RP, Batchelor HN, Lordy KA, Whitelaw BS, Stoessel MB, Bidlack JM, Brown E, Sur M, Majewska AK. Noradrenergic signaling in the wakeful state inhibits microglial surveillance and synaptic plasticity in the mouse visual cortex. Nat Neurosci. 22, 1782-1792 (2019).

15. Umpierre AD, Bystrom LL, Ying Y, Liu YU, Worrell G, Wu LJ. Microglial calcium signaling is attuned to neuronal activity in awake mice. Elife. 27, e56502 (2020).

16. Wang Z, Fei X, Liu X, Wang Y, Hu Y, Peng W, Wang YW, Zhang S, Xu M. REM sleep is associated with distinct global cortical dynamics and controlled by occipital cortex. Nat Commun. 13, 6896 (2022).

17. Zhang Y, Chen K, Sloan SA, Bennett ML, Scholze AR, O’Keeffe S, Phatnani HP, Guarnieri P, Caneda C, Ruderisch N, Deng S, Liddelow SA, Zhang C, Daneman R, Maniatis T, Barres BA, Wu JQ. An RNA-sequencing transcriptome and splicing database of glia, neurons, and vascular cells of the cerebral cortex. J Neurosci. 34, 11929–11947 (2014).

Author Response

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

Reviewer #1 (Public Review):

The Hedgehog (HH) protein family is important for embryonic development and adult tissue maintenance. Deregulation or even temporal imbalances in the activity of one of the main players in the HH field, sonic hedgehog (SHH), can lead to a variety of human diseases, ranging from congenital brain disorders to diverse forms of cancers. SHH activates the GLI family of transcription factors, yet the mechanisms underlying GLI activation remain poorly understood. Modification and activation of one of the main SHH signalling mediators, GLI2, depends on its localization to the tip of the primary cilium. In a previous study the lab had provided evidence that SHH activates GLI2 by stimulating its phosphorylation on conserved sites through Unc-51-like kinase 3 (ULK3) and another ULK family member, STK36 (Han et al., 2019). Recently, another ULK family member, ULK4, was identified as a modulator of the SHH pathway (Mecklenburg et al. 2021). However, the underlying mechanisms by which ULK4 enhances SHH signalling remained unknown. To address this question, the authors employed complex biochemistry-based approaches and localization studies in cell culture to examine the mode of ULK4 activity in the primary cilium in response to SHH. The study by Zhou et al. demonstrates that ULK4, in conjunction with STK36, promotes GLI2 phosphorylation and thereby SHH pathway activation. Further experiments were conducted to investigate how ULK4 interacts with SHH pathway components in the primary cilium. The authors show that ULK4 interacts with a complex formed between STK36 and GLI2 and hypothesize that ULK4 functions as a scaffold to facilitate STK36 and GLI2 interaction and thereby GLI2 phosphorylation by STK36. Furthermore, the authors provide evidence that ULK4 and STK36 co-localize with GLI2 at the ciliary tip of NIH 3T3 cells, and that ULK4 and STK36 depend on each other for their ciliary tip accumulation. Overall, the described ULK4-mediated mechanism of SHH pathway modulation is based on detailed and rigorous Co-IP experiments and kinase assays as well as confocal imaging localization studies. The authors used various mutated and wild-type constructs of STK36 and ULK4 to decipher the mechanisms underlying GLI2 phosphorylation at the tip of the primary cilium. These novel results on SHH pathway activation add valuable insight into the complexity of SHH pathway regulation. The data also provide possible new strategies for interfering with SHH signalling which has implications in drug development (e.g., cancer drugs).

However, it will be necessary to explore additional model systems, besides NIH3T3, HEK293 and MEF cell cultures, to conclude on the universality of the mechanisms described in this study. Ultimately, it needs to be addressed whether ULK4 modulates SHH pathway activity in vivo. Is there evidence that genetic ablation of ULK4 in animal models leads to less efficient SHH pathway induction? It also remains to be resolved how ULK3 and ULK4 act in distinct or common manners to promote SHH signalling. Another remaining question is, whether cell type- and tissue-specific features exist, that play a role in ULK3- versus ULK4-dependent SHH pathway modulation. In particular for the studies on ciliary tip localization of factors, relevant for SHH pathway transduction, a higher temporal resolution will be needed in the future as well as a deeper insight into tissue/ cell type-specific mechanisms. These caveats, mentioned here, don't have to be addressed in new experiments for the revision of this manuscript but could be discussed.

We agree with the reviewer that it would be important to investigate in the future the in vivo function Ulk4 in Shh signaling, the relationship between Ulk3 and Ulk4/Stk36, and possible cell type/tissue specificity of these two kinase systems. This will need the generation of single and double knockout mice and examine Hh related phenotypes in different tissues and developmental stages. The precise mechanism by which Ulk4 and Stk36 are translocated to the ciliary tip is also an important and unsolved issue. We include several paragraphs in the “discussion” section to address these outstanding questions for future study.

Reviewer #2 (Public Review):

The authors provide solid molecular and cellular evidence that ULK4 and STK36 not only interact, but that STK36 is targeted (transported?) to the cilium by ULK4. Their data helps generate a model for ULK4 acting as a scaffold for both STK36 and its substrate, Gli2, which appear to co-localise through mutual binding to ULK4. This makes sense, given the proposed role of most pseuodkinases as non-catalytic signaling hubs. There is also an important mechanistic analysis performed, in which ULK4 phosphorylation in an acidic consensus by STK36 is demonstrated using IP'd STK36 or an inactive 'AA' mutant, which suggests this phosphorylation is direct.

The major strength of the study is the well-executed combination of logical approaches taken, including expression of various deletion and mutation constructs and the careful (but not always quantified in immunoblot) effects of depleting and adding back various components in the context of both STK36 and ULK3, which broadens the potential impact of the work. The biochemical analysis of ULK4 phosphorylation appears to be solid, and the mutational study at a particular pair of phosphorylation sites upstream of an acidic residue (notably T2023) is further strong evidence of a functional interaction between ULK4/STK36. The possibility that ULK4 requires ATP binding for these mechanisms is not approached, though would provide significant insight: for example it would be useful to ask if Lys39 in ULK4 is involved in any of these processes, because this residue is likely important for shaping the ULK4 substrate-binding site as a consequence of ATP binding; this was originally shown in PMID 24107129 and discussed more recently in PMID: 33147475 in the context of the large amount of ULK4 proteomics data released.

The reviewer raised an interesting question of whether ATP binding to the pseudokinase domain of Ulk4 might be required for its function, i.e., by regulating the interaction with its binding partner. In a recent study (Preuss et al. 2020;PMID: 33147475), the critical Lys39 for ATP binding was converted to Arg (KR mutation); however, unlike in most kinases the KR mutation affect ATP binding, the K39R mutation in the Ulk4 pseudokinase did not affect ATP binding although it slightly increased ADP binding (PMID: 33147475). Another mutation made by Preuss et al(PMID: 33147475), N239L, affected protein stability, making it impossible to determine whether this mutation affect ATP binding. Therefore, in the absence of clear approach to perturb ATP binding without affecting the overall structure of Ulk4, it would be challenging to address whether ATP binding regulates the ability of Ulk4 to bind its substrates. Nevertheless, we discuss the possibility that ATP binding might regulate Ulk4/Stk36 interaction and Shh signaling.

The discussion is excellent, and raises numerous important future work in terms of potential transportation mechanisms of this complex. It also explains why the ULK4 pseudokinase domain is linked to an extended C-terminal region. Does AF2 predict any structural motifs in this region that might support binding to Gli2?

The extended C-terminal domain of Ulk4 contains Arm/HEAT repeats (protein-protein interacting domain), which are predicted by AF2 to form alpha helixes.

A weakness in the study, which is most evident in Figure 1, where Ulk4 siRNA is performed in the NIH3T3 model (and effects on Shh targets and Gli2 phosphorylation assessed), is that we do not know if ULK4 protein is originally present in these cells in order to actually be depleted. Also, we are not informed if the ULK4 siRNA has an effect on the 'rescue' by HA-ULK4; perhaps the HA-ULK4 plasmid is RNAi resistant, or if not, this explains why phosphorylation of Gli2 never reaches zero? Given the important findings of this study, it would be useful for the authors to comment on this, and perhaps discuss if they have tried to evaluate endogenous levels of ULK4 (and Stk36) in these cells using antibody-based approaches, ideally in the presence and absence of Shh. The authors note early on the large number of binding partners identified for ULK4, and siRNA may unwittingly deplete some other proteins that could also be involved in ULK4 transport/stability in their cellular model.

Due to the lack of reliable Ulk4 and Stk36 antibodies, we were unable to confirm knockdown efficiency by western blot analysis. Therefore, we relied on the measure Ulk4 and STk36 mRNA expression by RT-qPCR to estimate the knockdown efficiency (Fig 1- figure supplement 1). We used mouse Ulk4 shRNA to carry out the knockdown experiments in NIH3T3 and MEF cells while the human version of Ulk4 (hUlk4) was used for the rescue experiments (Fig 1- figure supplement 2; Fig. 8). We have confirmed that the mUlk4 shRNA targeting sequence is not conserved in hUlk4; therefore, the hULK4 construct is RNAi resistant. The rescue experiments strongly argue that the effect of Ulk4 RNAi on Shh signaling is due to loss of endogenous Ulk4. This argument is further strengthened by the observations that mutations that affected Ulk4 and Stk36 ciliary tip localization also affected Shh signaling such as Gli2 phosphorylation and Ptch1/Gli expression (Fig. 8).

The sequence of ULK4 siRNAs is not included in the materials and methods as far as I can see.

We have added the mouse Ulk4 RNAi target sequence in the revised version.

Reviewer #3 (Public Review):

In this manuscript, Zhou et al. demonstrate that the pseudokinase ULK4 has an important role in Hedgehog signaling by scaffolding the active kinase Stk36 and the transcription factor Gli2, enabling Gli2 to be phosphorylated and activated.

Through nice biochemistry experiments, they show convincingly that the N-terminal pseudokinase domain of ULK4 binds Stk36 and the C-terminal Heat repeats bind Gli2.

Lastly, they show that upon Sonic Hedgehog signaling, ULK4 localizes to the cilia and is needed to localize Stk36 and Gli2 for proper activation.

This manuscript is very solid and methodically shows the role of ULK4 and STK36 throughout the whole paper, with well controlled experiments. The phosphomimetic and incapable mutations are very convincing as well. I think this manuscript is strong and stands as is, and there is no need for additional experiments.

Overall, the strengths are the rigor of the methods, and the convincing case they bring for the formation of the ULK4-Gli2-Stk36 complex. There are no weaknesses noted. I think a little additional context for what is being observed in the immunofluorescence might benefit readers who are not familiar with these cell types and structures.

We thank this reviewer for the positive comments.

Recommendations For the Authors

Reviewer #1 (Recommendations For The Authors):

This elegant study has been thoroughly and thoughtfully designed and the dataset is solid. The biochemistry results are overall very convincing. Some data lack quantification and there needs to be more information on data analyses and statistics. The following suggestions and comments aim at strengthening the manuscript.

1. Please provide quantification normalized to input for IP experiments (Figures 1 E - F; Figure 8 C). More information on data analyses and statistics should be provided and included as information in the figure legends.

Thanks for the suggestions, we have done the quantification and statistics analyses for Figures 1E-G and Figure8 C as requested.

1. Did the authors investigate whether overexpressing hULK4 in the control NIH3T3 cells leads to an increase in pS230/232 (related to Figure 1E)? This would nicely support the notion of a promoting effect of ULK4 on GLI2 phosphorylation.

We did not. We speculated that overexpressing hULK4 may not significantly promote GLI2 phosphorylation because Ulk4 is a pseudokinase and endogenous Stk36 (the kinase partner of Ulk4) is limited.

1. The CO-IP experiments to show GLI2 activation were performed in NIH3T3 cells, whereas HEK293 cells were used for the experiments shown in Figure 2. Is there a specific reason for switching between cell lines also for experiments shown in Figures 3 C- I? Did the authors repeat some of the key experiments in both cell lines?

In mammalian cells, Shh-induced activation of GLI2 depends on primary cilia (Han et al., 2019). NIH3T3 cells form the primary cilia but HEK293T cells do not. Therefore, we used NIH3T3 cells to examine the processes that are regulated by the Shh treatment assay (e.g., the Shh-induced phosphorylation of GLI2 and STK36). The HEK293 cells were used to map binding domain between ULK4 and STK36/GLI2/SUFU due to the high transfection efficiency.

1. In Figure 2 D-E the authors nicely showed that hUlk4N-HA interacted with CFP-Stk36 but not with Myc-Gli2/Fg-Sufu whereas hUlk4C-HA formed a complex with Myc-Gli2/Fg-Sufu but not with CFP-Stk36. In Figure 4E the authors showed in their Co-IP experiments that Fg-Stk36 and Myc-Gli2 form a complex independent of SHH treatment. Did the authors see some pull down of Stk36, still in complex with Gli2, using hUlk4C IP and pull down of Gli2, still in complex with Stk36, using hUlk4N IP?

We did not test that. As we have shown in Figures 4A and 4E, knockdown of endogenous ULK4 nearly abolished the interaction between Myc-GLi2 and Fg-Stk36, suggesting that Ulk4 is the major scaffold to bring Skt36 and Gli2 together, and that there is little if any direct interaction between GLi2 and Stk36.

1. Another method to verify hULK4-Stk36-Gli2 complex formation (Figure 4) would be helpful. For example, proximity ligation assays, tripartite split GFP assays, or colocalization based on expansion STED immunofluorescence microscopy could be performed to temporally and spatially resolve localization of Ulk4, Stk36 and Gli2 upon SHH stimulation in the primary cilium

Thanks for the suggestions. We think that our current study using biochemical and cell biology approaches have provide sufficient evidence that Ulk4, Stk36 and Gli2 form complexes. We will keep in mind of those more sophisticated methods in our future endeavors.

1. Please provide more representative images of Ulk4, Stk36 and Gli2 localization in NIH3T3 cells or lower magnification overview images showing more than one cell (Figure 5).

We have provided more representative images in Figure 5- figure supplement 1A-F of the revised manuscript.

1. Confirmation of the results shown in Figure 5 in a second cell line would strengthen the data.

We have confirmed the results in MEFs (see Figure 5- figure supplement 1G-J)

1. Did the authors add immunofluorescence for tubulin as a ciliary base marker to ensure correct assignment of ciliary tip versus ciliary base localization for quantification experiments (Figures 5 - 8)?

It has been well documented that GLi2 is accumulated at the ciliary tip in respond to Shh treatment; therefore, we used Gli2 as a marker for ciliary tip where both Ulk4 and Stk36 were also accumulated. γ tubulin staining could be another marker to assign the ciliary tip vs base; however, the antibody combination we have did not allow us to simultaneously stain γ tubulin and acetylated tubulin (Ac-Tub).

1. SMO localization as a further readout of SHH pathway activation might be considered to be added for some of the key results (e.g., Figure 6). Is SMO trafficking affected after depletion or overexpression of ULK4?

Due to the lack of a workable antibody to detect endogenous Smo in our hands, we did not determine whether the trafficking of SMO is affected after depletion or overexpression of ULK4. However, we noticed that a recent study reported that the SHH-induced ciliary SMO accumulation was impaired in Ulk4 siRNA treated cells (Mecklenburg et al. 2021). We include this information and its implication in the discussion section

1. Do the authors see ULK4 only at the ciliary tip after SHH stimulation or is there also a dynamic time-dependent localization along the ciliary shaft? The image in Figure 6E (dKO + Stk36 WT) seems to show ULK4 also in the shaft.

Unlike Smo that is evenly distributed alone the axoneme of primary cilia, ULK4 is mainly accumulated at ciliary tips upon Shh stimulation. Ulk4 is also located at low levels outside the cilia and sometimes in the ciliary shaft during its transit to the ciliary tip (e.g., see Figure 5- figure supplement 1F1-2; J1-2).

1. Is the immunofluorescence signal for Ulk4 significantly reduced after shRNA treatment to deplete Ulk4 (Figure 6A)?

We constructed a cell line that stably expressed ULK4 shRNA. The knockdown efficiency was determined by measuring Ulk4 mRNA expression (Fig 1_figure supplement 1). Because we were unable to obtain a reliable ULK4 antibody for immunostaining, we did not examine by whether ULK4 signal was depleted by Ulk4 shRNA.

1. The labelled ciliary tip resembles in some cases images seen for ciliary abscission. The authors could use membrane/ciliary membrane markers to ensure "intraciliary" localization of the investigated factors.

Thanks for the suggestion. We will try that in our future experiments.

1. How many replicates were used in the three independent quantitative RT-PCR experiments (Figure 1 A-D)?

We used 3 replicates in each independent quantitative RT-PCR assay.

1. Please provide p values or statement on no significance for the comparison between Ulk3 single and Ulk3/Ulk4 double knockdown (Figure 1C) and between Stk36 single and Stk36/Ulk4 double knockdown (Figure 1D; Fig1_Figure Supplement 2).

Thanks for the suggestion, we have added the p value or “ns” as asked.

1. Figure legends in general are a bit short could have some more detailed information.

Thank you for the suggestion, we have revised the Figure legends as asked.

1. What do the asterisks present in Figure 4 C-D?

Thanks for the suggestion. The asterisks in Figure 4C-D indicated the full length STK36 and truncated form STK36N and STK36C fragments. We that included this information in the figure legend.

1. The authors state that a previous study described ULK4 as a genetic modifier for holoprosencephaly and that this raised the possibility that ULK4 may participate in HH signal transduction. Primary ciliary localization of ULK4 in mouse neuronal tissue and SHH pathway modulation by ULK4 in cell culture have been shown by Mecklenburg et al. 2021 before. Maybe the authors could rephrase their introduction and discussion accordingly.

Thanks for the suggestion, we have changed the introduction and discussion accordingly.

1. Overexpression studies in heterologous systems using tagged proteins can potentially have an influence on their subcellular localization and function. Please discuss this caveat.

We have mentioned this caveat in the “discussion” section of the revised manuscript. However, we have tried to express the transgene at low levels using the lentiviral vector containing a weak promoter to ensure that the exogenously expressed proteins are still regulated by Hh signaling. We have also confirmed that the tagged Ulk4 and Stk36 can rescue the loss of endogenous genes.

1. More details in the Methods section should be provided on the SHH induction in NIH3T3 cells, HEK293 cells and MEFs.

We have revised the methods section on Shh induction.

1. ULK4 is known to have at least three isoforms that exhibit varying abundance across developmental stages in mice and humans (Lang et al., 2014) (DOI:10.1242/jcs.137604). Can the authors speculate on potential common and distinct functions of the different ULK4 isoforms on SHH pathway modulation based on their present results?

It is interesting that Ulk4 has multiple isoforms in both mouse and human. Several short isoforms in both mouse and human lack the pseudokinase domain while one short isoform in mouse lacks the C-terminal region essential for Ulk4 ciliary tip localization. We speculate that the C-terminally deleted isoform may not have a function in the Shh pathway based on our results shown in Fig. 7 and 8 but might still have functions in other cellular processes.

Reviewer #2 (Recommendations For The Authors):

The paper is well written, and clear throughout, with excellent (up-to-date) citations to the field.

We thank reviewer #2 for the positive comments.

Reviewer #3 (Recommendations For The Authors):

My only quibble is that the immunofluorescence images are a little confusing, especially to people outside of the field. Please include an image of the whole field and improve the captions. Is that a single cell for each cilia? Why are there so few cilia? The DAPI makes it seem like What are we looking at? Are those multiple nuclei in Figure 6? They seem a little small if that's the 5 uM scale bar

We provide uncropped images of Figure 5E to show the entire cells (below). We have added some context to improve the captions. Most of the mammalian cells such as MEF and NIH3T3 cells contain a single primary cilium; however, mutilated cells do exist. The DAPI staining indicated the nuclei. The cells shown in Figure 6 have single nucleus (the scale should be 2 µM). Due to the unevenness of DAPI signals in the nuclei, only the strong signals (puncta) were shown for individual nuclei.

Author response image 1.

One small typo: GLL2 instead of GLI2 on line 363

Thanks, we have corrected this spelling mistake.

Author Response

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

We sincerely thank and express our appreciation to each of the reviewers for their thorough critique of our manuscript.

Reviewer #1 (Recommendations For The Authors):

1. The analysis of whole study comes from only 4 cells from L2/3 of ferret visual cortex; however, it is well known that there is a high level of functional heterogeneity within the cortical neurons. Do those four neurons have similar or different response properties? If the four neurons are functionally different, the weak or no correlation may result from heterogenous distribution pattern of mitochondria associated with heterogenous functionality.

This is an important consideration and often a limitation of CLEM studies. While cortical neurons do exhibit a high degree of functional heterogeneity (similar to spine activity), the 4 cells examined all had selective (OSI > 0.4) somatic responses to oriented gratings, although they differed in their exact orientation preference. Due to experimental limitations of recording from a large population of dendritic spines, we did not characterize other response properties for which their sensitivity might differ. We did not consider orientation preference a metric of study, but instead characterized the difference in preference from the somatic output, allowing comparisons across spines. In addition, our measurements were limited to proximal, basal dendrites rather than any location in the dendritic tree. Nonetheless, we attempted to address this concern by examining spines with functionally heterogenous visual responses within single cells, as reported in our manuscript: mitochondrial volume within a 1 µm radius was correlated with difference in orientation preference relative to the soma across all 4 cells, the mean r = 0.49 +/- 0.22 s.d.), suggesting that cell-to-cell variability has a minimal impact on our main conclusions.

Even with our limited measurements, there is a large amount of functional heterogeneity in dendritic spine responses (Extended Data Figure 2, Scholl et al. 2021), far greater than the small differences in somatic responses of these 4 cells (Figure 3, Scholl et al. 2021). Moreover, the individual dendrites from these 4 cells exhibited substantial heterogeneity in the distribution of mitochondria. We cannot rule out whether heterogeneity at various scales may obscure certain relationships or result in the weak correlations we observed. We also note that future advancements in volume electron microscopy should allow for greater sample sizes that can better address the role of functional (and structural) heterogeneity. We have added text in the Discussion section about the potential structure-function relationships that might be obscured or revealed by neuron heterogeneity.

1. The authors argued that "mitochondria are not necessarily positioned to support the energy needs of strong spines." However, the overall energy needs of a spine depend not only on the synaptic strength but also on the frequency of synaptic activity. Is there a correlation between the mitochondria volume around a spine and the overall activity of the spine? This data needs to be analyzed to confirm the distribution of mitochondria is independent of local energy needs.

The reviewer is correct, but our experimental paradigm was not optimally designed to measure the ‘frequency’ of synaptic activity in vivo. This could have been accomplished with flashed gratings or, perhaps, presenting drifting gratings at different temporal frequencies. For spines tuned to higher temporal frequencies in V1, we may expect greater energy needs, although as the reviewer suggests, energy needs will depend on synapse (and bouton) size. Because we are not able to directly measure activity frequency as carefully or beautifully as can be done ex vivo or in nerve fibers, we do not feel confident in attempting such analysis in this study. Instead, based on previous studies, we assumed that larger synapses might be able to transmit higher frequencies, and thus have higher energy demands. We believe future in vivo studies will more directly measure synaptic frequency for comparison with mitochondria.

We have added text in the Discussion about this caveat and potential future experiments.

1. In the results section, the authors briefly mentioned that "We also considered other spine response properties related to tuning preference; specifically, orientation selectivity and response amplitude at the preferred orientation. For either metric, we observed no relationship to mitochondria within 1 μm radius (selectivity: 1 μm: r = -0.081, p = 0.269, n = 60; max response amplitude: 1 μm: r = -0.179, p = 0.078, n = 64) but did see a weak, significant relationship to both at a 5 μm radius (selectivity: r = 0.175, p = 0.027, n = 121; max response amplitude: r =-0.166, p = 0.030, n = 129)." Here only statistic results were given while the data were not presented in the figure illustration. Based on the methods and Figure 3B, it seems that the preferred orientations were calculated based on the vector summation. How did the authors calculate the "response amplitude at the preferred orientation"? This needs to be clarified. In addition, given the huge variety of orientation selectivity, using the response amplitude at the preferred orientation may not be the best parameter to correlate with the mitochondria volume which is indicative of energy needs. It might be necessary to include the baseline activity without visual stimulation and the average response for visual stimuli of different orientations in the analysis.

We apologize for this oversight, as the details are present in our previous study (Scholl et al., 2021). Response amplitude and preferred orientation were calculated from a Gaussian curve fitting procedure with specific parameters describing those exact values (see Scholl et al. 2021 or Scholl et al. 2013). Only spines with selective responses (vector strength index > 0.1) and passing our SNR criterion were used for these analyses. We have now added this information to the Methods section and referred to it in the Results. With respect to the reviewer’s other concern, we also examined the average response amplitude (across all visual stimuli). There we found no relationship between the volume of mitochondria within 1 or 5 microns of a spine, however, because spines differ greatly in their selectivity (range = 0 – 0.8) the average response may not be an appropriate metric to compare across spines.

1. A continuation from the former point, do the spines with similar preferred orientation to the somatic Ca signal have similar Ca signal strength, orientation selectivity index and other characteristics to the spines with different preferred orientation? As shown in the examples (Figure 3B), the spine on the right with different orientation preference compared with its soma has considerably larger response in non-preferred orientation compared to the spine on the left. Thus, the overall activity of the spine on the right may be higher than the spine which has similar preferred orientation to the soma. The authors showed that a positive correlation between difference in orientation preference and mitochondria volume (Figure 3C). Could this be simply due to higher spine activity for non-preferred orientation or spontaneous activity? Thus, the mitochondria might be positioned to support spines with higher overall activity rather than diverse response property.

The reviewer brings up an interesting consideration. We examined the response properties of spines co-tuned (∆θpref < 22.5 deg) and differentially-tuned (∆θpref > 67.5 deg) to the soma. The orientation selectivity was not different between the two groups (p = 0.12, Wilcoxon ranksum test), although there was a small trend towards co-tuned inputs being more selective. We found that calcium response amplitudes for the preferred stimulus were also not different (p = 0.58, Wilcoxon ranksum test). These analyses are now included as a sentence in the Results.

We agree with the reviewer that higher spontaneous activity in non-preferred spines may help explain the mitochondrial relationship we observe. However, our current dataset does not have sufficiently long recordings to measure spontaneous synaptic activity. Further, when considering a non-anesthetized preparation, spontaneous activity is highly dependent on brain state and an animal’s self-driven brain activity, which all must be experimentally controlled or measured to accurately address this.

1. In addition, the information about the orientation selectivity of the soma is also missing. Do the four cells shown here all have similar level of orientation selectivity? Or some have relative weak orientation selectivity in the soma?

Yes, all 4 cells have a similar OSI (range = 0.4 – 0.57, mean = 0.46 +/ 0.08 s.d.). This has been added to the Results section.

1. This study focused on only a fraction of spines that are (1) responsive (2) osi > 0.1. However, in theory energy consumption is also related to non-responsive spines and spines with weak orientation tuning. What is the percentage of tuned and untuned spines? What's the correlation of mitochondria volume and spine activity level for untuned spines? I also recommend including the non-responsive spines into the analysis. For example, for each mitochondrion calculate the averaged overall activity of spines within certain distance from the mitochondrion, including the non-responsive spines. I would predict there may be more active spines and higher overall spine activity of dentritic segments near a mitochondrion than segments far from a mitochondrion.

A majority of spines were tuned for orientation (~91%), although we specifically chose to only analyze data from spines with verifiable, independent calcium events. All analyses except those involving measurements of orientation preference use all dendritic spines (i.e. tuned and untuned). We have clarified this in the Results.

These other ‘silent’ (i.e. without resolvable visual activity) spines may significantly contribute to energy demands of a dendrite too, as our methods (GCaMP6s expression) likely only capture synaptic events driving Ca+2 influx through NMDA receptors or VGCCs. We expect that glutamate imaging (e.g. iGlusnfr) may open the door to additional analyses to fully characterize functional relationship between spines and mitochondria.

1. The correlation coefficient for mitochondria volume and difference in orientation preference is relatively low (r=0.3150). With such weak correlation, the explanatory power of this data is limited.

We agree that while the correlation is significant, it is not particularly strong. To better represent the noise surrounding this measurement, we performed a bootstrap correlation analysis, sampling with replacement (1 micron: mean r = 0.31 +/- 0.11 s.e., 5 micron: mean r = 0.02 +/- 0.10 s.e.). We now include this in the Results.

1. Why do the numbers of spines in different figures vary? For example, n=60 for 1micron in Figure 3, 54 in Figure 3c, 31 in Figure 4b, 51 in Figure 4e and so on.

We apologize for the lack of clarity. Each analysis presented different requirements of the data. For example, orientation preference was computed only for selective (OSI > 1) spines (Fig. 3c), but this requirement did not apply to comparisons with selectivity or response amplitude (Fig. 3d). Similarly, as stated in the Results and Methods, measurements of local heterogeneity require a minimum number of neighboring spines (n > 2), limiting the number of usable spines for analysis (Fig. 4). We have clarified this in the text.

1. In Figure 6a, the sample sizes of mito+ spines and mito- spines are extremely unbalanced, which affects the stat power of the analysis. I recommend performing a randomization test.

We thank the reviewer for this suggestion. We ran permutation tests to compare the similarity in mean values between equally sampled values from each distribution. These tests supported our original analysis and conclusions. We have added these tests to the Results.

1. Ca signals are approximations of electrical signals. How well are spinal calcium signals correlated to synaptic strength and local depolarization? This should be put into discussion.

There is unlikely a simple, direct relationship between spine calcium signal and synaptic strength or membrane depolarization, and this has never been addressed in vivo. Koester and Johnston (2005) performed paired recordings in slice and showed that single presynaptic action potentials producing successful transmission generate widely different calcium amplitudes (Fig. 3). Another study from Sobczyk, Scheuss, and Svoboda (2005) used two-photon glutamate uncaging on single spines and showed that micro-EPSC’s evoked are uncorrelated with the spine calcium signal amplitude. We have added a note about this in the discussion.

1. In Figure 4i, the negative correlation may depend on the 4 data points on the right side. How influential are those data points?

Spearman’s correlation coefficient analysis is robust to outliers and it is highly unlikely these datapoints are critical with our sample size (n > 100 spines).

1. Raw data of Ca responses were missing.

Some data has been published with the parent publication (Scholl et al., 2021). As spine imaging data is difficult to obtain and highly unique, we prefer to provide raw data directly upon reasonable request of the corresponding author.

1. What is the temporal frequency of the drifting grating? Was it fixed or the speed of the grating was fixed?

This was fixed to 4 Hz and this is now included in the Methods.

Reviewer #2 (Recommendations For The Authors):

1. Most of the measurements were based on the distance from the base of the spine neck, and "only on spines with measurable mitochondrial volume at each radius" were analyzed. To better understand the causality, it may also be interesting to have an analysis based on the distance from mitochondria. Would the result be different if the measurements are not 1µm / 5µm from spine but 1µm / 5µm from mitochondria? (e.g. total spine volume in 1µm / 5µm from mitochondria).

In fact, our first iteration of this study focused on exactly this metric: measuring the distance to nearest mitochondria. However, after lengthy discussions between the authors, we ultimately decided this metric was inferior to a volumetric one. Our decision was based on several factors: (1) distance to mitochondrion is ill defined (e.g. distance to the a mitochondrion center or nearest membrane edge?), (2) the total amount of mitochondrial volume within a dendritic shaft should allow the greatest amount of energetic support (e.g. more cristae for ATP production, greater capacity for calcium buffering), and (3) we would not account for the geometry of individual mitochondria or their placement near a spine (e.g. when 2 different mitochondria are next to the same spine) We have added further clarification of our reasoning to the Results.

Nonetheless, we present the reviewer some of our original analyses correlating distance to mitochondria (from the base of the spine and including the spine neck length):

Author response image 1.

Here, we examined the relationship to spine head volume, spine-soma orientation preference difference, and the local orientation preference heterogeneity. No relationship showed any significant correlation. Again, this may not be surprising given the drawbacks of measuring ‘distance to mitochondria’.

1. Is there a selection criterion for the spine for the analysis? Are filopodia spines excluded in the analysis?

Spines were analyzed regardless of structural classification; however, they were only analyzed if they had a synaptic density with synaptic vesicle accumulation. In our dataset (including those visualized in vivo and reconstructed from the EM volume) we observed no filopodia.

1. The result states that "56.8% of spines had no mitochondria volume within 1 μm and 12.1% of spines had none within 5 μm.". In other words, around 43% of spines had mitochondria within 1 μm. It would be interesting to show whether there is a correlation between mitochondria size and spine density.

We agree that this is an interesting measurement. It has been reported that mitochondrial unit length along the dendrite co-varies with linear synapse density in the neocortical distal dendrites of mice (Turner et al., 2022). This was specifically true in distal portions of dendrites more than 60 µm from the soma, because mitochondria volume increases as a function of distance roughly up to this point, then remains relatively constant beyond this distance.

To investigate this possibility, we calculated the local spine density around an individual spine and compared to the mitochondria volume within 1 or 5 µm. We found no evidence of a correlation between local spine density and the volume of mitochondria (1 µm: Spearman r = -0.07447, p = 0.2859; 5 µm: r = -0.04447, p = 0.3141). However, the majority of our measurements are more proximal than 60 µm (our median distance of all spines = 49.4 µm, max = 114 µm) and this may be one reason why observe no correlation.

1. In Figure 3B, the drifting grating directions are examined from 0 to 315 degrees in the experiment. However, in Figure 3C and 3D, the spine-soma difference of orientation preference was limited to 0 to 90 degree in the graph. Is the graph trimmed, or is there a cause that limits the spine-soma difference of orientation preference to 90?

Ferret visual cortical neurons are highly sensitive to grating direction and the responses are fit by a double Gaussian curve which estimates the ‘orientation preference’ (0-180 deg). We then calculated the absolute difference in orientation preference and wrapped that value in circular space so the maximum difference possible is 90 deg (e.g. 135 deg -> 45 deg).

1. In Figure 4D-F, how is the temporal correlation of calcium activity determined? Is it based on stimulated activity or basal activity? A brief explanation may be helpful to the readers. Also, scale bars could be added to Fig 4D.

Temporal correlation is computed as the signal correlation between 2 spines over the entire imaging session at that field of view. Specifically, we measured the Pearson correlation between each spine’s ∆F/F trace. To measure the local spatiotemporal correlation, we computed correlations between all neighboring spines within 5 microns and took the average of those values. We have clarified this in the Results section.

1. Figure 3C and Figure 4D displayed a significant correlation in 1µm range and such correlation drastically diminished once the criterion changed to 5µm range. It would be interesting to include the criterion of intermediate ranges. It would be interesting to see if there is a trend or tendency or if there is a "cut-off" limit.

We agree with the reviewer that the drastic change in the correlations between 1 and 5 µm range was surprising to see. While these volumetric measurements are time consuming, we returned to our data and measured an intermediate point of 3 µm. Investigating relationships reported in our study, we found no significant trends for spine-soma similarity (Spearman’s r = -0.011, p = 0.54) or local heterogeneity (Spearman’s r = 0.11, p = 0.23). This suggests that a potential ‘critical distance’ might be less than 3 µm; however, far more additional measurements and analyses would be needed to attempt to identify exactly what this distance is.

1. In Figure 5, it is shown that spines having mitochondrion in the head or neck are larger. However, only 10 spines are found with mitochondria inside. In the current dataset, are mitochondria abundantly found in large spines? Further analysis or justification would be informative to address this.

In our dataset, mitochondria were found in ~5% of all spines. Spines with mitochondria have a median volume of approximately 0.6 µm3, roughly twice as large as than those without mitochondria, as the reviewer suggests. In the entire population of spines without mitochondria, a volume of 0.6 µm3 represents roughly the 82nd percentile. In other words, of the total population of 157 spines without mitochondria, only 29 had equal or greater volume than the median spine with a mitochondrion. We believe this trend is clearly shown in Figure 5A and is supported by our analysis, including new permutation tests suggested by Reviewer 1.

Reviewer #3 (Recommendations For The Authors):

1. The authors state that their unsupervised method "quickly and accurately identified mitochondria," but the methods section only says that segmentations were proofread. Was every segmentation examined and judged to be accurate, or was only a subset of the 324 mitochondria checked?

After deep learning-based extraction, each mitochondrion segmentation was manually proofread. For each dendrite segment, this was ~10-20 mitochondria, so it did not take long to manually inspect and edit each mitochondrion segmentation.

1. The EM image of the mitochondrion in the spine head in Figure 2C is low resolution and does not apply to the bulk of the data. Images more representative of the analyzed data should be added to supplement the cartoons.

Our primary rationale for including this specific image was to show that the mitochondria located within spines are small, round, and to include a view of the synapse as well as the mitochondrion. We now include enlarge and additional EM images to Figure 1C.

1. The majority of spines did not have any mitochondria within a 1 micron radius and were excluded from the correlation analyses, so most of the conclusions are based on a minority of spines. It would be interesting to see comparisons between spines with and without nearby mitochondria. Correlations between the absolute distance to any mitochondrion, synapse size, and mismatch to soma orientation would be especially interesting.

The reviewer brings up a good point. It is true that many spines were excluded from our analysis based on the fact that they did not have nearby mitochondria within 1 or 5 µm (56.8% of spines had no mitochondria volume within 1 μm and 12.1% of spines had none within 5 μm). We compared the distributions of synapse size, mismatch to soma, and orientation selectivity of two groups of spines – those with at least some mitochondria within 1 µm (n = 65) versus spines without any mitochondria within 5 µm (n = 19).

We found no difference in the distributions between spine volume (1 µm: median = 0.29 µm3, IQR = 0.41 µm3; no mitochondria within 5 µm: median = 0.40 µm3, IQR = 0.37 µm3; p = 0.67) or PSD area (1 µm: median = 0.26 µm2, IQR = 0.33 µm2; no mitochondria within 5 µm: median = 0.31 µm2, IQR = 0.36 µm2; p = 0.49). For functional measures, we also saw no difference in orientation selectivity (1 µm: median = 0.29, IQR = 0.28; no mitochondria within 5 µm: median = 0.28, IQR = 0.15; p = 0.74) or mismatch to soma orientation (1 µm: median = 0.54 deg, IQR = 0.86 deg; no mitochondria within 5 µm: median = 0.46 deg, IQR = 0.47 deg; p = 0.75). We now include analyses in the Results.

We also looked at the absolute distances to mitochondria and did not find any significant relationships to spine head volume, spine-soma orientation preference difference, or the local orientation preference heterogeneity (see our response to reviewer #2 for more information).

1. In Figure 1A the mitochondria appear to be taking up a substantial fraction of the dendritic shaft diameter, even for distal dendrites. It would be useful to know the absolute diameter of the dendrites and mitochondria, given that this is not rodent data and there is no reference for either in the ferret.

We agree with the reviewer’s point, although we would like to remind the reviewer that these are basal dendrites of layer 2/3 cells. Basal dendrites tend to be thinner than apical branches. Interestingly, in some cases, the dendrite even swells to accommodate a mitochondrion. We did not incorporate this measurement in our study because it is not trivial; dendrite diameter is variable and dendrites are not perfect cylinders. Although we did not make precise measurements across our dendrites, the diameter is comparable to what has been seen in mouse cortex (Turner et al., 2022), roughly 500-1000 nm, but as small as 100 nm at some pinch points. In terms of mitochondria, many were roughly spherical or oblong, therefore the maximum diameters we report are roughly similar to, if not a bit larger than, those of the cross-sectional diameter.

1. As a rule, PSD area is correlated with spine volume, which makes the observation that spines with mitochondria have larger volume but not PSD area surprising. With n=10 it is difficult to draw conclusions, but it would be interesting to know the PSD area-to-volume ratio of other spines of the same volume and synapse size.

We were also somewhat surprised to see this, but exactly as the reviewer mentioned, we believe it to be a limitation of the sample size. The difference in volume was large enough to be detected despite a small sample size. We saw a trend towards larger synapses when spines have mitochondria (the median was approximately 60% larger), and we would expect with a larger comparison that PSD area would be significantly greater in spines with mitochondria.

We calculated the PSD area-to-spine head volume ratio for spines with or without mitochondria. Spines with mitochondria had a significantly lower ratio compared to those without (Mann-Whitney test, p = 0.0056, mito - = 0.78, n = 10; mito + = 0.53, n = 157). As the reviewer mentions, it is somewhat difficult to draw a conclusion from this, but it appears that the PSD does not scale with the increased spine head size.

Author response image 2.

The only way to definitively address this is to increase the sample size, which is becoming easier to achieve with the progression of volume EM imaging and analysis techniques in recent times. We look forward to addressing this in the future.

1. Nothing is made of the significant fact that these data come from the visual system of a carnivore, not a mouse. Consideration of differences in visual physiology between rodents and carnivores would be worthwhile to put the function of these dendrites in context.

We thank the reviewer for this consideration and have added text to the Discussion.

Author Response

Reviewer #2 (Public Review):

Manassaro et al. present an extensive three-session study in which they aimed to change defensive responses (skin conductance; SCR) to an aversively conditioned stimulus by targeting medial prefrontal cortex (their words) using repetitive TMS prior to retrieval. They report that stimulating mPFC using TMS abolishes SCR responses to the conditioned stimulus, and that this effect is specific for the stimulated region and the specific CS-US association, given that SCR responses to a different modality US are not changed.

I like how the authors have clearly attempted to control for several potential confounds by including multiple stimulation sites, measured SCR responses to several unconditioned stimuli, and applied the experiment in multiple contexts. However, several conceptual and practical issues remain that I think limit the value of potential conclusions drawn from this work.

The first issue that I have with this study concerns the relationship between the TMS manipulation and the theoretical background the authors present in their rationale. In the introduction the authors sketch that what they call 'mPFC' is involved in regulation of threat responses. They make a convincing case, however, almost all of the evidence they present concerns the ventromedial part of the prefrontal cortex (refs 18-25). The authors then mention that no one has ever studied the effects of 'mPFC'-TMS on threat memories. That is not surprising given that stimulating vmPFC with TMS is very difficult, if not impossible. Simulation of the electrical field that develops as a consequence from the authors manipulation (using the same TMS coil and positioning the authors use) shows that vmPFC (or mPFC for that matter) is not stimulated. The authors then continue in the methods section stating that the region they aimed for was BA10. This region they presumably do stimulate, however, that does not follow logically from their argument. BA10 is anatomically, cytoarchitectonically and functionally a wholly different area than vmPFC and I wonder if their rationale would hold given that they stimulate BA10.

We would like to thank the Reviewer for highlighting this very important point. The Reviewer is right in stating that the Brodmann area 10 (BA 10) is anatomically, cytoarchitectonically, and functionally distinct from the ventromedial PFC. As we reported in the Methods section, the coil placement over the frontopolar midline electrode (Fpz) according to the international 10‒20 EEG coordinate system directly focused the stimulation over the medial portion of the BA 10. In the literature, the aPFC is also known as the “frontopolar cortex” or the “rostral frontal cortex” and encompasses the most anterior portion of the prefrontal cortex, which corresponds to the BA 10. In line with this observation, we have corrected “medial prefrontal cortex” (mPFC) with “medial anterior prefrontal cortex” (aPFC) throughout the manuscript. We also have corrected the theoretical background and the rationale in the Introduction section by mentioning several studies that: i) Reported the involvement of the aPFC in emotional down-regulation (Volman et al., 2013; Koch et al., 2018; Bramson et al., 2020). ii) Traced anatomical connections between the medial/lateral aPFC and the amygdala (Peng et al., 2018; Folloni et al., 2019; Bramson et al., 2020). iii) Detected functional connections between the aPFC and the vmPFC during fear down-regulation (Klumpers et al., 2010). iv) Found hypoactivation, reduced connectivity, and altered thickness of aPFC in PTSD patients (Lanius et al., 2005; Morey et al., 2008; Sadeh et al., 2015; Sadeh et al., 2016). v) Revealed that strong activation of the aPFC may promote a higher resilience against PTSD onset (Kaldewaij et al., 2021) and that enhanced aPFC activity and potentiated aPFC-vmPFC connectivity is detectable after effective therapy in PTSD patients (Fonzo et al., 2017). Furthermore, we discussed our results in light of this evidence in the Discussion section. We really thank the Reviewer for this key implementation of our study.

The second concern I have is that although I think the authors should be praised for including both sham and active control regions, the controls might not be optimally chosen to control for the potential confounds of their condition of interest (mPFC-TMS). Namely, TMS on the forehead can be unpleasant, if not painful, whereas sham-TMS or TMS applied to the back of the head or even over dlPFC is not (or less so at the very least). Given that the SCR results after mPFC TMS show exactly the same temporal pattern as the sham-TMS but with a lower starting point, one could wonder whether a painful stimulation prior to the retrieval might have already caused habituation to painful stimulation observed in SCR in consequent CS presentations. A control region that would have been more obvious to take is the lateral part of BA10, by moving the TMS coil several centimeters to the left or right, circumventing all things potentially called medial but giving similar unpleasant sensations (pain etc).

We would also like to thank the Reviewer for bringing to light this issue and allowing us to strengthen our results. The Reviewer is right in pointing out that rTMS application over the forehead can be subjectively perceived as unpleasant, relative to other head coordinates or sham stimulation. The question of whether an unpleasant stimulation prior to the retrieval might provoke habituation to discomfort sensations and lead to weaker SCRs in the consequent CS presentations is valid and reasonable. We also thank the Reviewer for advising us to stimulate the lateral part of BA 10 as an active control site. However, given the potential involvement of the lateral BA 10 in the fear network (see previous point) and the potential risks due to the anatomical proximity of lateral BA 10 with the temporal lobe, we reasoned to adopt an alternative approach to investigate whether “a painful stimulation prior to the retrieval might have already caused habituation to painful stimulation observed in SCR in consequent CS presentations”. We repeated the entire experiment in one further group (ctrl discomfort, n = 10) by replacing the rTMS procedure with a 10-min discomfort-inducing procedure over the same site of the forehead (Fpz) to mimic the rTMS-evoked unpleasant sensations in the absence of neural stimulation effects (see the new version of the Methods section). The electrical stimulation intensity was individually calibrated through a staircase procedure (0 = no discomfort; 10 = high discomfort). The shock amplitude was set at the current level corresponding to the mean rating of ‘4’ on the subjective scale because, in the new experiments that we performed targeting the aPFC with rTMS (n = 9), we collected participants’ rTMS-induced discomfort ratings obtaining a mean rating of 3.833 ± 0.589 SEM on the same scale. We found CS-evoked SCR levels not significantly different to those of the sham group during the test session as well as during the follow-up session, suggesting that the discomfort experienced during the rTMS procedure did not contribute to the reduction of electrodermal responses observed in the aPFC group. We reported the results of this experiment in the Results section and Figure 2-figure supplement 2.

My final concern is that the main analyses are performed on single trials of SCR responses, which is a relatively noise measure to use on single trials. This is also done in relatively small groups (n=21). I would have liked to see both the raw or at least averaged timeseries SCR data plotted, and a rationale explaining how the authors decided on the current sample sizes, if that was based on a power analyses one must have expected quite strong effects.

Following the Reviewer’s suggestion, we decided to remove the analysis on single trials, and we apologize for not including SCR timeseries. To quantify the amount of effect induced by the rTMS protocol, we have now added within-group comparisons (through 2 × 2 mixed ANOVAs) that show, for each group, the amount of change in CS-evoked SCRs from the conditioning phase to the test phase, as well as from the conditioning phase to the follow-up phase. Furthermore, to directly and simply depict these changes, in addition to dot plots, we have also represented them with line charts (Figs. 2C, 2H, 4C, 4H, 5C, 5H). To estimate the sample size, we had previously performed a power analysis through G*Power 3.1.9.2 and it had resulted in n = 21 per experimental group. However, by correcting data pre-processing procedures (in accordance with Reviewer 1), we obtained data that were not normally distributed. Thus, we reasoned to enlarge our sample width by re-performing a power analysis (with the new suggested statistical analyses) and then repeating the experiments. For the main statistics, i.e. mixed ANOVA (within-between interaction) with two groups and two measurements, with the following input parameters: α equal to 0.05, power (1-β) equal to 0.95, and a hypothesized effect size (f) equal to 0.25, the new estimated sample size resulted in n = 30 per experimental group.

eLife assessment

This study presents the useful observation that repetitive Transcranial Magnetic Stimulation (rTMS) over the medial prefrontal cortex (mPFC) is associated with immediate dampening effects of conditioned responses and generalization of these responses to similar cues. Additionally, the effects were still present one week later, in the absence of any stimulation. However, the evidence supporting the claims of the authors is incomplete. The main outcome data (skin conductance response) have been normalized and standardized in suboptimal ways and, most critically, no comparisons are being made with the strength of conditioned responses during acquisition. If the observations hold, when based on within-subject comparisons, the work will be of interest to psychologists and neuroscientists working on interventions into aberrant emotional memories.

Reviewer #1 (Public Review):

In this manuscript, Manessero and colleagues argue that the prefrontal cortex (PFC), given its exquisite capability to down-regulate down-stream regions central in driving emotional responses to threat, maybe a promising target to stimulate in order to reduce aberrant fear memory responses. They aim to differ from previous studies that tested the strengthening of extinction learning, by merely focusing on the expression of threat memory without extinction learning. Given that other studies have often focused on the dorsolateral prefrontal cortex as promising target to regulate fear responses, they also ran experiments to directly compare effectiveness of targeting the mPFC and dlPFC in reducing fear memory responses. These aims are all focused on what the authors describe as "implicit memory", but they also test the effects of the interventions on "explicit memory" of the presented cues. However, in the introduction, the authors do not explicitly describe what their aim or theoretical rationale to implement these tests was. Likewise, the authors implemented generalisation stimuli (i.e., cues similar to the original CS) in the implicit memory tests, but the aim of these tests is also not explained.

In order to test their hypotheses, the authors adopt a single-cue fear conditioning paradigm where participants learned to associate an auditory cue with the occurrence of short electrical stimulation across 15 repetitions of the CS-US pairing (80% reinforcement rate). One week later, for the second session, this cue was again presented 4 times, along with 2 types of generalization stimuli, that were each also presented four times. This test session took place in another environment. Conditioned skin conductance responses were measured as index of defensive responding. In the critical condition, during 10 minutes prior to these cue presentations, repetitive Transcranial Magnetic Stimulation (rTMS) was applied to specifically target the medial PFC. Another independent group of participants completed a two-alternative forced-choice (2AFC) explicit recognition test, to inquire to what extent they could recognize whether a given tone was presented during the conditioning phase (basically a source memory task). Finally, a two-alternative forced-choice (2AFC) perceptual discrimination test was presented, to ascertain that participants could discern the different tones presented. The second session was repeated yet another week later, but without any rTMS and in the original conditioning context again, to test whether any potential fear dampening effects were retained.

The observations are quite straightforward: compared to sham and an active control group, mPFC stimulation prior to fear memory retrieval resulted in an immediate reduction of conditioned responses, a difference that was consistent across all 4 test trials. Also conditioned responses to the generalization cues were reduced upon mPFC stimulation. These effects seemed to be specific for memories, since responding to novel unconditioned cues (loud female scream) were not affected by prior mPFC stimulation. Likewise, measures of explicit memory were unaffected. In separate experiments, stimulation of the mPFC also outperformed stimulation of the dlPFC. This pattern of results was again observed during the tests a week later.

The authors conclude that, since these outcomes were observed in the absence of extinction training, the rTMS procedure directly modulated the defensive responses activated by the threat memory trace. The fact that defensive responses to novel unconditioned stimuli were not affected are in line with earlier observations that the mPFC seems critical for the expression of conditioned but not innate fear. Given that dlPFC stimulation seemed less effective, the mPFC may be the most suitable candidate for future therapeutic interventions.

Major strengths:<br /> - Earlier work delving into the involvement of the prefrontal cortex in fear regulation has not only revealed a central role for the mPFC, but also for the dlPFC. An important strength of this study is that the authors therefore also directly compare groups that are targeted in either one of these regions, thereby revealing that even though stimulating the dlPFC results in some fear reduction, the effect is much stronger for mPFC. Another nice consequence of this extra group is that the earlier observations when targeting the mPFC are being replicated.

- It is important to test novel avenues to achieve enduring fear dampening effects of interventions. An intervention that only exerts immediate but transient effects does not bring much clinical value. So the fact that this study incorporates a follow-up test and then shows that the acute fear dampening effects are retained in the absence of any TMS stimulation certainly is important.

- It is only natural to show defensive responses to cues that previously have been paired with something aversive, like a shock. For this reason, generalized fear responses to cues that are similar to fear cues but in fact innocuous is considered maladaptive, and at the core of anxiety disorders. A strength of the paper is that the authors have added generalization tests in addition to (adaptive) fear retention, to ascertain that their intervention in fact also targets maladaptive responding.

Major weaknesses<br /> - There are two major weaknesses in this paper, that can have a potentially detrimental consequence for the robustness of the results and conclusions. First of all, even though comparing the effect of mPFC stimulation with other groups that have been stimulated in other brain regions is important, another comparison - perhaps an even more essential one - is lacking: is there a significant reduction in conditioned fear responses after targeting the mPFC as compared to that group's own fear acquisition (or at least the final phase of acquisition)? Instead, the authors compare fear responses with responses PRIOR to conditioning, which is not meaningful. The same goes for the long-term follow-up: also here, a comparison with fear responding prior to the intervention is lacking. Such a reduction in conditioned fear responding should be larger than any reduction (e.g., due to habituation or forgetting) in fear responding in the (sham) control groups (i.e., an overall interaction between group and fear responding should be present). Whether this is the case is unfortunately unknown, since the fear acquisition data (neither raw, nor pre-processed) are not to be found in the manuscript, and are therefore also not included in any of the analyses. Since there is also no safe control stimulus, the crucial comparison is made entirely between-subjects, and for such a comparison groups of n~20 are quite modest.

- Second, against commons practice, the authors commence by square root transforming all SCR data to normalize the data (while this should only be done in the final phase or preprocessing, if the variables entered in the statistical tests require so), only to then again normalize these obtained values by dividing by the unconditioned responses of the participants, that then are used to calculate differences scores with preconditioning. In these descriptions it is unclear which unconditioned stimulus it was (the original one, from conditioning?) and whether it was standardised to the highest response or an average of all the responses. Decisions that are taken in these early pre-processing steps can have a gigantic impact on the outcomes and conclusions, so this is not trivial. One may say that this cannot explain the group effects that have been observed, given the fact that all groups have been pre-processed in the same way. However, the mPFC group of interest seems to display relatively high unconditioned responses - standardising with these measures may result in relatively low conditioned responses in this particular group. This shortcoming is therefore closely related to the point made above: given that conditioning data would be standardised in the same vein, a test that included the within-subject comparison between acquisition and post-intervention is absolutely crucial to ascertain that the effects observed are not merely due to coincidences in pre-processing values and pre-existing group differences.

- In addition to the above-described main analyses, some other potential weaknesses concern the analysis strategy applied to the generalization tests. Several ANOVAS are being run, one to test for the pattern of generalization responses within-subjects (i.e., the CS, NS1 and NS2), and several ones to compare each of these between the three groups. But such analyses are not warranted in the absence of an overall interaction between the within subject factors and group factor. Such overall omnibus tests however are lacking, and the high number of separate anovas risks false positives (i.e., these comparisons should have been made with planned contrasts). The fact that the included factors and levels are not being described, makes it generally hard to gauge what variables exactly have been entered in every analysis.

Further remarks:<br /> - There is a possibility that a re-analysis of the data using properly preprocessed SCR data along with analyses that include comparisons with the conditioned responses during acquisition reveal a different pattern of results. Therefore, whether the authors truly achieved their aims and whether the results support their conclusions is as of yet undecided.

- Even if the pattern of results holds, then the claim that the long-term follow-up reductions of fear were achieved in the absence of any extinction cannot be made with confidence: after all, upon mPFC stimulation during the second session, the CS was presented four times, and so were each of the two generalization stimuli. So perhaps extinction was not complete, but almost certainly some extinction has taken place: it is well-known that the strongest extinction-learning typically takes place in the first trials (e.g., due to higher prediction errors). The authors do not give any alternative theoretical explanation for the enduring reduction of fear reduction, which would be interesting to learn their thoughts on.

- If the results hold and satisfiable reasons are provided as to why the effects remain visible in the follow-up, this study could be a valuable contribution to the field: it may refocus future studies to the mPFC as major target to not only promote acute fear regulation, but perhaps even more importantly form a clinical perspective, a route for enduring fear reductions.

Reviewer #2 (Public Review):

Manassaro et al. present an extensive three-session study in which they aimed to change defensive responses (skin conductance; SCR) to an aversively conditioned stimulus by targeting medial prefrontal cortex (their words) using repetitive TMS prior to retrieval. They report that stimulating mPFC using TMS abolishes SCR responses to the conditioned stimulus, and that this effect is specific for the stimulated region and the specific CS-US association, given that SCR responses to a different modality US are not changed.

I like how the authors have clearly attempted to control for several potential confounds by including multiple stimulation sites, measured SCR responses to several unconditioned stimuli, and applied the experiment in multiple contexts. However, several conceptual and practical issues remain that I think limit the value of potential conclusions drawn from this work.

The first issue that I have with this study concerns the relationship between the TMS manipulation and the theoretical background the authors present in their rationale. In the introduction the authors sketch that what they call 'mPFC' is involved in regulation of threat responses. They make a convincing case, however, almost all of the evidence they present concerns the ventromedial part of the prefrontal cortex (refs 18-25). The authors then mention that no one has ever studied the effects of 'mPFC'-TMS on threat memories. That is not surprising given that stimulating vmPFC with TMS is very difficult, if not impossible. Simulation of the electrical field that develops as a consequence from the authors manipulation (using the same TMS coil and positioning the authors use) shows that vmPFC (or mPFC for that matter) is not stimulated. The authors then continue in the methods section stating that the region they aimed for was BA10. This region they presumably do stimulate, however, that does not follow logically from their argument. BA10 is anatomically, cytoarchitectonically and functionally a wholly different area than vmPFC and I wonder if their rationale would hold given that they stimulate BA10.

The second concern I have is that although I think the authors should be praised for including both sham and active control regions, the controls might not be optimally chosen to control for the potential confounds of their condition of interest (mPFC-TMS). Namely, TMS on the forehead can be unpleasant, if not painful, whereas sham-TMS or TMS applied to the back of the head or even over dlPFC is not (or less so at the very least). Given that the SCR results after mPFC TMS show exactly the same temporal pattern as the sham-TMS but with a lower starting point, one could wonder whether a painful stimulation prior to the retrieval might have already caused habituation to painful stimulation observed in SCR in consequent CS presentations. A control region that would have been more obvious to take is the lateral part of BA10, by moving the TMS coil several centimeters to the left or right, circumventing all things potentially called medial but giving similar unpleasant sensations (pain etc).

My final concern is that the main analyses are performed on single trials of SCR responses, which is a relatively noise measure to use on single trials. This is also done in relatively small groups (n=21). I would have liked to see both the raw or at least averaged timeseries SCR data plotted, and a rationale explaining how the authors decided on the current sample sizes, if that was based on a power analyses one must have expected quite strong effects.

Author Response

Reviewer #1 (Public Review):

In this manuscript, the Authors implement a delayed feedback control method and use it for the first time in biological neuronal networks. They extend a well-established computational theory and expand it into the biological realm. With this, they obtain novel evidence, never considered before, that showcases the difference between simulated neuronal networks and biological ones. Furthermore, they optimize the DFC method to achieve optimal results in the control of cell excitability in the content of biological neuronal networks, taking advantage of a closed-loop stimulation setup that, by itself, is not trivial to build and operate and that will certainly have a positive impact the fields of cellular and network electrophysiology.

Regarding the results, it would be very constructive if the Authors could share the code for the quasi-real-time interface with the Multichannel Systems software (current and older hardware versions), as this represents likely a bottleneck preventing more researchers to implement such an experimental paradigm.

On the data focusing on the effects of the DFC algorithms on neuronal behavior, the evidence is very compelling, although more care should be devoted to the statistical analyses, since some of the applied statistical tests are not appropriate. In a more biological sense, further discussion and clarification of the experimental details would improve this manuscript, making it more accessible and clearer for researchers across disciplines (i.e., ranging from computational to experimental Neuroscience) and increasing the impact of this research.

In summary, this work represents a necessary bridge between recent advances in computational neuroscience and the biological implementation of neuronal control mechanisms.

Regarding sharing the control code, our application for closed-loop stimulation using aDFC, DFC and Poisson is now available in GitHub (https://github.com/NCN-Lab/aDFC). This was, in fact, our initial intention following the reviewing process. With this application, the user can run the developed algorithms with the MEA2100-256 System from Multi Channel Systems MCS GmbH.

Same with the data. The dataset with the spike data from all experiments is also now publicly available in Zenodo. The data can be found in https://doi.org/10.5281/zenodo.10138446.

Regarding the improvements in the statistical analysis, the tests are now performed following Reviewer #1 suggestions. Important to emphasize that this did not change the results/ conclusions of the work.

eLife assessment

Large populations of neurons are capable of entering pathological synchronous oscillations under a variety of conditions and work over many decades has found ways to disrupt such oscillations using stimulation in both open loop and closed loop configurations. This study adds useful results and methodology to this line of research, by providing solid evidence that delayed feedback control via electrical stimulation can, under certain conditions, terminate network level oscillations in cultured cortical neurons. The study provides analyses and simulation results that shed light on why some networks respond to such feedback control while others do not.

Reviewer #1 (Public Review):

In this manuscript, the Authors implement a delayed feedback control method and use it for the first time in biological neuronal networks. They extend a well-established computational theory and expand it into the biological realm. With this, they obtain novel evidence, never considered before, that showcases the difference between simulated neuronal networks and biological ones. Furthermore, they optimize the DFC method to achieve optimal results in the control of cell excitability in the content of biological neuronal networks, taking advantage of a closed-loop stimulation setup that, by itself, is not trivial to build and operate and that will certainly have a positive impact the fields of cellular and network electrophysiology.

Regarding the results, it would be very constructive if the Authors could share the code for the quasi-real-time interface with the Multichannel Systems software (current and older hardware versions), as this represents likely a bottleneck preventing more researchers to implement such an experimental paradigm.

On the data focusing on the effects of the DFC algorithms on neuronal behavior, the evidence is very compelling, although more care should be devoted to the statistical analyses, since some of the applied statistical tests are not appropriate. In a more biological sense, further discussion and clarification of the experimental details would improve this manuscript, making it more accessible and clearer for researchers across disciplines (i.e., ranging from computational to experimental Neuroscience) and increasing the impact of this research.

In summary, this work represents a necessary bridge between recent advances in computational neuroscience and the biological implementation of neuronal control mechanisms.

Reviewer #2 (Public Review):

This study applies a new neuromodulation algorithm, adaptive delayed feedback control (aDFC) to in vitro and in silico neuron populations to demonstrate its effectiveness at desynchronizing synchronous neural population activity. The study compares aDFC to other neuromodulation approaches such as non-adaptive DFC and random stimulation and demonstrates that in a subset of controllable networks, aDFC succeeds in reducing overall synchrony in the neural population. Further, when characterizing population firing bouts as asynchronous versus synchronous, aDFC increased the fraction of time that the neural population was in the asynchronous versus synchronous state (albeit in one network). Overall, this study is an impressive combination of computational and experimental work that details a promising new adaptive neuromodulation algorithm that may be relevant for neurological disorders where excessive synchronous brain activity is currently treated with conventional open-loop DBS.

Strengths: The authors build on existing work that has suggested DFC may be a viable algorithm for desynchronizing hyper-synchronous neural populations. They demonstrate by performing in vivo experiments that, contrary to the suggestions of previous work, DFC exacerbates oscillatory intensity. As a result, they develop a new adaptive DFC (aDFC) that updates the estimate of the population's periodicity, enabling superior desynchronization of the population. Further, aDFC enables more population spiking activity that is not just a response to the stimulation (Fig. S3), potentially making the approach conducive to reducing excessive synchronization while also being permissive to neural encoding.

Another innovation of this study is developing a framework for detecting which neural populations are controllable vs. uncontrollable, i.e. consistently responsive to stimulation vs. not consistently responsive. The authors find that populations with intermediate levels of synchrony and firing rate are controllable, whereas populations outside this regime are uncontrollable. These findings are substantiated with a neural network model, where a controllable regime is also detected. The controllable subspace in the in vivo networks and in silico networks also appear to roughly correspond (intermediate synchrony and firing rates) though a direct comparison is not made.

Finally, not only do the authors find that aDFC reduces synchrony, they further identify extended periods of time when the network is in an asynchronous state and find that aDFC can extend the amount of time that the network spends in this state. While these results are compelling, there is only a single network that is able to demonstrate this effect so it is unclear how general a property this is.

Overall, the study presents a novel closed loop neuromodulation algorithm and presents compelling data demonstrating that the algorithm reduces synchrony in in vitro and in silico neural populations.

Weaknesses: The authors point out Parkinson's disease, essential tremor, epilepsy, and dystonia as the neurological disorders that suffer from excessive neural synchronization. In two of these disorders the frequency of the neural synchronization is ~15-30 Hz (Parkinson's disease) and ~5-7 Hz (essential tremor). These frequencies are well above the ~1 Hz synchronization frequency observed in the in vitro population. While this study exhibits a nice proof of principle, how readily it would extend to populations that exhibit higher synchronization frequencies is unclear.

In addition, the study relies on computing population spiking activity of neurons. Current closed-loop neuromodulation devices are outfitted with large electrodes that can sense local field potentials. The impact of this study would have been higher and more readily translatable if the authors could have detected neural population synchronization using local field potential features.

Finally, since the authors were seeking to develop a closed-loop neuromodulation solution that exhibited an improvement over existing open-loop solutions, it would have strengthened the findings and relevance of this study to have done comparisons between aDFC and high frequency open-loop stimulation (~100-120 Hz). Without this comparison it is difficult to know how aDFC may differ from existing therapeutics.

Author Response

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

We want to thank the Editor and Reviewers for their thorough assessment of the manuscript as well as their constructive critiques. We have collated below the public review and recommendations from each Reviewer as well as our responses to them.

eLife assessment

This study by Verdikt et al. provided solid evidence demonstrating the potential impacts of Δ9-tetrahydrocannabinol (Δ9-THC) on early embryonic development using mouse embryonic stem cells (mESCs) and in vitro differentiation. Their results revealed that Δ9-THC enhanced mESCs proliferation and metabolic adaptation, possibly persisting through differentiation to Primordial Germ Cell-Like Cells (PGCLCs), though the evidence supporting this persistence was incomplete. Although the study is important, it was limited by being conducted solely in vitro and lacking parallel human model experiments.

Reviewer #1 (Public Review):

The authors investigated the metabolic effects of ∆9-THC, the main psychoactive component of cannabis, on early mouse embryonic cell types. They found that ∆9-THC increases proliferation in female mouse embryonic stem cells (mESCs) and upregulates glycolysis. Additionally, primordial germ cell-like cells (PGCLCs) differentiated from ∆9-THC-exposed cells also show alterations to their metabolism. The study is valuable because it shows that physiologically relevant ∆9-THC concentrations have metabolic effects on cell types from the early embryo, which may cause developmental effects. However, the claim of "metabolic memory" is not justified by the current data, since the effects on PGCLCs could potentially be due to ∆9-THC persisting in the cultured cells over the course of the experiment, even after the growth medium without ∆9-THC was added.

The study shows that ∆9-THC increases the proliferation rate of mESCs but not mEpiLCs, without substantially affecting cell viability, except at the highest dose of 100 µM which shows toxicity (Figure 1). Treatment of mESCs with rimonabant (a CB1 receptor antagonist) blocks the effect of 100 nM ∆9-THC on cell proliferation, showing that the proliferative effect is mediated by CB1 receptor signaling. Similarly, treatment with 2-deoxyglucose, a glycolysis inhibitor, also blocks this proliferative effect (Figure 4G-H). Therefore, the effect of ∆9-THC depends on both CB1 signaling and glycolysis. This set of experiments strengthens the conclusions of the study by helping to elucidate the mechanism of the effects of ∆9-THC.

Although several experiments independently showed a metabolic effect of ∆9-THC treatment, this effect was not dose-dependent over the range of concentrations tested (10 nM and above). Given that metabolic effects were observed even at 10 nM ∆9-THC (see for example Figure 1C and 3B), the authors should test lower concentrations to determine the dose-dependence and EC50 of this effect. The authors should also compare their observed EC50 with the binding affinity of ∆9-THC to cellular receptors such as CB1, CB2, and GPR55 (reported by other studies).

The study also profiles the transcriptome and metabolome of cells exposed to 100 nM ∆9-THC. Although the transcriptomic changes are modest overall, there is upregulation of anabolic genes, consistent with the increased proliferation rate in mESCs. Metabolomic profiling revealed a broad upregulation of metabolites in mESCs treated with 100 nM ∆9-THC.

Additionally, the study shows that ∆9-THC can influence germ cell specification. mESCs were differentiated to mEpiLCs in the presence or absence of ∆9-THC, and the mEpiLCs were subsequently differentiated to mPGCLCs. mPGCLC induction efficiency was tracked using a BV:SC dual fluorescent reporter. ∆9-THC treated cells had a moderate increase in the double positive mPGCLC population and a decrease in the double negative population. A cell tracking dye showed that mPGCLCs differentiated from ∆9-THC treated cells had undergone more divisions on average. As with the mESCs, these mPGCLCs also had altered gene expression and metabolism, consistent with an increased proliferation rate.

My main criticism is that the current experimental setup does not distinguish between "metabolic memory" vs. carryover of THC (or its metabolites) causing metabolic effects. The authors assume that their PGCLC induction was performed "in the absence of continuous exposure" but this assumption may not be justified. ∆9-THC might persist in the cells since it is highly hydrophobic. In order to rule out the persistence of ∆9-THC as an explanation of the effects seen in PGCLCs, the authors should measure concentrations of ∆9-THC and THC metabolites over time during the course of their PGCLC induction experiment. This could be done by mass spectrometry. This is particularly important because 10 nM of ∆9-THC was shown to have metabolic effects (Figure 1C, 3B, etc.). Since the EpiLCs were treated with 100 nM, if even 10% of the ∆9-THC remained, this could account for the metabolic effects. If the authors want to prove "metabolic memory", they need to show that the concentration of ∆9-THC is below the minimum dose required for metabolic effects.

Overall, this study is promising but needs some additional work in order to justify its conclusions. The developmental effects of ∆9-THC exposure are important for society to understand, and the results of this study are significant for public health.

*Reviewer #1 (Recommendations For The Authors):

This has the potential to be a good study, but it's currently missing two key experiments:

What is the minimum dose of ∆9-THC required to see metabolic effects?

We would like to thank Reviewer 1 for their insightful comments. We have included exposures to lower doses of ∆9-THC in Supplementary Figure 1. Our data shows that ∆9-THC induces mESCs proliferation from 1nM onwards. However, when ESCs and EpiLCs were exposed to 1nM of ∆9-THC, no significant change in mPGCLCs induction was observed (updated Figure 6B). Of note, in their public review, Reviewer 1 mentioned that “The authors should also compare their observed EC50 with the binding affinity of ∆9-THC to cellular receptors such as CB1, CB2, and GPR55 (reported by other studies).” According to the literature, stimulation of non-cannabinoid receptors and ion channels (including GPR18, GPR55, TRPVs, etc.) occurs at 40nM-10µM of ∆9-THC (Banister et al., 2019). We therefore expect that at the lower nanomolar range tested, CB1 is the main receptor stimulated by ∆9-THC, as we showed for the 100nM dose in our rimonabant experiments (Fig. 2).

Is the residual THC concentration during the PGCLC induction below this minimum dose? Even if the effects are due to residual ∆9-THC, this would not undermine the overall study. There would simply be a different interpretation of the results.

This experiment was particularly important to distinguish between a “true” ∆9-THC metabolic memory or residual ∆9-THC leftover during PGCLCs differentiation. Our mass spectrometry quantification revealed that no significant ∆9-THC could be detected in day 5 embryoid bodies compared to treated EpiLCs prior to differentiation (Supplementary Figure 13). These results support the existence of ∆9-THC metabolic memory across differentiation.

You also do not mention whether you tested your cells for mycoplasma. This is important since mycoplasma contamination is a common problem that can cause artifactual results. Please test your cells and report the results.

All cells were tested negative for mycoplasma by a PCR test (ATCC® ISO 9001:2008 and ISO/IEC 17025:2005 quality standards). This information has been added in the Material and Methods section.

Minor points:

1. I don't think it's correct to say that cannabis is the most commonly used psychoactive drug. Alcohol and nicotine are more commonly used. See: https://nida.nih.gov/research-topics/alcohol and https://www.cancer.gov/publications/dictionaries/cancer-terms/def/psychoactive-substance I looked at the UN drugs report [ref 1] and alcohol or nicotine were not included on that list of drugs, so the UN may use a different definition. This doesn't affect the importance or conclusions of this study, but the wording should be changed.

We agree and are now following the WHO description of cannabis (https://www.who.int/teams/mental-health-and-substance-use/alcohol-drugs-and-addictive-behaviours/drugs-psychoactive/cannabis) by referring to it as the “most widely used illicit drug in the world”. (Line 44).

1. It would be informative to use your RNA-seq data to examine the expression of receptors for ∆9-THC such as CB1, CB2, and GPR55. CB1 might be the main one, but I am curious to see if others are present.

We have explored the protein expression of several cannabinoid receptors, including CB2, GPR18, GPR55 and TRPV1 (Bannister et al., 2019). These proteins, except TRPV1, were lowly expressed in mouse embryonic stem cells compared to the positive control (mouse brain extract, see Author response image 1). Furthermore, our experiment with Rimonabant showed that the proliferative effects of ∆9-THC are mediated through CB1.

Author response image 1.

Cannabinoid receptors and non-cannabinoid receptors protein expression in mouse embryonic stem cells.

1. Make sure to report exact p-values. You usually do this, but there are a few places where it says p<0.0001. Also, report whether T-tests assumed equal variance (Student's) or unequal variance (Welch's). [In general, it's better to use unequal variance, unless there is good reason to assume equal variance.]

Prism, which was used for statistical analyses, only reports p-values to four decimal places. For all p-values that were p<0.0001, the exact decimals were calculated in Excel using the “=T.DIST.2T(t, df)” function, where the Student’s distribution and the number of degrees of freedom computed by Prism were inputted. Homoscedasticity was confirmed for all statistical analyses in Prism.

1. Figure 2A: An uncropped gel image should be provided as supplementary data. Additionally, show positive and negative controls (from cells known to either express CB1 or not express CB1)

The uncropped gel image is presented in Author response image 2. The antibody was validated on mouse brain extracts as a positive control as shown in Figure 1.

Author response image 2.

Uncropped gel corresponding to Fig. 2A where an anti-CB1 antibody was used.

1. Figure 6B: Please show a representative gating scheme for flow cytometry (including controls) as supplementary data. Also, was a live/dead stain used? What controls were used for compensation? These details should be reported.

The gating strategy is presented in Supplementary Figure 11. The Material and Methods section has also been expanded.

1. As far as I can tell, you only used female mESCs. It would be good to test the effects on male mESCs as well since these have some differences due to differences in X-linked gene expression (female mESCs have two active X chromosomes). I understand that you might not have a male BV:SC reporter line, so it would be acceptable to omit the mPGCLC experiments on male cells.

We have tested the 10nM-100µM dose range in the male R8 mESCs (Supplementary Figure 3). Similar results as with the female H18 cells were observed. Accordingly, PGCLCs induction was increased when R8 ESCs + EpiLCs were exposed to 100nM of ∆9-THC (Supplementary Figure 12). This is in line with ∆9-THC impact on fundamentally conserved metabolic pathways across species and sex, although it should be noted that one representative model of each sex is not sufficient to exclude sex-specific effects.

Reviewer #2 (Public Review):

In the study conducted by Verdikt et al, the authors employed mouse Embryonic Stem Cells (ESCs) and in vitro differentiation techniques to demonstrate that exposure to cannabis, specifically Δ9-tetrahydrocannabinol (Δ9-THC), could potentially influence early embryonic development. Δ9-THC was found to augment the proliferation of naïve mouse ESCs, but not formative Epiblast-like Cells (EpiLCs). This enhanced proliferation relies on binding to the CB1 receptor. Moreover, Δ9-THC exposure was noted to boost glycolytic rates and anabolic capabilities in mESCs. The metabolic adaptations brought on by Δ9-THC exposure persisted during differentiation into Primordial Germ Cell-Like Cells (PGCLCs), even when direct exposure ceased, and correlated with a shift in their transcriptional profile. This study provides the first comprehensive molecular assessment of the effects of Δ9-THC exposure on mouse ESCs and their early derivatives. The manuscript underscores the potential ramifications of cannabis exposure on early embryonic development and pluripotent stem cells. However, it is important to note the limitations of this study: firstly, all experiments were conducted in vitro, and secondly, the study lacks analogous experiments in human models.

Reviewer #2 (Recommendations For The Authors):

1. EpiLCs, characterized as formative pluripotent stem cells rather than primed ones, are a transient population during ESC differentiation. The authors should consider using EpiSCs and/or formative-like PSCs (Yu et al., Cell Stem Cell, 2021; Kinoshita et al., Cell Stem Cell, 2021), and amend their references to EpiLCs as "formative".

Indeed, EpiLCs are a transient pluripotent stem cell population that is “functionally distinct from both naïve ESCs and EpiSCs” and “enriched in formative phase cells related to pre-streak epiblast” (Kinoshita et al., Cell Stem Cell, 2021). Here, we used the differentiation system developed by M. Saitou and colleagues to derive PGCLCs (Hayashi et al, 2011). Since EpiSCs are refractory to PGCLCs induction (Hayashi et al, 2011), we used the germline-competent EpiLCs and took advantage of a well-established differentiation system to derive mouse PGCLCs. Most authors, however, agree that in terms of epigenetic and metabolic profiles, mouse EpiLCs represent a primed pluripotent state. We have added that PGCs arise in vivo “from formative pluripotent cells in the epiblast” on lines 85-86.

1. Does the administration of Δ9-THC, at concentrations from 10nM to 1uM, alter the cell cycle profiles of ESCs?

The proliferation of ESCs was associated with changes in the cell cycle, as presented in the new Supplementary Figure 2, which we discuss in lines 118-123.

1. Could Δ9-THC treatment influence the differentiation dynamics from ESCs to EpiLCs?

No significant changes were observed in the pluripotency markers associated with ESCs and EpiLCs (Supplementary Figure 9). We have added this information in lines 277-279.

1. The authors should consider developing knockout models of cannabinoid receptors in ESCs and EpiLCs (or EpiSCs and formative-like PSCs) for control purposes.

This is an excellent suggestion. Due to time and resource constraints, however, we focused our mechanistic investigation of the role of CB1 on the use of rimonabant which revealed a reversal of Δ9-THC-induced proliferation at 100nM.

1. Lines 134-136: "Importantly, SR141716 pre-treatment, while not affecting cell viability, led to a reduced cell count compared to the control, indicating a fundamental role for CB1 in promoting proliferation." Regarding Figure 2D, does the Rimonabant "+" in the "mock" group represent treatment with Rimonabant only? If that's the case, there appears to be no difference from the Rimonabant "-" mock. The authors should present results for Rimonabant-only treatment.

To be able to compare the effects +/- Rimonabant and as stated in the figure legend, each condition was normalized to its own control (mock with, or without Rimonabant). Author response image 3 is the unnormalized data showing the same effects of Δ9-THC and Rimonabant on cell number.

Author response image 3.

Unnormalized data corresponding to the Figure 2D.

1. In Figure 3, both ESCs and EpiLCs show a significant decrease in oxygen consumption and glycolysis at a 10uM concentration. Do these conditions slow cell growth? BrdU incorporation experiments (Figure 1) seem to contradict this. With compromised bioenergetics at this concentration, the authors should discuss why cell growth appears unaffected.

Indeed, we believe that cell growth is progressively restricted upon increasing doses of ∆9-THC (consider Supplementary Figure 2). In addition, oxygen consumption and glycolysis can be decoupled from cellular proliferation, especially considering the lower time ranges we are working with (44-48h).

1. Beyond Δ9-THC exposure prior to PGCLCs induction, it would be also interesting to explore the effects of Δ9-THC on PGCLCs during their differentiation.

We agree with the Reviewer. Our aim was to study whether exposure prior to differentiation could have an impact, and if so, what are the mediators of this impact. Full exposure during differentiation is another exposure paradigm that is relevant but would not have allowed us to show the metabolic memory of ∆9-THC exposure. Future work, however, will be dedicated to analyzing the effect of continuous exposure through differentiation.

1. As PGC differentiation involves global epigenetic changes, it would be interesting to investigate how Δ9-THC treatment at the ESCs/EpiLCs stage may influence PGCLCs' transcriptomes.

We also agree with the Reviewer. While this paper was not primarily focused on Δ9-THC’s epigenetic effects, we have explored the impact of Δ9-THC on more than 100 epigenetic modifiers in our RNA-seq datasets. These results are shown in Supplementary Table 1 and Supplementary Figure 10 and discussed in lines 301-316.

1. Lines 407-408: The authors should exercise caution when suggesting "potentially adverse consequences" based solely on moderate changes in PGCLCs transcriptomes.

We agree and have modified the sentence as follows: “Our results thus show that exposure to Δ9-THC prior to specification affects embryonic germ cells’ transcriptome and metabolome. This in turn could have adverse consequences on cell-cell adhesion with an impact on PGC normal development in vivo.“

1. Investigating the possible impacts of Δ9-THC exposure on cultured mouse blastocysts, implantation, post-implantation development, and fertility could yield intriguing findings.

We thank the Reviewer for this comment. We have amended our discussion to include these points in the last paragraph.

1. Given that naïve human PSCs and human PGCLCs differentiation protocols have been established, the authors should consider carrying out parallel experiments in human models.

We have performed Δ9-THC exposures in hESCs (Supplementary Figure 4 and Supplementary Figure 5), showing that Δ9-THC alters the cell number and general metabolism of these cells. We present these results in light of the differences in metabolism between mouse and human embryonic stem cells on lines 135-141 and 185-188. Implications of these results are discussed in lines 474-486.

Reviewer #3 (Public Review):

Verdikt et al. focused on the influence of Δ9-THC, the most abundant phytocannabinoid, on early embryonic processes. The authors chose an in vitro differentiation system as a model and compared the proliferation rate, metabolic status, and transcriptional level in ESCs, exposure to Δ9-THC. They also evaluated the change of metabolism and transcriptome in PGCLCs derived from Δ9-THC-exposed cells. All the methods in this paper do not involve the differentiation of ESCs to lineage-specific cells. So the results cannot demonstrate the impact of Δ9-THC on preimplantation developmental stages. In brief, the authors want to explore the impact of Δ9-THC on preimplantation developmental stages, but they only detected the change in ESCs and PGCLCs derived from ESCs, exposure to Δ9-THC, which showed the molecular characterization of the impact of Δ9-THC exposure on ESCs and PGCLCs.

Reviewer #3 (Recommendations For The Authors):

1. To demonstrate the impact of Δ9-THC on preimplantation developmental stages, ESCs are an appropriate system. They have the ability to differentiate three lineage-specific cells. The authors should perform differentiation experiments under Δ9-THC-exposure, and detect the influence of Δ9-THC on the differentiation capacity of ESCs, more than just differentiate to PGCLCs.

We apologize for the lack of clarity in our introduction. We specifically looked at the developmental trajectory of PGCs because of the sensitivity of these cells to environmental insults and their potential contribution to transgenerational inheritance. We have expanded on these points in our introduction and discussion sections (lines 89-91 and 474-486). Because our data shows the relevance of Δ9-THC-mediated metabolic rewiring in ESCs subsisting across differentiation, we agree that differentiation towards other systems (neuroprogenitors, for instance) would yield interesting data, albeit beyond the scope of the present study.

1. Epigenetics are important to mammalian development. The authors only detect the change after Δ9-THC-exposure on the transcriptome level. How about methylation landscape changes in the Δ9-THC-exposure ESCs?

We have explored the impact of Δ9-THC on more than 100 epigenetic modifiers in our RNA-seq datasets. These results are shown in Supplementary Table 1 and Supplementary Figure 10, discussed in lines 301-316. While indeed the changes in DNA methylation profiles appear relevant in the context of Δ9-THC exposure (because of Tet2 increased expression in EpiLCs), we highlight that other epigenetic marks (histone acetylation, methylation or ubiquitination) might be relevant for future studies.

1. In the abstract, the authors claimed that "the results represent the first in-depth molecular characterization of the impact of Δ9-THC exposure on preimplantation developmental stages." But they do not show whether the Δ9-THC affects the fetus through the maternal-fetal interface.

We have addressed the need for increased clarity and have modified the sentence as follows: “These results represent the first in-depth molecular characterization of the impact of Δ9-THC exposure on early stages of the germline development.”

1. To explore the impact of cannabis on pregnant women, the human ESCs may be a more proper system, due to the different pluripotency between human ESCs and mouse ESCs.

We have performed Δ9-THC exposures in hESCs (Supplementary Figure 4 and Supplementary Figure 5). These preliminary results show that Δ9-THC exposure negatively impacts the cell number and general metabolism of hESCs. With the existence of differentiation systems for hPGCLCs, future studies will need to assess whether Δ9-THC-mediated metabolic remodelling is also carried through differentiation in human systems. We discuss these points in the last paragraph of our discussion section.

1. All the experiments are performed in vitro, and the authors should validate their results in vivo, at least a Δ9-THC-exposure pregnant mouse model.

Our work is the first of its kind to show that exposure to a drug of abuse can alter the normal development of the embryonic germline. We agree with the Reviewer that to demonstrate transgenerational inheritance of the effects reported here, future experiments in an in vivo mouse model should be conducted. The metabolic remodeling observed upon cannabis exposure could also be directly studied in a human context, although these experiments would be beyond the scope of the present study. For instance, changes in glycolysis may be detected in pregnant women using cannabis, or directly measured in follicular fluid in a similar manner as done by Fuchs-Weizman and colleagues (Fuchs-Weizman et al., 2021). We hope that our work can provide the foundation to inform such in vivo studies.

Author Response

Reviewer #1 (Public Review):

The manuscript by Grove and colleagues analyzes the role of TEAD1 transcription factors in all events regulating PNS myelin formation and maintenance and regeneration. Throughout the manuscript, the authors compare the results obtained to those they previously described in YAP/TAZ double knockout mice. Strengths of the manuscript are combined in vivo analyses by generating mutants constitutively lacking TEAD1 expression in myelinating Schwann cells (P0Cre//TEAD1f/f mice: cKO) and mutants in which TEAD1 expression can be ablated after tamoxifen-mediated recombination is myelinating Schwann cells (PlpCreER//TEAD1f/f mice: iKO). Using this approach the authors were able to assess the role of TEAD1 in all aspects related to PNS myelin: formation as well as maintenance and remyelination after injury. By exploiting these models, they were able to define the role of TEAD1 in regulating Schwann cell proliferation as well as in the cholesterol biosynthetic pathway. Collectively, their data indicate that TEAD 1 has a composite role in PNS myelination being required for developmental myelination, but dispensable for myelin maintenance. Further, they also describe a role for TEAD1 in promoting PNS remyelination after an injury event.

Despite these strengths, there are some weaknesses that should be addressed by the authors:

1) The manuscript would benefit from better and more detailed analysis of the role of the other TEAD transcription factors, as they are likely redundant in function to TEAD1. For example, since in cKO mice some fibers can escape the sorting defect and eventually myelinate, albeit at a lower level, could they determine whether TEAD2-4 transcription factors might compensate for TEAD1 absence in this setting?

We speculate that other TEADs, most likely both TEAD2 and TEAD3, compensate TEAD1 in myelinating some developing axons. We also speculate that TEAD4 counteracts TEAD1, resulting in excessive proliferation of Schwann cells in Tead1 cKO. Unfortunately, because, unlike TEAD1, floxed/congenic alleles and IHC-compatible antibodies are not yet available for TEAD2-4, it is difficult to determine their roles. We attempted to knock down TEAD2-4 by injecting AAV-shRNAs into the sciatic nerves of WT and Tead1 iKO, but this intervention was not successful. Our future studies will determine compensatory and/or opposing roles of other TEADs during development and homeostasis and after nerve injury.

2) A striking result of the study is the morphological defects observed in the process of axonal sorting and in the Remak fibers formation of TEAD1 cKO mice. To explain the sorting defect, the authors correctly analyze Schwann cell proliferation. However, since axonal sorting is mediated by the interaction between the extracellular matrix and intracellular cytoskeleton rearrangement, they should address also these two aspects. As per the Remak bundles and the poly-axonal myelination they observe, it is difficult to reconcile this "abnormal" myelination with the fact that TEAD1 cKO mice have a very severe myelinating phenotype, which is persistent in adulthood.

It is noteworthy that we found radial sorting to be delayed, but not blocked, in Tead1 cKO, as we had previously reported for Yap/Taz cDKO mice in our earlier publication (Grove et al., eLIFE 2017). The primary reason that myelin development fails in Schwann cells lacking YAP/TAZ (or TEAD1 in the present report) is because they do not initiate myelination of sorted axons, not because of defective radial sorting. We showed that radial sorting was delayed in Schwann cells lacking YAP/TAZ because of their late S phase entry (Figure 4 in Grove et al., eLIFE 2017). In addition, our earlier report demonstrated that the key laminin receptor, integrin 6, is strongly downregulated but axons are nevertheless sorted out by Schwann cells in Yap/Taz cDKO (Figure 4-figure supplement 2 in Grove et al., eLIFE 2017). Our current view, therefore, is that extracellular matrix may contribute to reducing Schwann cell proliferation (Berti et al., 2011; Pellegatta et al., 2013; Yu, Feltri, Wrabetz, Strickland, & Chen, 2005), which helps to delay radial sorting, but that it is not required for Schwann cells lacking YAP/TAZ (or TEAD1) to sort axons (see the author response #2 in Grove et al., eLIFE 2017). Based on this information, we disagree with the reviewer that it is essential for us to address the role of extracellular matrix in delaying radial sorting in Tead1 cKO.

Regarding Remak bundles, ‘thinly’ myelinated Remak bundles are only ‘occasionally’ observed in Tead1 cKO mice. Given that some large axons are still myelinated in Tead1 cKO mice, likely due to compensation by other TEADs, we speculate that Remak bundles are occasionally myelinated by other TEADs in Tead1 cKO. We have clarified our description and expanded our discussion of TEAD1 regulation of Remak bundles, including abnormal polyaxonal myelination.

3) In the analyses of the cholesterol biosynthetic pathway, TEAD1 seems to be only partly involved. Again, which is the role of any of the other TEADs?

Examining cholesterol biosynthesis pathways (SREBP1 and 2) and their target enzymes (SCD1, HMGCR, FDPS, IDI1) in Tead1 cKO and Yap/Taz cDKO, we showed that TEAD1 is required for upregulating FDPS and IDI1. These data suggest that TEAD1 plays a major role in mediating YAP/TAZ-driven cholesterol synthesis by upregulating FDPS and IDI1. It is also important to note that FDPS and IDI1 levels are reduced in TEAD1 cKO as ‘greatly’ as those in Yap/Taz cDKO (Figure 5). We therefore speculate that other TEADs compensate TEAD1 modestly, if at all, in upregulating FDPS and IDI1. We do not rule out the possibility, however, that other TEADs fully compensate TEAD1 in ‘maintaining’ cholesterol synthesis in adult Schwann cells. We will address these important questions in the future when the key resources mentioned above become available to study TEAD2-4.

4) Why do cKO mice die before P60?

In accordance with IACUC guidelines, we humanely euthanized Tead1 cKO mice before P60 because, like Yap/Taz cKO mice, they develop severe peripheral neuropathy.

eLife assessment

This important study demonstrates that the transcription factor TEAD1 is required for the function of Yap/Taz in Schwann cells, with conditional mouse mutants having very similar dysmyelinated phenotypes. Convincing histological evidence is shown for the role of TEAD1 itself, leaving open the function of other TEAD proteins in this system. This study will nevertheless be of great interest to researchers in the field of peripheral nerve development.

Reviewer #1 (Public Review):

The manuscript by Grove and colleagues analyzes the role of TEAD1 transcription factors in all events regulating PNS myelin formation and maintenance and regeneration. Throughout the manuscript, the authors compare the results obtained to those they previously described in YAP/TAZ double knockout mice. Strengths of the manuscript are combined in vivo analyses by generating mutants constitutively lacking TEAD1 expression in myelinating Schwann cells (P0Cre//TEAD1f/f mice: cKO) and mutants in which TEAD1 expression can be ablated after tamoxifen-mediated recombination is myelinating Schwann cells (PlpCreER//TEAD1f/f mice: iKO). Using this approach the authors were able to assess the role of TEAD1 in all aspects related to PNS myelin: formation as well as maintenance and remyelination after injury. By exploiting these models, they were able to define the role of TEAD1 in regulating Schwann cell proliferation as well as in the cholesterol biosynthetic pathway.

Collectively, their data indicate that TEAD 1 has a composite role in PNS myelination being required for developmental myelination, but dispensable for myelin maintenance. Further, they also describe a role for TEAD1 in promoting PNS remyelination after an injury event.

Despite these strengths, there are some weaknesses that should be addressed by the authors:

1. The manuscript would benefit from better and more detailed analysis of the role of the other TEAD transcription factors, as they are likely redundant in function to TEAD1. For example, since in cKO mice some fibers can escape the sorting defect and eventually myelinate, albeit at a lower level, could they determine whether TEAD2-4 transcription factors might compensate for TEAD1 absence in this setting?

2. A striking result of the study is the morphological defects observed in the process of axonal sorting and in the Remak fibers formation of TEAD1 cKO mice. To explain the sorting defect, the authors correctly analyze Schwann cell proliferation. However, since axonal sorting is mediated by the interaction between the extracellular matrix and intracellular cytoskeleton rearrangement they should address also these two aspects. As per the Remak bundles and the poly-axonal myelination they observe, it is difficult to reconcile this "abnormal" myelination with the fact that TEAD 1 cKO mice have a very severe myelinating phenotype, which is persistent in adulthood.

3. In the analyses of the cholesterol biosynthetic pathway, TEAD1 seems to be only partly involved. Again, which is the role of any of the other TEADs?

4. Why do cKO mice die before P60?

Reviewer #2 (Public Review):

The manuscript addresses the role of TEAD1 in developmental myelination and nerve regeneration after nerve injury and establishes TEAD1 as a key component for YAP/TAZ-related Schwann cell biology. The authors use genetic and biochemical techniques, as well as immunostainings of tissues to address TEAD1's function in myelin biology. While the constitutive knockout of TEAD1 is convincing, the tamoxifen-induced variation requires some validation. Experimental procedures to study the effect of TEAD1 on myelin development and regeneration were properly performed. TEAD1 is believed to be the major driver of the TEAD family in regulating myelination in Schwann cells. However, the delineation of TEAD1 in myelin biology from the other TEAD family members TEAD2, 3, and 4 needs further verification. In particular, the biochemical techniques assessing the potentially competitive binding of TEAD1 versus TEAD2, 3, and 4 to YAP1 and TAZ (WWTR1) require a thorough functional validation. Overall, the identification of TEAD1 as the major driver of myelin in development and regeneration is a very important finding for Schwann cell biology.

Reviewer #3 (Public Review):

The Hippo signalling pathway has been implicated in organ growth through the regulation of cell proliferation and apoptosis. The main transcriptional effectors of this pathway, the Yap and Taz proteins, associate with members of the TEAD family of transcription factors to drive diverse transcriptional programs of proliferation, growth, and differentiation. It has previously been shown that YAP/TAZ are essentially required in Schwann cells for developmental myelination, homeostasis, and regenerative myelination in the peripheral nervous system. All four members of the TEAD family are expressed in the Schwann cell lineage, raising the possibility that different aspects of YAP/TAZ role in the Schwann cell lineage are underpinned by differential associations with TEAD transcription factors. In this study, Grove and colleagues provide convincing evidence that TEAD1 is the main transcription factor through which YAP/TAZ affects myelination in development and following nerve injury. A careful comparison between Schwann cell-specific and inducible Yap/Taz and TEAD1 knock out animals reveal unique and redundant roles for TEAD1 in myelination by regulating Schwann cell proliferation, Krox20-dependent myelin gene expression, and cholesterol biosynthesis. Interestingly, their study appears to reveal a YAP/TAZ independent role for TEAD1 in non-myelinating Schwann cell ensheathment of low calibre axons and Remak bundle formation. The conclusions of this study are based on rigorous biochemical, immune-histochemical, electron microscopic, and functional analysis of mutant and wild-type nerves at different stages of postnatal development and following crush nerve injury.

Perhaps the most surprising finding of this study is that TEAD1 can function independently from YAP/TAZ in one branch of the Schwann cell lineage (the authors had reported earlier that non-myelinating Schwann cells do not express YAP/TAZ).<br /> How TEAD1 transcriptional activity is modulated in these Remak Schwann cells is an interesting avenue of future research.

Author Response

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

Reviewer #1:

We thank the reviewer for the positive evaluation of our manuscript. We have closely examined the issues raised, and below we offer a point-by-point response to each comment. In the revised manuscript below, all the introduced changes are marked with red font.

1. There may be a general typo concerning micromolar and millimolar…

Response 1: The reviewer is correct, and during the reformatting of the manuscript, in some portions of the manuscript, the units used to indicate TPEN concentrations, always µM, were switched to mM. We have corrected those mistakes.

1. In Figure 1C/Lines 150-152, the authors use DTPA and EDTA as extracellular chelators for zinc… Was the amount of zinc in the media measured and determined to be below the amount of chelator used? Additionally, these chelators are not specific for zinc, but can bind other divalent cations including calcium. Even though zinc binds more tightly than calcium to these chelators, by mass action calcium and magnesium ions may outcompete DTPA and EDTA, leaving zinc availability unperturbed. How do the authors take these interactions into account to determine that chelation of extracellular zinc has no effect on intracellular calcium oscillations? The best way to test this is to use zinc responsive fluorescent probes in a sample of the calcium- and magnesium-replete medium and see if the addition of the DTPA or EDTA alters zinc fluorescence in the cuvette.

Response 2: We tested several conditions to determine the effect of chelators on the zinc concentration of the monitoring media using commercially available Zn2+ probes. The fluorescent zinc probe FluoZin3 added extracellularly shows high fluorescence, consistent with trace amounts of zinc and possibly non-specific bindings of other cations.

Further, the media tested was replete with the concentrations of Ca2+ and Mg2+ in TLHEPES. To establish if the non-permeable external chelators we used could bind external Zn2+ despite the high concentrations of Ca2+ and Mg2+, we followed the reviewer’s suggestion of adding the chelators to the complete media in the presence of FluoZin3. The addition of EDTA caused a protracted, ~5 min, but significant decrease in FluoZin3’s fluorescence, suggesting it is effective at removing external Zn2+ despite the presence of other divalent cations (Author response image 1A). We used a second approach where we added the chelator in the presence of nominal concentrations of Ca2+ and Mg2+ to increase the chelators’ chances to find and chelate Zn2+ (Author response image 1B). Then, we injected mPlcζ mRNA, which initiated persistent but low-frequency oscillations, as expected due to the lack of external Ca2+. Remarkably, upon restoring it, the responses became of high frequency, and upon increasing Mg2+, they acquired the regular pattern, consistent with Mg2+’s inhibition of channels that mediate Ca2+ influx. These results show that the chelation of extracellular zinc does not replicate TPEN’s effect, which suggests that TPEN’s abrupt and inhibiting ability on Ca2+ oscillations is most likely due to the 43 chelation of internal Zn2+.

Author response image 1.

Cell-impermeable chelators effectively reduce Zn2+ levels in external media but do prevent initiation or continuation of Ca2+ oscillations. (A) A representative trace of FluoZin3 fluorescence in replete monitoring media (TL-HEPES). The media was supplemented with cell-impermeable FluoZin-3, and after initiation of monitoring, the addition of EDTA (100 μM) occurred at the designated point (triangle). (B) The left black trace represents Ca2+ oscillations initiation by injection of mPlcζ mRNA (0.01 μg/μl). The oscillations were monitored in Ca2+ and Mg2+-free media and in the presence of EDTA (110 μM) to chelate residual divalent cations derived from the water source or reagents used to make the media. The right red trace represents the initiation of oscillations as above, but after a period indicated by the black and green bars, Ca2+ and Mg2+ were sequentially added back.

Noteworthy, low EDTA concentrations, 10-µM, have been used to enhance in vitro culture conditions of mammalian embryos. In fact, it is the key ingredient to overcome the two-cell block that initially prevented the in vitro development of zygotes srom inbred strains. It is unknown how EDTA mediates this effect, which is detectable in Ca2+ and Mg2+ replete media and is only effective when placed extracellularly, but it has been attributed to its ability to chelate toxic metals introduced as impurities by other media components; one study demonstrated that the Zn2+ present in the oil used to overlay the culture medium micro drops was the target (Erbach et al., Human Reproduction, 1995, 10, 3248-54). We included some of these points in the revised version of the manuscript and added this figure as Supplementary Figure 1.

1. The reviewer noted that while dKO eggs showed reduced labile zinc levels, the amount of total zinc is not determined. Further, the response to thapsigargin in dKO eggs didn’t phenocopy the profile in eggs treated with TPEN. The reviewer argued that without further experimentation, such as comparing polar body extrusion and egg activation rate between WT and dKO, it seems to be a stretch to state that these eggs are zinc deficient.

Response 3: We agree that the statement, ‘zinc deficient,’ is an overstatement without determining the total zinc levels and associated phenotypes. Therefore, in the revised version of the manuscript, we referred to dKO-derived eggs and embryos as “low-level labile Zn2+ eggs”. Our follow-up studies show that eggs from dKO females seem to undergo egg activation events, such as the timing and rate of second polar body extrusion and pronuclear formation, with a similar dynamic to WT females. Hence, we estimate that the labile Zn2+ levels in dKO eggs are not as low as those of WT eggs treated with TPEN. Consequently, these intermediate zinc levels may have subtle effects, such as changing the Thapsigargin-induced Ca2+ release through the IP3R1 without causing widespread inhibition of cellular events observed after TPEN. We would argue that this approach is significant because it can distinguish how the different cellular events and proteins and enzymes have distinct affinities or zinc requirements and, in this case, start uncovering the channel(s) present in oocytes and eggs that may contribute to regulating zinc homeostasis.

1. The reviewer pointed out that since zinc is not redox active, it is unclear how zinc could be modifying cysteine residues of IP3R1.The reviewer suggested the possibility that excess zinc is binding to the cysteines and preventing their oxidation leading to the inhibition of the IP3R1 by blocking the channel, thereby preventing calcium release.

Response 4: The reviewer correctly points out that the mechanism(s) whereby excess Zn2+ modifies the IP3R1 function is undetermined in our study. Further, our description of ‘modifying’ is ambiguous and could be misinterpreted. Data in the literature, some of which we cite in the manuscript, shows that “oxidation of cysteine residues enhances receptor’s sensitivity to ligands in various cell types”. Zn2+ preferentially binds to reduced cysteine residues, and thus, we agree with the proposed reviewer's suggestion that “excess zinc may occupy reduced cysteine residues, preventing their oxidization required to sensitize the receptor”. As noted by the reviewer, we cannot rule out that it might be directly blocking the IP3R1 channel. We have modified the corresponding paragraphs in the Discussion.

1. Line 80 and 411, there are three other reports demonstrate the zinc reallocation to the egg shell or ejection as the zinc spark; Zebrafish: Converse et al. in Sci. Reports 10, 15673 (2020); X. lavis: Seeler et al. in Nature Chem. 13, 683-691 (2021), C. elegans: Mendoza et al. in Biology of Reproduction 107(2):406-418 (2022).

Response 5: Thank you for pointing this out, and we have added these references.

1. Line 129, when discussing that Zn2+ concentrations are reduced after TPEN as visualized by FluoZin-3, the authors should cite the article in which FluoZin-3 was first reported and this result was demonstrated initially: "Detection and Imaging of Zinc Secretion from Pancreatic β-Cells Using a New Fluorescent Zinc Indicator" by Gee et al. J. Am. Chem. Soc 124, 5, 776-778.

Response 6: Thank you for pointing this out, and we have added this reference.

1. In Figure 1E/Table 1 the authors evaluated if TPEN supplementation affects meiosis and pronuclear formation; however, the timing of TPEN treatment is unclear. When was TPEN introduced? Were the eggs left in the same media containing TPEN following fertilization, or were they transferred to different media?

Response 7: Thank you for pointing this out, and we have noted the time of the addition in the figure and text.

1. Line 1011 and 1012, ZnTP should be ZnPT.

Response 8: Thank you for pointing this out, which is now corrected.

Reviewer #2:

1. The reviewer raises the question of whether a more complex relationship could exist between the levels of zinc in MII eggs by indicating, “a more active relationship such that zinc efflux associated with each calcium spike could be necessary for terminating the Ca spike by depleting cytoplasmic zinc.” The reviewer also states, “Perhaps, rather than simply a permissive role, the normal Zn fluxes during activation may be acutely changing IP3-R gating sensitivity.”

Response 1: We agree that the demonstration that TPEN dose-dependently delays and consistently terminates ongoing Ca2+ rises perhaps reflects a more nuanced relationship between cytoplasmic labile zinc concentrations, Ca2+ oscillations, and IP3R1 function. Uncovering the precise nature of this relationship would require additional studies, such as determining the impact of TPEN on IP3 binding to its cognate receptor, regulation of channel gating, and more in-depth functional-structural experiments. However, these studies will demand time and complex experimental design and are beyond the scope of the current work. Nevertheless, they are excellent suggestions for future studies.

We would argue against the reviewer’s suggestion that “zinc sparks directly contribute to shaping the oscillations.” Zn2+ released during the sparks is not labile, but Zn2+ bound to cortical granules-resident proteins, most of which are inaccessible to the cytosol and hence to IP3R1s and should not perturb its function. We examined (data not shown) that the levels of cytosolic labile Zn2+, as assessed with FluoZin3, remained steady for over three hours of Plcζ mRNA-initiated oscillations. Further, because the Zn2+ sparks cease after the third or fourth Ca2+ rise, it would mean, at the very least, that this mechanism only operates on the first few responses. Thus, while the change of cytosolic Ca2+ concentrations triggers the Zn2+ sparks, we argue that the opposite influence is unlikely to hold true.

1. The reviewer also pointed out that the role of Trpv3 and Trpm7 in Zn2+ homeostasis seems to be minor and that the effects of genetic deletion of those channels are not as clear as those obtained by TPEN. Given that dKO eggs make it to the MII and release more but not less calcium upon thapsigargin than control despite the lowered labile Zn2+ level, the reviewer speculated that the loss of those channels changes calcium gating independent of Zn2+ concentration.

Response 2: TRPV3, TRPM7, and Cav3.2 are the three channels identified to permeate Ca2+ during oocyte maturation and egg activation in mice. We and other groups have observed that in oocytes and eggs, these channels partly compensate for the absence of each other because the deletion of these channels individually has a limited effect on Ca2+ oscillations and fertility. Thus, in the case of oocytes from Trpv3 and Trpm7 dKO animals, the other plasma membrane channel(s), most likely Cav3.2, is plausibly compensating, and its enhanced function underlies the increased Ca2+ response to Thapsigargin.

Nevertheless, the slower time to the peak and the lesser steep rise of the Thapsigargin induced rise suggest a negative impact of the dKO environment on IP3R1’s ability to mediate Ca2+ release. Based on the rest of the results in the manuscript, we attribute this change to the lower levels of labile Zn2+ in dKO eggs.

1. Lastly, the reviewer noted the upregulation of the Fura-2AM following addition of ZnPT. The reviewer indicated that 0.05 uM ZnPT might not increase intracellular Zn2+ to change Fura-2 fluorescence, but it might be sufficient Zn2+ to enter the cell and keep the IP3R1 channels open causing a sustained rise in cytoplasmic calcium and preventing oscillations. Further, if this interpretation holds true, the inhibitory effects of high Zn2+ on IP3R1’s gating shown in figure 7 would be precluded.

Response 3: We acknowledge that the increased levels of Fura-2 fluorescence following the addition of ZnPT could be due to the increased Zn2+ levels acting on IP3R1, increasing its open probability, and elevating cytosolic Ca2+ levels. We have added this consideration to the discussion. Nevertheless, our evidence suggests that this is unlikely because, as shown in Figure 6 H, I, the ER-Ca2+ levels as assessed by D1ER recordings did not change following the addition of ZnPT, whereas Rhod-2 fluorescence did, suggesting that the two events are seemingly uncoupled. Further, constant leak from the ER and extended high cytosolic Ca2+ would lead to egg activation or cell death, neither of which changes were observed.

Reviewer #3:

The reviewer noted that the present study deepened the understanding of the role of zinc in regulating calcium channels and stores at fertilization beyond the previously known Zn2+ requirement in oocyte maturation and the cell cycle progression. We appreciate these comments.

1. Fig. 1. The reviewer wondered why we selected 10 μM TPEN for most of the experiments in the manuscript. The reviewer noted this concentration only stopped the Ca2+oscillations in just half of the eggs after ICSI.

Response 1: We used 10-μM TPEN throughout the study because it blocked ~50% of the oscillations of a robust trigger of Ca2+ responses such as ICSI and reduced the frequency in the remaining eggs. This concentration of TPEN abrogates and prevents the responses by milder stimuli, such as Acetylcholine and SrCl2. Importantly, thimerosal and Plcζ mRNA overcome the inhibition by 10μM but not 50-μM TPEN. However, 50μM TPEN inactivates Emi2, a Zn2+-dependent enzyme, causing parthenogenic activation and cell cycle progression, and we wanted to avoid this confounding factor. Therefore, we determined 10-μM is a “threshold” concentration and selected it for the remaining studies. We also reasoned that it would allow the detection of more subtle effects of reducing the levels of labile zinc, causing a milder inhibition of IP3R1 sensitivity and a progressive delay or modification of the responses to other agonists rather than fully abrogating them, which is the case with higher concentrations.

1. Line131 - no concentration of TPEN stated? Or 'the addition of different concentrations of TPEN"?

Response 2: We have corrected this. We have now added 50-100 µM concentrations.

1. Line 146 - instead of TPEN, all TPEN concentrations?

Response 3: We have added these corrections, as at the concentrations we tested here, 5μM TPEN and above, all caused a reduction in the baseline of Fura-2 fluorescence.

1. Line 1046 - 'We submit'? Propose?

Response 4: We have replaced the word submit for propose. Thank you for the suggestion.

Author Response

Reviewer #2 (Public Review):

In this paper, the authors discover that postsynaptic mitochondria in C. elegans govern glutamate receptor trafficking dynamics. The core results are two-fold. For one, they find that loss or inhibition of mcu-1 - the C. elegans mitochondrial calcium uniporter - increases GLR-1 glutamate receptor accumulation at the postsynaptic dendritic sites and enhances its trafficking dynamics. The authors hypothesize that this effect on glutamate receptors may have something to do with mitochondrial ROS production. This is because ROS is a by-product of normal oxidative phosphorylation, downstream of calcium import. Indeed, the generation of artificially high amounts of mitochondrial ROS has the opposite effect of mcu-1 loss: decreased glutamate receptor subunit accumulation. Collectively, the results support the idea that mitochondrial function can control receptor dynamics at synaptic sites. This is interesting because tight control of synaptic function likely combines several mitochondrial functions: energy production, calcium buffering, and (here) ROS signaling.

STRENGTHS

• The C. elegans genetic model is a strength because the authors are able to make refined conclusions by classical loss-of-function mutants (e.g., mcu-1) along with an impressive cytological toolkit to examine GLR-1 dynamics.

• The use of pharmacology as a second means to test those genetic conclusions is a strength.

• The authors' careful reagent verification of reporters (Ca2+, ROS, etc.) is a strength.

• The ability to link fundamental mitochondrial processes to GLR-1 exocytosis will expand how the field thinks about mitochondrial synapse function.

WEAKNESSES

For the most part, the data in the paper support the conclusions, and the authors were careful to try experiments in multiple ways. But please see below:

• (Main Point) The data are good, but they fall short of mechanism (e.g., Line 322). Figure 6 is accurate as drawn. But calcium and ROS are not abstract signals. They are likely exerting affirmative actions on specific targets. The Discussion does acknowledge this in terms of ROS and it speculates on possible targets.

We thank the reviewer for their analytical review of our manuscript. We agree that all molecular players involved in the proposed mechanism were not identified by the data presented, so we modified the text to remove overstatements. We also agree that Ca2+ and ROS signaling is not abstract. Rather, there are specific and diverse targets of both Ca2+ and ROS signaling. Follow-up experiments are underway to identify and provide evidence for the necessity of potential ROS/Ca2+ targets in this proposed mechanism. For the current manuscript, we have modified our verbiage in an attempt to not mislead or overstate what our results suggest (e.g., changes/additions to the beginning of the ‘Discussion’, lines 365-377 and 385-388) and updated the illustration of the proposed model to include dashed lines that, as mentioned in the figure legend, indicate indirect action by ROS and Ca2+ (see revised Figure 7).

The general idea seems to be that mitochondria import calcium through MCU-1 (and interacting factors). As a result, oxidative phosphorylation successfully occurs and mitochondrial ROS is a signaling by-product that signals glutamate receptors not to undergo exocytosis. But there are other interpretations of what might happen in between. In fact, if OXPHOS is disrupted, it is known that this can generate a lot more mitochondrial ROS than the normal by-product levels.

We do agree that an alternative explanation could be that genetic or pharmacological inhibition of mitochondrial Ca2+ uptake disrupts oxidative phosphorylation, and as a result, inefficiencies or uncoupling in the electron transport chain would lead to an even greater increase in mitochondrial ROS production. Although oxidative phosphorylation was not directly measured, one of our post hoc analyses of GLR-1 transport suggests ATP levels are comparable between controls, mcu-1 mutants, and with Ru360 treatment: the velocity of GLR-1 transport is unchanged between these experimental groups. The processivity of molecular motors (which dictates transport velocity) is highly sensitive to relative ATP abundance. Thus, if ATP levels were dramatically decreased in mcu-1 mutants or following Ru360 treatment, then one would expect a detectable change in GLR-1 transport velocities, but we observed no change (see revised Figure S2E and related discussion at lines 183-190). Although these results do not directly indicate whether ATP production is altered with loss or inhibition of MCU-1, it does suggest that basal ATP levels remain sufficient to support the metabolic demands of GLR-1 transport.

This reviewer wonders if excess ROS would cause an extreme response. Or alternatively, if scavenging ROS via pharmacological scavengers or SOD expression would reverse the effects.

These are good points, and we have previously published experiments that address each of them. First, we have seen that globally increasing ROS with various concentrations of H2O2 within the physiological range (<100 nM) decreased GLR-1 transport to a similar extent (PMID: 32847966) indicating that there is not a dose-dependent decrease in GLR-1 transport. We have also assessed GLR-1 transport after treatment with concentrations of H2O2 well above the physiological range (e.g., 500 nM), but these high concentrations obliterated all GLR-1 transport. Contrary to what one may expect, we showed that decreasing ROS via pharmacological or genetic means (probably below physiological range) decreased GLR-1 transport (PMID: 35622512) via a Ca2+ independent mechanism. In other words, ROS scavenging did not have the opposite effect on GLR-1 transport, but we have not combined ROS scavenging with optical induction of ROS production (e.g., via KillerRed) nor have we assessed the potential influence of ROS scavenging on synaptic recruitment. Although we agree that these are important follow-up experiments, they will require a more sensitive ROS indicator because current genetically encoded in vivo ROS sensors cannot detect decreases in ROS levels below the physiological range (< 10 nM) (PMID: 31586057).

Small Points

• 33.3 mHz - just making sure, do the authors mean once every 30 seconds? That would be more straightforward.

Yes, we do mean a 1-second pulse of light every 30 seconds. We have clarified this in the manuscript text (line 115).

• Figure 2 is confusing. The text says that the mcu-1 mutants have a GLR-1::GFP FRAP rate that is comparable to controls (Lines 165-167). But Figure 2E suggests that it is markedly less, which is the opposite result of the slight increase in rate resulting from Ru360 treatment. And is the explanation why the GLR-1::GFP results differ from the SEP::GLR-1 results a difference between total GFP vs. surface GFP?

The confusion is due to an incorrect statement in the results text. We have corrected this error and appreciate the reviewer for bringing it to our attention (lines 173-174).

• I could not watch Video 2 (not sure if it is the file or just the copy I downloaded).

We thank the reviewer for bringing this to our attention and we believe we have remedied the issue.

• It is good that the authors tried both optical stimulation and mechanical stimulation (dropping culture plates to stimulate the worms, Figure 3). Why was the mechanical stimulation set aside for further tests in the paper?

Mechanical stimulation consisted of dropping culture plates containing 2-3 C. elegans onto a lab bench every 30 seconds for 5 or 10 minutes. This mechanical stimulation paradigm was technically cumbersome and was less effective at inducing changes in mito-roGFP fluorescence that optical stimulation. This is likely due to habituation to the mechanical stimulus which has been well-characterized in C. elegans. The optical stimulation was therefore used as it is a more reliable and repeatable method for stimulating the AVA neuron.

• Does this process affect all kinds of transport, or is it just the glutamate receptors? Was anything else examined?

Transport of other proteins has not been examined in the context of mitoROS signaling. Our attempts at visualizing and quantifying the transport, synaptic delivery and exocytosis of other synaptic proteins in vivo has proven to be more technically challenging likely due to relatively lower expression in the C. elegans neurons suitable for transport analysis.

Reviewer #3 (Public Review):

Reactive oxygen species (ROS) have been previously shown to regulate glutamate receptor phosphorylation, long-distance transport, and delivery of glutamate receptors to synapses, however, the source of ROS is unclear. In this study, the authors test if mitochondria act as a signaling hub and produce ROS in response to neuronal activity in order to regulate glutamate receptor trafficking. The authors use a variety of optogenetic tools including the calcium reporter mitoGCaMP and the ROS reporter mito-roGFP to monitor changes in calcium and ROS, respectively, in mitochondria after activating neurons with ChRimson in the genetic model organism C. elegans. Repeated stimulation of interneurons called AVA with ChRimson leads to increased calcium uptake into mitochondria in dendrites and increased mitochondrial ROS production. The mitochondrial calcium uniporter mcu-1 is required for these effects because mcu-1 genetic loss of function or treatment with Ru360, a drug that inhibits mcu-1, inhibits the uptake of calcium into mitochondria and ROS production after neuronal activation. Mcu-1 genetic loss of function is correlated with an increase in exocytosis of glutamate receptors but a decrease in glutamate receptor transport and delivery to dendrites. This study suggests that mitochondria monitor neuronal activity by taking up calcium and downregulating glutamate receptor trafficking via ROS, as a means to negatively regulate excitatory synapse function.

Strengths

-The use of multiple optogenetic tools and approaches to monitor mitochondrial calcium, reactive oxygen species, and glutamate receptor trafficking in live organisms.

-Identifying a novel signaling role for dendritic mitochondria which is to monitor neuronal activity (via calcium uptake into mitochondria) and generate a signal (reactive oxygen species) that regulates glutamate receptors at synapses.

Weaknesses

-Although the use of KillerRed to generate ROS downstream of mcu-1 is a clever approach, the fact that activation of KillerRed results in reduced GLR-1 exocytosis, delivery, and transport raises the concern that KillerRed is generating a high level or ROS that might be toxic to cellular processes. Experiments showing that other cellular processes are not affected by KillerRed activation and testing if reduced ROS production mimics the effects of blocking mcu-1 would strengthen the conclusions in this study.

We thank the reviewer for their careful analyses of our findings. It is plausible that KillerRed could cause toxic levels of ROS, in fact, it was originally used to instigate oxidative stress-induced apoptosis to achieve cell-specific ablation. These cell ablation protocols required 20+ minutes of KillerRed activation with substantially higher levels of irradiation (e.g., 3.8 mW/mm [PMID: 24209746] vs. our light dosage of 25 µW/mm2). Additionally, our transgenic C. elegans strains expressing KillerRed were designed to have a relatively low KillerRed expression and were screened for low expression based on KillerRed’s fluorescence. Using these strains, we were able to minimally activate KillerRed in the AVA neuron resulting in ROS elevations at mitochondria that were comparable to neuronal activity-induced increases in mitochondrial ROS as measured by mito-roGFP. Specifically, we found that 10 minutes of mechano-stimulation and 5 minutes of ChRimson stimulation increased the fluorescence ratio (Fratio) of mito-roGFP nearly two-fold (Figure 4A-B and 4C-E). A 15-second pulse of light focused on a small region activating mitoKR in the AVA neurite also caused similar two-fold increase in the mito-roGFP Fratio (Figure 4C-E) comparable to what neuronal activity induced. Our 5-minute global KillerRed activation less effectively increased the mito-roGFP Fratio at mitochondria in the AVA neurite compared to neuronal activity (revised Figure 4B and 4H) but was sufficient in decreasing GLR-1 transport (revised Figure 5G-H). So, we decided to do all experiments with 5 minutes of global KillerRed activation since lower activation levels of KillerRed were more likely to achieve non-toxic, signaling levels of ROS. Since we strongly agree that this data is important for tool validation, we have reorganized the manuscript such that these data are now a primary figure (see revised Figure 4 and new results sub-section starting at line 252).

Additionally, we added supplemental transport velocity data. This data shows that local photoactivation as well as whole-cell activation of KillerRed does not alter transport velocity of GLR-1 vesicles within the neurite (revised Figure S4A and S4B and lines 272-276 and 287-289), which would be the case if ATP, microtubules, or actin dynamics were affected. This supports that our local and whole-cell activation protocol does not cause toxic levels of ROS production.

Lastly, the reviewer questions whether decreasing ROS alters GLR-1 transport, synaptic delivery and exocytosis in a similar fashion to loss or inhibition of mcu-1, and if so, would further support the proposed mechanism. We have decreased ROS via genetic (catalase overexpression) and pharmacological (using the mitochondria-targeted antioxidant MitoTEMPO) means and seen that diminished ROS levels decrease GLR-1 transport albeit to a lesser degree than that caused by loss/inhibition of mcu-1 (PMID: 35622512). To determine if decreased GLR-1 transport during diminished ROS levels involves mcu-1, we would need to assess GLR-1 transport in mcu-1 mutants while ROS is decreased (e.g., using MitoTEMPO treatment) to see if their combined effect phenocopies the effect of mcu-1(lf) or decreased ROS alone. However, as mentioned previously, we are unable to measure ROS levels below the sensitivity of roGFP but within physiological range so we cannot currently calibrate or validate our methods for scavenging ROS in vivo. This is why we have not yet analyzed synaptic delivery or exocytosis rates of GLR-1 in the context of decreased ROS, but these would be interesting follow-up experiments that may further support our model once more sensitive ROS sensors are available.

Reviewer #4 (Public Review):

Using optogenetic stimulation, the authors presented compelling evidence that neuronal activity increases mitochondrial calcium levels, facilitated by the mitochondrial uniporter MCU-1. Through ratiometric measurements, they showed that mitochondrial ROS levels also increase due to neuronal activity via MCU-1. Subsequent FRAP studies were employed to investigate the trafficking of the AMPA receptor, GLR-1. By integrating genetic and pharmacological methodologies, the recovery rate of GLR-1 was assessed. The authors concluded that increased mitochondrial ROS due to neuronal activity reduces the trafficking and exocytosis of AMPA receptors. They proposed that mitochondrial ROS serves as a homeostatic mechanism regulating AMPA receptor trafficking and abundance, thus maintaining synaptic strength. This research is crucial as it provides a direct link between mitochondrial signaling and AMPA receptor trafficking.

However, there are several significant concerns regarding the methodologies and quantifications employed in this manuscript. The authors utilized GLR-SEP to label surface AMPA receptors and relied on the "FRAP rate" as an indicator of the exocytosis rate. The absence of direct visualization of exocytosis using GLR-SEP, and the lack of direct measurements of exocytosis events, casts doubt on the conclusions about ROS's impact on AMPA receptor exocytosis. Furthermore, the "FRAP rate" determined in this study is a combination of recovery rates (incorporating both endosomal trafficking and diffusion) with the mobile fractions of AMPA receptors, potentially weakened interpretations of the findings. A more comprehensive discussion addressing the conflicting effects of MCU-1 and ROS on GLR-GFP FRAP recovery and dendritic trafficking would enable readers to grasp the intricate roles of mitochondrial calcium and ROS in modulating synaptic receptors.

We appreciate the reviewer’s attention to detail while reviewing our article. Their major concern about directly visualizing exocytosis events is valid since changes in exocytosis and endocytosis would dictate the amount of SEP::GLR-1 at the synaptic membrane. However, streaming imaging of SEP in vivo is technically difficult showing only few exocytosis events and provides short “snapshots” (1-2 minutes, longer streaming imaging causes photobleaching and photo-toxicity) which must be extrapolated to longer time frames. Our 16-minute SEP::GLR-1 FRAP protocol allows us to capture all plasma membrane recruitment and quantify the relative balance between exo- and endocytosis. It also allows for longer observational periods during which we can detect changes in GLR-1 recruitment to and retention at the synaptic membrane in genetic mutants and with drug treatments. In addition, our photobleaching approach involves photobleaching a ~40-60 µm region proximally and distally to the imaging region which limits the influence of receptor diffusion on the FRAP rate. The reviewer makes a valid point that receptor endocytosis rates would also influence the SEP::GLR-1 FRAP rate. We have now changed the text in the results and discussion to include this information (lines 155-161, and changing “exocytosis” to “synaptic recruitment” throughout the manuscript when discussing SEP::GLR-1 FRAP results [e.g, at lines 169, 208, and 321]).

eLife assessment

This study examines an interplay between synaptic mitochondria and glutamate receptor exocytosis in C. elegans. Collectively, the solid results support the idea that mitochondrial function influences receptor dynamics at postsynaptic sites. This is important because tight control of synaptic function likely integrates several mitochondrial functions: energy production, calcium buffering, and (here) reactive oxygen species signaling.

Reviewer #2 (Public Review):

In this paper, the authors discover that postsynaptic mitochondria in C. elegans govern glutamate receptor trafficking dynamics. The core results are two-fold. For one, they find that loss or inhibition of mcu-1 - the C. elegans mitochondrial calcium uniporter - increases GLR-1 glutamate receptor accumulation at the postsynaptic dendritic sites and enhances its trafficking dynamics. The authors hypothesize that this effect on glutamate receptors may have something to do with mitochondrial ROS production. This is because ROS is a by-product of normal oxidative phosphorylation, downstream of calcium import. Indeed, the generation of artificially high amounts of mitochondrial ROS has the opposite effect of mcu-1 loss: decreased glutamate receptor subunit accumulation. Collectively, the results support the idea that mitochondrial function can control receptor dynamics at synaptic sites. This is interesting because tight control of synaptic function likely combines several mitochondrial functions: energy production, calcium buffering, and (here) ROS signaling.

STRENGTHS

• The C. elegans genetic model is a strength because the authors are able to make refined conclusions by classical loss-of-function mutants (e.g., mcu-1) along with an impressive cytological toolkit to examine GLR-1 dynamics.

• The use of pharmacology as a second means to test those genetic conclusions is a strength.

• The authors' careful reagent verification of reporters (Ca2+, ROS, etc.) is a strength.

• The ability to link fundamental mitochondrial processes to GLR-1 exocytosis will expand how the field thinks about mitochondrial synapse function.

WEAKNESSES

For the most part, the data in the paper support the conclusions, and the authors were careful to try experiments in multiple ways. But please see below:

• (Main Point) The data are good, but they fall short of mechanism (e.g., Line 322). Figure 6 is accurate as drawn. But calcium and ROS are not abstract signals. They are likely exerting affirmative actions on specific targets. The Discussion does acknowledge this in terms of ROS and it speculates on possible targets.

The general idea seems to be that mitochondria import calcium through MCU-1 (and interacting factors). As a result, oxidative phosphorylation successfully occurs and mitochondrial ROS is a signaling by-product that signals glutamate receptors not to undergo exocytosis. But there are other interpretations of what might happen in between. In fact, if OXPHOS is disrupted, it is known that this can generate a lot more mitochondrial ROS than the normal by-product levels.

This reviewer wonders if excess ROS would cause an extreme response. Or alternatively, if scavenging ROS via pharmacological scavengers or SOD expression would reverse the effects.

Small Points

• 33.3 mHz - just making sure, do the authors mean once every 30 seconds? That would be more straightforward.

• Figure 2 is confusing. The text says that the mcu-1 mutants have a GLR-1::GFP FRAP rate that is comparable to controls (Lines 165-167). But Figure 2E suggests that it is markedly less, which is the opposite result of the slight increase in rate resulting from Ru360 treatment. And is the explanation why the GLR-1::GFP results differ from the SEP::GLR-1 results a difference between total GFP vs. surface GFP?

• I could not watch Video 2 (not sure if it is the file or just the copy I downloaded).

• It is good that the authors tried both optical stimulation and mechanical stimulation (dropping culture plates to stimulate the worms, Figure 3). Why was the mechanical stimulation set aside for further tests in the paper?

• Does this process affect all kinds of transport, or is it just the glutamate receptors? Was anything else examined?

Reviewer #3 (Public Review):

Reactive oxygen species (ROS) have been previously shown to regulate glutamate receptor phosphorylation, long-distance transport, and delivery of glutamate receptors to synapses, however, the source of ROS is unclear. In this study, the authors test if mitochondria act as a signaling hub and produce ROS in response to neuronal activity in order to regulate glutamate receptor trafficking. The authors use a variety of optogenetic tools including the calcium reporter mitoGCaMP and the ROS reporter mito-roGFP to monitor changes in calcium and ROS, respectively, in mitochondria after activating neurons with ChRimson in the genetic model organism C. elegans. Repeated stimulation of interneurons called AVA with ChRimson leads to increased calcium uptake into mitochondria in dendrites and increased mitochondrial ROS production. The mitochondrial calcium uniporter mcu-1 is required for these effects because mcu-1 genetic loss of function or treatment with Ru360, a drug that inhibits mcu-1, inhibits the uptake of calcium into mitochondria and ROS production after neuronal activation. Mcu-1 genetic loss of function is correlated with an increase in exocytosis of glutamate receptors but a decrease in glutamate receptor transport and delivery to dendrites. This study suggests that mitochondria monitor neuronal activity by taking up calcium and downregulating glutamate receptor trafficking via ROS, as a means to negatively regulate excitatory synapse function.

Strengths<br /> -The use of multiple optogenetic tools and approaches to monitor mitochondrial calcium, reactive oxygen species, and glutamate receptor trafficking in live organisms.<br /> -Identifying a novel signaling role for dendritic mitochondria which is to monitor neuronal activity (via calcium uptake into mitochondria) and generate a signal (reactive oxygen species) that regulates glutamate receptors at synapses.

Weaknesses<br /> -Although the use of KillerRed to generate ROS downstream of mcu-1 is a clever approach, the fact that activation of KillerRed results in reduced GLR-1 exocytosis, delivery, and transport raises the concern that KillerRed is generating a high level or ROS that might be toxic to cellular processes. Experiments showing that other cellular processes are not affected by KillerRed activation and testing if reduced ROS production mimics the effects of blocking mcu-1 would strengthen the conclusions in this study.

Reviewer #4 (Public Review):

Using optogenetic stimulation, the authors presented compelling evidence that neuronal activity increases mitochondrial calcium levels, facilitated by the mitochondrial uniporter MCU-1. Through ratiometric measurements, they showed that mitochondrial ROS levels also increase due to neuronal activity via MCU-1. Subsequent FRAP studies were employed to investigate the trafficking of the AMPA receptor, GLR-1. By integrating genetic and pharmacological methodologies, the recovery rate of GLR-1 was assessed. The authors concluded that increased mitochondrial ROS due to neuronal activity reduces the trafficking and exocytosis of AMPA receptors. They proposed that mitochondrial ROS serves as a homeostatic mechanism regulating AMPA receptor trafficking and abundance, thus maintaining synaptic strength. This research is crucial as it provides a direct link between mitochondrial signaling and AMPA receptor trafficking.

However, there are several significant concerns regarding the methodologies and quantifications employed in this manuscript. The authors utilized GLR-SEP to label surface AMPA receptors and relied on the "FRAP rate" as an indicator of the exocytosis rate. The absence of direct visualization of exocytosis using GLR-SEP, and the lack of direct measurements of exocytosis events, casts doubt on the conclusions about ROS's impact on AMPA receptor exocytosis. Furthermore, the "FRAP rate" determined in this study is a combination of recovery rates (incorporating both endosomal trafficking and diffusion) with the mobile fractions of AMPA receptors, potentially weakened interpretations of the findings. A more comprehensive discussion addressing the conflicting effects of MCU-1 and ROS on GLR-GFP FRAP recovery and dendritic trafficking would enable readers to grasp the intricate roles of mitochondrial calcium and ROS in modulating synaptic receptors.

Author Response

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

necessary clarifications on some of the reviewers' suggestions.

Reviewer #1 (Public Review):

Weaknesses:

• This is a pilot study with only 24 cases and 24 controls. Because the human microbiota entails individual variability, this work should be confirmed with a higher sample size to achieve enough statistical power.

Thank you for your suggestion. Unlike the high sparsity of 16s rRNA, the data density of metagenomic data is higher. Based on the experience of previous research, the sample size used this time can basically meet the requirements. However, your suggestion is very valuable, increasing the sample size allows better in-depth analysis. Due to limitations of objective factors, it is difficult for us to continue to increase the sample size in this study.

• The authors do not report here the use of blank controls. The use of this type of control is important to "subtract" the potential background from plasticware, buffer or reagents from the real signal. Lack of controls may lead to microbiome artefacts in the results. This can be seen in the results presented where the authors report some bacterial contaminants (Agrobacterium tumefaciensis, Aequorivita lutea, Chitinophagaceae, Marinobacter vinifirmus, etc) as part of the most common bacteria found in cervical samples.

Thank you for your suggestion. Applying blank controls in low biomass areas can effectively avoid contamination caused by the environment or kits. This opinion is consistent with that published by Raphael Eisenhofer et al. in Trends in Microbiology. When designing this study, we considered that this study described a biomass-rich site, and the abundance of dominant species was much higher than that of the possible 'kitome', so we did not set a blank control. On the other hand, our main discussion object in this study is high-abundance species, and the species filtering threshold for some analyzes was raised to 50%. Therefore, we believe that the absence of the blank control has little effect on the conclusions of this study. However, your opinion is spot on. Failure to set up a negative control will affect our future research on rare species. We will add a description in the Limitations section of the Discussion section.

• Samples used for this study were collected from the cervix. Why not collect samples from the uterine cavity and isthmocele fluid (for cases)? In their previous paper using samples from the same research protocol ((IRB no. 2019ZSLYEC-005S) they used endometrial tissue from the patients, so access to the uterine cavity was guaranteed.

Thank you for your suggestion. In Author response image 1 we show the approximate location of our cervical swab sampling. There are two main reasons for choosing cervical swabs:

1) The adsorption of swabs allows us to obtain sufficient nucleic acid for high-depth sequencing, while the isthmocele fluid varies greatly among patients, which will introduce unnecessary batch effects.

2) Since the female reproductive tract is a continuous whole, our sampling location is close to the lesion in the cervix, which can be effectively studied. On the other hand, the microbial biomass of the endometrium is probably two orders of magnitude lower than that of the cervix, and it is difficult to avoid contamination of the lower genital tract when sampling.

Based on the above reasons, we selected cervical swabs for our microbial data.

Author response image 1.

• Through the use of shotgun genomics, results from all the genomes of the organisms present in the sample are obtained. However, the authors have only used the metagenomic data to infer the taxonomical annotation of fungi and bacteria.

Thank you for your suggestion. The advantage of metagenomics is that it can obtain all the nucleic acid information of the entire environment. However, in the study of the female reproductive tract, the database of viruses and archaea is still immature, in order to ensure the accuracy of the results, we did not conduct the study. Looking forward to the emergence of a mature database in the future.

Reviewer #1 (Recommendations For The Authors):

• It would be interesting to use another series of functional data coming from the metagenomic analyses (not only taxonomic) to expand and reinforce the results presented.

Thank you for your suggestion. We have dissected the functional data of microbiota in the article.

• The authors have previously published the 16S rRNA sequencing and transcriptomic analysis of the same set of patients. It would be nice to see the integration of all the datasets produced.

Thank you for your suggestion. There is no doubt that integrating all the data will have more dimensional results. In our previous study we focused on microbe-host interactions. However, there is an unanswered question: What are the characteristics of the regulatory network within microbiota? Therefore, we answered this question in this study, exploring the complex interaction processes within microbial communities. In addition to direct effects, interactions between microbiota may also occur through special metabolite experiments. Therefore, we introduced the analysis of the untargeted metabolome. However, 16s rRNA can only provide bacterial information, so we did not integrate the data. In addition, the transcriptome provides host information and is not the focus of this study. However, your suggestion is very valuable, and we will integrate all the data in the next study on the exploration of treatment methods.

Reviewer #2 (Public Review):

Weaknesses: Methodological descriptions are minimal.

Some example:

*The CON group (line 147) has not been defined. I supposed it is the control group.

• There are no statistics related to shotgun sequencing. How many reads have been sequenced? How many have been removed from the host? How many are left to study bacteria and fungi? Are these reads proportional among the 48 samples? If not, what method has been used to normalise the data?

• ggClusterNet has numerous algorithms to better display the modules of the microbiome network. Which one has been used?

Thank you for your suggestion. We have added details to the method.

Reviewer #2 (Recommendations For The Authors):

I think the author should take into account the points described in the "Weaknesses" section. The lack of detail extends to almost all the analyses that have been included in the manuscript. Although the results are sound, I think it is important to understand what has been analysed and how it has been analysed. It is important that all work is reproducible and this requires vital information.

For example, what parameters have been used for bowtie2? has a local analysis been used? or end-to-end ? Some parameters like --very-sensitive are important for this kind of analysis. You can also use specific programs like kneaddata.

The Raw data preprocessing section should be more detailed.

The same with the "Taxa and functional annotation" section, how have the data been normalised? has any Zero-Inflated Gamma probabilistic model algorithm been taken into account? How were the 0 (no species detected) in the shallow samples treated?

Which algorithms have been used for LEfSe ? Kluskal-Wallis->(Wilcoxon)->LDA ?

Which p-value has been used as cut-off ? this p-value has been corrected for multiple testing?

• Information on ggClusterNet should be included and explained.

The first section of the results and Table 1 should be in the Materials and Methods.

Thank you for your suggestion. We have added details to the method.

In the fungi section, it is mentioned that 431 species have been found. They should be included in a supplementary table.

How many bacteria were found? Please include them also in a supplementary table.

Thank you for your suggestion. We have added the corresponding table.

Reviewer #3 (Public Review):

Major

1. Smoke or drink conditions, as well as diseases like hypertension and diabetes are important factors that could influence the metabolism of the host, thus the authors should add them in the exclusion criteria in the Methods.

Thanks to reviewer #3 for professional comments. We have made corresponding additions in the method section. We also followed this standard when recruiting subjects.

1. The sample size of this study is not large enough to draw a convincing conclusion.

Thank you for your suggestion. Unlike the high sparsity of 16s rRNA, the data density of metagenomic data is higher. Based on the experience of previous research, the sample size used this time can basically meet the requirements. However, your suggestion is very valuable, increasing the sample size allows better in-depth analysis. Due to limitations of objective factors, it is difficult for us to continue to increase the sample size in this study.

Reviewer #3 (Recommendations For The Authors):

Please recruit more samples.

In addition, there are many formatting and grammatical mistakes in the manuscript.

Minor

1. In Line 24-25 of the "Composition and characteristics of fungal communities", the format of "Goyaglycoside A and Janthitrem E." shouldn't be italic.

2. In Line 126 of the "Metabolite detection using liquid chromatography (LC) and mass spectrometry (MS)", the "10 ul" should be changed to "Ten ul". Beginning with arabic numerals in a sentence should be avoided.

3. In Line 170 of the "Composition and characteristics of bacterial communities", the "162 differential species" should be "One hundred and sixty-two differential species".

4. In Line 187 of the "Composition and characteristics of fungal communities", the "42 differential" should be "Forty-two differential".

Thanks to reviewer #3 for professional comments. We have completely revised the language of the article.

Author Response

Reviewer #1 (Public Review):

Payne et al. have investigated the neural basis of VOR adaptation with the goal of constraining sites and mechanisms of plasticity supporting cerebellar learning. This has been an area of intense debate for decades; previous competing models have argued extensively about the sites of plasticity and the strength of eye velocity feedback/ efference copy signals to Purkinje cells has been central to the debate. This paper nicely explores the consequences of varying the strength of this feedback and in so doing, provides a potential explanation for why Purkinje cell responses during VOR cancellation could exhibit stronger responses following learning, despite net depression of the strength of their vestibular inputs. In that sense it provides some reconciliation of existing models. The work appears to be well done and the paper is well written. The manuscript could be improved and the significance of the work clarified and enhanced by contextualizing the work more appropriately within the existing literature in this area.

We thank the reviewer for the nice summary of this work’s contribution to the long-standing debate regarding sites and mechanisms of plasticity underlying cerebellar learning.

We have revised the manuscript to address several key points raised by the reviewer. We now emphasize that the main evidence for weak feedback arises from interpreting our model in the context of the existing experimental evidence for plasticity rules in the cerebellar cortex, and we have clarified the commonalities and differences from the Miles-Lisberger model. Several missing references are now included. Additionally, we clarify the comparison of our model to data after learning, and explain how altered signaling through the visual pathways drives paradoxical changes in neural activity without requiring plasticity in the visual pathways. We hope that these changes better situate the work to be interpreted appropriately in the context of the existing literature.

Reviewer #2 (Public Review):

Payne et al. use a computational approach to predict the sites and directions of plasticity within the vestibular cerebellum that explain an unresolved controversy regarding the basis of VOR learning. Specifically, the conclusion by Miles and Lisberger (1981) that vestibular inputs onto Purkinje cells (PCs) must potentiate, rather than depress (as in the Marr/Albus/Ito model), following gain-increase learning because when the VOR is cancelled, PC firing increases rather than decreases. Payne et al. provide a novel model solution that recapitulates the results of Miles and Lisberger but, paradoxically, uses plasticity in the cerebellar cortex that weakens PC output rather than strengthens it. However, the model only succeeds when efference copy feedback to the cerebellar cortex is relatively weak thereby allowing a second feedback pathway to drive PC activity during VOR cancellation to counteract the learned change in gain. Because the model is biologically constrained, the findings are well supported. This work will likely benefit the field by providing a number of potentially experimentally testable conclusions. The findings will be of interest to a wider audience if the results can be extrapolated to other cerebellar-dependent learning behaviors rather then just VOR gain-increase learning. Overall, the manuscript is very well written with clearly delineated results and conclusions.

We appreciate the reviewer’s comments that the model is well-constrained and provides a solution to the long-standing debate surrounding sites and directions of plasticity underlying VOR learning.

The reviewer raises an important question: do our results generalize across the cerebellum? We note first that we are studying the cerebellum to illustrate a core problem in modeling systems throughout the brain, namely, how to disambiguate plasticity in the face of ubiquitous feedback loops, both within the brain and between the brain and the environment. Within the cerebellum, we focused on VOR learning due to the wealth of experimental data available. While the specific effect of feedback strength on plasticity will depend on the details of the relevant cerebellar circuit, our general approach can be applied to other areas, given sufficient data, in order to determine how plasticity is distributed in the face of potential feedback loops. Importantly, error-driven LTD of the parallel fiber-Purkinje cell synapse is a fundamental hypothesized mechanism for cerebellar learning which has been generally accepted elsewhere in the cerebellum, but was called into question for VOR learning in the flocculus by the Miles-Lisberger model. Thus, our study of VOR learning has broad implications for reconciling plasticity mechanisms across the cerebellum.

We also note that, even within the VOR circuit, the direction of plasticity and the relative dependence on plasticity at each site may depend on the timescale of learning. On longer timescales, there is thought to be consolidation of learning from a cerebellar cortical site to a brainstem site. Such consolidation from a faster-learning site to a slower-learning site is known as systems consolidation and has been shown theoretically to mitigate the ‘plasticity-stability dilemma’ of having fast learning without over-writing longer-term learning. Our model is compatible with both error-driven plasticity in the cerebellar cortex and a site of plasticity in the brainstem, with brainstem plasticity potentially mediating consolidation of earlier learned changes in the cerebellar cortex. We have now updated the text significantly to discuss the broader implications of the results and to address the reviewer’s specific comments.

Reviewer #3 (Public Review):

Summary: In this study, the authors attempt to determine what is the role (and strength) of feedback in a closed-loop (cerebellar) system.

Strengths:

1) By combining extensive data fitting of cerebellar experimental observations this study provides deep insights into existing questions and more broadly on the role of feedback and what are the limitations when inferring feedback in (plastic) neural circuits.

2) Another strength of this study is the gradual build-up of evidence by using models of different complexities to help build the argument that weak feedback is sufficient to explain experimental observations.

3) The paper is well-written and structured.

Weaknesses:

1) In principle feedback can (i) drive dynamics or/and (ii) drive learning directly. Throughout the paper, the authors refer to only the first case (i.e. dynamics). However, the role of feedback in learning is already implicitly assumed by the authors when jointly fitting the model before and after learning. Note that the general conclusion that feedback (in general) is weak may be to the first view (i.e. dynamics), but not the second. Given that a key conclusion of the paper is that no feedback is sufficient to explain the data, this suggests that feedback may instead be used for learning/plasticity.

We fully agree with the reviewer that our conclusions do not preclude an important role for many other types of feedback, including as an instructive signal for learning. Instead of explicitly considering feedback for learning in our model, we consider static snapshots before and after learning to infer plasticity, while remaining agnostic to the neural algorithm used to achieve such plasticity. A widely held hypothesis is that motor error signals carried by climbing fibers instruct LTD at co-active parallel fiber inputs to Purkinje cells; this is indeed a form of feedback, operating on a slower timescale than “feedback for dynamics.” This “feedback for learning” is not modeled here but is fully consistent with our results, as discussed in a new paragraph of our Discussion (end of Section 3.4.1 “Pathways undergoing plasticity”).

2) There are some potential limitations of the conclusions drawn due to the model inference methods used. The methods used (fmincon) can easily get stuck in local minima and more importantly they do not provide an overview of the likelihood of parameters given the data. A few studies have now shown that it is important to apply more powerful inference techniques both to infer plasticity (Bykowska et al. Frontiers 2019) and neural dynamics (Gonçalves et al. eLife 2020). As highlighted by Costa et al. Frontiers 2013 using more standard fitting methods can lead to misleading interpretations. Given the large range of experimental data used to constrain the model, this may not be an issue, but it is not explicitly shown.

The reviewer correctly points out that we used a deterministic model-fitting procedure. To address this concern, we complemented the full dynamic model with a simple analytic model ( Figure 5 ) for which we could fully derive the cost function landscape and analytically show that there is a line of parameters corresponding to a perfect degeneracy in the model. Thus, the challenge in the model we analyze is that there are too many solutions, rather than it being difficult to find a solution. Given this degeneracy, we chose to fix the level of efference copy feedback and then find the (now non-degenerate) solutions, and to then compare these different solutions with regards to their implications for the correlated strengths and changes in strengths of different pathways. We have edited the relevant section of the Discussion for clarity on this topic, and have added references to the additional strategies for model inference mentioned above, in Section 3.3 “Relation to other sloppy models”.

3) There is some lack of clarity on how the feedback pathways as currently presented should be interpreted in the brain.

We interpret this comment as referring to the questions of (1) whether our model includes a pathway for learning through feedback, (2) what is the anatomical implementation of the efference copy feedback pathway and visual pathways, and (3) how should the positive weights on the efference copy feedback pathway k PE be interpreted. We address these below.

(1) Feedback for learning was discussed in point 1 above.

(2) Anatomical implementation of efference copy pathway: We have edited the Discussion to clarify that there is anatomical evidence for efference copy input to the cerebellum, but that a key aspect of ‘feedback’ is that activity functionally loops back onto itself. Instead, neurons carrying eye movement commands (such as in the vestibular nucleus) could send signals to the cerebellum, without receiving output from the same cerebellar neurons – this would correspond to a ‘spiraling’ pathway that does not form a closed feedback loop (Figure 8). Thus we argue that the existence of the gross anatomical pathways does not necessitate a role for strong, functional, efference copy feedback (Discussion, Section 3.1, lines 481-491).

Anatomical implementation of visual pathway: The visual feedback pathways considered here are those that would receive visual motion information from the environment. This visual feedback is itself changed by eye movements, thus providing a net overall negative feedback loop that helps to stabilize gaze. This pathway has been proposed to involve cortical regions such as MST (discussed in Materials and Methods, Model Implementation, lines 769-774).

(3) Interpretation of positive feedback loop: In our model, the efference copy feedback filter, k PE , has positive weight. This corresponds to the positive net sign of the Purkinje cell to brainstem to Purkinje cell feedback loop. Specifically, the Purkinje cell to brainstem pathway is inhibitory (because Purkinje cells are inhibitory), the brainstem to eye velocity command pathway is inhibitory (to achieve counter-rotation of the eyes in response to head turns), and the feedback of this eye velocity command back to Purkinje cells (k PE ) is positive. Thus this loop in our model represents positive feedback. This is now clarified in Materials and Methods, Model Implementation, lines 748.

4) The functional benefits of having (or not) feedback could be better discussed (related to point 1 above).

Related to point 1 above, it is certainly the case that feedback is necessary for learning. We do not explicitly model the climbing fiber feedback thought to be involved in learning/plasticity of the parallel fiber pathway.

We instead focus on the role of efference copy feedback, and how it functionally impacts the required sites and signs of plasticity in the circuit. As shown in the paper, if the efference copy pathway is strong, then this is most consistent with learned changes in eye movements being driven primarily by plasticity in the brainstem pathway (as in the Miles-Lisberger hypothesis), whereas if the efference copy pathway is weak, then this is most consistent with learned changes in eye movements being driven by net depression in the parallel fiber to Purkinje cell pathway (as in the classic Marr-Albus-Ito model and as suggested by most cellular and molecular studies of parallel fiber-Purkinje cell plasticity), in addition to a role of plasticity in the brainstem pathway. We also note that, in the ‘Strong Feedback’ model, the feedback is so strong that the system is on the brink of instability – this has been argued to have the functional benefit of providing ‘inertia’ to eye movements that could help to maintain eye movements during smooth pursuit when a target goes behind an occluder, but it also has the disadvantage of placing the system at a level of positive feedback near the brink of instability. We also note that the visual feedback pathway through the environment, emphasized in this work, serves as a negative feedback loop that reduces deviations between the eye and target velocity. We have extensively re-written the first section of the Discussion (Section 3.1), in order to more clearly lay out the implications of each model for circuit plasticity and feedback.

5) Some of the key conclusions of the work are not described in the abstract, namely that feedback is weak in the cerebellar system.

Thank you for raising this point, we have added this key conclusion to the end of the abstract: “Our results address a long-standing debate regarding cerebellum-dependent motor learning, suggesting a reconciliation in which error-driven plasticity of synaptic inputs to Purkinje cells is compatible with seemingly oppositely directed changes in Purkinje cell activity. More broadly, the results demonstrate how learning-related changes in neural activity can appear to contradict the sign of the underlying plasticity when either internal feedback or feedback through the environment is present.”

Claims:

The argument is well-built throughout the paper, but there are some potential caveats with the general interpretation (see weaknesses).

Impact:

This work has the potential to bring important messages on how best to interpret and infer the role of feedback in neural systems. For the field of the cerebellum, it also proposes solutions to long-standing problems.

Reviewer #3 (Public Review):

Summary: In this study, the authors attempt to determine what is the role (and strength) of feedback in a closed-loop (cerebellar) system.

Strengths:

1. By combining extensive data fitting of cerebellar experimental observations this study provides deep insights into existing questions and more broadly on the role of feedback and what are the limitations when inferring feedback in (plastic) neural circuits.

2. Another strength of this study is the gradual build-up of evidence by using models of different complexities to help build the argument that weak feedback is sufficient to explain experimental observations.

3. The paper is well-written and structured.

Weaknesses:

1. In principle feedback can (i) drive dynamics or/and (ii) drive learning directly. Throughout the paper, the authors refer to only the first case (i.e. dynamics). However, the role of feedback in learning is already implicitly assumed by the authors when jointly fitting the model before and after learning. Note that the general conclusion that feedback (in general) is weak may be to the first view (i.e. dynamics), but not the second. Given that a key conclusion of the paper is that no feedback is sufficient to explain the data, this suggests that feedback may instead be used for learning/plasticity.

2. There are some potential limitations of the conclusions drawn due to the model inference methods used. The methods used (fmincon) can easily get stuck in local minima and more importantly they do not provide an overview of the likelihood of parameters given the data. A few studies have now shown that it is important to apply more powerful inference techniques both to infer plasticity (Bykowska et al. Frontiers 2019) and neural dynamics (Gonçalves et al. eLife 2020). As highlighted by Costa et al. Frontiers 2013 using more standard fitting methods can lead to misleading interpretations. Given the large range of experimental data used to constrain the model, this may not be an issue, but it is not explicitly shown.

3. There is some lack of clarity on how the feedback pathways as currently presented should be interpreted in the brain.

4. The functional benefits of having (or not) feedback could be better discussed (related to point 1 above).

5. Some of the key conclusions of the work are not described in the abstract, namely that feedback is weak in the cerebellar system.

Claims:

The argument is well-built throughout the paper, but there are some potential caveats with the general interpretation (see weaknesses).

Impact:

This work has the potential to bring important messages on how best to interpret and infer the role of feedback in neural systems. For the field of the cerebellum, it also proposes solutions to long-standing problems.

eLife assessment

Payne et al. present a novel model that predicts the sites and directions of plasticity within the vestibular cerebellum to explain the basis for learned adjustments to reflexive eye movements in monkeys. The work is solid; the model is well constrained by prior biological observations and makes an important prediction about the level of feedback available to the cerebellar cortex post-learning. Overall, a number of exciting and testable experiments will likely be motivated by this study.

Reviewer #1 (Public Review):

Payne et al. have investigated the neural basis of VOR adaptation with the goal of constraining sites and mechanisms of plasticity supporting cerebellar learning. This has been an area of intense debate for decades; previous competing models have argued extensively about the sites of plasticity and the strength of eye velocity feedback/ efference copy signals to Purkinje cells has been central to the debate. This paper nicely explores the consequences of varying the strength of this feedback and in so doing, provides a potential explanation for why Purkinje cell responses during VOR cancellation could exhibit stronger responses following learning, despite net depression of the strength of their vestibular inputs. In that sense it provides some reconciliation of existing models. The work appears to be well done and the paper is well written. The manuscript could be improved and the significance of the work clarified and enhanced by contextualizing the work more appropriately within the existing literature in this area.

Reviewer #2 (Public Review):

Payne et al. use a computational approach to predict the sites and directions of plasticity within the vestibular cerebellum that explain an unresolved controversy regarding the basis of VOR learning. Specifically, the conclusion by Miles and Lisberger (1981) that vestibular inputs onto Purkinje cells (PCs) must potentiate, rather than depress (as in the Marr/Albus/Ito model), following gain-increase learning because when the VOR is cancelled, PC firing increases rather than decreases. Payne et al. provide a novel model solution that recapitulates the results of Miles and Lisberger but, paradoxically, uses plasticity in the cerebellar cortex that weakens PC output rather than strengthens it. However, the model only succeeds when efference copy feedback to the cerebellar cortex is relatively weak thereby allowing a second feedback pathway to drive PC activity during VOR cancellation to counteract the learned change in gain. Because the model is biologically constrained, the findings are well supported. This work will likely benefit the field by providing a number of potentially experimentally testable conclusions. The findings will be of interest to a wider audience if the results can be extrapolated to other cerebellar-dependent learning behaviors rather then just VOR gain-increase learning. Overall, the manuscript is very well written with clearly delineated results and conclusions.

Author Response

Reviewer #1 (Public Review):

Summary:

Cyclic Nucleotide Binding (CNB) domains are pervasive structural components involved in signaling pathways across eukaryotes and prokaryotes. Despite their similar structures, CNB domains exhibit distinct ligand-sensing capabilities. The manuscript offers a thorough and convincing investigation that clarifies numerous puzzling aspects of nucleotide binding in Trypanosoma.

Strengths:

One of the strengths of this study is its multifaceted methodology, which includes a range of techniques including crystallography, ITC (Isothermal Titration Calorimetry), fluorimetry, CD (Circular Dichroism) spectroscopy, mass spectrometry, and computational analysis. This interdisciplinary approach not only enhances the depth of the investigation but also offers a robust cross-validation of the results.

Weaknesses:

None noticed.

Reviewer #2 (Public Review):

Summary:

This manuscript clearly shows that Trypanosoma PKA is controlled by nucleoside analogues rather than cyclic nucleotides, which are the primary allosteric effectors of human PKA and PKG. The authors demonstrate that the inosine, guanosine, and adenosine nucleosides bind with high affinity and activate PKA in the tropical pathogens T. brucei, T. cruzi and Leishmania. The underlying determinants of nucleoside binding and selectivity are dissected by solving the crystal structure of T. cruzi PKAR(200-503) and T. brucei PKAR(199-499) bound to inosine at 1.4 Å and 2.1 Å resolution and through comparative mutational analyses. Of particular interest is the identification of a minimal subset of 2-3 residues that controls nucleoside vs. cyclic nucleotide specificity.

Strengths:

The significance of this study lies not only in the structure-activity relationships revealed for important targets in several parasite pathogens but also in the understanding of CNB's evolutionary role.

Weaknesses:

The main missing piece is the model for activation of the kinetoplastid PKA which remains speculative in the absence of a structure for the trypanosomatid PKA holoenzyme complex. However, this appears to be beyond the scope of this manuscript, which is already quite dense.

We fully agree that insight into the activation mechanism and its possible deviation from the mammalian paradigm requires a holoenzyme structure revealing the details of R-C interaction. We have attempted Cryo-EM from LEXSY-produced holoenzyme, yet upscaling the purification procedures described in this manuscript have repeatedly failed in spite of numerous protocol changes and optimizations. Much more work is required to achieve this.

Reviewer #2 (Recommendations For The Authors):

Some minor points to consider for enhancing the impact of this interesting manuscript:

1) The nucleoside affinities measured are mainly for the regulatory subunits unbound to the kinase domain. How would nucleoside affinities change when the regulatory subunits are bound to the kinase domain, which is presumably the case under resting conditions? An estimation of this change in affinity is important because it more closely relates to the variations in cellular nucleoside concentrations needed for activation.

This is an important question and we have given an indirect answer in the manuscript, but not very explicit. The EC50 values for kinase activation of the purified holoenzyme complexes are very similar or almost identical to the kD values measured by ITC with free regulatory subunits. By inference, the binding kD for the holoenzyme and for the free R-subunit cannot be very different. In addition, we have recently determined the EC50 for PKA activation in vivo in trypanosomes using a bioluminescence complementation reporter assay. The values fit perfectly to the values obtained with purified holoenzyme (Wu et al. in preparation). A sentence in Results (lines 201-203) has been added.

2) The authors should point out that a major implication of nucleoside vs. cyclic nucleotide activation is in terms of signal termination. If phosphodiesterases (PDEs) are responsible for cAMP/cGMP signal termination, what terminates nucleoside-dependent signaling? Although the answer to this question may not be known at this stage, it is important to highlight this critical implication of the authors' study.

The mechanism of signal termination is indeed unknown so far. We speculate that some enzymes of the purine salvage pathways are differentially localized in subcellular compartments and thereby able to establish microdomains that enable nucleoside signaling. In addition, PKA subunit phosphorylations/dephosphorylations and/or protein turnover may also regulate signal termination. As an example, free PKAC1 is rapidly degraded upon depletion of the PKAR subunit by RNAi. We have now mentioned signal termination in Discussion and have revised the last part of Discussion (lines 567-602). A possible approach to monitor compartmentalized signaling would be using the FluoSTEPs technology (Tenner et al., Sci. Adv. 2021; 7: eabe4091), but adapting this to the trypanosome system will not be a short-term task.

Author Response

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

Assessment note: “Whereas the results and interpretations are generally solid, the mechanistic aspect of the work and conclusions put forth rely heavily on in vitro studies performed in cultured L6 myocytes, which are highly glycolytic and generally not viewed as a good model for studying muscle metabolism and insulin action.”

While we acknowledge that in vitro models may not fully recapitulate the complexity of in vivo systems, we believe L6 myotubes are appropriate for studying the mechanisms underlying muscle metabolism and insulin action. L6 myotubes possess many important characteristics relevant to our research, including high insulin sensitivity and a similar mitochondrial respiration sensitivity compared to primary muscle fibres. Furthermore, several studies have demonstrated the utility of L6 myotubes as a model for studying insulin sensitivity and metabolism, including our own previous work (PMID: 19805130, 31693893, 19915010) and work of others (PMID:12086937, 29486284, 15193147).

Importantly, our observations from the L6 myotube model are supported by in vivo data from both mice and humans. Chow (Figure 3J, K) and high-fat fed mice (new data - Supplementary Figure 4 H-I) demonstrated a reduction in mitochondrial Ceramide and an increase in CoQ9. Muscle biopsies from humans showed a strong negative correlation between mitochondrial C18:0 ceramide levels and insulin sensitivity (PMID: 29415895). Further, complex I and IV abundance was strongly correlated with both muscle insulin sensitivity and mitochondrial ceramide (CerC18:0) (Figure 6E, F). This is consistent with our observations in L6 myotubes (Figure 6H, I). These findings support the relevance of our in vitro results to in vivo muscle metabolism.

Points from reviewer 1

1. Although the authors' results suggest that higher mitochondrial ceramide levels suppress cellular insulin sensitivity, they rely solely on a partial inhibition (i.e., 30%) of insulin-stimulated GLUT4-HA translocation in L6 myocytes. It would be critical to examine how much the increased mitochondrial ceramide would inhibit insulin-induced glucose uptake in myocytes using radiolabeled deoxy-glucose. Another important question to be addressed is whether glycogen synthesis is affected in myocytes under these experimental conditions. Results demonstrating reductions in insulin-stimulated glucose transport and glycogen synthesis in myocytes with dysfunctional mitochondria due to ceramide accumulation would further support the authors' claim.

Response: We have now conducted additional experiments focusing on glycogen synthesis as a readout of insulin sensitivity, as it offers an orthogonal method for assessing GLUT4 translocation and glucose uptake. L6-myotubes overexpressing the mitochondrial-targeted ASAH1 construct (as described in Fig. 3) were challenged with palmitate and insulin stimulated glycogen synthesis was measured using 14C radiolabeled glucose. As shown below, palmitate suppressed insulin-induced glycogen synthesis, which was effectively prevented by overexpression of ASAH1 (N = 5, * p<0.05) supporting our previous observation using GLUT4 translocation as a readout of insulin sensitivity (Fig. 3). These results provide additional evidence highlighting the role of dysfunctional mitochondria in muscle cell glucose metabolism.

These data have now been added to Supplementary Figure 4K and the results modified as follows:

“...For this reason, several in vitro models have been employed involving incubation of insulin sensitive cell types with lipids such as palmitate to mimic lipotoxicity in vivo. In this study we have used cell surface GLUT4-HA abundance as the main readout of insulin response...”

“Notably, mtASAH1 overexpression protected cells from palmitate-induced insulin resistance without affecting basal insulin sensitivity (Fig. 3E). Similar results were observed using insulin-induced glycogen synthesis as an orthologous technique for Glut4 translocation. These results provide additional evidence highlighting the role of dysfunctional mitochondria in muscle cell glucose metabolism (Sup. Fig. 5K). Importantly, mtASAH1 overexpression did not rescue insulin sensitivity in cells depleted…”

Author response image 1.

Additionally, the following text was added to the method section:

“L6 myotubes overexpressing ASAH were grown and differentiated in 12-well plates, as described in the Cell lines section, and stimulated for 16 h with palmitate-BSA or EtOH-BSA, as detailed in the Induction of insulin resistance section.

On day seven of differentiation, myotubes were serum starved in DMEM for 3.5 h. After incubation for 1 h at 37 °C with 2 µCi/ml D-[U-14C]-glucose in the presence or absence of 100 nM insulin, glycogen synthesis assay was performed, as previously described (Zarini S. et al., J Lipid Res, 63(10): 100270, 2022).”

1. In addition, it would be critical to assess whether the increased mitochondrial ceramide and consequent lowering of energy levels affect all exocytic pathways in L6 myoblasts or just the GLUT4 trafficking. Is the secretory pathway also disrupted under these conditions?

Response: This is an interesting point raised by the reviewer that is aimed at the next phase of this work, to identify how ceramide induced mitochondrial dysfunction drives insulin resistance. Looking at energy deficiency in more detail as well as general trafficking is part of ongoing work, but given the complexity of this question, it is beyond the scope of the current study.

Points from reviewer 2

1. The mechanistic aspect of the work and conclusions put forth rely heavily on studies performed in cultured myocytes, which are highly glycolytic and generally viewed as a poor model for studying muscle metabolism and insulin action. Nonetheless, the findings provide a strong rationale for moving this line of investigation into mouse gain/loss of function models.

Response: We acknowledge that in vitro models may not fully mimic in vivo complexity as described above in the response to the “Assessment note”. We have now added to the Discussion:

“In this study, we mainly utilised L6-myotubes, which share many important characteristics with primary muscle fibres. Both types of cells exhibit high sensitivity to insulin and respond similarly to maximal doses of insulin, with GLUT4 translocation stimulated between 2 to 4 times over basal levels in response to 100 nM insulin (as shown in Fig. 1-4 and (46,47)). Additionally, mitochondrial respiration in L6-myotubes has a similar sensitivity to mitochondrial poisons, as observed in primary muscle fibres (as shown in Fig. 5 (48)). Finally, inhibiting ceramide production increases CoQ levels in both L6-myotubes and adult muscle tissue (as shown in Fig. 2-3). Therefore, L6-myotubes possess the necessary metabolic features to investigate the role of mitochondria in insulin resistance, and this relationship is likely applicable to primary muscle fibres”.

1. One caveat of the approach taken is that exposure of cells to palmitate alone is not reflective of in vivo physiology. It would be interesting to know if similar effects on CoQ are observed when cells are exposed to a more physiological mixture of fatty acids that includes a high ratio of palmitate, but better mimics in vivo nutrition.

Response: We appreciate the reviewer's comment. Previously, we reported that mitochondrial CoQ depletion occurs in skeletal muscle after 14 and 42 days of HFHSD feeding, coinciding with the onset of insulin resistance (PMID: 29402381, see figure below).

Author response image 2.

These data demonstrated that our in vitro model recapitulates the loss of CoQ in insulin resistance observed in muscle tissue in response to a more physiological mixture of fatty acids. Further, it has been reported that different fatty acids can induce insulin resistance via different mechanisms (PMID:20609972), which would complicate interpretation of the data. Saturated fatty acids such as palmitate increase ceramides in cell-lines and humans, but unsaturated FAs generally do not (PMID: 10446195,14592453,34704121). As such we conclude that palmitate is a cleaner model for studying the effects of ceramide on skeletal muscle function.

We have added to discussion:

“…These findings align with our earlier observations demonstrating that mice exposed to HFHSD exhibit mitochondrial CoQ depletion in skeletal muscle (Fazakerley et al. 2018).”

1. While the utility of targeting SMPD5 to the mitochondria is appreciated, the results in Figure 5 suggest that this manoeuvre caused a rather severe form of mitochondrial dysfunction. This could be more representative of toxicity rather than pathophysiology. It would be helpful to know if these same effects are observed with other manipulations that lower CoQ to a similar degree. If not, the discrepancies should be discussed.

Response: As the reviewer suggests many of these lipids can cause cell death (toxicity) if the dose is too high. We have previously found that low levels (0.15 mM) of palmitate were sufficient to trigger insulin resistance without any signs of toxicity (Hoehn, K, PNAS, 19805130). Using a similar approach, we show that mitochondrial membrane potential is maintained in SMPD5 overexpressing cells (Sup. Fig. 2J - and Author response image 2). Given that toxicity is associated with a loss of mitochondrial membrane potential (eg., 50uM Saclac; RH panel), these data suggest SMPD5 overexpression is not causing overt toxicity.

Author response image 3.

Furthermore, we conducted an overrepresentation analysis of molecular processes within our proteomic data from SMPD5-overexpressing cells. As depicted below, no signs of cell toxicity were observed in our model at the protein level. This data is now available in supplementary table 1.

Author response table 1.

Our results are therefore consistent with a pathological condition induced by elevated levels of ceramides independently of cellular toxicity. The following text has been added to the discussion:“...downregulation of the respirasome induced by ceramides may lead to CoQ depletion.

Despite the significant impact of ceramide on mitochondrial respiration, we did not observe any indications of cell damage in any of the treatments, suggesting that our models are not explained by toxicity and increased cell death (Sup. Fig. 2H & J).”

1. The conclusions could be strengthened by more extensive studies in mice to assess the interplay between mitochondrial ceramides, CoQ depletion and ETC/mitochondrial dysfunction in the context of a standard diet versus HF diet-induced insulin resistance. Does P053 affect mitochondrial ceramide, ETC protein abundance, mitochondrial function, and muscle insulin sensitivity in the predicted directions?

Response: We agree with the referee about the importance of performing in vivo studies to corroborate our in vitro data. We have now conducted extensive new studies in mice skeletal muscle using targeted metabolomic and lipidomic analyses to investigate the impact of ceramide depletion in CoQ levels in HF-fed mice. Mice were exposed to a HF-fed diet with or without the administration of P053 (selective inhibitor of CerS1) for 5 weeks. As illustrated in the figures below, the administration of P053 led to a reduction in ceramide levels (left panel), increase in CoQ levels (middle panel) and a negative correlation between these molecules (right panel), which is consistent with our in vitro findings.

Author response image 4.

Additional suggestions:

1. Figure 1: How does increased mitochondrial ceramide affect fatty acid oxidation (FAO) in L6-myocytes? As the accumulation of mitochondrial ceramide inhibits respirasome and mitochondrial activity in vitro, can reduce FAO in vivo, due to high mitochondrial ceramide, accounts for ectopic lipid deposition in skeletal muscle of obese subjects?

Response: We appreciate the reviewer for bringing up this intriguing point. We would like to emphasise that Complex II activity is vital for fatty acid oxidation. As shown in Fig. 5H, our results indicate that specifically Complex II mediated respiration was diminished in cells with SMPD5 overexpression, suggesting that ceramides hinder the mitochondria's capability to oxidise lipids. We agree that this mechanism may potentially play a role in the ectopic lipid accumulation seen in individuals with obesity.

We have added the following text to discussion:

“...the mitochondria to switch between different energy substrates depending on fuel availability, named “metabolic Inflexibility”...this mechanism may potentially play a role in the ectopic lipid accumulation seen in individuals with obesity, a condition linked with cardio-metabolic disease.”

1. Figure 2: Although the authors show that mtSMPD5 overexpression does not affect ceramide abundance in whole cell lysate, it would be critical to examine the abundance of this lipid in other cellular membranes and organelles, particularly plasma membrane. What is the effect of mtSMPD5 overexpression on plasma membrane lipids composition? Does that affect GLUT4-containing vesicles fusion into the plasma membrane, possibly due to depletion of v-SNARE or tSNARE?

Response: While we acknowledge the importance of this point we strongly feel that measuring lipids in purified membranes has its limitations because it is impossible to purify specific membranes without contamination from other kinds of membranes. For example, we have done proteomics on purified plasma membranes from different cell types and we always observe considerable mitochondrial contamination with these membranes (e.g. PMID 21928809). This was the main factor that led us to use the mitochondrial targeting approach.

Nevertheless we do acknowledge that there is a possibility that ceramides that are produced in the mitochondria in SMPD5 cells could leak out of mitochondria into other membranes and this could influence other aspects of GLUT4 trafficking and insulin action. However, we believe that the studies using mito targeted ASAH mitigate against this problem. Thus, we have now included a statement in the revised manuscript as follows: “It is also possible that ceramides generated within mitochondria in SMPD5 cells leak out from the mitochondria into other membranes (e.g. PM and Glut4 vesicles) affecting other aspects of Glut4 trafficking and insulin action. However, the observation that ASAH1 overexpression reversed IR without affecting whole cell ceramides argues against this possibility.”.

1. Figure 4: One critical piece of information missing is the effect (if any) of mitochondrial ceramide accumulation on the mRNAs encoding the ETC components affected by this lipid. Although the ETC protein's lower stability may account for the effect of increased ceramide, transcriptional inhibition can't be ruled out without checking the mRNA expression levels for these ETC components.

Response: To address this point, we have quantified the mRNA abundance of nine complex I subunits that exhibit downregulation in our proteomic dataset subsequent to mtSMPD5 overexpression (as depicted in Figure 4G).

Induction of mtSMPD5 expression with doxycycline (below - Left hand panel) had no effect on the mRNA levels of the Complex I subunits (below - right hand panel).. This is consistent with our initial hypothesis that the reduction in electron transport chain (ETC) components, caused by heightened ceramide levels, primarily arises from alterations in protein stability rather than gene expression. While we acknowledge the possibility that certain subunits might be regulated at the transcriptional level, the absence of mRNA downregulation across our data strongly suggests that, at the very least, a portion of the observed protein depletion is attributed to diminished protein stability. We have incorporated this dataset into Supplementary Figure 6J and added the following text to the results:

Author response image 5.

“Importantly, CI downregulation was not associated with reduction in gene expression as shown in Sup. Fig. 6J.”

Additionally, we have added the following text to discussion:

“In addition, the absence of mRNA downregulation in mtSMPD5 overexpressing cells strongly suggests that at least a portion of the observed protein depletion within CI is attributed to diminished protein stability.”

1. Figure 3: The authors state that neither palmitate nor mtASAH1 overexpression affected insulin-dependent Akt phosphorylation. However, the results in Figure 3F-G do not support this conclusion, as the overexpression of mtASAH1 does enhance the insulin-stimulated AKT (thr-308) phosphorylation. They need to clarify this issue.

Response: We have now analysed these data in a manner that preserves the control variance, consistent with the other figures in the manuscript and there is no significant change in Akt phosphorylation in ASAH over-expressing cells.

Author response image 6.

1. Figure S2: A functional assessment of mitochondrial function in HeLa cells would be helpful to validate the small effect of Saclac treatment on CI NDUFB8.

Response: Mitochondrial respiration was conducted in cells treated with Saclac (2 µM and 10 µM) for 24 hours. As shown below, in Hela cells, we did not detect any mitochondrial respiratory impairments at low doses, but only at high doses of Saclac. This suggests that the minor effect of Saclac on CI NDUFB8 is insufficient to alter mitochondrial function.

Author response image 7.

Reviewer #2 (Recommendations For The Authors):

Additional questions and comments for consideration:

1. The working model links ceramide-induced CoQ depletion to a reduction in ETC proteins and accompanying deficits in OxPhos capacity. The idea that mitochondrial dysfunction necessarily precedes and causes insulin resistance has been heavily debated for years because many animal and human studies have found no overt changes in ETC proteins and/or mitochondrial respiratory capacity during the early phases of insulin resistance. How do the investigators reconcile their work in the context of this controversy?

Response: We acknowledge this controversy in our revised manuscript more clearly now as follows on page 21: “We present evidence that mitochondrial dysfunction precedes insulin resistance. However, previous studies have failed to observe changes in mitochondrial morphology, respiration or ETC components during early stages of insulin resistance (72). However, in many cases such studies fail to document changes in insulin-dependent glucose metabolism in the same tissue as was used for assessment of mitochondrial function. This is crucial because we and others do not observe impaired insulin action in all muscles from high fat fed mice for example. In addition, surrogate measures such as insulin-stimulated Akt phosphorylation may not accurately reflect tissue specific insulin action as demonstrated in figure 1C. Thus, further work is required to clarify some of these inconsistencies''.

1. While the utility of targeting SMPD5 to the mitochondria is appreciated, the results in Figure 5 suggest that this manoeuvre caused a rather severe form of mitochondrial dysfunction. Is this representative of pathophysiology or toxicity?

Response: We believe we have addressed this in point 3 above (Principal comments, reviewer 1, point 3)

1. How did this affect other mitochondrial lipids (e.g. cardiolipin)?

Response: As shown in the supplementary figure 3, SMPD5 overexpression did not affect other lipids species such as cardiolipin (D-J). We have added to results:

“Importantly, mtSMPD5 overexpression did not affect ceramide abundance in the whole cell lysate nor other lipid species inside mitochondria such as cardiolipin, cholesterol and DAGs (Sup. Fig. 3 A, D-J)”

1. Are these severe effects rescued by CoQ supplementation?

Response: We have performed additional experiments to address this point. As shown below, mitochondrial ceramide accumulation induced by palmitate was not reversed by CoQ supplementation, as demonstrated in Figure 1F. We have added to results:

“Addition of CoQ9 had no effect on control cells but overcame insulin resistance in palmitate treated cells (Fig. 1A). Notably, the protective effect of CoQ9 appears to be downstream of ceramide accumulation, as it had no impact on palmitate-induced ceramide accumulation (Fig. 1E-F). Strikingly, both myriocin and CoQ9…”

Additionally, we assessed mitochondrial respiration by using SeaHorse in cells with SMPD5 overexpression treated with or without CoQ supplementation. Our results, depicted below, indicate that CoQ supplementation reversed the ceramide-induced decrease in basal and ATP linked mitochondrial respiration. We have modified Fig.5.

Author response image 8.

We have added to results:

“Respiration was assessed in intact mtSMPD5-L6 myotubes treated with CoQ9 by Seahorse extracellular flux analysis. mtSMPD5 overexpression decreased basal and ATP-linked mitochondrial respiration (Fig. 5 A, B &C), as well as maximal, proton-leak and non-mitochondrial respiration (Fig. 5 A, D, E & F) suggesting that mitochondrial ceramides induce a generalised attenuation in mitochondrial function. Interestingly, CoQ9 supplementation partially recovered basal and ATP-linked mitochondrial respiration, suggesting that part of the mitochondrial defects are induced by CoQ9 depletion. The attenuation in mitochondrial respiration is consistent with a depletion of the ETC subunits observed in our proteomic dataset (Fig. 4)...”

1. Are these same effects observed with other manipulations that lower CoQ to a similar degree?

Response: As mentioned in point 5 (additional suggestions from Reviewer 1), we conducted mitochondrial respiration measurements on HeLa cells treated with Saclac (2 µM and 10 µM) for 24 hours. Our findings showed no signs of mitochondrial respiratory impairments at low doses of Saclac in HeLa cells, despite observing CoQ depletion at this dose (Fig. Sup. 2C). We believe that this variation could be due to the varying sensitivity of mitochondrial respiration/ETC abundance to ceramide-induced CoQ depletion in different cell lines. Alternatively, it is possible that reduced mitochondrial respiration is a secondary event to other mitochondrial/cellular defects such as mitochondrial fragmentation or deficient nutrient transport inside mitochondria.

*Author response image 9.

1. The mitochondrial concentrations of CoQ required to maintain insulin sensitivity in L6 myocytes seem to vary from experiment to experiment. Is it the absolute concentration that matters and/or the change relative to a baseline condition?

Response: This is an excellent observation. The findings indicate that the absolute concentration of CoQ is the determining factor for insulin sensitivity, rather than the relative depletion of CoQ compared to basal conditions. We have added to discussion: “Finally, mtASAH1 overexpression increased CoQ levels. In both control and mtASAH1 cells, palmitate induced a depletion of CoQ, however the levels in palmitate treated mtASAH1 cells remained similar to control untreated cells (Fig. 3I). This suggests that the absolute concentration of CoQ is crucial for insulin sensitivity, rather than the relative depletion compared to basal conditions, thus supporting the causal role of mitochondrial ceramide accumulation in reducing CoQ levels in insulin resistance”

1. Considering that CoQ has been shown to have antioxidant properties, does the rescue observed after a 16 h treatment require the prolonged exposure, or alternatively, are similar effects observed during short-term exposures (~1-2 h), which might imply a different or additional mechanism.

Response: This is an excellent point that we have long considered. The problem is how to address the question in a way that will be definitive and we are concerned that the experiment suggested by the referee will not generate definitive data. A major issue is that CoQ has low solubility and needs to reach the right compartment. As such if short term treatment (as suggested) does not rescue, it would be difficult to make any definite conclusions as this might just be because insufficient CoQ is delivered to mitochondria. Conversely, if short term treatment does rescue this could be either because CoQ does get into mitochondria and regulate ETC or because of its general antioxidant function. So, even if we observe a rescue after 1 hour of incubation with CoQ, it will not clarify whether this is due to the antioxidant effect or simply because 1 hour is adequate to boost mitoCoQ levels. Thus, in our view this experiment might not get us any closer to the answer. Nevertheless, we do feel this is an important point and we have added the following statement to our revised manuscript to acknowledge this: “Because CoQ can accumulate in various intracellular compartments, it's important to consider that its impact on insulin resistance might be due to its overall antioxidant properties rather than being limited to a mitochondrial effect”

1. In Figure 1, CoQ depletion due to 4NB treatment resulted in increased ceramide levels. Could this be due to impaired palmitate oxidation leading to rerouting of intracellular palmitate to the ceramide pathway? This could be tested using stable isotope tracers.

Response: We have added the statement below to the manuscript to address this point. We feel that while an interesting experiment to perform it is somewhat outside of the major focus of this study.

“One possibility is that CoQ directly controls ceramide turnover (35). An alternate possibility is that CoQ inside mitochondria is necessary for fatty acid oxidation (12) and CoQ depletion triggers lipid overload in the cytoplasm promoting ceramide production (36). Future studies are required to determine how CoQ depletion promotes Cer accumulation. Regardless, these data indicate that ceramide and CoQ have a central role in regulating cellular insulin sensitivity.”

1. To a similar point, it would be helpful to know if the C2 ceramide analog is sufficient to cause elevated mito-ceramide and/or CoQ depletion. If not, the results might imply mitochondrial uptake of palmitate is required.

Response: We feel this point is analogous to Point 7 above in that this experiment is not definitive enough to make any clear conclusions as it may or may not work for many different reasons. For example, C2 ceramide may not work simply because it has the wrong chain length.

Moreover, it is clear that C2 ceramide has effects that clearly differ from those observed with palmitate most notably the inhibitory effect on Akt signalling. For these reasons we do not agree with the logic of this experiment.

We have mentioned in the results section:

“Based on these data we surmise that C2-ceramide does not faithfully recapitulate physiological insulin resistance, in contrast to that seen with incubation with palmitate”.

1. Likewise, does inhibition of CPT1 ameliorate or exacerbate palmitate-induced insulin resistance?

Response: This experiment has been performed by a number of different labs. For instance, muscle specific CPT1 overexpression is protective against high fat diet induced insulin resistance in mice (Bruce C, PMID19073774), CPT1 overexpression protects L6E9 muscle cells from fatty acid-induced insulin resistance (Sebastian D, PMID17062841) and increased beta-oxidation in muscle cells enhances insulin stimulated glucose metabolism and is protective against lipid induced insulin resistance (Perdomo G, PMID15105415). We have now cited all of these studies in our revised manuscript in the discussion: “In fact, increased fatty acid oxidation is protective against insulin resistance in several model organisms (37–39)”

1. Does the addition of palmitate to the cells treated with mtSMPD5 further reduce CoQ9 (Figure 2I and 2J)?

Response: This intriguing observation, as highlighted by the referee, has prompted us to conduct additional experiments to investigate the effects of palmitate and SMPD5 overexpression on Coenzyme Q (CoQ) levels in L6 myotubes. As demonstrated in the figures presented below, both palmitate and SMPD5 overexpression independently resulted in the depletion of CoQ9, with no observed additive effects suggesting that they shared a common pathway driving CoQ9 deficiency. One plausible hypothesis is that ceramides may trigger the depletion of a specific CoQ9 pool localised within the inner mitochondrial membrane, likely the pool associated with Complex I (CI) in the Electron Transport Chain (ETC). This hypothesis is supported by previous studies indicating that approximately ~25 - 35 % of CoQ binds to CI (PMID: 33722627) and our data demonstrating that ceramide induces a selective depletion of CI in L6 myotubes (Fig. 4).

We have added this result to Fig. 2I in the main section.

Author response image 10.

We have added to the result section:

“Mitochondrial CoQ levels were depleted in both palmitate-treated and mtSMPD5-overexpressing cells without any additive effects. This suggests that these strategies to increase ceramides share a common mechanism for inducing CoQ depletion in L6 myotubes (Fig. 2I).”

We have added to the discussion section:

“...These are known to form supercomplexes or respirasomes where ~25 - 35 % of CoQ is localised in mammals (58,16).…The observation that both palmitate and SMPD5 overexpression trigger CoQ depletion without additive effects support the notion that ceramides may trigger the depletion of a specific CoQ9 pool localised within the inner mitochondrial membrane.”

1. Some of the cell-based experiments appear to be underpowered and therefore confidence in the interpretations might benefit from additional repeats. For example, in Figure 3i, it appears that palmitate still causes a substantial reduction of CoQ in the cells treated with mtASAH1, even though mito-ceramide levels are restored to baseline. Please specify if these and other results are representative of multiple cell culture experiments or a single experiment.

Response: All data were derived from a minimum of 3-4 independent experiments from at least two separate cultures of L6 cells. Separate batches of drug treatments were prepared for each experiment. We have previously compared metabolic parameters between batches of cells differentiated at different times (i.e. at least weeks apart) in a previous study (Krycer PMID 31744882) and found variations of <20% for insulin-stimulated glucose oxidation. With an expected variance of 20% and a type I error rate of 0.05, this is sufficient to detect a 40% difference with a power of 0.8. As the reviewer has indicated this is likely underpowered in situations where variance is unexpectedly high or if a small difference needs to be detected.

In terms of Fig3, the reviewer raises an interesting point. As discussed in point 6, the fact that palmitate still appears to cause a depletion of CoQ in mtASAH1 cells likely indicates that the absolute concentration of CoQ is the determining factor for insulin sensitivity, rather than the relative depletion of CoQ compared to basal conditions. We have added to the discussion:

“Finally, mtASAH1 overexpression increased CoQ levels. In both control and mtASAH1 cells, palmitate induced a depletion of CoQ, but this effect was less pronounced in the mtASAH1 cell line (Fig. 3I). Our results suggest that the absolute concentration of CoQ is crucial for insulin sensitivity, rather than the relative depletion compared to basal conditions, thus supporting the causal role of mitochondrial ceramide accumulation in reducing CoQ levels in insulin resistance”

1. The color scheme of 2E is inconsistent with other panels in the figure.

Response: Corrected

1. It would be helpful if the axis labels for CoQ graphs were labeled as "Mito-CoQ" for clarity.

Response: Corrected

eLife assessment

This important work combines DNA contact analysis and controlled genome rearrangements to investigate the processes that organize the E. coli chromosome, with a particular focus on how the SMC-related complex MukBEF is regulated. The evidence supporting the conclusions is compelling, with time-resolved experiments and analysis of mutant strains. The work will be of broad interest to chromosome biologists and bacterial cell biologists.

Author Response

The following is the authors’ response to the previous reviews

We appreciate the positive comments from the editors and reviewers. The followings are the point to point responses to the questions and comments of the Reviewers:

Reviewer #1 (Public Review):

In this study, Jiamin Lin et al. investigated the potential positive feedback loop between ZEB2 and ACSL4, which regulates lipid metabolism and breast cancer metastasis. They reported a correlation between high expression of ZEB2 and ACSL4 and poor survival of breast cancer patients, and showed that depletion of ZEB2 or ACSL4 significantly reduced lipid droplets abundance and cell migration in vitro. The authors also claimed that ZEB2 activated ACSL4 expression by directly binding to its promoter, while ACSL4 in turn stabilized ZEB2 by blocking its ubiquitination. While the topic is interesting, there are several concerns with the study:

1. My concern regarding the absence of appropriate thresholds or False Discovery Rate (FDR) adjustments for the RNA-seq analysis has not been addressed, leading to incorrect thresholds and erroneous identification of significant signals.

Response: We thank the reviewer for the concern about the RNA-seq analysis. RNA-seq data was analyzed by the Benjamini and Hochberg’s approach for controlling the false discovery rate. The procedure of RNA-seq bioinformatic analysis is as follows: For data analysis, raw data of fastq format were firstly processed through in-house perl scripts. In this step, clean data were obtained by removing reads containing adapter, reads containing N base and low quality reads from raw data. All the downstream analyses were based on the clean data with high quality. Index of the reference genome was built using Hisat2 v2.0.5 and paired-end clean reads were aligned to the reference genome using Hisat2 v2.0.5. FeatureCounts v1.5.0-p3 was used to count the reads numbers mapped to each gene, and then FPKM of each gene was calculated based on the length of the gene and reads count mapped to this gene. Differential expression analysis of two conditions/groups was performed using the DESeq2 R package (1.20.0). The resulting P-values were adjusted using the Benjamini and Hochberg’s approach for controlling the false discovery rate. Genes with an adjusted P-value (<0.05) found by DESeq2 were assigned as differentially expressed.

1. In Figure 3B and C, it appears that the knockdown efficiency of ACSL4 is inadequate in these cells, which contradicts the Western blot results presented in Figure 2F.

Response: We thank the reviewer for the concern. In figure 3B and 3C, we use the shRNA for the knockdown experiment and in Figure 2F we use siRNA for the knockdown experiment, so the efficiency of them were different.

1. Regarding Figure 6, the discovery of consensus binding sequences (CACCT) for ZEB2 alone is insufficient evidence to support the direct binding of ZEB2 to the ACSL4 promoter.

Response: We thank the reviewer for the concern. We performed chromatin immunoprecipitation (ChIP), which examines the direct interaction between DNA and protein, to test if ZEB2 directly binds to the ACSL4 promoter. The results showed that the primer set 1, which covered -184 to -295 of ACSL4 promoter region exhibited apparent ZEB2 binding (Fig. 6F). Moreover, the mutant sequence (AAAA) of ACSL4 promoter showed significant decreased luciferase activity (Fig. 7H). All these evidences suggest that ZEB2 directly bond to the consensus sequence of ACSL4 promoter.

1. For Figure 7E, there are multiple bands present, and it appears that ZEB2-HA has been cropped, which should ideally be presented with unaltered raw data. Please provide the uncropped raw data.

Response: We thank the reviewer for the concern. The raw data of the figure 7E ZEB2-HA is shown in Author response image 1:

Author response image 1.

1. In Figure 7C, the author claimed to have used 293T cells for the ubiquitin assay, which are not breast cancer cells. Moreover, the efficiency of over-expression differs between ZEB2 and ACSL4 in 293T cell lines. Performing the experiment in an unrelated cell line to justify an important interaction is not acceptable.

Response: We thank the reviewer for the concern. We also performed the ubiquitination assay in MDA-MB-231 cells in Fig 7D (Author response image 2), The results confirm that knockdown of ACSL4 obviously enhanced the ubiqutination of ZEB2. We also have performed the IP experiment in MDA-MB-231 cells in Author response image 3 (Fig 7F). The results confirmed the interaction between ZEB2 and ACSL4:

Author response image 2.

Author response image 3.

Reviewer #2 (Public Review):

In this study, the authors validated a positive feedback loop between ZEB2 and ACSL4 in breast cancer, which regulates lipid metabolism to promote metastasis.

Overall, the study is original, well structured, and easy to read.

We appreciate the positive comments from the reviewer.

Reviewer #3 (Public Review):

The manuscript by Lin et al. reveals a novel positive regulatory loop between ZEB2 and ACSL4, which promotes lipid droplets storage to meet the energy needs of breast cancer metastasis.

We appreciate the positive comments from the reviewer.

Reviewer #2 (Recommendations For The Authors):

I still have some points that should be addressed by the Authors:

The interaction between ACSL4 and ZEB2 is still not convincing, due to the cellular localization of ACSL4 and ZEB2 is different. The authors should consider utilizing the Duolink experiment to more accurately determine the interaction location of these two proteins in cells.

Response: We appreciate the reviewer’s suggestion. We performed GST pull-down assay to examine whether ZEB2 and ACSL4 form a complex. GST pull-down assay confirmed the interaction of ZEB2 and ACSL4 (Supplementary Fig. S10). We also performed immunofluorescence assay and found that ZEB2 was co-localized with ACSL4 in some certain regions of the cytoplasm in Author response image 5 (Supplementary Fig. S11):

Author response image 4.

Author response image 5.

In Figure S4, the authors showed both "shACSL4" and "siACSL4", which is a description error.

Response: We appreciate the reviewer to point out the mistake. We have corrected the "siACSL4" into "shACSL4".

Author response image 6.

Reviewer #3 (Recommendations For The Authors):

The manuscript is improved.

We appreciate the positive comments from the reviewer.

Author Response

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

Reviewer #1 (Public Review):

In this study, the authors attempt to describe alterations in gene expression, protein expression, and protein phosphorylation as a consequence of chronic adenylyl cyclase 8 overexpression in a mouse model. This model is claimed to have resilience to cardiac stress.

Major strengths of the study include 1) the large dataset generated which will have utility for further scientific inquiry for the authors and others in the field, 2) the innovative approach of using cross-analyses linking transcriptomic data to proteomic and phosphoproteomic data. One weakness is the lack of a focused question and clear relevance to human disease. These are all critical biological pathways that the authors are studying and essentially, they have compiled a database that could be surveyed to generate and test future hypotheses.

Thank you for your efforts to review our manuscript, we are delighted to learn that you found our approach to link transcriptomic, proteomic and phosphoproteome data in our analysis to be innovative. Your comment that we have not focused on a question with clear relevance to human disease is “right on point!”

During chronic pathophysiologic states e.g., chronic heart failure (CHF) in humans, AC/cAMP/PKA/Ca2+ signaling increases progressively the degree of heart failure progresses, leading to cardiac inflammation, mediated in part, by cyclic-AMP- induced up- regulation of renin-angiotensin system (RAS) signaling. Standard therapies for CHF include β-adrenoreceptor blockers and RAS inhibitors, which although effective, are suboptimal in amelioration of heart failure progression. One strategy to devise novel and better therapies for heart failure, would be to uncover the full spectrum of concentric cardio- protective adaptations that becomes activated in response to severe, chronic AC/cAMP/PKA/Ca2+ -induced cardiac stress.

We employed unbiased omics analyses, in our prior study (https://elifesciences.org/articles/80949v1) of the mouse harboring cardiac specific overexpression of adenylyl cyclase type 8 (TGAC8), and identified more than 2,000 transcripts and proteins, comprising a broad array of biological processes across multiple cellular compartments, that differed in TGAC8 left ventricle compared to WT. These bioinformatic analyses revealed that marked overexpression of AC8 engages complex, concentric adaptation "circuity" that has evolved in mammalian cells to confer resilience to stressors that threaten health or life. The main human disease category identified in these analyses was Organismal Injury and Abnormalities, suggesting that defenses against stress were activated as would be expected, in response to cardiac stress. Specific concentric signaling pathways that were enriched and activated within the TGAC8 protection circuitry included cell survival initiation, protection from apoptosis, proliferation, prevention of cardiac-myocyte hypertrophy, increased protein synthesis and quality control, increased inflammatory and immune responses, facilitation of tissue damage repair and regeneration and increased aerobic energetics. These TGAC8 stress response circuits resemble many adaptive mechanisms that occur in response to the stress of disease states and may be of biological significance to allow for proper healing in disease states such as myocardial infarction or failure of the heart. The main human cardiac diseases identified in bioinformatic analyses were multiple types cardiomyopathies, again suggesting that mechanisms that confer resilience to the stress of chronic increased AC-PKA-Ca2+ signaling are activated in the absence of heart failure in the super-performing TGAC8 heart at 3-months of age.

In the present study, we performed a comprehensive in silico analysis of transcription, translation, and post-translational patterns, seeking to discover whether the coordinated transcriptome and proteome regulation of the adaptive protective circuitry within the AC8 heart that is common to many types of cardiac disease states identified in our previous study (https://elifesciences.org/articles/80949v1) extends to the phosphoproteome.

Reviewer #2 (Public Review):

In this study, the investigators describe an unbiased phosphoproteomic analysis of cardiac-specific overexpression of adenylyl cyclase type 8 (TGAC8) mice that was then integrated with transcriptomic and proteomic data. The phosphoproteomic analysis was performed using tandem mass tag-labeling mass spectrometry of left ventricular (LV) tissue in TGAC8 and wild-type mice. The initial principal component analysis showed differences between the TGAC8 and WT groups. The integrated analysis demonstrated that many stress-response, immune, and metabolic signaling pathways were activated at transcriptional, translational, and/or post-translational levels.

The authors are to be commended for a well-conducted study with quality control steps described for the various analyses. The rationale for following up on prior transcriptomic and proteomic analyses is described. The analysis appears thorough and well-integrated with the group's prior work. Confirmational data using Western blot is provided to support their conclusions. Their findings have the potential of identifying novel pathways involved in cardiac performance and cardioprotection.

Thank you for your efforts to review our manuscript, we are delighted to learn that you found our approach to link transcriptomic, proteomic and phosphoproteome data in our analysis. We are delighted that you found our work to be well-conducted, to have been well performed, and that our analysis was thorough and well-integrated with our prior work in this arena and that are findings have the potential of identifying novel pathways involved in cardiac performance and cardioprotection.

Reviewer #1 (Recommendations For The Authors):

I humbly suggest that the authors reconsider the title, as it could be more clear as to what they are studying. Are the authors trying to highlight pathways related to cardiac resilience? Resilience might be a clearer word than "performance and protection circuitry".

Thank you for this important comment. We have revised the title accordingly: Reprogramming of cardiac phosphoproteome, proteome and transcriptome confers resilience to chronic adenylyl cyclase-driven stress.

Perhaps the text can be reviewed in detail by a copy-editor, as there are many grammatically 'awkward' elements (for example, line 56: "mammalians" instead of mammals), inappropriate colloquialisms (for example, line 73: "port-of-call"), and stylistic unevenness that make it difficult to read.

We have reviewed the text in detail, with the assistance of a copy editor, in order to identify and correct awkward elements and to search for other colloquialisms. Finally, although “stylistic unevenness” to which you refer may be difficult for us to identify during our re-edits, we have tried our best to identify and revise them.

The best-written sections are the first few paragraphs of the discussion section, which finally clarify why the TGAC8 mouse is important in understanding cardiac resilience to stress and how the present study leverages this model to disentangle the biological processes underlying the resilience. I wish this had been presented in this manner earlier in the paper, (in the abstract and introduction) so I could have had a clearer context in which to interpret the data. It would also be helpful to point out whether the TGAC8 mouse has any correlates with human disease.

Thank you for this very important comment. Well put! In addition to recasting the title to include the concept of resilience, we have revised both the abstract and introduction to feature what you consider to be important to the understanding of cardiac resilience to stress, and how the present study leverages this model to disentangle the biological processes underlying the resilience.

Reviewer #2 (Recommendations For The Authors):

1. How were the cutoffs determined to distinguish between upregulated/downregulated phosphoproteins and phosphopeptides?

Thank you for this important question. We used the same criteria to distinguish differences between TGAC8 and WT for unnormalized and normalized phosphoproteins, -log10(p-value) > 1.3, and log2FoldChange <= -0.4 (down) or log2FoldChange >= 0.4 (up), as stated in the methods section, main text and figure legend. The results were consistent across all analyses and selectively verified by experiments.

1. Were other models assessed for correlation between transcriptome and phosphoproteome other than a linear relationship of log2 fold change?

Thank you for this comment. In addition to a linear relationship of log2 fold change of molecule expression, we also compared protein activities, e.g., Fig 4F, and pathways enriched from different omics, e.g., Fig 3D, 5J, 6B and 6F.

1. Figures 1A and 5G seem to show outliers. How many biological and technical replicates would be needed to minimize error?

Thank you for the question. Figures 1A and 5G were PCA plots which, as expected, manifested some genetic variability among the same genotypes. The PCA plots, however, are useful in determining how the identified items separated, both within and among genotypes. For bioinformatics analysis such as ours, 4-5 samples are sufficient to accomplish this, as demonstrated by separation, by genotype, of samples in PCA. Thus, in addition to discovery of true heterogeneity among the samples, our results are still able to robustly discover the true differences between the genotypes.

1. Were the up/downregulated genes more likely to be lowly expressed (which would lead to larger log2 changes identified)?

In response to your query, we calculated the average expression of phosphorylation levels across all samples to observe whether they were expressed in low abundance in all samples. We also generated the MA plots, an application of a Bland–Altman plot, to create a visual representation of omics data. The MA plots in Author response image 1 illustrate that the target molecules with significantly changed phosphorylation levels did not aggregate within the very low abundance. To confirm this conclusion, we adopted two sets of cutoffs: (1) change: -log10(p-value) > 1.3, and log2FoldChange < 0 (down) or log2FoldChange > 0 (up); and (2) change_2: -log10(p-value) > 1.3, and log2FoldChange <= -0.4 (down) or log2FoldChange >= 0.4 (up).

Author response image 1.

1. "We verified some results through wet lab experiments" in the abstract is vague.

Thank you for the good suggestion. What we meant to indicate here was that identified genotypic differences in selected proteins, phosphoproteins and RNAs discovered in omics were verified by western blots, protein synthesis detection, proteosome activity detection, and protein soluble and insoluble fractions detection. However, we have deleted the reference to the wet lab experiments in the revised manuscript.

1. There are minor syntactical errors throughout the text.

Thank you very much for the suggestion. As noted in our response, we have edited and revised those errors throughout the text.

eLife assessment

This manuscript describes an important web resource for kinases connected to cytokines. The compelling information will be used by researchers across a number of fields including analysts, modelers, wet lab experimentalists and clinician-researchers, who are looking to improve our understanding of pathologies and means to correct them through modulating the immune response.

Reviewer #1 (Public Review):

Summary:

Kinase inhibitors represent a highly valuable class of drugs as evidenced by their continued clinical success. The target landscape of kinase targeting small molecules can be leveraged to alter multiple phenotypes with increasing complexity that broadly aligns with increasing target promiscuity. This 'tools and resources' contribution provides a starting point for researchers interested in aligning kinase inhibitor activity with cytokine/chemokine stimulated signal transduction networks.

Strengths:

KinCytE is a forward-thinking database that yields hypothesis-generating options for researchers interested in pharmacologically modulating cytokine/chemokine signaling.

Weaknesses:

As a 'tools and resources' contribution, the primary (potential) weakness will be the authors' willingness to update and improve the tool. KinCytE will require frequent updating to better inform users in terms of contextual cytokine/chemokine stimulated signaling and the target landscape of those agents that are included as options.

Reviewer #2 (Public Review):

Summary:

In this manuscript, "KinCytE- a Kinase to Cytokine Explorer to Identify Molecular Regulators and Potential Therapeutic", the authors present a web resource, KinCytE, that lets researchers search for kinase inhibitors that have been shown to affect cytokine and chemokine release and signaling networks. I think it's a valuable resource that has a lot of potential and could be very useful in deciding on statistical analysis that might precede lab experiments.

Opportunities:

With the release of the manuscript and the code base in place, I hope the authors continue to build upon the platform, perhaps by increasing the number of cell types that are probed (beyond macrophages). Additionally, when new drug-response data becomes available, perhaps it can be used to further validate the findings. Overall, I see this as a great project that can evolve.

Strengths:

The site contains valuable content, and the structure is such that growing that content should be possible.

Weaknesses:

Only based on macrophage experiments, would be nice to have other cell types investigated, but I'm sure that will be remedied with some time.

Reviewer #1 (Public Review):

Summary:<br /> Tuberculous meningitis (TBM) is one of the most severe forms of extrapulmonary TB. TBM is especially prevalent in people who are immunocompromised (e.g. HIV-positive). Delays in diagnosis and treatment could lead to severe disease or mortality. In this study, the authors performed the largest-ever host whole blood transcriptomics analysis on a cohort of 606 Vietnamese participants. The results indicated that TBM mortality is associated with increased neutrophil activation and decreased T and B cell activation pathways. Furthermore, increased angiogenesis was also observed in HIV-positive patients who died from TBM, whereas activated TNF signaling and down-regulated extracellular matrix organisation were seen in the HIV-negative group. Despite similarities in transcriptional profiles between PTB and TBM compared to healthy controls, inflammatory genes were more active in HIV-positive TBM. Finally, 4 hub genes (MCEMP1, NELL2, ZNF354C, and CD4) were identified as strong predictors of death from TBM.

Strengths:<br /> This is a really impressive piece of work, both in terms of the size of the cohort which took years of effort to recruit, sample, and analyse, and also the meticulous bioinformatics performed. The biggest advantage of obtaining a whole blood signature is that it allows an easier translational development into a test that can be used in the clinical with a minimally invasive sample. Furthermore, the data from this study has also revealed important insights into the mechanisms associated with mortality and the differences in pathogenesis between HIV-positive and HIV-negative patients, which would have diagnostic and therapeutic implications.

Weaknesses:<br /> The data on blood neutrophil count is really intriguing and seems to provide a very powerful yet easy-to-measure method to differentiate survival vs. death in TBM patients. It would be quite useful in this case to perform predictive analysis to see if neutrophil count alone, or in combination with gene signature, can predict (or better predict) mortality, as it would be far easier for clinical implementation than the RNA-based method. Moreover, genes associated with increased neutrophil activation and decreased T cell activation both have significantly higher enrichment scores in TBM (Figure 9) and in morality (Figure 8). While I understand the basis of selecting hub genes in the significant modules, they often do not represent these biological pathways (at least not directly associated in most cases). If genes were selected based on these biologically relevant pathways, would they have better predictive values?

Reviewer #2 (Public Review):

Summary:<br /> This manuscript describes the analysis of blood transcriptomic data from patients with TB meningitis, with and without HIV infection, with some comparison to those of patients with pulmonary tuberculosis and healthy volunteers. The objectives were to describe the comparative biological differences represented by the blood transcriptome in TBM associated with HIV co-infection or survival/mortality outcomes and to identify a blood transcriptional signature to predict these outcomes. The authors report an association between mortality and increased levels of acute inflammation and neutrophil activation, but decreased levels of adaptive immunity and T/B cell activation. They propose a 4-gene prognostic signature to predict mortality.

Strengths:<br /> -Biological evaluations of blood transcriptomes in TB meningitis and their relationship to outcomes have not been extensively reported previously.<br /> -The size of the data set is a major strength and is likely to be used extensively for secondary analyses in this field of research.

Weaknesses:<br /> The bioinformatic analysis is limited to a descriptive narrative of gene-level functional annotations curated in GO and KEGG databases. This analysis can not be used to make causal inferences. In addition, the functional annotations are limited to 'high-level' terms that fail to define biology very precisely. At best, they require independent validation for a given context. As a result, the conclusions are not adequately substantiated. The identification of a prognostic blood transcriptomic signature uses an unusual discovery approach that leverages weighted gene network analysis that underpins the bioinformatic analyses. However, the main problem is that authors seem to use all the data for discovery and do not undertake any true external validation of their gene signature. As a result, the proposed gene signature is likely to be overfitted to these data and not generalisable. Even this does not achieve significantly better prognostic discrimination than the existing clinical scoring.

Author Response

Reviewer #1 (Public Review):

Summary:

The investigators have performed a state-of-the art systematic review and meta-analysis of studies that may help to answer the research question: if administration of multiple antibiotics simultaneously prevents antibiotic resistance development in individuals. The amount of studies eligible for analysis is very low, and within that low number, there is huge variability in bug-drug combinations studied and most studies had a high risk of bias, further limiting the capability of meta-analysis to answer the research question. In addition, based on I2 values there is also huge statistical heterogeneity between outcomes of studies compared, further limiting the predictive value of meta-analysis. In fact, the only 2 studies meeting all eligibility criteria addressed the treatment of mycobacterium tuberculosis, for which the research question is hardly applicable. The authors, therefore, conclude that "our analysis could not identify any benefit or harm of using a higher or a lower number of antibiotics regarding within-patient resistance development." Apart from articulating this knowledge gap, the findings will not have consequences for patient care, but may stimulate the scientific community to better address this research question in future studies.

Strengths:

The systematic and rigorous approach for the review and meta-analysis.

Weaknesses:

None identified.

We thank the reviewer for this thoughtful and positive appraisal of our work.

Reviewer #2 (Public Review):

Summary:

The authors performed a systematic review and meta-analysis to investigate whether the frequency of emergence of resistance is different if combination antibiotic therapy is used compared to fewer antibiotics. The review shows that there is currently insufficient evidence to reach a conclusion due to the limited sample size. High-quality studies evaluating appropriate antimicrobial resistance endpoints are needed.

Strengths:

The strengths of the manuscript are that the article addresses a relevant research question that is often debated. The article is well-written and the methodology used is valid. The review shows that there is currently insufficient evidence to reach a conclusion due to the limited sample size. High-quality studies evaluating appropriate antimicrobial resistance endpoints are needed. I have several comments and suggestions for the manuscript.

Weaknesses:

Weaknesses of the manuscript are the large clinical and statistical heterogeneity and the lack of clear definitions of acquisition of resistance. Both these weaknesses complicate the interpretation of the study results.

We thank the reviewer for the positive comments and pointing out where our work can be improved.

Major comments:

My main concern about the manuscript is the extent of both clinical and statistical heterogeneity, which complicates the interpretation of the results. I don't understand some of the antibiotic comparisons that are included in the systematic review. For instance the study by Paul et al (50), where vancomycin (as monotherapy) is compared to co-trimoxazole (as combination therapy). Emergence (or selection) of co-trimoxazole in S. aureus is in itself much more common than vancomycin resistance. It is logical and expected to have more resistance in the co-trimoxazole group compared to the vancomycin group, however, this difference is due to the drug itself and not due to co-trimoxazole being a combination therapy. It is therefore unfair to attribute the difference in resistance to combination therapy. Another example is the study by Walsh (71) where rifampin + novobiocin is compared to rifampin + co-trimoxazole. There is more emergence of resistance in the rifampin + co-trimoxazole group but this could be attributed to novobiocin being a different type of antibiotic than co-trimoxazole instead of the difference being attributed to combination therapy. To improve interpretation and reduce heterogeneity my suggestion would be to limit the primary analyses to regimens where the antibiotics compared are the same but in one group one or more antibiotic(s) are added (i.e. A versus A+B). The other analyses are problematic in their interpretation and should be clearly labeled as secondary and their interpretation discussed.

We acknowledge the presence of statistical and clinical heterogeneity in our overall analysis. The decision to pursue this comprehensive examination was predefined in our previously published study protocol (PROSPERO CRD42020187257) and driven by our interest whether, despite some differences, we could either identify an overarching effect of combination therapy on resistance or identify factors that explain potential differences of the effect of combination therapy across pathogens/drugs. We indeed, find that heterogeneity is high, however identifying the driving factors of this heterogeneity is difficult as evidence is limited.

We carried out several subgroup analyses, e.g. explicitly focusing on specific pathogen groups and medical conditions or exploring heterogeneity in treatment arms (figure 3, supplementary materials section 6). However, it is important to highlight that the number of studies available for these subgroup analyses was low. Additionally, recognizing the high heterogeneity within treatment arms, we performed a subgroup analysis focusing solely on resistances of antibiotics common to both arms (supplementary material section 6.1.8; which would avoid comparisons such as the one between vancomycin and co-trimoxazole raised by the reviewer). Unfortunately, this also revealed substantial heterogeneity. While we aimed to address heterogeneity through these subgroup analyses, limitations arose due to the number of studies meeting specific criteria and the nature of data provided by these studies.

Moreover, regarding the concern on interpretation of co-trimoxazole as combination therapy, we acknowledge the confusion surrounding its classification as one or two antibiotics. Despite the common contemporary view of co-trimoxazole as a single antibiotic, we chose to consider it as two antibiotics due to historical practices, as observed in Black et al. (1982), where trimethoprim was compared to trimethoprim and sulfamethoxazole. We recognize that this decision may lead to confusion and we consider conducting a further sensitivity analysis in the future version of this manuscript, exploring the possibility of considering co-trimoxazole as a single antibiotic. We agree that the slight trend of less antibiotics performing better overserved for MRSA, should not be over interpreted as this is driven by the two studies Walsh et al 1993 and Paul et al 2015 as pointed out by the reviewer. In lines 183-186 we discuss this issue that for better evaluation of antibiotic combination therapy, more studies which use identical antibiotics (i.e. A versus A+B) are needed. We will try to clarify and highlight this in the future version of the manuscript.

Another concern is about the definition of acquisition of resistance, which is unclear to me. If for example meropenem is administered and the follow-up cultures show Enterococcus species (which is intrinsically resistant to meropenem), does this constitute acquisition of resistance? If so, it would be misleading to determine this as an acquisition of resistance, as many people are colonized with Enterococci and selection of Enterococci under therapy is very common. If this is not considered as the acquisition of resistance please include how the acquisition of resistance is defined per included study.

Thank you for pointing out this potential ambiguity. Our definition of “acquisition of resistance” is agnostic to bacterial species and hence intrinsically resistant species can be included if they were only detected during the follow-up culture by the studies. We will clarify this in the definition of “acquisition of the resistance” in the manuscript (see l. 259-260). However, it was not always clear from the studies which pathogens were acquired or whether intrinsically resistant species were not reported. Therefore, we rely on the studies' specifications of resistant and non-resistant without further classifying data into intrinsic and non-intrinsic resistance. The outcome “acquisition of resistance” can be seen more of a risk assessment for having any resistant bacterium during or after treatment. In contrast, the outcome “emergence of resistance” is more rigorous, demanding the same species to be measured as more resistant during or after treatment.

Table S1 is not sufficiently clear because it often only contains how susceptibility testing was done but not which antibiotics were tested and how a strain was classified as resistant or susceptible.

In Table S1, we omitted the listing of antibiotics for which susceptibility testing was performed, as this information is already presented in the main text (Table 1). However, we agree that linking this information better in a future version would benefit the understanding. Given the variability in methods used to assess resistance and the variability in drugs, the comparability of breakpoints is limited. Hence, we decided not to provide further details on this aspect so far.

Line 85: "Even though within-patient antibiotic resistance development is rare, it may contribute to the emergence and spread of resistance."

Depending on the bug-drug combination, there is great variation in the propensity to develop within-patient antibiotic resistance. For example: within-patient development of ciprofloxacin resistance in Pseudomonas is fairly common while within-patient development of methicillin resistance in S. aureus is rare. Based on these differences, large clinical heterogeneity is expected and it is questionable where these studies should be pooled.

We agree that our formulation neglects differences in prevalence of within-host resistance emergence depending on bug-drug combinations. We will correct this in our upcoming version. (i.e. we will correct our statement to: “Within-patient antibiotic resistance development, even if rare, can contribute to the emergence and spread of resistance.”)

Line 114: "The overall pooled OR for acquisition of resistance comparing a lower number of antibiotics versus a higher one was 1.23 (95% CI 0.68 - 2.25), with substantial heterogeneity between studies (I2=77.4%)"

What consequential measures did the authors take after determining this high heterogeneity? Did they explore the source of this large heterogeneity? Considering this large heterogeneity, do the authors consider it appropriate to pool these studies?

Thank you for highlighting this lack of clarity. In our upcoming version, we will emphasize the sub-analyses conducted to explore heterogeneity (i.e., figure 3 and supplementary materials section 6). Nevertheless, these analyses faced limitations due to the scarcity of evidence and the data provided by the studies. Given the lack of appropriate evidence, it is hard to identify the source of heterogeneity. The decision to pool all studies was pre-specified in our previously published study protocol (PROSPERO CRD42020187257) and was motivated by the question whether there is a general effect of combination therapy on resistance development or identify factors that explain potential differences of the effect of combination therapy across bug-drug combinations.

eLife assessment

This is a methodologically state-of-the-art systematic review and meta-analysis of studies that addressed the question of whether the administration of multiple antibiotics simultaneously prevents antibiotic resistance development in individuals. The findings are solid. Rather than providing a precise answer, the synthesis of studies eligible for analysis leads to the conclusion that "our analysis could not identify any benefit or harm of using a higher or a lower number of antibiotics regarding within-patient resistance development." This article is important as it articulates the existing knowledge gap, but also serves as an example to careful future use of the meta-analysis methodology, when existing data just don't allow conclusions.

Reviewer #1 (Public Review):

Summary:<br /> The investigators have performed a state-of-the art systematic review and meta-analysis of studies that may help to answer the research question: if administration of multiple antibiotics simultaneously prevents antibiotic resistance development in individuals. The amount of studies eligible for analysis is very low, and within that low number, there is huge variability in bug-drug combinations studied and most studies had a high risk of bias, further limiting the capability of meta-analysis to answer the research question. In addition, based on I2 values there is also huge statistical heterogeneity between outcomes of studies compared, further limiting the predictive value of meta-analysis. In fact, the only 2 studies meeting all eligibility criteria addressed the treatment of mycobacterium tuberculosis, for which the research question is hardly applicable. The authors, therefore, conclude that "our analysis could not identify any benefit or harm of using a higher or a lower number of antibiotics regarding within-patient resistance development." Apart from articulating this knowledge gap, the findings will not have consequences for patient care, but may stimulate the scientific community to better address this research question in future studies.

Strengths:<br /> The systematic and rigorous approach for the review and meta-analysis.

Weaknesses:<br /> None identified.

Reviewer #2 (Public Review):

Summary:<br /> The authors performed a systematic review and meta-analysis to investigate whether the frequency of emergence of resistance is different if combination antibiotic therapy is used compared to fewer antibiotics. The review shows that there is currently insufficient evidence to reach a conclusion due to the limited sample size. High-quality studies evaluating appropriate antimicrobial resistance endpoints are needed.

Strengths:<br /> The strengths of the manuscript are that the article addresses a relevant research question that is often debated. The article is well-written and the methodology used is valid. The review shows that there is currently insufficient evidence to reach a conclusion due to the limited sample size. High-quality studies evaluating appropriate antimicrobial resistance endpoints are needed. I have several comments and suggestions for the manuscript.

Weaknesses:<br /> Weaknesses of the manuscript are the large clinical and statistical heterogeneity and the lack of clear definitions of acquisition of resistance. Both these weaknesses complicate the interpretation of the study results.

Major comments:<br /> My main concern about the manuscript is the extent of both clinical and statistical heterogeneity, which complicates the interpretation of the results. I don't understand some of the antibiotic comparisons that are included in the systematic review. For instance the study by Paul et al (50), where vancomycin (as monotherapy) is compared to co-trimoxazole (as combination therapy). Emergence (or selection) of co-trimoxazole in S. aureus is in itself much more common than vancomycin resistance. It is logical and expected to have more resistance in the co-trimoxazole group compared to the vancomycin group, however, this difference is due to the drug itself and not due to co-trimoxazole being a combination therapy. It is therefore unfair to attribute the difference in resistance to combination therapy. Another example is the study by Walsh (71) where rifampin + novobiocin is compared to rifampin + co-trimoxazole. There is more emergence of resistance in the rifampin + co-trimoxazole group but this could be attributed to novobiocin being a different type of antibiotic than co-trimoxazole instead of the difference being attributed to combination therapy. To improve interpretation and reduce heterogeneity my suggestion would be to limit the primary analyses to regimens where the antibiotics compared are the same but in one group one or more antibiotic(s) are added (i.e. A versus A+B). The other analyses are problematic in their interpretation and should be clearly labeled as secondary and their interpretation discussed.

Another concern is about the definition of acquisition of resistance, which is unclear to me. If for example meropenem is administered and the follow-up cultures show Enterococcus species (which is intrinsically resistant to meropenem), does this constitute acquisition of resistance? If so, it would be misleading to determine this as an acquisition of resistance, as many people are colonized with Enterococci and selection of Enterococci under therapy is very common. If this is not considered as the acquisition of resistance please include how the acquisition of resistance is defined per included study. Table S1 is not sufficiently clear because it often only contains how susceptibility testing was done but not which antibiotics were tested and how a strain was classified as resistant or susceptible.

Line 85: "Even though within-patient antibiotic resistance development is rare, it may contribute to the emergence and spread of resistance."<br /> Depending on the bug-drug combination, there is great variation in the propensity to develop within-patient antibiotic resistance. For example: within-patient development of ciprofloxacin resistance in Pseudomonas is fairly common while within-patient development of methicillin resistance in S. aureus is rare. Based on these differences, large clinical heterogeneity is expected and it is questionable where these studies should be pooled.

Line 114: "The overall pooled OR for acquisition of resistance comparing a lower number of antibiotics versus a higher one was 1.23 (95% CI 0.68 - 2.25), with substantial heterogeneity between studies (I2=77.4%)"<br /> What consequential measures did the authors take after determining this high heterogeneity? Did they explore the source of this large heterogeneity? Considering this large heterogeneity, do the authors consider it appropriate to pool these studies?

eLife assessment

This very interesting study by Vuond and colleagues reports on the kinetics of viremia in a large set of individuals from Vietnam. In the large cohort, all 4 dengue serotypes are represented and the authors try to correlate viraemia measured at various days from illness onset with thrombocytopaenia and severe dengue, according to the WHO 2009 classification scheme. These are fundamental findings and provide compelling evidence of the importance of measuring viremia early in the phase of the disease. These data will help to inform the design of studies of antiviral drugs against dengue.

Reviewer #1 (Public Review):

Summary:

This manuscript by Vuong and colleagues reports a study that pooled data from 3 separate longitudinal studies that collectively spanned an observation period of over 15 years. The authors examined for correlation between viraemia measured at various days from illness onset with thrombocytopaenia and severe dengue, according to the WHO 2009 classification scheme. The motivation for this study is both to support the use of viraemia measurement as a prognostic indicator of dengue and also when an antiviral drug becomes licensed for use, to guide the selection of patients for antiviral therapy. They found that the four DENVs show differences in peak and duration of viraemia and that viraemia levels before day 5 but not those after from illness onset correlated with platelet count and plasma leakage at day 7 onwards. They concluded that the viraemia kinetics call for early measurement of viraemia levels in the early febrile phase of illness.

Strengths:

This is a unique study due to the large sample size and longitudinal viraemia measurements in the study subjects. The data addresses a gap in information in the literature, where although it has been widely indicated that viraemia levels are useful when collected early in the course of illness, this is the first time anyone has systematically examined this notion.

Weaknesses:

The study only analysed data from dengue patients in Vietnam. Moreover, the majority of these patients had DENV-1 infection; few had DENV-4 infection. The data could thus be skewed by the imbalance in the prevalence of the different types of DENV during the period of observation. The use of patient-reported time of symptom onset as a reference point for viraemia measurement is pragmatic although there is subjectivity and thus noise in the data.

Reviewer #2 (Public Review):

Summary:

This manuscript highlights very important findings in the field, especially in designing clinical trials for the evaluation of antivirals.

Strengths:

The study shows significant differences between the kinetics of viral loads between serotypes, which is very interesting and should be taken into account when designing trials for antivirals.

Weaknesses:

The kinetics of the viral loads based on disease severity throughout the illness are not described, and it would be important if this could be analyzed.

eLife assessment

This solid study investigates the transdifferentiation of chicken embryonic fibroblasts into muscle and fat cells in 3D to create whole-cut meat mimics. The study is important and provides a method to control muscle, fat, and collagen content within the 3D meat mimics and thus provides a new avenue for customized cultured meat production. Limitations of this study include the use of transgene for transdifferentiation and thus the creation of GMO food.

Reviewer #1 (Public Review):

Summary:

The authors presented here a novel 3D fibroblast culture and transdifferentiation approach for potential meat production with GelMA hydrogel.

Strengths:

1. Reduced serum concentration for 3D chicken fibroblast culture and transdifferentiation is optimized.<br /> 2. Efficient myogenic transdifferentiation and lipogenesis as well as controlled fat deposition are achieved in the 3D GelMA.

Weaknesses:

1. While the authors stated the rationale of using fibroblasts instead of myogenic/adipogenic stem cells for meat production, the authors did not comment on the drawbacks/disadvantages of genetic engineering (e.g., forced expression of MyoD) in meat production.<br /> 2. While the authors cited one paper to state the properties and applications of GelMA hydrogel in tissue engineering and food processing, concerns/examples of the food safety with GelMA hydrogel are not discussed thoroughly.<br /> 3. In Fig. 4C, there seems no significant difference in the Vimentin expression between Fibroblast_MyoD and Myofibroblast. The conclusion of "greatly reduced in the myogenic transdifferentiated cells" is overstated.<br /> 4. The presented cell culture platform is only applied to chicken fibroblasts and should be tested in other species such as pigs and fish.

Reviewer #2 (Public Review):

The manuscript by Ma et al. tries to develop a protocol for cell-based meat production using chicken fibroblasts as three-dimensional (3D) muscle tissues with fat accumulation. The authors used genetically modified fibroblasts which can be forced to differentiate into muscle cells and formulated 3D tissues with these cells and a biphasic material (hydrogel). The degrees of muscle differentiation and lipid deposition in culture were determined by immunohistochemical, biochemical, and molecular biological evaluations. Notably, the protocol successfully achieved the process of myogenic and lipogenic stimulation in the 3D tissues.

Overall, the study is reasonably designed and performed including adequate analysis. The manuscript is clearly written with well-supported figures. While it presents valuable results in the field of cultivated meat science and skeletal muscle biology, some critical concerns were identified. First, it is unclear whether some technical approaches were really the best choice for cell-based meat production. Next, more careful evaluations and justifications would be required to properly explain biological events in the results. These points include additional evaluations and considerations with regard to myocyte alignment and lipid accumulation in the differentiated 3D tissues. The present data are very suggestive in general, but further clarifications and arguments would properly support the findings and conclusions.

Author Response

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

We confirm that that “count-down” parameter, mentioned by reviewer 1, is indeed counted from the first lockdown day and increases continuously, even when we do not have any data – and that this is clearly written in the manuscript.

---

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

Reviewer 1:

(Note, while these authors do reference Derryberry et al., I thought that there could have been much more direct comparison between the results of the two approaches).

We added some more discussion of the differences between the papers.

One important drawback of the approach, which potentially calls into question the authors' conclusions, is that the acoustic sampling only occurred during the pandemic: for several lockdown periods and then for a period of 10 days immediately after the end of the final lockdown period in May of 2020. Several relevant things changed from March to May of 2020, most notably the shift from spring to summer, and the accompanying shift into and through the breeding season (differing for each of the three focal species). Although the statistical methods included an attempt to address this, neither the inclusion of the "count down" variable nor the temperature variable could account for any non-linear effects of breeding phenology on vocal activity. I found the reliance on temperature particularly troubling, because despite the authors' claims that it was "a good proxy of seasonality", an examination of the temperature data revealed a considerable non-linear pattern across much of the study duration. In addition, using a period immediately after the lockdowns as a "no-lockdown" control meant that any lingering or delayed effects of human activity changes in the preceding two months could still have been relevant (not to mention the fact that despite the end of an official lockdown, the pandemic still had dramatic effects on human activity during late May 2020).

In general, the reviewer is correct, and we reformulated some of the text to more carefully address these points. However, we would like to note two things: (1) Changes occurred rapidly with birds rapidly changing their behavior – this is one of the main conclusions of our study, i.e., that urban dwelling animals are highly plastic in behavior. So that lingering effects were unlikely. (2) Changes occurred in both directions, and thus seasonality (which is expected to have a uni-directional effect) cannot explain everything we observed. We are not sure what the reviewer means by ‘considerable non-linear patterns’ when referring to the temperature. Except for ~5 days with temperatures that exceeded the expected average by 3-4 degrees, the temperature increased approximately linearly during the period as expected from seasonality (see Author response image 1). Following the reviewer’s comment, we tested whether exclusion of data from these days changes the results and found no change.

We would like to note that in terms of breeding, all birds were within the same state during both the lockdown and the non-lockdown periods. Parakeets and crows have a long breeding season Feb-end of June with one cycle. They will stay around the nest throughout this season and especially in the peak of the season March-May. Prinias start slightly later at the beginning of March with 2-3 cycles till end of June.

Regarding the comment about human activity, as we now also note in the manuscript, reality in Israel was actually the opposite of the reviewer’s suggestion with people returning to normal behavior towards the end of the lockdown (even before its official removal). We believe that this added noise to our results, and that the effect of the lockdown was probably higher than we observed.

Author response image 1.

Another weakness of the current version of the manuscript is the use of a supposed "contradiction" in the existing literature to create the context for the present study. Although the various studies cited do have many differences in their results, those other papers lay out many nuanced hypotheses for those differences. Almost none of the studies cited in this manuscript actually reported blanket increases or decreases in urban birds, as suggested here, and each of those papers includes examples of species that showed different responses. To suggest that they are on opposite sides of a supposed dichotomy is a misrepresentation. Many of those other studies also included a larger number of different species, whereas this study focused on three. Finally, this study was completed at a much finer spatial scale than most others and was examining micro-habitat differences rather than patterns apparent across landscapes. I believe that highlighting differences in scale to explain nuanced differences among studies is a much better approach that more accurately adds to the body of literature.

We thank the reviewer for this good feedback and revised the manuscript, accordingly, placing more emphasis on the micro-scale of this study.

Finally a note on L244-247: I would recommend against discounting the possibility that lockdowns resulted in changes to the birds' vocal acoustics, as Derryberry et al. 2020 found, especially while suggesting that their results were the effects of signal processing artifacts. Audio analysis is not my area of expertise, but isn't it possible that the birds did increase call intensity, but were simply not willing (or able) to increase it to the same degree as the additional ambient noise?

This is an important question. The fact is that when ambient noise increases (at the relevant frequency channels), then the measured vocalizations will also increase. There is no way to separate the two effects. Thus, as scientists, when we cannot measure an effect, it is safer not to suggest an effect. Unfortunately, most studies that claim an increase in vocalizations’ intensity in noise, do not account for this potential artifact (and most of them do not estimate noise at a species-specific level as we have done). This has created a lot of “noise” in the field. We do not want to criticize the Derryberry results without analyzing the data, but from reading their methods it does not seem like they took the noise into account in their acoustic measurements. But if you look at their figure 4A you will see a lot of variability in measuring the minimum frequency – which could be strongly affected by ambient noise.

In light of the above, we thus prefer to be careful and not to state changes that are probably false. We added some of this information to the manuscript. We also added the linear equations to the graph (in the caption of figure 3) where it can be seen that the slope is always <=1.

Reviewer 2:

The explanation of methods can be improved. For example, it is not clear if data were low-pass filtered before resampling to avoid aliasing.

We edited the methods and hopefully they are clearer now. Regarding the specific question – yes, an LPF was applied to prevent aliasing before the resampling. This information was added to the manuscript.

It is quite possible that birds move into the trees and further from the recorders with human activity. Since sound level decreases by the square of the distance of the source from the recorders, this could significantly affect the data. As indicated in the Discussion, this is a significant parameter that could not be controlled.

The reviewer is correct, and we addressed this point. Such biases could arise with any type of surveying including manual transects (except for perhaps, placing tags on the animals). We note that we only analyzed high SNR signals and that the species we selected somewhat overcome this bias – both crows and parakeets are not shy and Prinias are anyway shy and prefer to not be out in the open. We would also expect to see a stronger effect for human speech if this was a central phenomenon, and we did not see this, but of course this might have affected our results.

In interpreting the data, the authors mention the effect of human activity on bird vocalizations in the context of inter-species predator-prey interactions; however, the presence of humans could also modify intraspecies interactions by acting as triggers for communication of warning and alarm, and/or food calls (as may sometimes be the case) to conspecifics. Along the same lines, it is important to have a better understanding of the behavioral significance of the syllables used to monitor animal activity in the present study.

We agree with this point and added more discussion of both this potential bias and the type of syllables that were analyzed.

Another potential effect that may influence the results but is difficult to study, relates to the examination of vocalizations near to the ambient noise level. This is the bandwidth of sound levels where most significant changes may occur, for example, due to the Lombard effect demonstrated in bird and bat species. However, as indicated, these are also more difficult to track and quantify. Moreover, human generated noise, other than speech, may be a more relevant factor in influencing acoustic activity of different bird species. Speech, per se, similar to the vocalizations of many other species, may simply enrich the acoustic environment so that the effects observed in the present study may be transient without significant long-term consequences.

We note that we already included a noise parameter (in addition to human speech) in the original manuscript. Following the reviewer’s comment, we examined another factor, namely we replaced the previous ambient noise parameter with an estimate of ambient noise under 1kHz which should reflect most anthropogenic noise (not restricted to human speech). This model gave very similar results to the previous one (which is not very surprising as noise is usually correlated). We added this information to the revised manuscript, and we now also added examples of anthropogenic noise to the supplementary materials (Fig. S8). In general, we accept the comments made by the reviewer, but would like to emphasize that we only analyze high SNR vocalization (and not vocalizations that were close to the noise level). This strategy should have overcome biases that resulted from slight changes in ambient noise.

In general, the authors achieved their aim of illustrating the complexity of the effect of human activity on animal behavior. At the same time, their study also made it clear that estimating such effects is not simple given the dynamics of animal behavior. For example, seasonality, temperature changes, animal migration and movement, as well as interspecies interactions, such as related to predator-prey behavior, and inter/intra-species competition in other respects can all play into site-specific changes in the vocal activity of a particular species.

We completely agree and tried to further emphasize this in the revised manuscript. This is one of the main conclusions of this study – we should be careful when reaching conclusions.

Although the methods used in the present study are statistically rigorous, a multivariate approach and visualization techniques afforded by principal components analysis and multidimensional scaling methods may be more effective in communicating the overall results.

Following this comment, we ran a discriminant function analysis with the parameters of the best model (site category, ambient noise, human activity, temperature and lockdown state) with the task of classifying the level of bird activity. The DFA analysis managed to classify activity significantly above chance and the weights of the parameters revealed some insight about their relative importance. We added this information to the revised manuscript

Suggestions for improvement:

In Figure 2, the labeling of the Y-axis in the right panel should be moved to the left, similar to A and C. This will provide clear separation between the two side-to-side panels.

Revised

In Figure 3, it will be good to see the regression lines (as dashed lines) separately for the lockdown and no-lockdown conditions in addition to the overall effect.

Revised

Editor:

Limitations

Scale: The study's limited spatial and temporal scale was not addressed by the authors, which contrasts with the broader scope of other cited studies. To enhance the significance of the study, acknowledging and clearly highlighting this limitation, along with its potential caveats, modifications in the language used throughout the text would be beneficial. Furthermore, although the authors examined slight variations in habitat, it is important to note that all sites were primarily located within an urban landscape.

We revised the manuscript accordingly.

Control period: The control period is significantly shorter than the lockdown treatment period and occurs at a different time of year, potentially impacting the vocalization patterns of birds due to different annual cycle stages. It is crucial to consider that the control period falls within the pandemic timeframe despite being shortly after the lockdowns ended.

Revised – we included a control comparison to periods of equal length within the lockdown. People gradually stopped obeying the lockdown regulations before its removal so in fact, the official removal date is probably an overestimate for the effect of the lockdown. We now explain this.

Recommendations

Human-generated noise, beyond speech, might have a greater influence on the acoustic activity of various bird species, but previous studies lacked detailed human activity data. Instead of solely noting the number of human talkers, the authors could quantify other aspects of human activity such as vehicles or overall anthropogenic noise volume. Exploring the relationships between these factors and bird activity at a fine scale, while disentangling them from bird detection, would be compelling. It is important to consider the potential difficulty in resolving other anthropogenic sounds within a specific bandwidth, which could be demonstrated to readers through spectrograms and potential post-pandemic changes. Such information, including daily coefficient of variation/fluctuation rather than absolute frequency spectra, could provide valuable insights.

We note that we have already included an ambient noise factor (in addition to human speech) in the previous version. Following the reviewers’ comments, we examined another factor, namely we replaced the current ambient noise parameter with the ambient noise under 1kHz which should reflect most of anthropogenic noise (not restricted to human speech). This model gave very similar results to the previous one (which is not surprising as noise is usually correlated). We also added several spectrograms in the Supplementary material that show examples of different types of noise.

Authors should limit their data interpretation to the impact of lockdown on behavioral responses within small-scale variations in habitat. A key critique is the assumption that activity changes solely resulted from the lockdown, disregarding other environmental factors and phenology.

Following the editor comment we realized that our conclusion\assertations were not clear. We never claimed that activity changes solely resulted from the lockdown. While revsing the mansucirpt we ensurred that we show a significant effect of temperature, ambient noise and human activity – all of which are not dependent on lockdown. We made an effort to emphasize the complexity of the system. We show that the lockdown seemed to have an additional impact, but we never claimed it was the only factor.

To address this, the authors could compare acoustic monitoring data within a shorter timeframe before and after the lockdown (20 days), while also controlling for temperature effects, to strengthen the validity of their claims. They would need to explain in their discussion, however, that such a comparison may still be confounded by any carry-over effects from the 10 days of treatment.

This analysis would be difficult because although the lockdown was officially removed at a specific date, it was gradually less respected by the citizens and thus the last period of the lockdown was somewhere between lockdown and no-lockdown. This is why we chose the approach of taking 10 days randomly from within the lockdown period and comparing them with the 10 post-lockdown days. We now clarify the reason better.

An option is that authors could frame their analysis as a study of the behavior of wildlife coming out of a lockdown, to draw a distinction from other studies that compared pre-pandemic data to pandemic data.

Good idea – revised.

Author Response

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

We thank all three Reviewers for their comments and have revised the manuscript accordingly.

Reviewer #1 (Public Review):

The main objective of this paper is to report the development of a new intramuscular probe that the authors have named Myomatrix arrays. The goal of the Myomatrix probe is to significantly advance the current technological ability to record the motor output of the nervous system, namely fine-wire electromyography (EMG). Myomatrix arrays aim to provide large-scale recordings of multiple motor units in awake animals under dynamic conditions without undue movement artifacts and maintain long-term stability of chronically implanted probes. Animal motor behavior occurs through muscle contraction, and the ultimate neural output in vertebrates is at the scale of motor units, which are bundles of muscle fibers (muscle cells) that are innervated by a single motor neuron. The authors have combined multiple advanced manufacturing techniques, including lithography, to fabricate large and dense electrode arrays with mechanical features such as barbs and suture methods that would stabilize the probe's location within the muscle without creating undue wiring burden or tissue trauma. Importantly, the fabrication process they have developed allows for rapid iteration from design conception to a physical device, which allows for design optimization of the probes for specific muscle locations and organisms. The electrical output of these arrays is processed through a variety of means to try to identify single motor unit activity. At the simplest, the approach is to use thresholds to identify motor unit activity. Of intermediate data analysis complexity is the use of principal component analysis (PCA, a linear second-order regression technique) to disambiguate individual motor units from the wide field recordings of the arrays, which benefits from the density and numerous recording electrodes. At the highest complexity, they use spike sorting techniques that were developed for Neuropixels, a large-scale electrophysiology probe for cortical neural recordings. Specifically, they use an estimation code called kilosort, which ultimately relies on clustering techniques to separate the multi-electrode recordings into individual spike waveforms.

The biggest strength of this work is the design and implementation of the hardware technology. It is undoubtedly a major leap forward in our ability to record the electrical activity of motor units. The myomatrix arrays trounce fine-wire EMGs when it comes to the quality of recordings, the number of simultaneous channels that can be recorded, their long-term stability, and resistance to movement artifacts.

The primary weakness of this work is its reliance on kilosort in circumstances where most of the channels end up picking up the signal from multiple motor units. As the authors quite convincingly show, this setting is a major weakness for fine-wire EMG. They argue that the myomatrix array succeeds in isolating individual motor unit waveforms even in that challenging setting through the application of kilosort.

Although the authors call the estimated signals as well-isolated waveforms, there is no independent evidence of the accuracy of the spike sorting algorithm. The additional step (spike sorting algorithms like kilosort) to estimate individual motor unit spikes is the part of the work in question. Although the estimation algorithms may be standard practice, the large number of heuristic parameters associated with the estimation procedure are currently tuned for cortical recordings to estimate neural spikes. Even within the limited context of Neuropixels, for which kilosort has been extensively tested, basic questions like issues of observability, linear or nonlinear, remain open. By observability, I mean in the mathematical sense of well-posedness or conditioning of the inverse problem of estimating single motor unit spikes given multi-channel recordings of the summation of multiple motor units. This disambiguation is not always possible. kilosort's validation relies on a forward simulation of the spike field generation, which is then truth-tested against the sorting algorithm. The empirical evidence is that kilosort does better than other algorithms for the test simulations that were performed in the context of cortical recordings using the Neuropixels probe. But this work has adopted kilosort without comparable truth-tests to build some confidence in the application of kilosort with myomatrix arrays.

Kilosort was developed to analyze spikes from neurons rather than motor units and, as Reviewer #1 correctly points out, despite a number of prior validation studies the conditions under which Kilosort accurately identifies individual neurons are still incompletely understood. Our application of Kilosort to motor unit data therefore demands that we explain which of Kilosort’s assumptions do and do not hold for motor unit data and explain how our modifications of the Kilosort pipeline to account for important differences between neural and muscle recording, which we summarize below and have included in the revised manuscript.

Additionally, both here and in the revised paper we emphasize that while the presented spike sorting methods (thresholding, PCA-based clustering, and Kilosort) robustly extract motor unit waveforms, spike sorting of motor units is still an ongoing project. Our future work will further elaborate how differences between cortical and motor unit data should inform approaches to spike sorting as well as develop simulated motor unit datasets that can be used to benchmark spike sorting methods.

For our current revision, we have added detailed discussion (see “Data analysis: spike sorting”) of the risks and benefits of our use of Kilosort to analyze motor unit data, in each case clarifying how we have modified the Kilosort code with these issues in mind:

“Modification of spatial masking: Individual motor units contain multiple muscle fibers (each of which is typically larger than a neuron’s soma), and motor unit waveforms can often be recorded across spatially distant electrode contacts as the waveforms propagate along muscle fibers. In contrast, Kilosort - optimized for the much more local signals recorded from neurons - uses spatial masking to penalize templates that are spread widely across the electrode array. Our modifications to Kilosort therefore include ensuring that Kilosort search for motor unit templates across all (and only) the electrode channels inserted into a given muscle. In this Github repository linked above, this is accomplished by setting parameter nops.sigmaMask to infinity, which effectively eliminates spatial masking in the analysis of the 32 unipolar channels recorded from the injectable Myomatrix array schematized in Supplemental Figure 1g. In cases including chronic recording from mice where only a single 8-contact thread is inserted into each muscle, a similar modification can be achieved with a finite value of nops.sigmaMask by setting parameter NchanNear, which represents the number of nearby EMG channels to be included in each cluster, to equal the number of unipolar or bipolar data channels recorded from each thread. Finally, note that in all cases Kilosort parameter NchanNearUp (which defines the maximum number of channels across which spike templates can appear) must be reset to be equal to or less than the total number of Myomatrix data channels.”

“Allowing more complex spike waveforms: We also modified Kilosort to account for the greater duration and complexity (relative to neural spikes) of many motor unit waveforms. In the code repository linked above, Kilosort 2.5 was modified to allow longer spike templates (151 samples instead of 61), more spatiotemporal PCs for spikes (12 instead of 6), and more left/right eigenvector pairs for spike template construction (6 pairs instead of 3). These modifications were crucial for improving sorting performance in the nonhuman primate dataset shown in Figure 3, and in a subset of the rodent datasets (although they were not used in the analysis of mouse data shown in Fig. 1 and Supplemental Fig. 2a-f).”

Furthermore, as the paper on the latest version of kilosort, namely v4, discusses, differences in the clustering algorithm is the likely reason for kilosort4 performing more robustly than kilosort2.5 (used in the myomatrix paper). Given such dependence on details of the implementation and the use of an older kilosort version in this paper, the evidence that the myomatrix arrays truly record individual motor units under all the types of data obtained is under question.

We chose to modify Kilosort 2.5, which has been used by many research groups to sort spike features, rather than the just-released Kilosort 4.0. Although future studies might directly compare the performance of these two versions on sorting motor unit data, we feel that such an analysis is beyond the scope of this paper, which aims primarily to introduce our electrode technology and demonstrate that a wide range of sorting methods (thresholding, PCA-based waveform clustering, and Kilosort) can all be used to extract single motor units. Additionally, note that because we have made several significant modifications to Kilosort 2.5 as described above, it is not clear what a “direct” comparison between different Kilosort versions would mean, since the procedures we provide here are no longer identical to version 2.5.

There is an older paper with a similar goal to use multi-channel recording to perform sourcelocalization that the authors have failed to discuss. Given the striking similarity of goals and the divergence of approaches (the older paper uses a surface electrode array), it is important to know the relationship of the myomatrix array to the previous work. Like myomatrix arrays, the previous work also derives inspiration from cortical recordings, in that case it uses the approach of source localization in large-scale EEG recordings using skull caps, but applies it to surface EMG arrays. Ref: van den Doel, K., Ascher, U. M., & Pai, D. K. (2008). Computed myography: three-dimensional reconstruction of motor functions from surface EMG data. Inverse Problems, 24(6), 065010.

We thank the Reviewer for pointing out this important prior work, which we now cite and discuss in the revised manuscript under “Data analysis: spike sorting” [lines 318-333]:

“Our approach to spike sorting shares the same ultimate goal as prior work using skin-surface electrode arrays to isolate signals from individual motor units but pursues this goal using different hardware and analysis approaches. A number of groups have developed algorithms for reconstructing the spatial location and spike times of active motor units (Negro et al. 2016; van den Doel, Ascher, and Pai 2008) based on skin-surface recordings, in many cases drawing inspiration from earlier efforts to localize cortical activity using EEG recordings from the scalp (Michel et al. 2004). Our approach differs substantially. In Myomatrix arrays, the close electrode spacing and very close proximity of the contacts to muscle fibers ensure that each Myomatrix channel records from a much smaller volume of tissue than skin-surface arrays. This difference in recording volume in turn creates different challenges for motor unit isolation: compared to skin-surface recordings, Myomatrix recordings include a smaller number of motor units represented on each recording channel, with individual motor units appearing on a smaller fraction of the sensors than typical in a skin-surface recording. Because of this sensordependent difference in motor unit source mixing, different analysis approaches are required for each type of dataset. Specifically, skin-surface EMG analysis methods typically use source-separation approaches that assume that each sensor receives input from most or all of the individual sources within the muscle as is presumably the case in the data. In contrast, the much sparser recordings from Myomatrix are better decomposed using methods like Kilosort, which are designed to extract waveforms that appear only on a small, spatially-restricted subset of recording channels.”

The incompleteness of the evidence that the myomatrix array truly measures individual motor units is limited to the setting where multiple motor units have similar magnitude of signal in most of the channels. In the simpler data setting where one motor dominates in some channel (this seems to occur with some regularity), the myomatrix array is a major advance in our ability to understand the motor output of the nervous system. The paper is a trove of innovations in manufacturing technique, array design, suture and other fixation devices for long-term signal stability, and customization for different muscle sizes, locations, and organisms. The technology presented here is likely to achieve rapid adoption in multiple groups that study motor behavior, and would probably lead to new insights into the spatiotemporal distribution of the motor output under more naturally behaving animals than is the current state of the field.

We thank the Reviewer for this positive evaluation and for the critical comments above.

Reviewer #2 (Public Review):

Motoneurons constitute the final common pathway linking central impulse traffic to behavior, and neurophysiology faces an urgent need for methods to record their activity at high resolution and scale in intact animals during natural movement. In this consortium manuscript, Chung et al. introduce highdensity electrode arrays on a flexible substrate that can be implanted into muscle, enabling the isolation of multiple motor units during movement. They then demonstrate these arrays can produce high-quality recordings in a wide range of species, muscles, and tasks. The methods are explained clearly, and the claims are justified by the data. While technical details on the arrays have been published previously, the main significance of this manuscript is the application of this new technology to different muscles and animal species during naturalistic behaviors. Overall, we feel the manuscript will be of significant interest to researchers in motor systems and muscle physiology, and we have no major concerns. A few minor suggestions for improving the manuscript follow.

We thank the Reviewer for this positive overall assessment.

The authors perhaps understate what has been achieved with classical methods. To further clarify the novelty of this study, they should survey previous approaches for recording from motor units during active movement. For example, Pflüger & Burrows (J. Exp. Biol. 1978) recorded from motor units in the tibial muscles of locusts during jumping, kicking, and swimming. In humans, Grimby (J. Physiol. 1984) recorded from motor units in toe extensors during walking, though these experiments were most successful in reinnervated units following a lesion. In addition, the authors might briefly mention previous approaches for recording directly from motoneurons in awake animals (e.g., Robinson, J. Neurophys. 1970; Hoffer et al., Science 1981).

We agree and have revised the manuscript to discuss these and other prior use of traditional EMG, including here [lines 164-167]:

“The diversity of applications presented here demonstrates that Myomatrix arrays can obtain highresolution EMG recordings across muscle groups, species, and experimental conditions including spontaneous behavior, reflexive movements, and stimulation-evoked muscle contractions. Although this resolution has previously been achieved in moving subjects by directly recording from motor neuron cell bodies in vertebrates (Hoffer et al. 1981; Robinson 1970; Hyngstrom et al. 2007) and by using fine-wire electrodes in moving insects (Pfluger 1978; Putney et al. 2023), both methods are extremely challenging and can only target a small subset of species and motor unit populations. Exploring additional muscle groups and model systems with Myomatrix arrays will allow new lines of investigation into how the nervous system executes skilled behaviors and coordinates the populations of motor units both within and across individual muscles…

For chronic preparations, additional data and discussion of the signal quality over time would be useful. Can units typically be discriminated for a day or two, a week or two, or longer?

A related issue is whether the same units can be tracked over multiple sessions and days; this will be of particular significance for studies of adaptation and learning.

Although the yields of single units are greatest in the 1-2 weeks immediately following implantation, in chronic preparations we have obtained well-isolated single units up to 65 days post-implant. Anecdotally, in our chronic mouse implants we occasionally see motor units on the same channel across multiple days with similar waveform shapes and patterns of behavior-locked activity. However, because data collection for this manuscript was not optimized to answer this question, we are unable to verify whether these observations actually reflect cross-session tracking of individual motor units. For example, in all cases animals were disconnected from data collection hardware in between recording sessions (which were often separated by multiple intervening days) preventing us from continuously tracking motor units across long timescales. We agree with the reviewer that long-term motor unit tracking would be extremely useful as a tool for examining learning and plan to address this question in future studies.

We have added a discussion of these issues to the revised manuscript [lines 52-59]:

“…These methods allow the user to record simultaneously from ensembles of single motor units (Fig. 1c,d) in freely behaving animals, even from small muscles including the lateral head of the triceps muscle in mice (approximately 9 mm in length with a mass of 0.02 g 23). Myomatrix recordings isolated single motor units for extended periods (greater than two months, Supp. Fig. 3e), although highest unit yield was typically observed in the first 1-2 weeks after chronic implantation. Because recording sessions from individual animals were often separated by several days during which animals were disconnected from data collection equipment, we are unable to assess based on the present data whether the same motor units can be recorded over multiple days.”

Moreover, we have revised Supplemental Figure 3 to show an example of single motor units recorded >2 months after implantation:

Author response image 1.

Longevity of Myomatrix recordings In addition to isolating individual motor units, Myomatrix arrays also provide stable multi-unit recordings of comparable or superior quality to conventional fine wire EMG…. (e) Although individual motor units were most frequently recorded in the first two weeks of chronic recordings (see main text), Myomatrix arrays also isolate individual motor units after much longer periods of chronic implantation, as shown here where spikes from two individual motor units (colored boxes in bottom trace) were isolated during locomotion 65 days after implantation. This bipolar recording was collected from the subject plotted with unfilled black symbols in panel (d).

It appears both single-ended and differential amplification were used. The authors should clarify in the Methods which mode was used in each figure panel, and should discuss the advantages and disadvantages of each in terms of SNR, stability, and yield, along with any other practical considerations.

We thank the reviewer for the suggestion and have added text to all figure legends clarifying whether each recording was unipolar or bipolar.

Is there likely to be a motor unit size bias based on muscle depth, pennation angle, etc.?

Although such biases are certainly possible, the data presented here are not well-suited to answering these questions. For chronic implants in small animals, the target muscles (e.g. triceps in mice) are so small that the surgeon often has little choice about the site and angle of array insertion, preventing a systematic analysis of this question. For acute array injections in larger animals such as rhesus macaques, we did not quantify the precise orientation of the arrays (e.g. with ultrasound imaging) or the muscle fibers themselves, again preventing us from drawing strong conclusions on this topic. This question is likely best addressed in acute experiments performed on larger muscles, in which the relative orientations of array threads and muscle fibers can be precisely imaged and systematically varied to address this important issue.

Can muscle fiber conduction velocity be estimated with the arrays?

We sometimes observe fiber conduction delays up to 0.5 msec as the spike from a single motor unit moves from electrode contact to electrode contact, so spike velocity could be easily estimated given the known spatial separation between electrode contacts. However (closely related to the above question) this will only provide an accurate estimate of muscle fiber conduction velocity if the electrode contacts are arranged parallel to fiber direction, which is difficult to assess in our current dataset. If the arrays are not parallel, this computation will produce an overestimate of conduction velocity, as in the extreme case where a line of electrode contacts arranged perpendicular to the fiber direction might have identical spike arrival times, and therefore appear to have an infinite conduction velocity. Therefore, although Myomatrix arrays can certainly be used to estimate conduction velocity, such estimates should be performed in future studies only in settings where the relative orientation of array threads and muscle fibers can be accurately measured.

The authors suggest their device may have applications in the diagnosis of motor pathologies. Currently, concentric needle EMG to record from multiple motor units is the standard clinical method, and they may wish to elaborate on how surgical implantation of the new array might provide additional information for diagnosis while minimizing risk to patients.

We thank the reviewer for the suggestion and have modified the manuscript’s final paragraph accordingly [lines 182-188]:

“Applying Myomatrix technology to human motor unit recordings, particularly by using the minimally invasive injectable designs shown in Figure 3 and Supplemental Figure 1g,i, will create novel opportunities to diagnose motor pathologies and quantify the effects of therapeutic interventions in restoring motor function. Moreover, because Myomatrix arrays are far more flexible than the rigid needles commonly used to record clinical EMG, our technology might significantly reduce the risk and discomfort of such procedures while also greatly increasing the accuracy with which human motor function can be quantified. This expansion of access to high-resolution EMG signals – across muscles, species, and behaviors – is the chief impact of the Myomatrix project.”

Reviewer #3 (Public Review):

This work provides a novel design of implantable and high-density EMG electrodes to study muscle physiology and neuromotor control at the level of individual motor units. Current methods of recording EMG using intramuscular fine-wire electrodes do not allow for isolation of motor units and are limited by the muscle size and the type of behavior used in the study. The authors of Myomatrix arrays had set out to overcome these challenges in EMG recording and provided compelling evidence to support the usefulness of the new technology.

Strengths:

They presented convincing examples of EMG recordings with high signal quality using this new technology from a wide array of animal species, muscles, and behavior.

• The design included suture holes and pull-on tabs that facilitate implantation and ensure stable recordings over months.

• Clear presentation of specifics of the fabrication and implantation, recording methods used, and data analysis.

We thank the Reviewer for these comments.

Weaknesses:

The justification for the need to study the activity of isolated motor units is underdeveloped. The study could be strengthened by providing example recordings from studies that try to answer questions where isolation of motor unit activity is most critical. For example, there is immense value for understanding muscles with smaller innervation ratio which tend to have many motor neurons for fine control of eyes and hand muscles.

We thank the Reviewer for the suggestion and have modified the manuscript accordingly [lines 170-174]:

“…how the nervous system executes skilled behaviors and coordinates the populations of motor units both within and across individual muscles. These approaches will be particularly valuable in muscles in which each motor neuron controls a very small number of muscle fibers, allowing fine control of oculomotor muscles in mammals as well as vocal muscles in songbirds (Fig. 2g), in which most individual motor neurons innervate only 1-3 muscle fibers (Adam et al. 2021).”

Reviewer #1 (Recommendations for The Authors):

I would urge the authors to consider a thorough validation of the spike sorting piece of the workflow. Barring that weakness, this paper has the potential to transform motor neuroscience. The validation efforts of kilosort in the context of Neuropixels might offer a template for how to convince the community of the accuracy of myomatrix arrays in disambiguating individual motor unit waveforms.

I have a few minor detailed comments, that the authors may find of some use. My overall comment is to commend the authors for the precision of the work as well as the writing. However, exercising caution associated with kilosort could truly elevate the paper by showing where there is room for improvement.

We thank the Reviewer for these comments - please see our summary of our revisions related to Kilosort in our reply to the public reviews above.

L6-7: The relationship between motor unit action potential and the force produced is quite complicated in muscle. For example, recent work has shown how decoupled the force and EMG can be during nonsteady locomotion. Therefore, it is not a fully justified claim that recording motor unit potentials will tell us what forces are produced. This point relates to another claim made by the authors (correctly) that EMG provides better quality information about muscle motor output in isometric settings than in more dynamic behaviors. That same problem could also apply to motor unit recordings and their relationship to muscle force. The relationship is undoubtedly strong in an isometric setting. But as has been repeatedly established, the electrical activity of muscle is only loosely related to its force output and lacks in predictive power.

This is an excellent point, and our revised manuscript now addresses this issue [lines 174-176]:

“…Of further interest will be combining high-resolution EMG with precise measurement of muscle length and force output to untangle the complex relationship between neural control, body kinematics, and muscle force that characterizes dynamic motor behavior. Similarly, combining Myomatrix recordings with high-density brain recordings….”

L12: There is older work that uses an array of skin mounted EMG electrodes to solve a source location problem, and thus come quite close to the authors' stated goals. However, the authors have failed to cite or provide an in-depth analysis and discussion of this older work.

As described above in the response to Reviewer 1’s public review comments, we now cite and discuss these papers.

L18-19: "These limitations have impeded our understanding of fundamental questions in motor control, ..." There are two independently true statements here. First is that there are limitations to EMG based inference of motor unit activity. Second is that there are gaps in the current understanding of motor unit recruitment patterns and modification of these patterns during motor learning. But the way the first few paragraphs have been worded makes it seem like motor unit recordings is a panacea for these gaps in our knowledge. That is not the case for many reasons, including key gaps in our understanding of how muscle's electrical activity relates to its force, how force relates to movement, and how control goals map to specific movement patterns. This manuscript would in fact be strengthened by acknowledging and discussing the broader scope of gaps in our understanding, and thus more precisely pinpointing the specific scientific knowledge that would be gained from the application of myomatrix arrays.

We agree and have revised the manuscript to note this complexity (see our reply to this Reviewer’s other comment about muscle force, above).

L140-143: The estimation algorithms yields potential spikes but lacking the validation of the sorting algorithms, it is not justifiable to conclude that the myomatrix arrays have already provided information about individual motor units.

Please see our replies to Reviewer #1s public comments (above) regarding motor unit spike sorting.

L181-182: "These methods allow very fine pitch escape routing (<10 µm spacing), alignment between layers, and uniform via formation." I find this sentence hard to understand. Perhaps there is some grammatical ambiguity?

We have revised this passage as follows [lines 194-197]:

"These methods allow very fine pitch escape routing (<10 µm spacing between the thin “escape” traces connecting electrode contacts to the connector), spatial alignment between the multiple layers of polyimide and gold that constitute each device, and precise definition of “via” pathways that connect different layers of the device.”

L240: What is the rationale for choosing this frequency band for the filter?

Individual motor unit waveforms have peak energy at roughly 0.5-2.0 kHz, although units recorded at very high SNR often have voltage waveform features at higher frequencies. The high- and lowpass cutoff frequencies should reflect this, although there is nothing unique about the 350 Hz and 7,000 Hz cutoffs we describe, and in all recordings similar results can be obtained with other choices of low/high frequency cutoffs.

L527-528: There are some key differences between the electrode array design presented here and traditional fine-wire EMG in terms of features used to help with electrode stability within the muscle. A barb-like structure is formed in traditional fine-wire EMG by bending the wire outside the canula of the needle used to place it within the muscle. But when the wire is pulled out, it is common for the barb to break off and be left behind. This is because of the extreme (thin) aspect ratio of the barb in fine wire EMG and low-cycle fatigue fracture of the wire. From the schematic shown here, the barb design seems to be stubbier and thus less prone to breaking off. This raises the question of how much damage is inflicted during the pull-out and the associated level of discomfort to the animal as a result. The authors should present a more careful statement and documentation with regard to this issue.

We have updated the manuscript to highlight the ease of inserting and removing Myomatrix probes, and to clarify that in over 100 injectable insertions/removal there have been zero cases of barbs (or any other part) of the devices breaking off within the muscle [lines 241-249]:

“…Once the cannula was fully inserted, the tail was released, and the cannula slowly removed. After recording, the electrode and tail were slowly pulled out of the muscle together. Insertion and removal of injectable Myomatrix devices appeared to be comparable or superior to traditional fine-wire EMG electrodes (in which a “hook” is formed by bending back the uninsulated tip of the recording wire) in terms of both ease of injection, ease of removal of both the cannula and the array itself, and animal comfort. Moreover, in over 100 Myomatrix injections performed in rhesus macaques, there were zero cases in which Myomatrix arrays broke such that electrode material was left behind in the recorded muscle, representing a substantial improvement over traditional fine-wire approaches, in which breakage of the bent wire tip regularly occurs (Loeb and Gans 1986).”

Reviewer #2 (Recommendations For The Authors):

The Abstract states the device records "muscle activity at cellular resolution," which could potentially be read as a claim that single-fiber recording has been achieved. The authors might consider rewording.

The Reviewer is correct, and we have removed the word “cellular”.

The supplemental figures could perhaps be moved to the main text to aid readers who prefer to print the combined PDF file.

After finalizing the paper we will upload all main-text and supplemental figures into a single pdf on biorXiv for readers who prefer a single pdf. However, given that the supplemental figures provide more technical and detailed information than the main-text figures, for the paper on the eLife site we prefer the current eLife format in which supplemental figures are associated with individual main-text figures online.

Reviewer #3 (Recommendations For The Authors):

• The work could be strengthened by showing examples of simultaneous recordings from different muscles.

Although Myomatrix arrays can indeed be used to record simultaneously from multiple muscles, in this manuscript we have decided to focus on high-resolution recordings that maximize the number of recording channels and motor units obtained from a single muscle. Future work from our group with introduce larger Myomatrix arrays optimized for recording from many muscles simultaneously.

• The implantation did not include mention of testing the myomatrix array during surgery by using muscle stimulation to verify correct placement and connection.

As the Reviewer points out electrical stimulation is a valuable tool for confirming successful EMG placement. However we did not use this approach in the current study, relying instead on anatomical confirmation of muscle targeting (e.g. intrasurgical and postmortem inspection in rodents) and by implanting large, easy-totarget arm muscles (in primates) where the risk of mis-targeting is extremely low. Future studies will examine both electrical stimulation and ultrasound methods for confirming the placement of Myomatrix arrays.

References cited above

Adam, I., A. Maxwell, H. Rossler, E. B. Hansen, M. Vellema, J. Brewer, and C. P. H. Elemans. 2021. 'One-to-one innervation of vocal muscles allows precise control of birdsong', Curr Biol, 31: 3115-24 e5.

Hoffer, J. A., M. J. O'Donovan, C. A. Pratt, and G. E. Loeb. 1981. 'Discharge patterns of hindlimb motoneurons during normal cat locomotion', Science, 213: 466-7.

Hyngstrom, A. S., M. D. Johnson, J. F. Miller, and C. J. Heckman. 2007. 'Intrinsic electrical properties of spinal motoneurons vary with joint angle', Nat Neurosci, 10: 363-9.

Loeb, G. E., and C. Gans. 1986. Electromyography for Experimentalists, First edi (The University of Chicago Press: Chicago, IL).

Michel, C. M., M. M. Murray, G. Lantz, S. Gonzalez, L. Spinelli, and R. Grave de Peralta. 2004. 'EEG source imaging', Clin Neurophysiol, 115: 2195-222.

Negro, F., S. Muceli, A. M. Castronovo, A. Holobar, and D. Farina. 2016. 'Multi-channel intramuscular and surface EMG decomposition by convolutive blind source separation', J Neural Eng, 13: 026027.

Pfluger, H. J.; Burrows, M. 1978. 'Locusts use the same basic motor pattern in swimming as in jumping and kicking', Journal of experimental biology, 75: 81-93.

Putney, Joy, Tobias Niebur, Leo Wood, Rachel Conn, and Simon Sponberg. 2023. 'An information theoretic method to resolve millisecond-scale spike timing precision in a comprehensive motor program', PLOS Computational Biology, 19: e1011170.

Robinson, D. A. 1970. 'Oculomotor unit behavior in the monkey', J Neurophysiol, 33: 393-403.

van den Doel, Kees, Uri M Ascher, and Dinesh K Pai. 2008. 'Computed myography: three-dimensional reconstruction of motor functions from surface EMG data', Inverse Problems, 24: 065010.

Author Response

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

Firstly, we must take a moment to express our sincere gratitude to editorial board for allowing this work to be reviewed, and to the peer reviewers for taking the time and effort to review our manuscript. The reviews are thoughtful and reflect the careful work of scientists who undoubtedly have many things on their schedule. We cannot express our gratitude enough. This is not a minor sentiment. We appreciate the engagement.

Allow us to briefly highlight some of the changes made to the revised manuscript, most on behalf of suggestions made by the reviewers:

1) A supplementary figure that includes the calculation of drug applicability and variant vulnerability for a different data set–16 alleles of dihydrofolate reductase, and two antifolate compounds used to treat malaria–pyrimethamine and cycloguanil.

2) New supplementary figures that add depth to the result in Figure 1 (the fitness graphs): we demonstrate how the rank order of alleles changes across drug environments and offer a statistical comparison of the equivalence of these fitness landscapes.

3) A new subsection that explains our specific method used to measure epistasis.

4) Improved main text with clarifications, fixed errors, and other addendums.

5) Improved referencing and citations, in the spirit of better scholarship (now with over 70 references).

Next, we’ll offer some general comments that we believe apply to several of the reviews, and to the eLife assessment. We have provided the bulk of the responses in some general comments, and in response to the public reviews. We have also included the suggestions and made brief comments to some of the individual recommendations.

On the completeness of our analysis

In our response, we’ll address the completeness issue first, as iterations of it appear in several of the reviews, and it seems to be one of the most substantive philosophical critiques of the work (there are virtually no technical corrections, outside of a formatting and grammar fixes, which we are grateful to the reviewers for identifying).

To begin our response, we will relay that we have now included an analysis of a data set corresponding to mutants of a protein, dihydrofolate reductase (DHFR), from Plasmodium falciparum (a main cause of malaria), across two antifolate drugs (pyrimethamine and ycloguanil). We have also decided to include this new analysis in the supplementary material (see Figure S4).

Author response image 1.

Drug applicability and variant vulnerability for 16 alleles of dihydrofolate reductase.

Here we compute the variant vulnerability and drug applicability metrics for two drugs, pyrimethamine (PYR) and cycloguanil (CYC), both antifolate drugs used to treat malaria. This is a completely different system than the one that is the focus of the submitted paper, for a different biomedical problem (antimalarial resistance), using different drugs, and targets. Further, the new data provide information on both drugs of different kinds, and drug concentrations (as suggested by Reviewer #1; we’ve also added a note about this in the new supplementary material). Note that these data have already been the subject of detailed analyses of epistatic effects, and so we did not include those here, but we do offer that reference:

● Ogbunugafor CB. The mutation effect reaction norm (mu-rn) highlights environmentally dependent mutation effects and epistatic interactions. Evolution. 2022 Feb 1;76(s1):37-48.

● Diaz-Colunga J, Sanchez A, Ogbunugafor CB. Environmental modulation of global epistasis is governed by effective genetic interactions. bioRxiv. 2022:202211.

Computing our proposed metrics across different drugs is relatively simple, and we could have populated our paper with suites of similar analyses across data sets of various kinds. Such a paper would, in our view, be spread too thin–the evolution of antifolate resistance and/or antimalarial resistance are enormous problems, with large literatures that warrant focused studies. More generally, as the reviewers doubtlessly understand, simply analyzing more data sets does not make a study stronger, especially one like ours, that is using empirical data to both make a theoretical point about alleles and drugs and offer a metric that others can apply to their own data sets.

Our approach focused on a data set that allowed us to discuss the biology of a system: a far stronger paper, a far stronger proof-of-concept for a new metric. We will revisit this discussion about the structure of our study. But before doing so, we will elaborate on why the “more is better” tone of the reviews is misguided.

We also note that study where the data originate (Mira et al. 2015) is focused on a single data set of a single drug-target system. We should also point out that Mira et al. 2015 made a general point about drug concentrations influencing the topography of fitness landscapes, not unlike our general point about metrics used to understand features of alleles and different drugs in antimicrobial systems.

This isn’t meant to serve as a feeble appeal to authority – just because something happened in one setting doesn’t make it right for another. But other than a nebulous appeal to the fact that things have changed in the 8 years since that study was published, it is difficult to argue why one study system was permissible for other work but is somehow “incomplete” in ours. Double standards can be appropriate when they are justified, but in this case, it hasn’t been made clear, and there is no technical basis for it.

Our study does what countless other successful ones do: utilizes a biological system to make a general point about some phenomena in the natural world. In our case, we were focused on the need for more evolution-inspired iterations of widely used concepts like druggability. For example, a recent study of epistasis focused on a single set of alleles, across several drugs, not unlike our study:

● Lozovsky ER, Daniels RF, Heffernan GD, Jacobus DP, Hartl DL. Relevance of higher-order epistasis in drug resistance. Molecular biology and evolution. 2021 Jan;38(1):142-51.

Next, we assert that there is a difference between an eagerness to see a new metric applied to many different data sets (a desire we share, and plan on pursuing in the future), and the notion that an analysis is “incomplete” without it. The latter is a more serious charge and suggests that the researcher-authors neglected to properly construct an argument because of gaps in the data. This charge does not apply to our manuscript, at all. And none of the reviewers effectively argued otherwise.

Our study contains 7 different combinatorially-complete datasets, each composed of 16 alleles (this not including the new analysis of antifolates that now appear in the revision). One can call these datasets “small” or “low-dimensional,” if they choose (we chose to put this front-and-center, in the title). They are, however, both complete and as large or larger than many datasets in similar studies of fitness landscapes:

● Knies JL, Cai F, Weinreich DM. Enzyme efficiency but not thermostability drives cefotaxime resistance evolution in TEM-1 β-lactamase. Molecular biology and evolution. 2017 May 1;34(5):1040-54.

● Lozovsky ER, Daniels RF, Heffernan GD, Jacobus DP, Hartl DL. Relevance of higher-order epistasis in drug resistance. Molecular biology and evolution. 2021 Jan;38(1):142-51.

● Rodrigues JV, Bershtein S, Li A, Lozovsky ER, Hartl DL, Shakhnovich EI. Biophysical principles predict fitness landscapes of drug resistance. Proceedings of the National Academy of Sciences. 2016 Mar 15;113(11):E1470-8.

● Ogbunugafor CB, Eppstein MJ. Competition along trajectories governs adaptation rates towards antimicrobial resistance. Nature ecology & evolution. 2016 Nov 21;1(1):0007.

● Lindsey HA, Gallie J, Taylor S, Kerr B. Evolutionary rescue from extinction is contingent on a lower rate of environmental change. Nature. 2013 Feb 28;494(7438):463-7.

These are only five of very many such studies, some of them very well-regarded.

Having now gone on about the point about the data being “incomplete,” we’ll next move to the more tangible comment-criticism about the low-dimensionality of the data set, or the fact that we examined a single drug-drug target system (β lactamases, and β-lactam drugs).

The criticism, as we understand it, is that the authors could have analyzed more data,

This is a common complaint, that “more is better” in biology. While we appreciate the feedback from the reviewers, we notice that no one specified what constitutes the right amount of data. Some pointed to other single data sets, but would analyzing two different sets qualify as enough? Perhaps to person A, but not to persons B - Z. This is a matter of opinion and is not a rigorous comment on the quality of the science (or completeness of the analysis).

● Should we analyze five more drugs of the same target (beta lactamases)? And what bacterial orthologs?

● Should we analyze 5 antifolates for 3 different orthologs of dihydrofolate reductase?

● And in which species or organism type? Bacteria? Parasitic infections?

● And why only infectious disease? Aren’t these concepts also relevant to cancer? (Yes, they are.)

● And what about the number of variants in the aforementioned target? Should one aim for small combinatorially complete sets? Or vaster swaths of sequence space, such as the ones generated by deep mutational scanning and other methods?

I offer these options in part because, for the most part, were not given an objective suggestion for appropriate level of detail. This is because there is no answer to the question of what size of dataset would be most appropriate. Unfortunately, without a technical reason why a data set of unspecified size [X] or [Y] is best, then we are left with a standard “do more work” peer review response, one that the authors are not inclined to engage seriously, because there is no scientific rationale for it.

The most charitable explanation for why more datasets would be better is tied to the abstract notion that seeing a metric measured in different data sets somehow makes it more believable. This, as the reviewers undoubtedly understand, isn’t necessarily true (in fact, many poor studies mask a lack of clarity with lots of data).

To double down on this take, we’ll even argue the opposite: that our focus on a single drug system is a strength of the study.

The focus on a single-drug class allows us to practice the lost art of discussing the peculiar biology of the system that we are examining. Even more, the low dimensionality allows us to discuss–in relative detail–individual mutations and suites of mutations. We do so several times in the manuscript, and even connect our findings to literature that has examined the biophysical consequences of mutations in these very enzymes.

(For example: Knies JL, Cai F, Weinreich DM. Enzyme efficiency but not thermostability drives cefotaxime resistance evolution in TEM-1 β-lactamase. Molecular biology and evolution. 2017 May 1;34(5):1040-54.)

Such detail is only legible in a full-length manuscript because we were able to interrogate a system in good detail. That is, the low-dimensionality (of a complete data set) is a strength, rather than a weakness. This was actually part of the design choice for the study: to offer a new metric with broad application but developed using a system where the particulars could be interrogated and discussed.

Surely the findings that we recover are engineered for broader application. But to suggest that we need to apply them broadly in order to demonstrate their broad impact is somewhat antithetical to both model systems research and to systems biology, both of which have been successful in extracting general principles for singular (often simple) systems and models.

An alternative approach, where the metric was wielded across an unspecified number of datasets would lend to a manuscript that is unfocused, reading like many modern machine learning papers, where the analysis or discussion have little to do with actual biology. We very specifically avoided this sort of study.

To close our comments regarding data: Firstly, we have considered the comments and analyzed a different data set, corresponding to a different drug-target system (antifolate drugs, and DHFR). Moreover, we don’t think more data has anything to do with a better answer or support for our conclusions or any central arguments. Our arguments were developed from the data set that we used but achieve what responsible systems biology does: introduces a framework that one can apply more broadly. And we develop it using a complete, and well-vetted dataset. If the reviewers have a philosophical difference of opinion about this, we respect it, but it has nothing to do with our study being “complete” or not. And it doesn’t speak to the validity of our results.

Related: On the dependence of our metrics on drug-target system

Several comments were made that suggest the relevance of the metric may depend on the drug being used. We disagree with this, and in fact, have argued the opposite: the metrics are specifically useful because they are not encumbered with unnecessary variables. They are the product of rather simple arithmetic that is completely agnostic to biological particulars.

We explain, in the section entitled “Metric Calculations:

“To estimate the two metrics we are interested in, we must first quantify the susceptibility of an allelic variant to a drug. We define susceptibility as $1 - w$, where w is the mean growth of the allelic variant under drug conditions relative to the mean growth of the wild-type/TEM-1 control. If a variant is not significantly affected by a drug (i.e., growth under drug is not statistically lower than growth of wild-type/TEM-1 control, by t-test P-value < 0.01), its susceptibility is zero. Values in these metrics are summaries of susceptibility: the variant vulnerability of an allelic variant is its average susceptibility across drugs in a panel, and the drug applicability of an antibiotic is the average susceptibility of all variants to it.”

That is, these can be animated to compute the variant vulnerability and drug applicability for data sets of various kinds. To demonstrate this (and we thank the reviewers for suggesting it), we have analyzed the antifolate-DHFR data set as outlined above.

Finally, we will make the following light, but somewhat cynical point (that relates to the “more data” more point generally): the wrong metric applied to 100 data sets is little more than 100 wrong analyses. Simply applying the metric to a wide number of datasets has nothing to do with the veracity of the study. Our study, alternatively, chose the opposite approach: used a data set for a focused study where metrics were extracted. We believe this to be a much more rigorous way to introduce new metrics.

On the Relevance of simulations

Somewhat relatedly, the eLife summary and one of the reviewers mentioned the potential benefit of simulations. Reviewer 1 correctly highlights that the authors have a lot of experience in this realm, and so generating simulations would be trivial. For example, the authors have been involved in studies such as these:

● Ogbunugafor CB, Eppstein MJ. Competition along trajectories governs adaptation rates towards antimicrobial resistance. Nature ecology & evolution. 2016 Nov 21;1(1):0007.

● Ogbunugafor CB, Wylie CS, Diakite I, Weinreich DM, Hartl DL. Adaptive landscape by environment interactions dictate evolutionary dynamics in models of drug resistance. PLoS computational biology. 2016 Jan 25;12(1):e1004710.

● Ogbunugafor CB, Hartl D. A pivot mutation impedes reverse evolution across an adaptive landscape for drug resistance in Plasmodium vivax. Malaria Journal. 2016 Dec;15:1-0.

From the above and dozens of other related studies, we’ve learned that simulations are critical for questions about the end results of dynamics across fitness landscapes of varying topography. To simulate across the datasets in the submitted study would be be a small ask. We do not provide this, however, because our study is not about the dynamics of de novo evolution of resistance. In fact, our study focuses on a different problem, no less important for understanding how resistance evolves: determining static properties of alleles and drugs, that provide a picture into their ability to withstand a breadth of drugs in a panel (variant vulnerability), or the ability of a drug in a panel to affect a breadth of drug targets.

The authors speak on this in the Introduction:

“While stepwise, de novo evolution (via mutations and subsequent selection) is a key force in the evolution of antimicrobial resistance, evolution in natural settings often involves other processes, including horizontal gene transfer and selection on standing genetic variation. Consequently, perspectives that consider variation in pathogens (and their drug targets) are important for understanding treatment at the bedside. Recent studies have made important strides in this arena. Some have utilized large data sets and population genetics theory to measure cross-resistance and collateral sensitivity. Fewer studies have made use of evolutionary concepts to establish metrics that apply to the general problem of antimicrobial treatment on standing genetic variation in pathogen populations, or for evaluating the utility of certain drugs’ ability to treat the underlying genetic diversity of pathogens”

That is, the proposed metrics aren’t about the dynamics of stepwise evolution across fitness landscapes, and so, simulating those dynamics don’t offer much for our question. What we have done instead is much more direct and allows the reader to follow a logic: clearly demonstrate the topography differences in Figure 1 (And Supplemental Figure S2 and S3 with rank order changes).

Author response image 2.

These results tell the reader what they need to know: that the topography of fitness landscapes changes across drug types. Further, we should note that Mira et al. 2015 already told the basic story that one finds different adaptive solutions across different drug environments. (Notably, without computational simulations).

In summary, we attempted to provide a rigorous, clean, and readable study that introduced two new metrics. Appeals to adding extra analysis would be considered if they augmented the study’s goals. We do not believe this to be the case.

Nonetheless, we must reiterate our appreciation for the engagement and suggestions. All were made with great intentions. This is more than one could hope for in a peer review exchange. The authors are truly grateful.

eLife assessment

The work introduces two valuable concepts in antimicrobial resistance: "variant vulnerability" and "drug applicability", which can broaden our ways of thinking about microbial infections through evolution-based metrics. The authors present a compelling analysis of a published dataset to illustrate how informative these metrics can be, study is still incomplete, as only a subset of a single dataset on a single class of antibiotics was analyzed. Analyzing more datasets, with other antibiotic classes and resistance mutations, and performing additional theoretical simulations could demonstrate the general applicability of the new concepts.

The authors disagree strongly with the idea that the study is ‘incomplete,” and encourage the editors and reviewers to reconsider this language. Not only are the data combinatorially complete, but they are also larger in size than many similar studies of fitness landscapes. Insofar as no technical justification was offered for this “incomplete” summary, we think it should be removed. Furthermore, we question the utility of “theoretical simulations.” They are rather easy to execute but distract from the central aims of the study: to introduce new metrics, in the vein of other metrics–like druggability, IC50, MIC–that describe properties of drugs or drug targets.

Public Reviews:

Reviewer #1 (Public Review):

The manuscript by Geurrero and colleagues introduces two new metrics that extend the concept of "druggability"- loosely speaking, the potential suitability of a particular drug, target, or drug-target interaction for pharmacological intervention-to collections of drugs and genetic variants. The study draws on previously measured growth rates across a combinatoriality complete mutational landscape involving 4 variants of the TEM-50 (beta lactamase) enzyme, which confers resistance to commonly used beta-lactam antibiotics. To quantify how growth rate - in this case, a proxy for evolutionary fitness - is distributed across allelic variants and drugs, they introduce two concepts: "variant vulnerability" and "drug applicability".

Variant vulnerability is the mean vulnerability (1-normalized growth rate) of a particular variant to a library of drugs, while drug applicability measures the mean across the collection of genetic variants for a given drug. The authors rank the drugs and variants according to these metrics. They show that the variant vulnerability of a particular mutant is uncorrelated with the vulnerability of its one-step neighbors and analyze how higher-order combinations of single variants (SNPs) contribute to changes in growth rate in different drug environments.

The work addresses an interesting topic and underscores the need for evolutionbased metrics to identify candidate pharmacological interventions for treating infections. The authors are clear about the limitations of their approach - they are not looking for immediate clinical applicability - and provide simple new measures of druggability that incorporate an evolutionary perspective, an important complement to the orthodoxy of aggressive, kill-now design principles. I think the ideas here will interest a wide range of readers, but I think the work could be improved with additional analysis - perhaps from evolutionary simulations on the measured landscapes - that tie the metrics to evolutionary outcomes.

The authors greatly appreciate these comments, and the proposed suggestions by reviewer 1. We have addressed most of the criticisms and suggestions in our comments above.

Reviewer #2 (Public Review):

The authors introduce the notions of "variant vulnerability" and "drug applicability" as metrics quantifying the sensitivity of a given target variant across a panel of drugs and the effectiveness of a drug across variants, respectively. Given a data set comprising a measure of drug effect (such as growth rate suppression) for pairs of variants and drugs, the vulnerability of a variant is obtained by averaging this measure across drugs, whereas the applicability of a drug is obtained by averaging the measure across variants.

The authors apply the methodology to a data set that was published by Mira et al. in 2015. The data consist of growth rate measurements for a combinatorially complete set of 16 genetic variants of the antibiotic resistance enzyme betalactamase across 10 drugs and drug combinations at 3 different drug concentrations, comprising a total of 30 different environmental conditions. For reasons that did not become clear to me, the present authors select only 7 out of 30 environments for their analysis. In particular, for each chosen drug or drug combination, they choose the data set corresponding to the highest drug concentration. As a consequence, they cannot assess to what extent their metrics depend on drug concentration. This is a major concern since Mira et al. concluded in their study that the differences between growth rate landscapes measured at different concentrations were comparable to the differences between drugs. If the new metrics display a significant dependence on drug concentration, this would considerably limit their usefulness.

The authors appreciate the point about drug concentration, and it is one that the authors have made in several studies.

The quick answer is that whether the metrics are useful for drug type-concentration A or B will depend on drug type-concentration A or B. If there are notable differences in the topography of the fitness landscape across concentration, then we should expect the metrics to differ. What Reviewer #2 points out as a “major concern,” is in fact a strength of the metrics: it is agnostic with respect to type of drug, type of target, size of dataset, or topography of the fitness landscape. And so, the authors disagree: no, that drug concentration would be a major actor in the value of the metrics does not limit the utility of the metric. It is simply another variable that one can consider when computing the metrics.

As discussed above, we have analyzed data from a different data set, in a different drug-target problem (DHFR and antifolate drugs; see supplemental information). These demonstrate how the metric can be used to compute metrics across different drug concentrations.

As a consequence of the small number of variant-drug combinations that are used, the conclusions that the authors draw from their analysis are mostly tentative with weak statistical support. For example, the authors argue that drug combinations tend to have higher drug applicability than single drugs, because a drug combination ranks highest in their panel of 7. However, the effect profile of the single drug cefprozil is almost indistinguishable from that of the top-ranking combination, and the second drug combination in the data set ranks only 5th out of 7.

We reiterate our appreciation for the engagement. Reviewer #2 generously offers some technical insight on measurements of epistasis, and their opinion on the level of statistical support for our claims. The authors are very happy to engage in a dialogue about these points. We disagree rather strongly, and in addition to the general points raised above (that speak to some of this), will raise several specific rebuttals to the comments from Reviewer #2.

For one, the Reviewer #2 is free to point to what arguments have “weak statistical support.” Having read the review, we aren’t sure what this is referring to. “Weak statistical support” generally applies to findings built from underpowered studies, or designs constructed in manner that yield effect sizes or p-values that give low confidence that a finding is believable (or is replicable). This sort of problem doesn’t apply to our study for various reasons, the least of which being that our findings are strongly supported, based on a vetted data set, in a system that has long been the object of examination in studies of antimicrobial resistance.

For example, we did not argue that magnetic fields alter the topography of fitness landscapes, a claim which must stand up to a certain sort of statistical scrutiny. Alternatively, we examined landscapes where the drug environment differed statistically from the non-drug environment and used them to compute new properties of alleles and drugs.

We can imagine that the reviewer is referring to the low-dimensionality of the fitness landscapes in the study. Again: the features of the dataset are a detail that the authors put into the title of the manuscript. Further, we emphasize that it is not a weakness, but rather, allows the authors to focus, and discuss the specific biology of the system. And we responsibly explain the constraints around our study several times, though none of them have anything to do with “weak statistical support.”

Even though we aren’t clear what “weak statistical support” means as offered by Reviewer 2, the authors have nonetheless decided to provide additional analyses, now appearing in the new supplemental material.

We have included a new Figure S2, where we offer an analysis of the topography of the 7 landscapes, based on the Kendall rank order test. This texts the hypothesis that there is no correlation (concordance or discordance) between the topographies of the fitness landscapes.

Author response image 3.

Kendall rank test for correlation between the 7 fitness landscapes.

In Figure S3, we test the hypothesis that the variant vulnerability values differ. To do this, we calculate a paired t-test. These are paired by haplotype/allelic variant, so the comparisons are change in growth between drugs for each haplotype.

Author response image 4.

Paired t-tests for variant vulnerability.

To this point raised by Reviewer #2:

“For example, the authors argue that drug combinations tend to have higher drug applicability than single drugs, because a drug combination ranks highest in their panel of 7. However, the effect profile of the single drug cefprozil is almost indistinguishable from that of the top-ranking combination, and the second drug combination in the data set ranks only 5th out of 7.”

Our study does not argue that drug combinations are necessarily correlated with a higher drug applicability. Alternatively, we specifically highlight that one of the combinations does not have a high drug applicability:

“Though all seven drugs/combinations are β-lactams, they have widely varying effects across the 16 alleles. Some of the results are intuitive: for example, the drug regime with the highest drug applicability of the set—amoxicillin/clavulanic acid—is a mixture of a widely used β-lactam (amoxicillin) and a β-lactamase inhibitor (clavulanic acid) (see Table 3). We might expect such a mixture to have a broader effect across a diversity of variants. This high applicability is hardly a rule, however, as another mixture in the set, piperacillin/tazobactam, has a much lower drug applicability (ranking 5th out of the seven drugs in the set) (Table 3).”

In general, we believe that the submitted paper is responsible with regards to how it extrapolates generalities from the results. Further, the manuscript contains a specific section that explains limitations, clearly and transparently (not especially common in science). For that reason, we’d encourage reviewer #2 to reconsider their perspective. We do not believe that our arguments are built on “weak” support at all. And we did not argue anything particular about drug combinations writ large. We did the opposite— discussed the particulars of our results in light of the biology of the system.

Thirdly, to this point:

“To assess the environment-dependent epistasis among the genetic mutations comprising the variants under study, the authors decompose the data of Mira et al. into epistatic interactions of different orders. This part of the analysis is incomplete in two ways. First, in their study, Mira et al. pointed out that a fairly large fraction of the fitness differences between variants that they measured were not statistically significant, which means that the resulting fitness landscapes have large statistical uncertainties. These uncertainties should be reflected in the results of the interaction analysis in Figure 4 of the present manuscript.”

The authors are uncertain with regards to the “uncertainties” being referred to, but we’ll do our best to understand: our study utilized the 7 drug environments from Mira et al. 2015 with statistically significant differences between growth rates with and without drug. And so, this point about how the original set contained statistically insignificant treatments is not relevant here. We explain this in the methods section:

“The data that we examine comes from a past study of a combinatorial set of four mutations associated with TEM-50 resistance to β-lactam drugs [39 ]. This past study measured the growth rates of these four mutations in combination, across 15 different drugs (see Supplemental Information).”

We go on to say the following:

“We examined these data, identifying a subset of structurally similar β-lactams that also included β-lactams combined with β-lactamase inhibitors, cephalosporins and penicillins. From the original data set, we focus our analyses on drug treatments that had a significant negative effect on the growth of wild-type/TEM-1 strains (one-tailed ttest of wild-type treatment vs. control, P < 0.01). After identifying the data from the set that fit our criteria, we were left with seven drugs or combinations (concentration in μg/ml): amoxicillin 1024 μg/ ml (β-lactam), amoxicillin/clavulanic acid 1024 μg/m l (βlactam and β-lactamase inhibitor) cefotaxime 0.123 μg/ml (third-generation cephalosporin), cefotetan 0.125 μg/ml (second-generation cephalosporins), cefprozil 128 μg/ml (second-generation cephalosporin), ceftazidime 0.125 μg/ml (third-generation cephalosporin), piperacillin and tazobactam 512/8 μg/ml (penicillin and β-lactamase inhibitor). With these drugs/mixtures, we were able to embody chemical diversity in the panel.”

Again: The goal of our study was to develop metrics that can be used to analyze features of drugs and targets and disentangle these metrics into effects.

Second, the interpretation of the coefficients obtained from the epistatic decomposition depends strongly on the formalism that is being used (in the jargon of the field, either a Fourier or a Taylor analysis can be applied to fitness landscape data). The authors need to specify which formalism they have employed and phrase their interpretations accordingly.

The authors appreciate this nuance. Certainly, how to measure epistasis is a large topic of its own. But we recognize that we could have addressed this more directly and have added text to this effect.

In response to these comments from Reviewer #2, we have added a new section focused on these points (reference syntax removed here for clarity; please see main text for specifics):

“The study of epistasis, and discussions regarding the means to detect and measure now occupies a large corner of the evolutionary genetics literature. The topic has grown in recent years as methods have been applied to larger genomic data sets, biophysical traits, and the "global" nature of epistatic effects. We urge those interested in more depth treatments of the topic to engage larger summaries of the topic.”

“Here will briefly summarize some methods used to study epistasis on fitness landscapes. Several studies of combinatorially-complete fitness landscapes use some variation of Fourier Transform or Taylor formulation. One in particular, the Walsh-Hadamard Transform has been used to measure epistasis across a wide number of study systems. Furthermore, studies have reconciled these methods with others, or expanded upon the Walsh-Hadamard Transform in a way that can accommodate incomplete data sets. These methods are effective for certain sorts of analyses, and we strongly urge those interested to examine these studies.”

“The method that we've utilized, the LASSO regression, determines effect sizes for all interactions (alleles and drug environments). It has been utilized for data sets of similar size and structure, on alleles resistant to trimethoprim. Among many benefits, the method can accommodate gaps in data and responsibly incorporates experimental noise into the calculation.”

As Reviewer #2 understands, there are many ways to examine epistasis on both high and low-dimensional landscapes. Reviewer #2 correctly offers two sorts of formalisms that allow one to do so. The two offered by Reviewer #2, are not the only means of measuring epistasis in data sets like the one we have offered. But we acknowledge that we could have done a better job outlining this. We thank Reviewer #2 for highlighting this, and believe our revision clarifies this.

Reviewer #3 (Public Review):

The authors introduce two new concepts for antimicrobial resistance borrowed from pharmacology, "variant vulnerability" (how susceptible a particular resistance gene variant is across a class of drugs) and "drug applicability" (how useful a particular drug is against multiple allelic variants). They group both terms under an umbrella term "drugability". They demonstrate these features for an important class of antibiotics, the beta-lactams, and allelic variants of TEM-1 beta-lactamase.

The strength of the result is in its conceptual advance and that the concepts seem to work for beta-lactam resistance. However, I do not necessarily see the advance of lumping both terms under "drugability", as this adds an extra layer of complication in my opinion.

Firstly, the authors greatly appreciate the comments from Reviewer #3. They are insightful, and prescriptive. And allow us to especially thank reviewer 3 for supplying a commented PDF with some grammatical and phrasing suggestions/edits. This is much appreciated. We have examined all these suggestions and made changes.

In general, we agree with the spirit of many of the comments. In addition to our prior comments on the scope of our data, we’ll communicate a few direct responses to specific points raised.

I also think that the utility of the terms could be more comprehensively demonstrated by using examples across different antibiotic classes and/or resistance genes. For instance, another good model with published data might have been trimethoprim resistance, which arises through point mutations in the folA gene (although, clinical resistance tends to be instead conferred by a suite of horizontally acquired dihydrofolate reductase genes, which are not so closely related as the TEM variants explored here).

1. In our new supplemental material, we now feature an analysis of antifolate drugs, pyrimethamine and cycloguanil. We have discussed this in detail above and thank the reviewer for the suggestion.

2. Secondly, we agree that the study will have a larger impact when the metrics are applied more broadly. This is an active area of investigation, and our hope is that others apply our metrics more broadly. But as we discussed, such a desire is not a technical criticism of our own study. We stand behind the rigor and insight offered by our study.

The impact of the work on the field depends on a more comprehensive demonstration of the applicability of these new concepts to other drugs.

The authors don’t disagree with this point, which applies to virtually every potentially influential study. The importance of a single study can generally only be measured by its downstream application. But this hardly qualifies as a technical critique of our study and does not apply to our study alone. Nor does it speak to the validity of our results. The authors share this interest in applying the metric more broadly.

Reviewer #1 (Recommendations For The Authors):

• The main weakness of the work, in my view, is that it does not directly tie these new metrics to a quantitative measure of "performance". The metrics have intuitive appeal, and I think it is likely that they could help guide treatment options-for example, drugs with high applicability could prove more useful under particular conditions. But as the authors note, the landscape is rugged and intuitive notions of evolutionary behavior can sometimes fail. I think the paper would be much improved if the authors could evaluate their new metrics using some type of quantitative evolutionary model. For example, perhaps the authors could simulate evolutionary dynamics on these landscapes in the presence of different drugs. Is the mean fitness achieved in the simulations correlated with, for example, the drug applicability when looking across an ensemble of simulations with the same drug but varied initial conditions that start from each individual variant? Similarly, if you consider an ensemble of simulations where each member starts from the same variant but uses a different drug, is the average fitness gain captured in some way by the variant vulnerability? All simulations will have limitations, of course, but given that the landscape is fully known I think these questions could be answered under some conditions (e.g. strong selection weak mutation limit, where the model could be formulated as a Markov Chain; see 10.1371/journal.pcbi.1004493 or doi: 10.1111/evo.14121 for examples). And given the authors' expertise in evolutionary dynamics, I think it could be achieved in a reasonable time. With that said, I want to acknowledge that with any new "metrics", it can be tempting to think that "we need to understand it all" before it is useful, and I don't want to fall into that trap here.

The authors respect and appreciate these thoughtful comments.

As Reviewer #1 highlighted, the authors are experienced with building simulations of evolution. For reasons we have outlined above, we don’t believe they would add to the arc of the current story and may encumber the story with unnecessary distractions. Simulations of evolution can be enormously useful for studies focused on particulars of the dynamics of evolution. This submitted study is not one of those. It is charged with identifying features of alleles and drugs that capture an allele’s vulnerability to treatment (variant vulnerability) and a drug’s effectiveness across alleles (drug applicability). Both features integrate aspects of variation (genetic and environmental), and as such, are improvements over both metrics used to describe drug targets and drugs.

• The new metrics rely on means, which is a natural choice. Have the authors considered how variance (or other higher moments) might also impact evolutionary dynamics? I would imagine, for example, that the ultimate outcome of a treatment might depend heavily on the shape of the distribution, not merely its mean. This is also something one might be able to get a handle on with simulations.

These are relevant points, and the authors appreciate them. Certainly, moments other than the mean might have utility. This is the reason that we computed the one-step neighborhood variant vulnerability–to see if the variant vulnerability of an allele was related to properties of its mutational neighborhood. We found no such correlation. There are many other sorts of properties that one might examine (e.g., shape of the distribution, properties of mutational network, variance, fano factor, etc). As we don’t have an informed reason to pursue any of this in lieu of others, we are pleased to investigate this in the future.

Also, while we’ve addressed general points about simulations above, we want to note that our analysis of environmental epistasis does consider the variance. We urge Reviewer #1 to see our new section on “Notes on Methods Used to Measure Epistasis” where we explain some of this and supply references to that effect.

• As I understand it, the fitness measurements here are measures of per capita growth rate, which is reasonable. However, the authors may wish to briefly comment on the limitations of this choice-i.e. the fact that these are not direct measures of relative fitness values from head-to-head competition between strains.

Reviewer #1 is correct: the metrics are computed from means. As Reviewer 1 definitely understands, debates over what measurements are proper proxies for fitness go back a long time. We added a slight acknowledgement about the existence of multiple fitness proxies in our revision.

• The authors consider one-step variant vulnerability. Have the authors considered looking at 2-step, 3-step, etc analogs of the 1-step vulnerability? I wonder if these might suggest potential vulnerability bottlenecks associated with the use of a particular drug/drug combo or trajectories starting from particular variants.

This is an interesting point. We provided one-step values as a means of interrogating the mutational neighborhood of alleles in the fitness landscape. While there could certainly be other pattern-relationships between the variant vulnerability and features of a fitness landscape (as the reviewer recognizes), we don’t have a rigorous reason to test them, other than an appeal to “I would be curious if [Blank].” As in, attempting to saturate the paper with these sorts of examinations might be fun, could turn up an interesting result, but this is true for most studies.

To highlight just how serious we are about future questions along these lines, we’ll offer one specific question about the relationship between metrics and other features of alleles or landscapes. Recent studies have examined the existence of “evolvabilityenhancing mutations,” that propel a population to high-fitness sections of a fitness landscape:

● Wagner, A. Evolvability-enhancing mutations in the fitness landscapes of an RNA and a protein. Nat Commun 14, 3624 (2023). https://doi.org/10.1038/s41467023-39321-8

One present and future area of inquiry involves whether there is any relationship between metrics like variant vulnerability and these sorts of mutations.

We thank Reviewer 1 for engagement on this issue.

• Fitness values are measured in the presence of a drug, but it is not immediately clear how the drug concentrations are chosen and, more importantly, how the choice of concentration might impact the landscape. The authors may wish to briefly comment on these effects, particularly in cases where the environment involves combinations of drugs. There will be a "new" fitness landscape for each concentration, but to what extent do the qualitative features changes-or whatever features drive evolutionary dynamics--change?

This is another interesting suggestion. We have analyzed a new data set for dihydrofolate reductase mutants that contains a range of drug concentrations of two different antifolate drugs. The general question of how drug concentrations change evolutionary dynamics has been addressed in prior work of ours:

● Ogbunugafor CB, Wylie CS, Diakite I, Weinreich DM, Hartl DL. Adaptive landscape by environment interactions dictate evolutionary dynamics in models of drug resistance. PLoS computational biology. 2016 Jan 25;12(1):e1004710.

● Ogbunugafor CB, Eppstein MJ. Competition along trajectories governs adaptation rates towards antimicrobial resistance. Nature ecology & evolution. 2016 Nov 21;1(1):0007.

There are a very large number of environment types that might alter the drug availability or variant vulnerability metrics. In our study, we used an established data set composed of different alleles of a Beta lactamase, with growth rates measured across a number of drug environments. These drug environments consisted of individual drugs at certain concentrations, as outlined in Mira et al. 2015. For our study, we examined those drugs that had a significant impact on growth rate.

For a new analysis of antifolate drugs in 16 alleles of dihydrofolate reductase (Plasmodium falciparum), we have examined a breadth of drug concentrations (Supplementary Figure S4). This represents a different sort of environment that one can use to measure the two metrics (variant vulnerability or drug applicability). As we suggest in the manuscript, part of the strength of the metric is precisely that it can incorporate drug dimensions of various kinds.

• The metrics introduced depend on the ensemble of drugs chosen. To what extent are the chosen drugs representative? Are there cases where nonrepresentative ensembles might be advantageous?

The authors thank the reviewer for this. The general point has been addressed in our comments above. Further, the general question of how a study of one set of drugs applies to other drugs applies to every study of every drug, as no single study interrogates every sort of drug ensemble. That said, we’ve explained the anatomy of our metrics, and have outlined how it can be directly applied to others. There is nothing about the metric itself that has anything to do with a particular drug type – the arithmetic is rather vanilla.

Reviewer #2 (Recommendations For The Authors):

1. Regarding my comment about the different formalisms for epistatic decomposition analysis, a key reference is

Poelwijk FJ, Krishna V, Ranganathan R (2016). The Context-Dependence of Mutations: A Linkage of Formalisms. PLoS Comput Biol 12(6): e1004771.

The authors appreciate this, are fans of this work, and have cited it in the revision.

An example where both Fourier and Taylor analyses were carried out and the different interpretations of these formalisms were discussed is

Unraveling the causes of adaptive benefits of synonymous mutations in TEM-1 βlactamase. Mark P. Zwart, Martijn F. Schenk, Sungmin Hwang, Bertha Koopmanschap, Niek de Lange, Lion van de Pol, Tran Thi Thuy Nga, Ivan G. Szendro, Joachim Krug & J. Arjan G. M. de Visser Heredity 121:406-421 (2018)

The authors are grateful for these references. While we don’t think they are necessary for our new section entitled “Notes on methods used to detect epistasis,” we did engage them, and will keep them in mind for other work that more centrally focuses on methods used to detect epistasis. As the author acknowledges, a full treatment of this topic is too large for a single manuscript, let alone a subsection of one study. We have provided a discussion of it, and pointed the readers to longer review articles that explore some of these topics in good detail:

● C. Bank, Epistasis and adaptation on fitness landscapes, Annual Review of Ecology, Evolution, and Systematics 53 (1) (2022) 457–479.

● T. B. Sackton, D. L. Hartl, Genotypic context and epistasis in individuals and populations, Cell 166 (2) (2016) 279–287.

● J. Diaz-Colunga, A. Skwara, J. C. C. Vila, D. Bajic, Á. Sánchez, Global epistasis and the emergence of ecological function, BioRxviv

1. Although the authors label Figure 4 with the term "environmental epistasis", as far as I can see it is only a standard epistasis analysis that is carried out separately for each environment. The analysis of environmental epistasis should instead focus on which aspects of these interactions are different or similar in different environments, for example, by looking at the reranking of fitness values under environmental changes [see Ref.[26] as well as more recent related work, e.g. Gorter et al., Genetics 208:307-322 (2018); Das et al., eLife9:e55155 (2020)]. To some extent, such an analysis was already performed by Mira et al., but not on the level of epistatic interaction coefficients.

The authors have provided a new analysis of how fitness value rankings have changed across drug environments, often a signature of epistatic effects across environments (Supplementary Figure S1).

We disagree with the idea that our analysis is not a sort of environmental epistasis; we resolve coefficients between loci across different environments. As with every interrogation of G x E effects (G x G x E in our case), what constitutes an “environment” is a messy conversation. We have chosen the route of explaining very clearly what we mean:

“We further explored the interactions across this fitness landscape and panels of drugs in two additional ways. First, we calculated the variant vulnerability for 1-step neighbors, which is the mean variant vulnerability of all alleles one mutational step away from a focal variant. This metric gives information on how the variant vulnerability values are distributed across a fitness landscape. Second, we estimated statistical interaction effects on bacterial growth through LASSO regression. For each drug, we fit a model of relative growth as a function of M69L x E104K x G238S x N276D (i.e., including all interaction terms between the four amino acid substitutions). The effect sizes of the interaction terms from this regularized regression analysis allow us to infer higher-order dynamics for susceptibility. We label this calculation as an analysis of “environmental epistasis.”

As the grammar for these sorts of analyses continues to evolve, the best one can do is be clear about what they mean. We believe that we communicated this directly and transparently.

1. As a general comment, to strengthen the conclusions of the study, it would be good if the authors could include additional data sets in their analysis.

The authors appreciate this comment and have given this point ample treatment. Further, other main conclusions and discussion points are focused on the biology of the system that we examined. Analyzing other data sets may demonstrate the broader reach of the metrics, but it would not alter the strength of our own conclusions (or if they would, Reviewer #2 has not told us how).

1. There are some typos in the units of drug concentrations in Section 2.4 that should be corrected.

The authors truly appreciate this. It is a great catch. We have fixed this in the revised manuscript.

Reviewer #3 (Recommendations For The Authors):

I would suggest demonstrating the concepts for a second drug class, and suggest folA variants and trimethoprim resistance, for which there is existing published data similar to what the authors have used here (e.g. Palmer et al. 2015, https://doi.org/10.1038/ncomms8385)

The authors appreciate this insight. As previously described, we have analyzed a data set of folA mutants for the Plasmodium falciparum ortholog of dihydrofolate reductase, and included these results in new supplemental material. Please see the supplementary material.

There are some errors in formatting and presentation that I have annotated in a separate PDF file (https://elife-rp.msubmit.net/eliferp_files/2023/04/11/00117789/00/117789_0_attach_8_30399_convrt.pdf), as the absence of line numbers makes indicating specific things exceedingly difficult.

The authors apologize for the lack of line numbers (an honest oversight), but moreover, are tremendously grateful for this feedback. We have looked at the suggested changes carefully and have addressed many of them. Thank you.

One thing to note: we have included a version of Figure 4 that has effects on the same axes. It appears in the supplementary material (Figure S4).

In closing, the authors would like to thank the editors and three anonymous reviewers for engagement and for helpful comments. We are confident that the revised manuscript qualifies as a substantive revision, and we are grateful to have had the opportunity to participate.

Author Response

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

Public Reviews:

Reviewer #1 (Public Review):

The regulation of motor autoinhibition and activation is essential for efficient intracellular transport. This manuscript used biochemical approaches to explore two members in the kinesin-3 family. They found that releasing UNC-104 autoinhibition triggered its dimerization whereas unlocking KLP-6 autoinhibition is insufficient to activate its processive movement, which suggests that KLP-6 requires additional factors for activation, highlighting the common and diverse mechanisms underlying motor activation. They also identified a coiled-coil domain crucial for the dimerization and processive movement of UNC-104. Overall, these biochemical and single-molecule assays were well performed, and their data support their statements. The manuscript is also clearly written, and these results will be valuable to the field.

Thank you very much!

Ideally, the authors can add some in vivo studies to test the physiological relevance of their in vitro findings, given that the lab is very good at worm genetic manipulations. Otherwise, the authors should speculate the in vivo phenotypes in their Discussion, including E412K mutation in UNC-104, CC2 deletion of UNC-104, D458A in KLP-6.

1. We have shown the phenotypes unc-104(E412K) mutation in C. elegans (Niwa et al., Cell Rep, 2016) and described about it in discussion (p.14 line 3-4). The mutant worm showed overactivation of the UNC-104-dependent axonal transport, which is consistent with our biochemical data showing that UNC-104(1-653)(E412K) is prone to form a dimer and more active than wild type.

2. It has been shown that L640F mutation induces a loss of function phenotype in C. elegans (Cong et al., 2021). The amount of axonal transport is reduced in unc-104(L640F) mutant worms. L640 is located within the CC2 domain. To show the importance of CC2-dependent dimerization in the axonal transport in vivo, we biochemically investigated the impact of L640F mutation.

By introducing L640F into UNC-104(1-653)(E412K), we performed SEC analysis. The result shows that UNC-104(1-653)(E412K,L640F) failed to form stable dimers despite the release of their autoinhibition (new Figure S8). This result strongly suggests the importance of the CC2 domain in the axonal transport in vivo. Based on the result, we discussed it in the revised manuscript (p.13 line 6-8).

1. Regarding KLP-6(D458A), we need a genetic analysis using genome editing and we would like to reserve it for a future study. We speculate that the D458A mutation could lead to an increase in transport activity in vivo similar to unc-104(E412K). This is because the previous study have shown that wild-type KLP-6 was largely localized in the cell body, while KLP-6(D458A) was enriched at the cell periphery in the N2A cells (Wang et al., 2022). We described it in discussion (p.14 line 13-14).

While beyond the scope of this study, can the author speculate on the candidate for an additional regulator to activate KLP-6 in C. elegans?

The heterodimeric mechanoreceptor complex, comprising LOV-1 and PKD-2, stands as potential candidates for regulating KLP-6 dimerization. We speculate the heterodimerization property is suitable for the enhancement of KLP-6 dimerization. On the other hand, it's noteworthy that KLP-6 can undergo activation in Neuro 2a cells upon the release of autoinhibition (Wang et al., 2022). This observation implies the involvement of additional factors which are not present in sf9 cells may be able to induce dimerization. Post-translational modifications would be one of the candidates. We discussed it in p14 line 7-14.

The authors discussed the differences between their porcine brain MTs and chlamydonomas axonemes in UNC-104 assays. However, the authors did not really retest UNC-104 on axonemes after more than two decades, thereby not excluding other possibilities.

We thought that comparing different conditions used in different studies is essential for the advancement of the field of molecular motors. Therefore, we newly performed single-molecule assay using Chlamydomonas axonemes and compared the results with brain MTs (Fig. S6). Just as observed in the study by Tomoshige et al., we were also unable to observe the processive runs of UNC-104(1-653) on Chlamydomonas axonemes (Fig. S6A). Furthermore, we found that the landing rate of UNC-104(1-653) on Chlamydomonas axonemes was markedly lower in comparison to that on purified porcine microtubules (Fig. S6B).

Reviewer #1 (Recommendations For The Authors):

More discussion as suggested above would improve the manuscript.

We have improved our manuscript as described above.

Reviewer #2 (Public Review):

The Kinesin superfamily motors mediate the transport of a wide variety of cargos which are crucial for cells to develop into unique shapes and polarities. Kinesin-3 subfamily motors are among the most conserved and critical classes of kinesin motors which were shown to be self-inhibited in a monomeric state and dimerized to activate motility along microtubules. Recent studies have shown that different members of this family are uniquely activated to undergo a transition from monomers to dimers.

Niwa and colleagues study two well-described members of the kinesin-3 superfamily, unc104 and KLP6, to uncover the mechanism of monomer to dimer transition upon activation. Their studies reveal that although both Unc104 and KLP6 are both self-inhibited monomers, their propensities for forming dimers are quite different. The authors relate this difference to a region in the molecules called CC2 which has a higher propensity for forming homodimers. Unc104 readily forms homodimers if its self-inhibited state is disabled while KLP6 does not.

The work suggests that although mechanisms for self-inhibited monomeric states are similar, variations in the kinesin-3 dimerization may present a unique form of kinesin-3 motor regulation with implications on the forms of motility functions carried out by these unique kinesin-3 motors.

Thank you very much!

Reviewer #2 (Recommendations For The Authors):

The work is interesting but the process of making constructs and following the transition from monomers to dimers seems to be less than logical and haphazard. Recent crystallographic studies for kinesin-3 have shown the fold and interactions for all domains of the motor leading to the self-inhibited state. The mutations described in the manuscript leading to disabling of the monomeric self-inhibited state are referenced but not logically explained in relation to the structures. Many of the deletion constructs could also present other defects that are not presented in the mutations. The above issues prevent wide audience access to understanding the studies carried out by the authors.

We appreciate this comment. We improved it as described bellow.

Suggestions: Authors should present schematic, or structural models for the self-inhibited and dimerized states. The conclusions of the papers should be related to those models. The mutations should be explained with regard to these models and that would allow the readers easier access. Improving access to the readers in and outside the motor field would truly improve the impact of the manuscript on the field.

The structural models illustrating the autoinhibited state have been included in new Figure S4, accompanied by an explanation of the correlation between the mutations and these structures in the figure legend. Additionally, schematic models outlining the dimerization process of both UNC-104 and KLP-6 have been provided in Figure S9 to enhance reader comprehension of the process.

Reviewer #3 (Public Review):

In this work, Kita et al., aim to understand the activation mechanisms of the kinesin-3 motors KLP-6 and UNC-104 from C. elegans. As with many other motor proteins involved in intracellular transport processes, KLP-6 and UNC-104 motors suppress their ATPase activities in the absence of cargo molecules. Relieving the autoinhibition is thus a crucial step that initiates the directional transport of intracellular cargo. To investigate the activation mechanisms, the authors make use of mass photometry to determine the oligomeric states of the full-length KLP-6 and the truncated UNC-104(1-653) motors at sub-micromolar concentrations. While full-length KLP-6 remains monomeric, the truncated UNC-104(1-653) displays a sub-population of dimeric motors that is much more pronounced at high concentrations, suggesting a monomer-to-dimer conversion. The authors push this equilibrium towards dimeric UNC-104(1-653) motors solely by introducing a point mutation into the coiled-coil domain and ultimately unleashing a robust processivity of the UNC-104 dimer. The authors find that the same mechanistic concept does not apply to the KLP-6 kinesin-3 motor, suggesting an alternative activation mechanism of the KLP-6 that remains to be resolved. The present study encourages further dissection of the kinesin-3 motors with the goal of uncovering the main factors needed to overcome the 'self-inflicted' deactivation.

Thank you very much!

Reviewer #3 (Recommendations For The Authors):

126-128: It is surprising that surface-attachment does not really activate the full-length KLP6 motor (v=48 {plus minus} 42 nm/s). Can the authors provide an example movie of the gliding assay for the FL KLP6 construct? Gliding assays are done by attaching motors via their sfGFP to the surface using anti-GFP antibodies. Did the authors try to attach the full-length KLP-6 motor directly to the surface? If the KLP-6 motor sticks to the surface via its (inhibitory) C-terminus, this attachment would be expected to activate the motor in the gliding assay, ideally approaching the in vivo velocities of the activated motor.

We have included an example kymograph showing the gliding assay of KLP-6FL (Fig. S1A). When we directly attached KLP-6FL to the surface, the velocity was 0.15 ± 0.02 µm/sec (Fig. S1B), which is similar to the velocity of KLP-6(1-390). While the velocity observed in the direct-attachment condition is much better than those observed in GFP-mediated condition, the observed velocity remains considerably slower than in vivo velocities. Firstly, we think this is because dimerization of KLP-6 is not induced by the surface attachment. Previous studies have shown that monomeric proteins are generally slower than dimeric proteins in the gliding assay (Tomishige et al., 2002). These are consistent with our observation that KLP-6 remains to be monomeric even when autoinhibition is released. Secondly, in vitro velocity of motors is generally slower than in vivo velocity.

156-157: It seems that the GCN4-mediated dimerization induces aggregation of the KLP6 motor domains as seen in the fractions under the void volume in Figure 3B (not seen with the Sf9 expressed full-length constructs, see Figure 1B). Also, the artificially dimerized motor construct does not fully recapitulate the in vivo velocity of UNC-104. Did the authors analyze the KLP-6(1-390)LZ with mass photometry and is it the only construct that is expressed in E. coli?

KLP-6::LZ protein is not aggregating. We have noticed that DNA and RNA from E. coli exists in the void fraction and they occasionally trap recombinant kinesin-3 proteins in the void fraction. To effectively remove these nucleic acids from our protein samples, we employed streptomycin sulfate as a purification method (Liang et al., Electrophoresis, 2009). Please see Purification of recombinant proteins in Methods. In the size exclusion chromatography analysis, we observed that KLP-6(1-393)LZ predominantly eluted in the dimer fraction (New Figure 3). Subsequently, we reanalyzed the motor's motility using a total internal reflection fluorescence (TIRF) assay, as shown in the revised Figure 3. Even after these efforts, the velocity was not changed significantly. The velocity of KLP-6LZ is about 0.3 µm/sec while that of cellular KLP-6::GFP is 0.7 µm/sec (Morsci and Barr, 2011). Similar phenomena, "slower velocity in vitro", has been observed in other motor proteins.

169: In Wang et al., (2022) the microtubule-activated ATPase activities of the mutants were measured in vitro as well, with the relative activities of the motor domain and the D458A mutant being very similar. The D458A mutation is introduced into the full-length motor in Wang et al., while in the present work, the mutation is introduced into the truncated KLP-6(1-587) construct. Can the authors explain their reasoning for the latter?

(1) Kinesins are microtubule-stimulated ATPases. i.e. The ATPase activity is induced by the binding with a microtubule.

(2) Previous studies have shown that the one-dimensional movement of the monomeric motor domain of kinesin-3 depends on the ATPase activity even when the movement does not show clear plus-end directionality (Okada et al., Science, 1998).

(3) While KLP-6(1-587) does not bind to microtubules, both KLP-6(1-390) (= the monomeric motor domain) and KLP-6(1-587)(D458A) similarly bind to microtubules and show one dimensional diffusion on microtubules (Fig. 4E and S2B).

Therefore, the similar ATPase activities of the motor domain(= KLP-6(1-390)) and KLP-6(D458A) observed by Wang et al. is because both proteins similarly associate with and hydrolyze ATP on microtubules, which is consistent with our observation. On the other hand, because KLP-6(wild type) cannot efficiently bind to microtubules, the ATPase activity is low.

Can the authors compare the gliding velocities of the KLP-6(1-390)LZ vs KLP-6(1-587) vs KLP-6(1-587)(D458A) constructs to make sure that the motors are similarly active?

We conducted a comparative analysis of gliding velocities involving KLP-6(1-390), KLP-6(1-587), and KLP-6(1-587)(D458A) (Fig. S1C). We used KLP-6(1-390) instead of KLP-6(1-390)LZ, aligning with the protein used by Wang et al.. We demonstrated that both KLP-6(1-587) and KLP-6(1-587) (D458A) exhibited activity levels comparable to that of KLP-6(1-390). The data suggests that the motor of all recombinant proteins are similarly active.

Please note that, unlike full length condition (Fig. 1D and S1A and S1B), the attachment to the surface using the anti-GFP antibody can activates KLP-6(1-587). The data suggests that, due to the absence of coverage by the MBS and MATH domain (Wang et al., Nat. Commun., 2022), the motor domain of KLP-6(1-587) to some extent permits direct binding to microtubules under gliding assay conditions.

Are the monomeric and dimeric UNC-104(1-653) fractions in Figure 5B in equilibrium? Did the authors do a re-run of the second peak of UNC-104(1-653) (i.e. the monomeric fraction with ~100 kDa) to assess if the monomeric fraction re-equilibrates into a dimer-monomer distribution?

We conducted a re-run of the second peak of UNC-104(1-653) and verified its re-equilibration into a distribution of dimers and monomers after being incubated for 72 hours at 4°C (Fig. S5).

UNC-104 appears to have another predicted coiled-coiled region around ~800 aa (e.g. by NCoils) that would correspond to the CC3 in the mammalian homolog KIF1A. This raises the question if the elongated UNC-104(1-800) would dimerize more efficiently than UNC-104(1-653) (authors highlight the sub-population of dimerized UNC-104(1-653) at low concentrations in Figure 5C) and if this dimerization alone would suffice to 'match' the UNC-104(1-653)E412K mutant (Figure 5D). Did the authors explore this possibility? This would mean that dimerization does not necessarily require the release of autoinhibition.

We have tried to purify UNC-104(1-800) and full-length UNC-104 using the baculovirus system. However, unfortunately, the expression level of UNC-104(1-800) and full length UNC-104 was too low to perform in vitro assays even though codon optimized vectors were used. Instead, we have analyzed full-length human KIF1A. We found that full-length KIF1A is mostly monomeric, not dimeric (Please look at the Author response image 1). The property is similar to UNC-104(1-653) (Figure 5A-C). Therefore, we think CC3 does not strongly affect dimerization of KIF1A, and probably its ortholog UNC-104. Moreover, a recent study has shown that CC2 domain, but not other CC domains, form a stable dimer in the case of KIF1A (Hummel and Hoogenraad, JCB, 2021). Given the similarity in the sequence of KIF1A and UNC-104, we anticipate that the CC2 domain of UNC-104 significantly contributes to dimerization, potentially more than other CC domains. We explicitly describe it in the Discussion in the revised manuscript.

Author response image 1.

Upper left, A representative result of size exclusion chromatography obtained from the analysis of full-length human KIF1A fused with sfGFP.

Upper right, A schematic drawing showing the structure of KIF1A fused with sfGFP and a result of SDS-PAGE recovered from SEC analysis. Presumable dimer and monomer peaks are indicated.

Lower left, Presumable dimer fractions in SEC were collected and analyzed by mass photometry. The result confirms that the fraction contains considerable amount of dimer KIF1A.

Lower right, Presumable monomer fractions were collected and analyzed by mass photometry. The result confirms that the fraction mainly consists of monomer KIF1A.

Note that these results obtained from full-length KIF1A protein are similar to those of UNC-104(1-653) protein shown in Figure 5A-C.

Author Response

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

eLife assessment

The authors describe a method to decouple the mechanisms supporting pancreatic progenitor self-renewal and expansion from feed-forward mechanisms promoting their differentiation. The findings are important because they have implications beyond a single subfield. The strength of evidence is solid in that the methods, data and analyses broadly support the claims with only minor weaknesses.

We are grateful for the substantial effort that reviewers put into reading our manuscript and providing such a detailed feedback. We have strived to address, as much as possible, all comments and criticisms. Thanks to the feedback, we believe that we have now a significantly improved manuscript. Below, there is a point-bypoint response.

Reviewer #1 (Public Review)

In this manuscript, the authors are developing a new protocol that aims at expanding pancreatic progenitors derived from human pluripotent stem cells under GMP-compliant conditions. The strategy is based on hypothesis-driven experiments that come from knowledge derived from pancreatic developmental biology.

The topic is of major interest in the view of the importance of amplifying human pancreatic progenitors (both for fundamental purposes and for future clinical applications). There is indeed currently a major lack of information on efficient conditions to reach this objective, despite major recurrent efforts by the scientific community.

Using their approach that combines stimulation of specific mitogenic pathways and inhibition of retinoic acid and specific branches of the TGF-beta and Wnt pathways, the authors claim to be able, in a highly robust and reproducible manner) to amplify in 10 passages the number of pancreatic progenitors (PP) by 2,000 folds, which is really an impressive breakthrough.

The work is globally well-performed and quite convincing. I have however some technical comments mainly related to the quantification of pancreatic progenitor amplification and to their differentiation into beta-like cells following amplification.

We thank the reviewer for the positive assessment. Below we provide a point-by-point response to specific comments and criticisms.

Reviewer #1 (Recommendations For The Authors)

Figure 1:

Panel A: What is exactly counted in Fig. 1A? Is it the number of PP (as indicated in the title) or the total number of cells? If it is PPs, was it done following PDX1/NKX6.1/SOX9 staining and FACS quantification? This question applies to a number of Figures and the authors should be clear on this point.

We now define ‘PP cells’ as ‘PP-containing cells’ (PP cells) the first time we use the term in the RESULTS section.

Panel D: I do not understand the source of TGFb1, GDF11, FGF18, PDGFA. Which cell type(s) express such factors in culture? I was not convinced that the signals are produced by PP and act through an autocrine loop. I have the same type of questions for the receptors: PDGFR on the second page of the results; RARs and RXRs on the third page.

We refer to these factors/receptors as components of a tentative autocrine loop. We agree we do not prove it and we now comment on this in the discussion section.

Figure 2:

FACS plots are very difficult to analyze for two reasons: I do not understand the meaning of the y axes (PDX1/SOX9). Does that mean that 100% of the cells were PDX1+/SOX9+? The authors should show the separated FACS plots. More importantly, the x axes indicate that NKX6.1 FACS staining is very weak. This is by far different from what can be read in publications performing the same types of experiments (publications by Millman, Otonkoski...as examples). How was quantification performed when it is so difficult to properly define positive vs negative populations? It is necessary to present proper "negative controls" for FACS experiments and to clearly indicate how positive versus cells were defined

We now explain the gating strategy better in the results section, all controls are included in figure S2.

Figure 3:

What is the exact "phenotype" of the cells that incorporated EdU: It would be really instructive to add PDX1/NKX6.1/SOX9 staining on top of EdU. I am also surprised that 20% of the cells stain positive for Annexin V. This is a huge fraction. Does that mean that many cells (20%) are dying and if the case, how amplification can take place under such deleterious conditions?

This is an interesting mechanistic point but performing these experiments would delay the publication of the final manuscript for too long. These assays were done at p3 in order to catch CINI cells that do not expand in most cases. It is important to note that cell death also appears higher in CINI cells. It is likely that the combination of these effects results in reproducible expansion under C5. We comment on the possibilities in the discussion section.

Figure 4:

On FACS plots the intensity at the single cell level (see x-axis of the figure) of the NKX6.1 staining is found to increase in Fig. 4G by 50-100 folds when compared to Fig. 4E. Is it expected? This should be discussed in the text. Do the authors observe the same increase by immunocytochemistry?

The apparent difference is actually 10-fold (from 2x102 to 2x103). We think that the most likely reason for this apparent increase is that at p0 we typically used very few cells for the FC in order to keep as many as possible for the subsequent expansion. If we had used more, we would be able to also detect cells with higher expression. As we mention in the bioinformatics analysis, NKX6 expression does increase with passaging and therefore it is also possible that at least part of this increase is real. However, we don’t have suitable data (same number of cells analyzed at each passage) to address this in a reliable manner.

Figure 5

Previous data from the scientific literature indicate that in vitro, by default, PP gives rise to duct-like cells. This is a bit described in the result section and supplementary figures taking into account the expression of transcription factors. However the data are not clearly explained and described in quite a qualitative manner. They should appear in a quantitative fashion (and the main figures), adding additional duct cell markers such as Carbonic anhydrase, SPP1, CFTR, and others. I assume that the authors can easily use their transcriptomic data to produce a Figure to be described and discussed in detail.

We think it can be misleading to use such markers (other than TFs and the latter only as a collective) because specific markers of terminal differentiation are more often than not expressed during development in multipotent progenitors, the most conspicuous example been CPA1. To illustrate the point, we used the RNA Seq data of and plotted the expression values of a panel of duct genes in isolated human fetal progenitors (Ramond et al., 2017) together with their expression in p0 PP and ePP cells from all three different procedure (please see below). All raw RNA Seq data were processed together to enable direct comparison. According to the analysis of Ramond et al the A population corresponds to MPCs, C to early endocrine progenitors (EP), D to late endocrine progenitors and, by inference and gene expression pattern B to BPs. Expression levels of all these markers were very similar suggesting that these markers cannot be used to distinguish between duct cells and progenitor cells. Importantly, SC-islets derived from either dPP or ePP cells express extremely low and similar levels of KRT19, a marker of duct cells. This latter information is now included in the last part of the results (Figure S7).

Author response image 1.

Fig. 7:<br /> The figure is a bit disappointing for 2 reasons. In A and B, the quality of INS, GCG, and SST staining is really poor. In E, GSIS is really difficult to interpret. They should not be presented as stimulatory indexes. The authors should present independently: INS content; INS secretion at low glucose; INS secretion at high glucose; INS secretion with KCL. Finally, the authors should indicate that glucose poorly (around 2 fold) activates insulin/C-Pept secretion in their stem-cell-derived islets.

We disagree with the quality assessment of the immunofluorescence. Stimulation indexes are also used very widely but we now provide data for actual C-peptide secretion normalized for DNA content of the SC-islets. For technical reasons we do not have normalized C-peptide secretion for human islets. However, we provide a direct comparison to the stimulation index of human islets assayed under the same conditions (2.7 mM glucose / 16.7 mM glucose / 16.7 mM glucose + 30 mM KCl) without presenting SC-islets separately and tweaking the glucose basal (lowering) and stimulation (increasing) levels to inflate the stimulation index. This is unfortunately common. In any case, we do not claim an improvement in the differentiation conditions and our S5-S7 steps may not be optimal but this is not the subject of this work.

Reviewer #2 (Public Review)

Summary

The paper presents a novel approach to expand iPSC-derived pdx1+/nkx6.1+ pancreas progenitors, making them potentially suitable for GMP-compatible protocols. This advancement represents a significant breakthrough for diabetes cell replacement therapies, as one of the current bottlenecks is the inability to expand PP without compromising their differentiation potential. The study employs a robust dataset and state-of-the-art methodology, unveiling crucial signaling pathways (eg TGF, Notch...) responsible for sustaining pancreas progenitors while preserving their differentiation potential in vitro.

Strengths

This paper has strong data, guided omics technology, clear aims, applicability to current protocols, and beneficial implications for diabetes research. The discussion on challenges adds depth to the study and encourages future research to build upon these important findings.

We thank the reviewer for the positive assessment. Below we provide a point-by-point response to general comments and criticisms.

Weaknesses

The paper does have some weaknesses that could be addressed to improve its overall clarity and impact. The writing style could benefit from simplification, as certain sections are explained in a convoluted manner and difficult to follow, in some instances, redundancy is evident. Furthermore, the legends accompanying figures should be self-explanatory, ensuring that readers can easily understand the presented data without the need to be checking along the paper for information.

We have simplified the text in several places and removed redundancies, particularly in the discussion. We revisited the figure legends and made minor corrections to increase clarity. However, regarding the figure legends, we think that adding the interpretation of the results would be redundant to the main text.

The culture conditions employed in the study might benefit from more systematic organization and documentation, making them easier to follow.<br /> There is a comparative Table (Table S1) where all conditions are summarized. We refer to this Table every time that we introduce a new condition. We also have a Table (Table S4) which presents all different media and components used it the differentiation procedure.

Another important aspect is the functionality of the expanded cells after differentiation. While the study provides valuable insights into the expansion of pancreas progenitors in vitro and does the basic tests to measure their functionality after differentiation the paper could be strengthened by exploring the behavior and efficacy of these cells deeper, and in an in vivo setting.

This will be done in a future study where we will also introduce a number of modifications in S5-S7

Quantifications for immunofluorescence (IF) data should be displayed.

We have not conducted quantifications of IFs because FC is much more objective and accurate. We have not conducted FC for CDX2 and AFP because all other data strongly favor C6 anyway. It should be noted that CDX2 and AFP expression is generally not addressed at all presumably because it raises uncomfortable questions and, to our knowledge, we are the first to address this so exhaustively.

Some claims made in the paper may come across as somewhat speculative.

We have now indicated so where applicable.

Additionally, while the paper discusses the potential adaptability of the method to GMP-compatible protocols, there is limited elaboration on how this transition would occur practically or any discussion of the challenges it might entail.

We have now added a paragraph discussing this in the discussion section.

Reviewer #2 (Recommendations For The Authors)

Related to Figure 1:

• Unclear if CINI or SB431542 + CINI was used (first paragraph of results...)

The paragraph was unclear and it is now rewritten

• Was the differentiation to PP similar between the different attempts? A basic QC for each Stem Cell technology differentiation would be good to include.

We added (Figure 1B) a comparison of expression data of general genes (QC) in PP cells showing very comparable patterns of expression. Some of these PP cells went on to expand and most did not but there is no apparent correlation of this with the gene expression data.

• qPCR data - relative fold? over what condition? (indicate on axis label)

We added a label as well as an explanation on p0 values in the figure legend

• FGF18/ PDGFA - worth including background in pancreas development as in the other factors.

Background information has been added

• Bioinformatics is a bit biased with a few genes selected - what are the DEGs / top enriched pathways? Maybe worth showing a volcano plot of the DEGs for example.

We have done all these standard analyses but we think that they did not contribute anything else useful to the study with the exception of pointing to the finding that the TGFb pathway is negatively correlated with expansion, and this is included in the study. The ‘unbiased’ analysis that the reviewer suggests did not turn out something else useful to exploit for the expansion. This does not mean that our approach is biased – in our view it is hypothesis-driven. As we also write in the manuscript, if in a certain pathway a key gene fails to be expressed, the pathway will not show up in any GO or GSEA analyses. However, the pathway will still be regulated. The RA and FGF18 cases clearly illustrate this. We realize that these analyses have become a standard but we think that it is not the only way to approach genomics data and these approaches did not offer much in the context of this study.

• The E2F part is very speculative

The pathway came up as a result of ‘unbiased’ GSEA analyses. However, we do agree and rephrased.

• The authors claim ' the negative correlation of TGFb signalling with expansion retrospectively justifies the use of A83 '. However, p0 is not treated with A83 - how can they tell that there is a correlation between TGFb signalling and expansion?

The correlation came from the RNA Seq data analysis during expansion. We have rephrased slightly to convey the message more clearly.

• Typo with TGFbeta inhibitor name is mispelled (A3801)

Corrected

• Page 5 - last paragraph - Table S3? (isnt it refering to S2?)

Since Table S2 is the list of the regulated genes and S3 is the list of the regulated signaling pathway components both are relevant here, we now refer to both.

• In the text Figure 2G should read Figure 1G (page 7, end of 1st paragraph).

Corrected

• 'Autocrine loop' existence – speculative

Added the phrase ‘we speculated’. We refer to this only as a tentative interpretation. We also elaborate in the discussion now.

Related to Figure 2:

• I am not sure if I would refer to chemical "activation/inhibition" of pathways as 'gain/loss of function'. Maybe this term is more adequate for genetic modifications.

For genetic manipulations, these terms are (supposed to be) accompanied by the adjective ‘genetic’ but to avoid misinterpretations we changed the terms to activation and inhibition as suggested.

• It would be good to include a summary of the different conditions as a schematic in one of the figures, to make it very clear to the reader what the conditions are.

We tried this in an early version of the manuscript but, in our view, it was adding complexity, rather than simplifying things. The problem is that as such the Table cannot be integrated in any figure if eg in Figure 2 it would be too early, if in Figure 4 it would be too late and so on. All conditions show up in detail in Table S1.

• Nkx6.1 - is the image representative? It looks like Nkx6.1 decreases over the passages.

We do mention in the text that ‘… even though expansion (in C5) appeared to somewhat reduce the number of NKX6.1+ cells. (Figure 2E-G). As we mentioned, this was one of the reasons to continue with other conditions (C6-C8).

• Upregulation of AFP/ CDX2 is a bit concerning - the IF for C5 p5 shows a high proportion of CDX2+ cells (Fig S2I). perhaps it would be good to quantify the IF.

It was concerning – this is why we then tested conditions C6-8. Since it is C6 that we propose at the end, it would be, in our view, extraneous to quantify CDX2 in C5.

• How do C5/C1/C0 compare to CINI?

We now remind the reader in the results section that CINI was not reproducible - so any other comparison would be extraneous.

Related to Figure 3:

• There is a 'Lore Ipsum' label above B

Corrected

Related to Figure 4:

• It is good that AFP expression is reduced at p10, but there seems to be a high proportion of AFP at p5. IF/FACS should be quantified.

We think that this would not add significantly since there are several other criteria, particularly the increase of the PDX1+/SOX9+/NKX6.1+ that clearly show that the C6 condition is preferable. Further elaboration of C6 could use such additional criteria. We comment on CDX2 / AFP in the discussion.

• CDX2 should be quantified by IF / FACS.

We think that this would not add significantly since there are several other criteria, particularly the increase of the PDX1+/SOX9+/NKX6.1+ that clearly show that the C6 condition is preferable. Further elaboration of C6 could use such additional criteria. We comment on CDX2 / AFP in the discussion.

• Karyotype analysis is good but not very precise when analyzing genetic micro alterations... what does a low-pass sequencing of the expanding lines look like? Are there any micro-deletions in the expanding lines?

This is an unusual request. Microdeletions may occur at any point – during passaging of hPS cells, differentiation as well as well as expansion but such data are so far not shown in publications – and reasonably so in our opinion. Thus, we have not done this analysis but it certainly would be appropriate in a clinical setting as part of QC.

• Data supporting that the cells can be cryopreserved and recovered with >85% survival rate is not provided.

We now provide data for the C6-mediated expansion (Figure 4J). The freezing procedure was developed during the time we were testing C5 and we don’t have sufficient data to show reliably the survival of the cells during C5 expansion. Thus, we have now removed the reference in the C5 part of the manuscript.

Related to Figure 5:

-Figure 5C - perhaps worth commenting on the different pathways that are enriched when cells undergo expansion and show some of the genes that are up/down regulated.

This is indeed of interest but since it will not address any specific question in the context of this work (eg is the endocrine program repressed?) and since it would not be followed by additional experiments we think that it would burden the manuscript unnecessarily. The data are accessible for any type of analysis through the GEO database.

• Figure S5D shows in vitro clustering away from in vivo PP - it would be good to explain how in vitro generated PP differs from their in vivo counterparts instead of restricting the comparison to the in vitro protocol.

We have added a possible interpretation of this observation in the results section and discuss, how one could go properly about this comparison.

• Quantification of Fig5F should be included. Is GP2 expression detectable by IF at p5 too?

We have quantified GP2 expression by FC at p10 but not at earlier stages. We include now the FC data in Fig5F

• Validation of Fig5G by qPCR would be good. PDX1 did not seem reduced by IF in Figure 4.

The purpose of Fig5G is to compare the expression of the same genes across different expansion approaches. Therefore, in our view, qPCRs would not be appropriate since we do not have samples from the other approaches. We did not claim a reduction in PDX1 expression.

• How can the authors explain the NGN3 expression at PP?

In our view, differentiation is a dynamic process and not all cells are synchronized at the same cell type, this is true in vivo and in vitro. Sc-RNA Seq data indeed show a small population of cells at PP that are NEUROG3+ (our unpublished data). We have now included this in the discussion.

Related to Figure 6:

• How do the different lines differ? Any statistical comparison between lines?

There is a paragraph dealing with the comparison of PP and ePP cells (p5 and p10) from different lines at the level of gene expression and the data are in Figure S6A-G. Then there is a paragraph addressing this at the level of PDX1/SOX9/NKX6.1 expression by FC. We have now expanded and rewrote the latter to include statistical comparisons across PPs from different lines at p0, p5 an p10

Related to Figure 7:

• Mention the use of micropatterned

Micropatterned wells - not really correct. They use Aggrewells, micropatterned plates are something else.

We changed ‘micropatterned wells’ into ‘microwells’

• Figure 7D, those are qPCR data. The label is inconsistent, why did they call it fold induction instead of fold change? Also, not sure if plotting the fold change to hPSC is the best here.

We use fold change when comparing the expression of the same gene at different passages but fold induction when comparing to its expression in hPS cells. We made sure it is also explained in the figure legends.

• Absolute values should be shown for the GSIS to determine basal insulin secretion. Also, sequential stimulation to address if the cells are able to respond to multiple glucose stimulations.

We include now the secreted amounts of human C-peptide under the different conditions (Figure S7) normalized for cell numbers using their DNA content for the normalization. The many parameters we have used suggest that dPP and ePP SC-islets are very similar. If we were claiming a better S5-S7 procedure, such an assay would have been necessary but in this context, we think it is not absolutely necessary.

• In vivo data would have strengthened the story. It is not clear if, in vivo, the cells will behave as the nonexpanded iPSC-derived beta cells.

We agree and these studies are under way but we do not expect to complete them soon. We feel that it is important that this work appears sooner rather than later.

Reviewer #3 (Public Review)

Summary:

In this work, Jarc et al. describe a method to decouple the mechanisms supporting progenitor self-renewal and expansion from feed-forward mechanisms promoting their differentiation.

The authors aimed at expanding pancreatic progenitor (PP) cells, strictly characterized as PDX1+/SOX9+/NKX6.1+ cells, for several rounds. This required finding the best cell culture conditions that allow sustaining PP cell proliferation along cell passages, while avoiding their further differentiation. They achieve this by comparing the transcriptome of PP cells that can be expanded for several passages against the transcriptome of unexpanded (just differentiated) PP cells.

The optimized culture conditions enabled the selection of PDX1+/SOX9+/NKX6.1+ PP cells and their consistent, 2000-fold, expansion over ten passages and 40-45 days. Transcriptome analyses confirmed the stabilization of PP identity and the effective suppression of differentiation. These optimized culture conditions consisted of substituting the Vitamin A containing B27 supplement with a B27 formulation devoid of vitamin A (to avoid retinoic acid (RA) signaling from an autocrine feed-forward loop), substituting A38-01 with the ALK5 II inhibitor (ALK5i II) that targets primarily ALK5, supplementation of medium with FGF18 (in addition to FGF2) and the canonical Wnt inhibitor IWR-1, and cell culture on vitronectin-N (VTN-N) as a substrate instead of Matrigel.

Strengths:

The strength of this work relies on a clever approach to identify cell culture modifications that allow expansion of PP cells (once differentiated) while maintaining, if not reinforcing, PP cell identity. Along the work, it is emphasized that PP cell identity is associated with the co-expression of PDX1, SOX9, and NKX6.1. The optimized protocol is unique (among the other datasets used in the comparison shown here) in inducing a strong upregulation of GP2, a unique marker of human fetal pancreas progenitors. Importantly GP2+ enriched hPS cell-derived PP cells are more efficiently differentiating into pancreatic endocrine cells (Aghazadeh et al., 2022; Ameri et al., 2017).

The unlimited expansion of PP cells reported here would allow scaling-up the generation of beta cells, for the cell therapy of diabetes, by eliminating a source of variability derived from the number of differentiation procedures to be carried out when starting at the hPS cell stage each time. The approach presented here would allow the selection of the most optimally differentiated PP cell population for subsequent expansion and storage. Among other conditions optimized, the authors report a role for Vitamin A in activating retinoic acid signaling in an autocrine feed-forward loop, and the supplementation with FGF18 to reinforce FGF2 signaling.

This is a relevant topic in the field of research, and some of the cell culture conditions reported here for PP expansion might have important implications in cell therapy approaches. Thus, the approach and results presented in this study could be of interest to researchers working in the field of in vitro pancreatic beta cell differentiation from hPSCs. Table S1 and Table S4 are clearly detailed and extremely instrumental to this aim.

We thank the reviewer for the positive assessment. Below we provide a point-by-point response to general comments and criticisms.

Weaknesses

The authors strictly define PP cells as PDX1+/SOX9+/NKX6.1+ cells, and this phenotype was convincingly characterized by immunofluorescence, RT-qPCR, and FACS analysis along the work. However, broadly defined PDX1+/SOX9+/NKX6.1+ could include pancreatic multipotent progenitor cells (MPC, defined as PDX1+/SOX9+/NKX6.1+/PTF1A+ cells) or pancreatic bipotent progenitors (BP, defined as PDX1+/SOX9+/NKX6.1+/PTF1A-) cells. It has been indeed reported that Nkx6.1/Nkx6.2 and Ptf1a function as antagonistic lineage determinants in MPC (Schaffer, A.E. et al. PLoS Genet 9, e1003274, 2013), and that the Nkx6/Ptf1a switch only operates during a critical competence window when progenitors are still multipotent and can be uncoupled from cell differentiation. It would be important to define whether culturing PDX1+/SOX9+/NKX6.1+ PP (as defined in this work) in the best conditions allowing cell expansion is reinforcing either an MPC or BP phenotype. Data from Figure S2A (last paragraph of page 7) suggests that PTF1A expression is decreased in C5 culture conditions, thus more homogeneously keeping BP cells in this media composition. However, on page 15, 2nd paragraph it is stated that "the strong upregulation of NKX6.2 in our procedure suggested that our ePP cells may have retracted to an earlier PP stage". Evaluating the co-expression of the previously selected markers with PTF1A (or CPA2), or the more homogeneous expression of novel BP markers described, such as DCDC2A (Scavuzzo et al. Nat Commun 9, 3356, 2018), in the different culture conditions assayed would more shield light into this relevant aspect.

This is certainly an interesting point. The RNA Seq data suggest that ePP cells resemble BP cells rather than MPCs and that this occurs during expansion. We have now added a new paragraph in the results section to illustrate this and added graphs of CPA2, PTF1A and DCDC2A expression during expansion in Figure 5, S5 as well as data in Table S5. In summary, we favor the interpretation that expanded cells are close but not identical to the BP identity and refer to that in the discussion. We have also amended the statement on page 15 stating the strong upregulation of NKX6.2 in our procedure suggested that our ePP cells may have retracted to an earlier PP stage.

In line with the previous comment, it would be extremely insightful if the authors could characterize or at least discuss a potential role for YAP underlying the mechanistic effects observed after culturing PP in different media compositions. It is well known that the nuclear localization of the co-activator YAP broadly promotes cell proliferation, and it is a key regulator of organ growth during development. Importantly in this context, it has been reported that TEAD and YAP regulate the enhancer network of human embryonic pancreatic progenitors and disruption of this interaction arrests the growth of the embryonic pancreas (Cebola, I. et al. Nat Cell Biol 17, 615-26, 2015). More recently, it has also been shown that a cell-extrinsic and intrinsic mechanotransduction pathway mediated by YAP acts as gatekeeper in the fate decisions of BP in the developing pancreas, whereby nuclear YAP in BPs allows proliferation in an uncommitted fate, while YAP silencing induces EP commitment (Mamidi, A. et al. Nature 564, 114-118, 2018; Rosado-Olivieri et al. Nature Communications 10, 1464, 2019). This mechanism was further exploited recently to improve the in vitro pancreatic beta cell differentiation protocol (Hogrebe et al., Nature Protocols 16, 4109-4143, 2021; Hogrebe et al, Nature Biotechnology 38, 460-470, 2020). Thus, YAP in the context of the findings described in this work could be a key player underlying the proliferation vs differentiation decisions in PP.

We do refer to these publications now and refer to the YAP pathway in the introduction and results sections as well as in the discussion. We have not investigated more because the kinetics of the different components of the pathway are complex and do not give an indication of whether the pathway becomes more or less active – please see below.

Author response image 2.

Regarding the improvements made in the PP cell culture medium composition to allow expansion while avoiding differentiation, some of the claims should be better discussed and contextualized with current stateof-the-art differentiation protocols. As an example, the use of ALK5 II inhibitor (ALK5i II) has been reported to induce EP commitment from PP, while RA was used to induce PP commitment from the primitive gut tube cell stage in recently reported in vitro differentiation protocols (Hogrebe et al., Nature Protocols 16, 41094143, 2021; Rosado-Olivieri et al. Nature Communications 10, 1464, 2019). In this context, and to the authors' knowledge, is Vitamin A (triggering autocrine RA signaling) usually included in the basal media formulations used in other recently reported state-of-the-art protocols? If so, at which stages? Would it be advisable to remove it?

These points and our views are now included in the discussion

In this line also, the supplementation of cell culture media with the canonical Wnt inhibitor IWR-1 is used in this work to allow the expansion of PP while avoiding differentiation. A role for Wnt pathway inhibition during endocrine differentiation using IWR1 has been previously reported (Sharon et al. Cell Reports 27, 22812291.e5, 2019). In that work, Wnt inhibition in vitro causes an increase in the proportion of differentiated endocrine cells. It would be advisable to discuss these previous findings with the results presented in the current work. Could Wnt inhibition have different effects depending on the differential modulation of the other signaling pathways?

These points are now included in the discussion together with the points above

Reviewer #3 (Recommendations For The Authors)

Recommendations for improving the writing and presentation and minor comments on the text and figures:

• In the Introduction (page 3, line 1) it is stated: "Diabetes is a global epidemic affecting > 9% of the global population and its two main forms result from .....". The authors could rephrase/remove "global" repeated twice.

Corrected

• On page 4 of the introduction, in the context of "Unlimited expansion of PP cells in vitro will require disentangling differentiation signals from proliferation/maintenance signals. Several pathways have been implicated in these processes..." the authors are advised to consider mentioning the YAP mediated mechanisms as another key aspect underlying MPC phenotype (Cebola, I. et al. Nat Cell Biol 17, 615-26, 2015) and the BP to endocrine progenitor (EP) commitment (Mamidi, A. et al. Nature 564, 114-118, 2018; Rosado-Olivieri et al. Nature Communications 10, 1464, 2019). This should be better discussed in the context of the Weaknesses mentioned in the Public Review. It would be worth considering adding effectors and other molecules involved in YAP and Hippo pathway signaling to Table S3.

We have added the role of the Hippo/YAP pathway in the introduction and mentioned in the results the finding that components of the pathway are generally not regulated except two that are now added in Table S3

• In page 4, paragraph 3, near "and SB431542, another general (ALK4/5/7) TGFβ inhibitor", consider removing "another". SB431542 is the same inhibitor mentioned in the other protocols at the beginning of the paragraph.

The paragraph is rewritten because it was not clear – we used A83-01 and not SB431542. Other approaches had used SB431542.

• Page 5, Table S2 is cited after Table S3, please consider reordering.

In fact, both S2 and S3 are relevant there, therefore we quote both now.

• Page 8, 2nd paragraph, near "Expression of both AFP and CDX2 increased transiently upon expansion, at p5 (Figure S2H-J)." How do you explain results in FigS2C, D and FigS2E (AFP/CDX2)? RT-qPCR data does not suggest transient downregulation.

AFP and CDX2 were – wrongly – italicized in the quoted passage. Therefore, in one case we refer to the protein and in the other to the transcript levels. We corrected and added the qualifier ‘appeared’. The difference is most likely due to translational regulation but we did not elaborate since we do not know. In any case, we have used the, less favorable but more robust, gene expression levels as the main criterion.

• Page 9, end of 2nd paragraph, Figure 5A is cited but it looks like this should be Figure 4A.

Corrected

• Page 9, 3rd paragraph, when stating "C5 ePP cells of the same passage no..." please replace "no" with a number or a suitable abbreviation.

Corrected

• Page 9, 3rd paragraph. Expressing the values in the Y axis in a consistent manner for FigS2B-D and FigS4A would make a comparison easier.

We strive to keep sections autonomous so that the reader would not have to flip between figures and sections – this is why we think that figure S4A is preferable as it is; it is a direct comparison of C6 to C5 for the different markers and has the additional advantage that one needs not to include p0 levels.

• Page 9, 3rd paragraph. Green dots in FigS4A stand for p5 cells? if so, shouldn't these average 1 for all assayed genes?

No, because the baseline (average 1) is the C5 expression at the corresponding passage no. We changed the y-axis label, hopefully it is clearer now.

• Page 10 3rd paragraph, please include color labels in Fig. 5G.

The different colors here correspond to the different expansion procedures that are compared. The samples are labelled on the x axis.

• Page 10 3rd paragraph, Figure 6G is cited but it looks like this should be Figure 5G.

Corrected

• Page 11, 1st paragraph, at "TF genes such as FOXA2 and RBJ remained comparable", please double check if "RBJ" should be "RBPJ".

Corrected

• Page 11, end of 1st paragraph, when stating "Of note, expression of PTF1A was also undetectable in all ePP cells (Table S5)", is PTF1A expression level close to 1000 (which units?) in Table S5 considered undetectable?

This statement regarding ‘undetectable PTF1A expression’ refers to expanded PP cells (ePP), not PP cells at p0. For the latter, expression is indeed close to 1000 in normalized RNA-sequence counts as mentioned in the Table legend.

-Page 11, 4th paragraph, "In summary, the comparative transcriptome analyses suggested that our C6 expansion procedure is more efficient at strengthening the PP identity". In the context of comments made in the Public Review, more accuracy needs to be put when defining PP identity. Are these MPC or BP?

The RNA Seq data suggest that expansion promotes a MPC  BP transition. We have added a paragraph in the corresponding results section and comment in the discussion.

• Page 15, 2nd paragraph, the sentence "expression of PTF1A, recently shown to promote endocrine differentiation of hPS cells (Miguel-Escalada et al., 2022)" is confusing. Please double-check sentence syntax and reference. Does PTF1A expression "promote" or "create epigenetic competence" for endocrine differentiation?

Its role is in the MPCs and it prepares the epigenetic landscape to allow for duct and endocrine specification later, thus it ‘creates epigenetic competence’. The paper was cited out of context and we have now corrected it.

Additional recommendations by the Reviewing Editor:

An insufficient number of experimental repetitions have been used for the following data: (Figure 1A, n = 2; Figures 2B-D, p10, n = 2; Figures 6A and B, VTN-N, n = 1).

This is true but we do not draw quantitative conclusions from or do comparisons with these data.

Author Response

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

We would like to thank the reviewers for their thoughtful evaluation of our manuscript. We considered all the comments and prepared the revised version. The following are our responses to the reviewers’ comments. All references, including those in the original manuscript are included at the end of this point-by-point response.

Reviewer #1 (Public Review):

Weaknesses:

1) The authors should better review what we know of fungal Drosophila microbiota species as well as the ecology of rotting fruit. Are the microbiota species described in this article specific to their location/setting? It would have been interesting to know if similar species can be retrieved in other locations using other decaying fruits. The term 'core' in the title suggests that these species are generally found associated with Drosophila but this is not demonstrated. The paper is written in a way that implies the microbiota members they have found are universal. What is the evidence for this? Have the fungal species described in this paper been found in other studies? Even if this is not the case, the paper is interesting, but there should be a discussion of how generalizable the findings are.

The reviewer inquires as to whether the microbial species described in this article are ubiquitously associated with Drosophila or not. Indeed, most of the microbes described in this manuscript are generally recognized as species associated with Drosophila spp. For example, yeasts such as Hanseniaspora uvarum, Pichia kluyveri, and Starmerella bacillaris have been detected in or isolated from Drosophila spp. collected in European countries as well as the United States and Oceania (Chandler et al., 2012; Solomon et al., 2019). As for bacteria, species belonging to the genera Pantoea, Lactobacillus, Leuconostoc, and Acetobacter have also previously been detected in wild Drosophila spp. (Chandler et al., 2011). These statements have been incorporated into our revised manuscript (lines 391-397). Nevertheless, the term “core” in the manuscript and title may lead to misunderstanding, as the generality does not ensure the ubiquitous presence of these microbial species in every individual fly. Considering this point, we replaced the “core” with “key,” a term that is more appropriate to our context.

2) Can the authors clearly demonstrate that the microbiota species that develop in the banana trap are derived from flies? Are these species found in flies in the wild? Did the authors check that the flies belong to the D. melanogaster species and not to the sister group D. simulans?

Can the authors clearly demonstrate that the microbiota species that develop in the banana trap are derived from flies? Are these species found in flies in the wild?

The reviewer asked whether the microbial species detected from the fermented banana samples were derived from flies. To address this question, additional experiments under more controlled conditions would be needed, such as artificially introducing wild flies onto fresh bananas in the laboratory. Nevertheless, the microbes potentially originate from wild flies, as supported by the literature cited in our response to the Weakness 1).

Alternative sources of microbes also merit consideration. For example, microbes may have been introduced to unfermented bananas by penetration through peel injuries (lines 1300-1301). In addition, they could be introduced by insects other than flies, given that rove beetles (Staphylinidae) and sap beetles (Nitidulidae) were observed in some of the traps. The explanation of these possibilities have been incorporated into DISCUSSION (lines 414427) of our revised manuscript.

Did the authors check that the flies belong to the D. melanogaster species and not to the sister group D. simulans?

Our sampling strategy was designed to target not only D. melanogaster but also other domestic Drosophila species, such as D. simulans, that inhabit human residential areas. For the traps where adult flies were caught, we identified the species of the drosophilids as shown in Table S1, thereby showing the presence of either or both D. melanogaster and D. simulans. We added these descriptions in MATERIALS AND METHODS (lines 511-512 and 560-562), and DISCUSSION (lines 378-379).

3) Did the microarrays highlight a change in immune genes (ex. antibacterial peptide genes)? Whatever the answer, this would be worth mentioning. The authors described their microarray data in terms of fed/starved in relation to the Finke article. They should clarify if they observed significant differences between species (differences between species within bacteria or fungi, and more generally differences between bacteria versus fungi).

Did the microarrays highlight a change in immune genes (ex. antibacterial peptide genes)? Whatever the answer, this would be worth mentioning.

Regarding the antimicrobial peptide genes, statistical comparisons of our RNA-seq data across different conditions were impracticable because most of the genes showed low expression levels. The RNA-seq data of the yeast-fed larvae is shown in Author response Table 1. While a subset of genes exhibited significantly elevated expression in the nonsupportive conditions relative to the supportive ones, this can be due to intra-sample variability rather than the difference in the nutritional conditions. Similar expression profiles were observed in the bacteria-fed larvae as well (data not shown). Therefore, it is difficult to discuss a change in immune genes in the paper. Additionally, the previous study that conducted larval microarray analysis (Zinke et al., 2002) did not explicitly focus on immune genes.

Author response table 1.

Antimicrobial peptide genes are not up-regulated by any of the microbes. Antimicrobial peptides gene expression profiles of whole bodies of first-instar larvae fed on yeasts. TPM values of all samples and comparison results of gene expression levels in the larvae fed on supportive and non-supportive yeasts are shown. Antibacterial peptide genes mentioned in Hanson and Lemaitre, 2020 are listed. NA or na, not available.

They should clarify if they observed significant differences between species (differences between species within bacteria or fungi, and more generally differences between bacteria versus fungi).

We did not observe significant differences in the gene expression profiles of the larvae fed on different microbial species within bacteria or fungi, or between those fed on bacteria and those fed on fungi. For example, the gene expression profiles of larvae fed on the various supportive microbes showed striking similarities to each other, as evidenced by the heat map showing the expression of all genes detected in larvae fed either yeast or bacteria (Author response image 1). Similarities were also observed among larvae fed on various nonsupportive microbes.

Only a handful of genes showed different expression patterns between larvae fed on yeast and those fed on bacteria. Thus, it is challenging to discuss the potential differential impacts of yeast and bacteria on larval growth, if any.

Author response image 1.

Gene expression profiles of larvae fed on the various supporting microbes show striking similarities to each other. Heat map showing the gene expression of the first-instar larvae that fed on yeasts or bacteria. Freshly hatched germ-free larvae were placed on banana agar inoculated with each microbe and collected after 15 h feeding to examine gene expression of the whole body. Note that data presented in Figures 3A and 4C in the original manuscript, which are obtained independently, are combined to generate this heat map. The labels under the heat map indicate the microbial species fed to the larvae, with three samples analyzed for each condition. The lactic acid bacteria (“LAB”) include Lactiplantibacillus plantarum and Leuconostoc mesenteroides, while the lactic acid bacterium (“AAB”) represents Acetobacter orientalis. “LAB + AAB” signifies mixtures of the AAB and either one of the LAB species. The asterisks in the label highlight “LAB + AAB” or “LAB” samples clustered separately from the other samples in those conditions; “” indicates a sample in a “LAB + AAB” condition (Lactiplantibacillus plantarum + Acetobacter orientalis), and “*” indicates a sample in a “LAB” condition (Leuconostoc mesenteroides). Brown abbreviations of scientific names are for the yeast-fed conditions. H. uva, Hanseniaspora uvarum; K. hum, Kazachstania humilis; M. asi, Martiniozyma asiatica; Sa. cra, Saccharomycopsis crataegensis; P. klu, Pichia kluyveri; St. bac, Starmerella bacillaris; BY4741, Saccharomyces cerevisiae BY4741 strain.

4) The whole paper - and this is one of its merits - points to a role of the Drosophila larval microbiota in processing the fly food. Are these bacterial and fungal species found in the gut of larvae/adults? Are these species capable of establishing a niche in the cardia of adults as shown recently in the Ludington lab (Dodge et al.,)? Previous studies have suggested that microbiota members stimulate the Imd pathway leading to an increase in digestive proteases (Erkosar/Leulier). Are the microbiota species studied here affecting gut signaling pathways beyond providing branched amino acids?

The whole paper - and this is one of its merits - points to a role of the Drosophila larval microbiota in processing the fly food. Are these bacterial and fungal species found in the gut of larvae/adults? Are these species capable of establishing a niche in the cardia of adults as shown recently in the Ludington lab (Dodge et al.,)?

Although we did not investigate the microbiota in the gut of either larvae or adults, we did compare the microbiota within surface-sterilized larvae or adults with the microbiota in food samples. We found that adult flies and early-stage foods, as well as larvae and late-stage foods, harbored similar microbial species (Figure 1F). Additionally, previous studies examining the gut microbiota in wild adult flies have detected microbes belonging to the same species or taxa as those isolated from our foods (Chandler et al., 2011; Chandler et al., 2012). We have elaborated on this in our response to Weakness 1).

While we did not investigate whether these species are capable of establishing a niche in the cardia of adults, we have cited the study by Dodge et al., 2023 in our revised manuscript and discussed the possibility that predominant microbes in adult flies may show a propensity for colonization (lines 410-413).

Previous studies have suggested that microbiota members stimulate the Imd pathway leading to an increase in digestive proteases (Erkosar/Leulier). Are the microbiota species studied here affecting gut signaling pathways beyond providing branched amino acids?

The reviewer inquires whether the supportive microbes in our study stimulate gut signaling pathways and induce the expression of digestive protease genes, as demonstrated in a previous study (Erkosar et al., 2015). Based on our RNA-seq data, this is unlikely. The aforementioned study demonstrated that seven protease genes are upregulated through Imd pathway stimulation by a bacterium that promotes the larval growth. In our RNA-seq analysis, these seven genes did not exhibit a consistent upregulation in the presence of the supportive microbes (H. uva or K. hum in Author response table 2A; Le. mes + A. ori in Author response table 2B). Rather, they exhibited a tendency to be upregulated by the presence of non-supportive microbes (St. bac or Pi. klu in Author response table 2A; La. pla in Author Response Table 2B).

Author response table 2.

Most of the peptidase genes reported by Erkosar et al., 2015 are more highly expressed under the non-supportive conditions than the supportive conditions. Comparison of the expression levels of seven peptidase genes derived from the RNA-seq analysis of yeast-fed (A) or bacteria-fed (B) first-instar larvae. A previous report demonstrated that the expression of these genes is upregulated upon association with a strain of Lactiplantibacillus plantarum, and that the PGRP-LE/Imd/Relish signaling pathway, at least partially, mediates the induction (Erkosar et al., 2015). H. uva, Hanseniaspora uvarum; K. hum, Kazachstania humilis; P. klu, Pichia kluyveri; S. bac, Starmerella bacillaris; La. pla, Lactiplantibacillus plantarum; Le. mes, Leuconostoc mesenteroides; A. ori, Acetobacter orientalis; ns, not significant.

Reviewer #2 (Public Review):

Weaknesses:

The experimental setting that, the authors think, reflects host-microbe interactions in nature is one of the key points. However, it is not explicitly mentioned whether isolated microbes are indeed colonized in wild larvae of Drosophila melanogaster who eat bananas. Another matter is that this work is rather descriptive and a few mechanical insights are presented. The evidence that the nutritional role of BCAAs is incomplete, and molecular level explanation is missing in "interspecies interactions" between lactic acid bacteria (or yeast) and acetic acid bacteria that assure their inhabitation. Apart from these matters, the future directions or significance of this work could be discussed more in the manuscript.

The experimental setting that, the authors think, reflects host-microbe interactions in nature is one of the key points. However, it is not explicitly mentioned whether isolated microbes are indeed colonized in wild larvae of Drosophila melanogaster who eat bananas.

The reviewer asks whether the isolated microbes were colonized in the larval gut. Previous studies on microbial colonization associated with Drosophila have predominantly focused on adults (Pais et al. PLOS Biology, 2018), rather than larval stages. Developing larvae continually consume substrates which are already subjected to microbial fermentation and abundant in live microbes until the end of the feeding larval stage. Therefore, we consider it difficult to discuss microbial colonization in the larval gut. We have mentioned this point in DISCUSSION of the revised manuscript (lines 408-410).

Another matter is that this work is rather descriptive and a few mechanical insights are presented. The evidence that the nutritional role of BCAAs is incomplete, and molecular level explanation is missing in "interspecies interactions" between lactic acid bacteria (or yeast) and acetic acid bacteria that assure their inhabitation.

While we recognize the importance of comprehensive mechanistic analysis, elucidation of more detailed molecular mechanisms lies beyond the scope of this study and will be a subject of future research.

Regarding the nutritional role of BCAAs, the incorporation of BCAAs enabled larvae fed with the non-supportive yeast to grow to the second-instar stage. This observation implies that consumption of BCAAs upregulates diverse genes involved in cellular growth processes in larvae. We mentioned a previously reported interaction between lactic acid bacteria (LAB) and acetic acid bacteria (AAB) in the manuscript (lines 433-436). LAB may facilitate lactate provision to AAB, consequently enhancing the biosynthesis of essential nutrients such as amino acids. To test this hypothesis, future experiments will include the supplementation of lactic acid to AAB culture plates, and the co-inoculation of AAB with LAB mutant strains defective in lactate production to assess both larval growth and continuous larval association with AAB. With respect to AAB-yeast interactions, metabolites released from yeast cells might benefit AAB growth, and this possibility will be investigated through the supplementation of AAB culture plates with candidate metabolites identified in the cell suspension supernatants of the late-stage yeasts.

Apart from these matters, the future directions or significance of this work could be discussed more in the manuscript.

We appreciate the reviewer's recommendations. The explanation of the universality of our findings has been included in the revised DISCUSSION (lines 391-397). We have also added descriptions on the implication of compositional shifts occurring in adult microbiota (lines 404413), possible inoculation routes of different microbes (lines 414-427), and hypotheses on the mechanism of larval growth promotion by yeasts (lines 469-476), all of which could be the focus of our future study.

Reviewer #3 (Public Review):

Weaknesses:

Despite describing important findings, I believe that a more thorough explanation of the experimental setup and the steps expected to occur in the exposed diet over time, starting with natural "inoculation" could help the reader, in particular the non-specialist, grasp the rationale and main findings of the manuscript. When exactly was the decision to collect earlystage samples made? Was it when embryos were detected in some of the samples? What are the implications of bacterial presence in the no-fly traps? These samples also harbored complex microbial communities, as revealed by sequencing. Were these samples colonized by microbes deposited with air currents? Were they the result of flies that touched the material but did not lay eggs? Could the traps have been visited by other insects? Another interesting observation that could be better discussed is the fact that adult flies showed a microbiome that more closely resembles that of the early-stage diet, whereas larvae have a more late-stage-like microbiome. It is easy to understand why the microbiome of the larvae would resemble that of the late-stage foods, but what about the adult microbiome? Authors should discuss or at least acknowledge the fact that there must be a microbiome shift once adults leave their food source. Lastly, the authors should provide more details about the metabolomics experiments. For instance, how were peaks assigned to leucine/isoleucine (as well as other compounds)? Were both retention times and MS2 spectra always used? Were standard curves produced? Were internal, deuterated controls used?

When exactly was the decision to collect early-stage samples made? Was it when embryos were detected in some of the samples?

We collected traps and early-stage samples 2.5 days after setting up the traps. This duration was determined from pilot experiments. A shorter collection time resulted in a lower likelihood of obtaining traps visited by adult flies, whereas a longer collection time caused overcrowding of larvae as well as deaths of adults from drowning in the liquid seeping out of the fruits. These procedural details have been included in the MATERIALS AND METHODS section of the revised manuscript (lines 523-526).

What are the implications of bacterial presence in the no-fly traps? These samples also harbored complex microbial communities, as revealed by sequencing. Were these samples colonized by microbes deposited with air currents? Were they the result of flies that touched the material but did not lay eggs? Could the traps have been visited by other insects?

We assume that the origins of the microbes detected in the no-fly trap foods vary depending on the species. For instance, Colletotrichum musae, the fungus that causes banana anthracnose, may have been present in fresh bananas before trap placement. The filamentous fungi could have originated from airborne spores, but they could also have been introduced by insects that feed on these fungi. We have included these possibilities in the DISCUSSION section of the revised manuscript (lines 417-421).

Another interesting observation that could be better discussed is the fact that adult flies showed a microbiome that more closely resembles that of the early-stage diet, whereas larvae have a more late-stage-like microbiome. It is easy to understand why the microbiome of the larvae would resemble that of the late-stage foods, but what about the adult microbiome? Authors should discuss or at least acknowledge the fact that there must be a microbiome shift once adults leave their food source.

We are grateful for the reviewer's insightful suggestion regarding shifts in the adult microbiome. We have included in the DISCUSSION section of the revised manuscript the possibility that the microbial composition may change substantially during pupal stages or after adult eclosion (lines 404-413).

Lastly, the authors should provide more details about the metabolomics experiments. For instance, how were peaks assigned to leucine/isoleucine (as well as other compounds)? Were both retention times and MS2 spectra always used?

In this metabolomic analysis, LC-MS/MS with triple quadrupole MS monitors the formation of fragment ions from precursor ions specific to each target compound. The use of PFPP columns, which provide excellent separation of amino acids and nucleobases, allows chromatographic peaks of many structural isomers to be separated into independent peaks. In addition, all measured compounds are compared with data from a standard library to confirm retention time agreement. Structural isomers were separated either by retention time on the column or by compound-specific MRM signals (in fact, leucine and isoleucine have both unique MRM channels and column separations). Detailed MRM conditions are identical to the previously published study (Oka et al., 2017). These have been included in the revised ‘LC-MS/MS measurement’ section in MATERIALS AND METHODS (lines 810-824).

Were standard curves produced?

Since relative quantification of metabolite amounts was performed in this study, no standard curve was generated to determine absolute concentrations. However, a standard compound of known concentration (single point) was measured to confirm retention time and relative area values.

Were internal, deuterated controls used?

Internal standards for deuterium-labeled compounds were not used in this study. This is because it is not realistic to obtain deuterium-labeled compounds for all compounds since a large number of compounds are measured. However, an internal standard (L-methionine sulfone) is added to the extraction solvent to calculate the recovery rate. This has been included in the revised ‘LC-MS/MS measurement’ section in MATERIALS AND METHODS (lines 824-825).

Reviewer #1 (Recommendations For The Authors):

Additional comments 1. The authors should do a better job of presenting their data. It took me quite a while to understand the protocol of Figure 1. Panel 1A, B, C could be improved. For instance, 1A suggests that flies are transferred to the lab while this is in fact the banana trap. Indicate 'Banana trap colonized by flies' rather 'wild-type flies in the trap'. 1C: should indicate that the food suspension comes from the banana trap. 1B,D,D: do not use pale color as legend. Avoid the use of indices in Figure 2 (Y1 rather than Y1). Grey colors are difficult to distinguish in Figure 2. Etc. It is a pain for reviewers that figure legends are on the verso of each figure and not just below.

We thank the reviewer for the detailed suggestions to improve the clarity and comprehensibility of our figures. We have improved the figures according to the suggestions. As for the figure legends, we have placed them below each respective figure whenever possible.

1. Clarify in the text if 'sample' means food substratum or flies/larvae (ex. line 116 and elsewhere).

We have revised the word “sample” throughout our manuscript and eliminated the confusion.

1. Line 170 - clarify what you mean by fermented food.

We have replaced the “fermented larval foods” with “fermented bananas” in our revised manuscript (line 165).

1. Line 199 - what is the meaning of 'stocks'.

We have replaced the “stocks” with “strains” (line 195).

1. Line 320 - explain more clearly what the yeast-conditioned banana-agar plate and cell suspension supernatant are, and what the goals of using these media are. This will help in understanding the subsequent text.

We have added a supplemental figure illustrating the sample preparation for the metabolomic analysis (Figure S6), with the following legend describing the procedure (lines 1335-1346): “Sample preparation process for the metabolomic analysis. We suspected that the supportive live yeast cells may release critical nutrients for larval growth, whereas the non-supportive yeasts may not. To test this possibility, we made three distinct sample preparations of individual yeast strains (yeast cells, yeast-conditioned banana-agar plates, and cell suspension supernatants). Yeast cells were for the analysis of intracellular metabolites, whereas yeast-conditioned banana-agar plates and cell suspension supernatants were for that of extracellular metabolites. The samples were prepared as the following procedures. Yeasts were grown on banana-agar plates for 2 days at 25°C, and then scraped from the plates to obtain “yeast cells.” Next, the remaining yeasts on the resultant plates were thoroughly removed, and a portion from each plate was cut out (“yeast-conditioned banana agar”). In addition, we suspended yeast cells from the agar plates into sterile PBS, followed by centrifugation and filtration to eliminate the yeast cells, to prepare “cell suspension supernatants.”

1. Figure 5 is difficult to understand. Provide more explanation. Consider moving the 'all metabolites panel' to Supp. Better explain what this holidic medium is.

The holidic medium is a medium that has been commonly used in the Drosophila research community, which contains ~40 known nutrients, and supports the larval development to pupariation (Piper et al., 2014; Piper et al., 2017). We have introduced this explanation to the RESULTS section of the manuscript (lines 322-327). However, the scope of our research reaches beyond the analysis of the holidic medium components, because feeding the holidic medium alone causes a significant delay in larval growth, suggesting a lack of nutritional components (Piper et al., 2014). Thus, we believe the "All Metabolites" panels should be placed alongside the corresponding “The holidic medium components” panels.

1. I could not access Figure 6 when downloading the PDF. The page is white and an error message appears - it is problematic to review a paper lacking a figure.

We regret any inconvenience caused, perhaps due to a system error. Please refer to the Author response image 2, which is identical to Figure 6 of our original manuscript.

Author response image 2.

Supportive yeasts facilitate larval growth by providing nutrients, including branched-chain amino acids, by releasing them from their cells (Figure 6 from the original manuscript). (A and B) Growth of larvae feeding on yeasts on banana agar supplemented with leucine and isoleucine. (A) The mean percentage of the live/dead individuals in each developmental stage. n=4. (B) The percentage of larvae that developed into second instar or later stages. The “Not found” population in Figure 6A was omitted from the calculation. Each data point represents data from a single tube. Unique letters indicate significant differences between groups (Tukey-Kramer test, p < 0.05). (C) The biosynthetic pathways for leucine and isoleucine with S. cerevisiae gene names are shown. The colored dots indicate enzymes that are conserved in the six isolated species, while the white dots indicate those that are not conserved. Abbreviations of genera are given in the key in the upper right corner. LEU2 is deleted in BY4741. (D-G) Representative image of Phloxine B-stained yeasts. The right-side images are expanded images of the boxed areas. The scale bar represents 50 µm. (H) Summary of this study. H. uvarum is predominant in the early-stage food and provides Leu, Ile, and other nutrients that are required for larval growth. In the late-stage food, AAB directly provides nutrients, while LAB and yeasts indirectly contribute to larval growth by enabling the stable larva-AAB association. The host larva responds to the nutritional environment by dramatically altering gene expression profiles, which leads to growth and pupariation. H. uva, Hanseniaspora uvarum; K. hum, Kazachstania humilis; Pi. klu, Pichia kluyveri; St. bac, Starmerella bacillaris; GF, germ-free.

1. Line 323 - Consider rewriting this sentence (too long, explain what the holidic medium is and why this is interesting). "In the yeast-conditioned banana-agar plates, which were anticipated to contain yeast-derived nutrients, many well-known nutrients included in a chemically defined synthetic (holidic) medium for Drosophila melanogaster (Piper et al., 2014, 2017) were not increased compared to the sterile banana-agar plates; instead, they exhibited drastic decreases irrespective of the yeast species."

We thank the reviewer's suggestion to improve the readability of our manuscript. We have rewritten the sentence in the revised manuscript (lines 320-328) as follows: “The yeastconditioned banana-agar plates were expected to contain yeast-derived nutrients. On the contrary, the result revealed a depletion of various metabolites originally present in the sterile banana agar (Figure 5A). This result prompted us to focus on the metabolites in the chemically defined (holidic) medium for Drosophila melanogaster Piper et al., 2014; Piper et al., 2017. This medium contains ~40 known nutrients, and supports the larval development to pupariation, albeit at the half rate compared to that on a yeast-containing standard laboratory food Piper et al., 2014; Piper et al., 2017. Therefore, the holidic medium could be considered to contain the minimal essential nutrients required for larval growth. Our analysis indicated a substantial reduction of these known nutrients in the yeast-conditioned plates compared to their original quantities (Figure 5B).”

Reviewer #2 (Recommendations For The Authors):

Suggestions for improved or additional experiments, data or analyses.

1. It should be clearly shown (or stated) that isolated microbes, such as H. uvarum and Pa. agglomerans, are indigenous microbes in wild Drosophila melanogaster in their outdoor sampling.

We thank the reviewer for the suggestions. Addressing the presence of isolated microbes within wild D. melanogaster adults is important, but cannot be feasible with our data for the following reasons. Our microbiota analysis of adults was conducted using pooled individuals of multiple Drosophila species, rather than using D. melanogaster exclusively. Moreover, the microbial isolation and the analysis of adult microbiota were carried out in two independent samplings (Figures 1A and 1E in the original manuscript, respectively). As a result, the microbial species detected in the adults were slightly different from those isolated from the food samples collected in the previous sampling. Nevertheless, it is worth noting that H. uvarum dominated in 2 out of the 3 adult samples, constituting >80% of the fungal composition. Pantoea agglomerans was not detected in the adults, although Enterobacterales accounted for >59% in 2 out of the 3 samples. Therefore, these isolated microbial species, or at least their phylogenetically related species, are presumed to be indigenous to wild D. melanogaster.

If the reviewer’s suggestion was to state the dominance of H. uvarum and Pantoea agglomerans in early-stage foods, we have added a supplemental figure showing the species-level microbial compositions corresponding to Figure 1B of the original manuscript (Figure S1), and further revised the manuscript (lines 180-186).

1. The reviewer supposes that the indigenous microbes of flies may differ from what they usually eat. In this study, the authors use banana-based food, but is it justified in terms of the natural environment of the places where those microbes were isolated? In other words, did sampled wild flies eat bananas outside the laboratory at Kyoto University?

Drosophila spp. inhabit human residential areas and feed on various fermented fruits and vegetables. In the areas surrounding Kyoto University, they can be found in garbage in residential dwellings as well as supermarkets. In this regard, fruits are natural food sources of wild Drosophila in the area.

Among various fruits, bananas were selected based on the following two reasons. Firstly, bananas were commonly used in previous Drosophila studies as a trap bait or a component of Drosophila food (Anagnostou et al., 2010; Stamps et al., 2012; Consuegra et al., 2020). Secondly, and rather practically, bananas can be obtained in Japan all year at a relatively low cost. Previous studies have used various fruits such as grapes (Quan and Eisen, 2018), figs (Pais et al., 2018), and raspberries (Cho and Rohlfs, 2023). However, these fruits are only available during limited seasons and are more expensive per volume than bananas. Thus, they were not practical for our study, which required large amounts of fruit-based culture media. We have included a brief explanation regarding this point in MATERIALS AND METHODS (lines 514-518).

1. In Fig. 6B, the Leu and Ile experiment, is the added amount of those amino acids appropriate in the context that they mention "...... supportive yeasts had concentrations of both leucine and isoleucine that were at least four-fold higher than those of non-supportive yeasts"?

We acknowledge that the supplementation should be carried out ideally in a quantity equivalent to the difference between the released amounts of supportive and non-supportive species. However, achieving this has been highly challenging. Previous studies determined the amount of amino acid supplementation by quantifying their concentration in the bacteriaconditioned media (Consuegra et al., 2020; Henriques et al., 2020). However, we found that quantifying the exact concentrations of the amino acids is not feasible with our yeasts. As shown in Figure 5B in the original manuscript, the amino acid contents were markedly reduced in the yeast-conditioned banana agar compared to the agar without yeasts, presumably because of the uptake by the yeasts. Thus, the amino acids released from yeast cells on the banana-agar plate are not expected to accumulate in the medium. As this reviewer pointed out, in the cell suspension supernatants of the supportive yeasts, concentrations of both leucine and isoleucine were at least four-fold higher compared to those of non-supportive yeasts (Figures 5G-H in the original submission), However, this measurement does not give the absolute amount of either amino acid available for larvae. Given these constraints, we opted for the amino acid concentrations in the holidic medium, which support larval growth under axenic conditions (Piper et al., 2014). We also showed that the supplementation of the amino acids at that concentration to the bananaagar plate was not detrimental to larval growth (Figures 6A-B in the original manuscript). These rationales have been included in the revised ‘Developmental progression with BCAA supplementation’ section in MATERIALS AND METHODS of our manuscript (lines 840-847).

1. In addition to the above, it can be included other amino acids or nutrients as control experiments.

As mentioned in our manuscript (lines 365-368), we did supplement other amino acids, lysine and asparagine, which failed to rescue the larval growth.

1. In the experiment of Fig. 2E, how about examining larval development using heat-killed LAB or yeast with live AAB? The reviewer speculates that one possibility is that AAB needs nutrients from LAB.

We did not feed larvae with heat-killed LAB and live AAB for the following reasons. LAB grows very poorly on banana agar compared to yeasts, and preparation of LAB required many banana-agar plates even when we fed live bacteria to larvae. Adding dead LAB to banana-agar tubes would require far more plates, but this preparation is impractical. Furthermore, heat-killing may not allow the investigation of the contribution of heat-unstable or volatile compounds.

As for the reviewer's suggestion regarding the addition of heat-killed yeast with AAB, heat-killed yeast itself promotes larval growth, as shown in Figures 4G and 4H in the original manuscript, so the contribution of yeast cannot be examined using this method.

Recommendations for improving the writing and presentation.

1. It would be good to mention that during sample collection, other insects (other than Drosophila species) were not found in the food if this is true.

Insects other than Drosophila spp. were found in several traps in the sampling shown in Figures 1C-F. These insects, rove beetles (Staphylinidae) and sap beetles (Nitidulidae), seemed to share a niche with Drosophila in nature. Therefore, we believe that the contamination of these insects did not interfere with our goal of obtaining larval food samples. We added these descriptions and explanations to MATERIALS AND METHODS (lines 527531).

1. There are many different kinds of bananas. It should be mentioned the detailed information.

We had included the information on the banana in MATERIALS AND METHODS section (line 622).

1. Concerning the place of sample collection, detailed longitude, and latitude information can be provided (this is easily obtained from Google Maps). When the collection was performed should also be mentioned. This may suggest the environment of the "wild flies" they collected.

We added a table listing the dates of our collections, along with the longitude and latitude of each sampling place (Table S1A).

1. The reviewer could not find how the authors conducted heat killing of yeast.

We added the following procedure to the ‘Quantification of larval development’ section in MATERIALS AND METHODS (lines 680-688). “When feeding heat-killed yeasts to larvae, yeasts were added to the banana-agar tubes and subsequently heated as following procedures. The yeasts were revived from frozen stocks on banana-agar plates, incubated at 25°C, and then streaked on fresh agar plates. After 2-day incubation, yeast cells were scraped from the plates and suspended in PBS at the concentration of 400 mg of yeast cells in 500 µL of PBS. 125 µL of the suspensions were added to banana-agar tubes prepared as described, and after centrifugation at 3,000 x g for 5 min, the supernatants were removed. The amount of cells in each tube is ~50x compared to that when feeding live yeasts, which compensates for the reduced amount due to their inability to proliferate. The tubes were subsequently heated at 80°C for 30 min before adding germ-free larvae.”

1. The reviewer prefers that all necessary information on how to see figures be provided in figure legends. For example, an explanation of some abbreviations is missing.

We carefully re-examined the figure legends and added necessary information.

1. Many of the figures are not kind to readers, i.e., one needs to refer to the legends and main text very frequently. Adding subheadings (titles) to each figure may help.

We added subheadings to our figures to improve the comprehensibility.

Reviewer #3 (Recommendations For The Authors):

I have some minor questions/suggestions about the manuscript that, if addressed, may increase the clarity and quality of the work.

1. Please, when referring to microbial species in the abbreviated form, use only the first letter of the genus. For example, P. agglomerans should be used, not Pa. agglomerans.

We are concerned about the potential confusion caused by using only the first letter of genera, since several genera mentioned in our work share the first letters, such as P (Pichia and Pantoea), S (Starmerella, Saccharomyces, and Saccharomycopsis), or L (Lactiplantibacillus and Leuconostoc). Therefore, we used only the unabbreviated form of the above seven genera in our revised manuscript. We have also made every effort to avoid abbreviations in our figures and tables, but found it necessary to retain two-letter abbreviations when spaces are particularly limiting.

1. In lines 294-298, how exactly was the experiment where yeasts were killed by anti-fungal agents performed? If these agents killed the yeast, how was the microbial growth on plates required to have biomass for fly inoculation obtained? Please, clarify this section.

The yeasts were grown on normal banana-agar plates before the addition onto the anti-fungal agents-containing banana agar. We added the following procedure to MATERIALS AND METHODS (lines 689-695). “When feeding yeasts on banana agar supplemented with antifungal agents, the yeasts were individually grown on normal banana agar twice before being suspended in PBS at the concentration of 400 mg of yeast cells in 500 µL of PBS. 125 µL of the suspensions was introduced onto the anti-fungal agents (10 mL/L 10% p-hydroxybenzoic acid in 70% ethanol and 6 mL/L propionic acid, following the concentration described in Kanaoka et al., 2023)-containing banana agar in 1.5 mL tubes. After centrifugation, the supernatants were removed. The amount of cells in each tube is ~50x compared to that when feeding live yeasts.”

1. In lines 557-558, please clarify how rDNA copy numbers can be calculated in this way.

Considering the results of the ITS and 16S sequencing analysis, it was highly likely that rDNAs from bananas and Drosophila were amplified along with microbial rDNA in this qPCR. To estimate the microbial rDNA copy number, we assumed that the proportion of microbial rDNA within the total amplification products remains consistent between the qPCR and the corresponding sequencing analysis, because the template DNA samples and amplified regions were shared between the analyses. Based on this, the copy number of microbial rDNA was estimated by multiplying the qPCR results with the microbial rDNA ratio observed in the ITS or 16S sequencing analysis of each sample. This methodology has been detailed in the MATERIALS AND METHODS section (lines 609-615).

1. In lines 609-611, how did you check for cells left from the previous day? Microscopy? Or do you mean that if there was liquid still in the sample you would not add more bacterial cultures? Please, clarify.

We observed with the naked eye from outside the tubes to determine if additional AAB should be introduced. Since we placed AAB on the banana agar in a lump, we examined whether the lumps were gone or not. We have added these procedures in MATERIALS AND METHODS (lines 671-673).

1. In Figure 2A, it is hard to differentiate between the gray tones. Please, improve this.

We have distinguished the plots for different conditions by changing the shape of the markers on the graphs.

1. In the legend of Figure 4, line 1101, I believe the panel letters are incorrect.

We have corrected the manuscript (lines 1241-1242) from “heat-killed yeasts on banana agar (H and I) or live yeasts on a nutritionally rich medium (J and K)” to “heat-killed yeasts on banana agar (G and H) or live yeasts on a nutritionally rich medium (I and J).”

1. In Figure S1, authors showed that bananas that were not inoculated still had detectable rDNA signal. Is this really because bacteria can penetrate the peel? Or could this be the “reagent microbiome”? Alternatively, could these microbes have been introduced during sample prep, such as cutting the bananas?

The detection of rDNA in bananas that were not inoculated with microbes was unlikely to be due to microbial contamination during experimental manipulation. The reviewer pointed out the possibility that the “reagent microbiome”, presumably the microbes in PBS, are detected from the uninoculated bananas. This seems to be unlikely, considering the PBS was sterilized by autoclaving before use. To ensure that no viable microbe was left in the autoclaved PBS, we applied a portion of the PBS onto a banana-agar plate and confirmed no colony was formed after incubation for a few days. DNA derived from dead microbes might be present in the PBS, but the PBS-added bananas were incubated for 4 days, so it is also unlikely that a detectable amount of DNA remained until sample collection. Furthermore, we believe that no contamination occurred during sample preparation. Banana peels were treated with 70% ethanol before removing them extremely carefully to avoid touching the fruit inside. All tools were sterilized before use. Taking all of these into account, we speculate that the microbes were already present in the bananas before peeling. We added the details of the sample preparation processes in MATERIALS AND METHODS (lines 518-521 and 540).

Other major revisions

1. We deposited our yeast genome annotation data in the DDBJ Annotated/Assembled Sequences database, and the accession numbers have been added to the ‘Data availability’ section in MATERIALS AND METHODS (lines 868-873).

2. The bacterial composition data in Figure 1B was corrected, because in the original version, the data for Place 3 and Place 4 was plotted in reverse. The original and revised plots are shown side by side in Author response image 3. We hope that the reviewers agree that this replacement of the plots does not affect our conclusion (p5, lines 117-120).

Author response image 3.

Comparison of the original and revised version of bacterial composition graph in Figure 1B. Comparison of the original (left) and revised (right) version of the graph at the bottom of Figure 1B, which shows the result of bacterial composition analysis. The color key, which is unmodified, is placed below the revised version.

1. The plot data and labels in the RNA-seq result heatmaps (Figures 3A and 4C) have been corrected. In these figures, row Z-scores of log2(TPM + 1) were to be plotted, as indicated by the key in each figure. However, in the original version, row Z-scores of TPM was erroneously plotted. Thus, Figures 3A and 4C of the original version have been replaced with the correct plots, and the original and revised plots are shown side by side in Author response images 4A and 4B. We hope that the reviewers agree that this replacement of the plots does not affect our conclusion (p7, lines 222-226 and p9, lines 277-281).

Author response image 4.

Comparison of the original and revised version of Figures 3A and 4C. (A and B) Comparison of the original (left) and revised (right) version of Figures 3A (A) or 4C (B).

1. The keys in the original Figures 3D and 4F indicate that log2(fold change) was used to plot all data. However, when plotting the data from the previous study (Zinke et al., 2002), their “fold change value” was used. We have corrected the keys, plots, and legend of Figure 3D to reflect the different nature of the data from our RNA-seq analysis and those from microarray analysis by Zinke et al. The original and revised plots are shown side by side in Author response image 5. We hope that the reviewers agree that this replacement of the plots does not affect our conclusion (p7, lines 228230 and p9, 277-284).

Author response image 5.

Comparison of the original and revised version of Figures 3D and 4F. (A and B) Comparison of the original (left) and revised (right) version of Figures 3D (A) or 4F (B).

1. The labels in Figure S5C and S5D (Figure S4C and S4D in the original version) have been corrected (they are "Pichia kluyveri > Supportive" and "Starmerella bacillaris > Supportive" rather than "Non-support. > H. uva" and "Non-support. > K. hum"). Additionally, we have reintroduced the circle indicating the number of “dme04070: Phosphatidylinositol signaling system” DEGs in Figure S5D, which was missing in Figure S4D of the original version. The original and revised figures are shown in Author response image 6.

Author response image 6.

Comparison of the original and revised version of Figures S5C and S5D. (A and B) Comparison of the original (left) and revised (right) versions of Figures S5C (A) or S5D (B). The original figures corresponding to the aforementioned figures were Figures S4C and S4D, respectively.

1. The "Fermentation stage" column in Table 1, which indicated whether each microbe was considered an early-stage microbe or a late-stage microbe, has been removed to avoid confusion. This is because some of the microbes (Hanseniaspora uvarum, Pichia kluyveri, and Pantoea agglomerans) were employed in both of the feeding experiments using the microbes detected from the early-stage foods (Figures 2A, 2B, S2A, and S2B) and those from the late-stage foods (Figures 2C, 2D, S2C, and S2D).

2. The leftmost column in Table S7 has been edited to indicate species names rather than “Sample IDs,” because the IDs were not used in anywhere else in the paper.

Reference

Chandler, J. A., Lang, J., Bhatnagar, S., Eisen, J. A. and Kopp, A. (2011). Bacterial communities of diverse Drosophila species: Ecological context of a host-microbe model system. PLoS Genetics 7, e1002272.

Chandler, J. A., Eisen, J. A. and Kopp, A. (2012). Yeast communities of diverse Drosophila species: Comparison of two symbiont groups in the same hosts. Applied and Environmental Microbiology 78, 7327–7336.

Cho, H. and Rohlfs, M. (2023). Transmission of beneficial yeasts accompanies offspring production in Drosophila—An initial evolutionary stage of insect maternal care through manipulation of microbial load? Ecology and Evolution 13, e10184.

Consuegra, J., Grenier, T., Akherraz, H., Rahioui, I., Gervais, H., da Silva, P. and Leulier, F. (2020). Metabolic Cooperation among Commensal Bacteria Supports Drosophila Juvenile Growth under Nutritional Stress. iScience 23, 101232.

Dodge, R., Jones, E. W., Zhu, H., Obadia, B., Martinez, D. J., Wang, C., Aranda-Díaz, A., Aumiller, K., Liu, Z., Voltolini, M., et al. (2023). A symbiotic physical niche in Drosophila melanogaster regulates stable association of a multi-species gut microbiota. Nat Commun 14, 1557.

Erkosar, B., Storelli, G., Mitchell, M., Bozonnet, L., Bozonnet, N. and Leulier, F. (2015). Pathogen Virulence Impedes Mutualist-Mediated Enhancement of Host Juvenile Growth via Inhibition of Protein Digestion. Cell Host & Microbe 18, 445–455.

Hanson, M. A. and Lemaitre, B. (2020). New insights on Drosophila antimicrobial peptide function in host defense and beyond. Current Opinion in Immunology 62, 22–30.

Henriques, S. F., Dhakan, D. B., Serra, L., Francisco, A. P., Carvalho-Santos, Z., Baltazar, C., Elias, A. P., Anjos, M., Zhang, T., Maddocks, O. D. K., et al. (2020). Metabolic cross-feeding in imbalanced diets allows gut microbes to improve reproduction and alter host behaviour. Nat Commun 11, 4236.

Oka, M., Hashimoto, K., Yamaguchi, Y., Saitoh, S., Sugiura, Y., Motoi, Y., Honda, K., Kikko, Y., Ohata, S., Suematsu, M., et al. (2017). Arl8b is required for lysosomal degradation of maternal proteins in the visceral yolk sac endoderm of mouse embryos. Journal of Cell Science jcs.200519.

Pais, I. S., Valente, R. S., Sporniak, M. and Teixeira, L. (2018). Drosophila melanogaster establishes a species-specific mutualistic interaction with stable gut-colonizing bacteria. PLOS Biology 16, e2005710.

Piper, M. D. W., Blanc, E., Leitão-Gonçalves, R., Yang, M., He, X., Linford, N. J., Hoddinott, M. P., Hopfen, C., Soultoukis, G. A., Niemeyer, C., et al. (2014). A holidic medium for Drosophila melanogaster. Nature Methods 11, 100–105.

Piper, M. D. W., Soultoukis, G. A., Blanc, E., Mesaros, A., Herbert, S. L., Juricic, P., He, X., Atanassov, I., Salmonowicz, H., Yang, M., et al. (2017). Matching Dietary Amino Acid Balance to the In Silico-Translated Exome Optimizes Growth and Reproduction without Cost to Lifespan. Cell Metab 25, 610–621.

Quan, A. S. and Eisen, M. B. (2018). The ecology of the drosophila-yeast mutualism in wineries. PLOS ONE 13, e0196440.

Solomon, G. M., Dodangoda, H., McCarthy-Walker, T. T., Ntim-Gyakari, R. R. and Newell, P. D. (2019). The microbiota of Drosophila suzukii influences the larval development of Drosophila melanogaster. PeerJ 7, e8097.

Zinke, I., Schütz, C. S., Katzenberger, J. D., Bauer, M. and Pankratz, M. J. (2002). Nutrient control of gene expression in Drosophila: microarray analysis of starvation and sugar-dependent response. The EMBO Journal 21, 6162–6173.

Author Response

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

We would like to thank the reviewers for their thoughtful comments and constructive suggestions. Point-by-point responses to comments are given below:

Reviewer #1 (Recommendations For The Authors):

This manuscript provides an important case study for in-depth research on the adaptability of vertebrates in deep-sea environments. Through analysis of the genomic data of the hadal snailfish, the authors found that this species may have entered and fully adapted to extreme environments only in the last few million years. Additionally, the study revealed the adaptive features of hadal snailfish in terms of perceptions, circadian rhythms and metabolisms, and the role of ferritin in high-hydrostatic pressure adaptation. Besides, the reads mapping method used to identify events such as gene loss and duplication avoids false positives caused by genome assembly and annotation. This ensures the reliability of the results presented in this manuscript. Overall, these findings provide important clues for a better understanding of deep-sea ecosystems and vertebrate evolution.

Reply: Thank you very much for your positive comments and encouragement.

However, there are some issues that need to be further addressed.

1. L119: Please indicate the source of any data used.

Reply: Thank you very much for the suggestion. All data sources used are indicated in Supplementary file 1.

1. L138: The demographic history of hadal snailfish suggests a significant expansion in population size over the last 60,000 years, but the results only show some species, do the results for all individuals support this conclusion?

Reply: Thank you for this suggestion. The estimated demographic history of the hadal snailfish reveals a significant population increase over the past 60,000 years for all individuals. The corresponding results have been incorporated into Figure 1-figure supplements 8B.

Author response image 1.

(B) Demographic history for 5 hadal snailfish individuals and 2 Tanaka’s snailfish individuals inferred by PSMC. The generation time of one year for Tanaka snailfish and three years for hadal snailfish.

1. Figure 1-figure supplements 8: Is there a clear source of evidence for the generation time of 1 year chosen for the PSMC analysis?

Reply: We apologize for the inclusion of an incorrect generation time in Figure 1-figure supplements 8. It is important to note that different generation times do not change the shape of the PSMC curve, they only shift the curve along the axis. Due to the absence of definitive evidence regarding the generation time of the hadal snailfish, we have referred to Wang et al., 2019, assuming a generation time of one year for Tanaka snailfish and three years for hadal snailfish. The generation time has been incorporated into the main text (lines 516-517): “The generation time of one year for Tanaka snailfish and three years for hadal snailfish.”.

1. L237: Transcriptomic data suggest that the greatest changes in the brain of hadal snailfish compared to Tanaka's snailfish, what functions these changes are specifically associated with, and how these functions relate to deep-sea adaptation.

Reply: Thank you for this suggestion. Through comparative transcriptome analysis, we identified 3,587 up-regulated genes and 3,433 down-regulated genes in the brains of hadal snailfish compared to Tanaka's snailfish. Subsequently, we conducted Gene Ontology (GO) functional enrichment analysis on the differentially expressed genes, revealing that the up-regulated genes were primarily associated with cilium, DNA repair, protein binding, ATP binding, and microtubule-based movement. Conversely, the down-regulated genes were associated with membranes, GTP-binding, proton transmembrane transport, and synaptic vesicles, as shown in following table (Supplementary file 15). Previous studies have shown that high hydrostatic pressure induces DNA strand breaks and damage, and that DNA repair-related genes upregulated in the brain may help hadal snailfish overcome these challenges.

Author response table 1.

GO enrichment of expression up-regulated and down-regulated genes in hadal snailfish brain.

We have added new results (Supplementary file 15) and descriptions to show the changes in the brains of hadal snailfish (lines 250-255): “Specifically, there are 3,587 up-regulated genes and 3,433 down-regulated genes in the brain of hadal snailfish compared to Tanaka snailfish, and Gene Ontology (GO) functional enrichment analyses revealed that up-regulated genes in the hadal snailfish are associated with cilium, DNA repair, and microtubule-based movement, while down-regulated genes are enriched in membranes, GTP-binding, proton transmembrane transport, and synaptic vesicles (Supplementary file 15).”

1. L276: What is the relationship between low bone mineralization and deep-sea adaptation, and can low mineralization help deep-sea fish better adapt to the deep sea?

Reply: Thank you for this suggestion. The hadal snailfish exhibits lower bone mineralization compared to Tanaka's snailfish, which may have facilitated its adaptation to the deep sea. On one hand, this reduced bone mineralization could have contributed to the hadal snailfish's ability to maintain neutral buoyancy without excessive energy expenditure. On the other hand, the lower bone mineralization may have also rendered their skeleton more flexible and malleable, enhancing their resilience to high hydrostatic pressure. Accordingly, we added the following new descriptions (lines 295-300): “Nonetheless, micro-CT scans have revealed shorter bones and reduced bone density in hadal snailfish, from which it has been inferred that this species has reduced bone mineralization (M. E. Gerringer et al., 2021); this may be a result of lowering density by reducing bone mineralization, allowing to maintain neutral buoyancy without expending too much energy, or it may be a result of making its skeleton more flexible and malleable, which is able to better withstand the effects of HHP.”

1. L293: The abbreviation HHP was mentioned earlier in the article and does not need to be abbreviated here.

Reply: Thank you for the correction. We have corrected the word. Line 315.

1. L345: It should be "In addition, the phylogenetic relationships between different individuals clearly indicate that they have successfully spread to different trenches about 1.0 Mya".

Reply: Thank you for the correction. We have corrected the word. Line 374.

1. It is curious what functions are associated with the up-regulated and down-regulated genes in all tissues of hadal snailfish compared to Tanaka's snailfish, and what functions have hadal snailfish lost in order to adapt to the deep sea?

Reply: Thank you for this suggestion. We added a description of this finding in the results section (lines 337-343): “Next, we identified 34 genes that are significantly more highly expressed in all organs of hadal snailfish in comparison to Tanaka’s snailfish and zebrafish, while only seven genes were found to be significantly more highly expressed in Tanaka’s snailfish using the same criterion (Figure 5-figure supplements 1). The 34 genes are enriched in only one GO category, GO:0000077: DNA damage checkpoint (Adjusted P-value: 0.0177). Moreover, five of the 34 genes are associated with DNA repair.” This suggests that up-regulated genes in all tissues in hadal snailfish are associated with DNA repair in response to DNA damage caused by high hydrostatic pressure, whereas down-regulated genes do not show enrichment for a particular function.

Overall, the functions lost in hadal snailfish adapted to the deep sea are mainly related to the effects of the dark environment, which can be summarized as follows (lines 375-383): “The comparative genomic analysis revealed that the complete absence of light had a profound effect on the hadal snailfish. In addition to the substantial loss of visual genes and loss of pigmentation, many rhythm-related genes were also absent, although some rhythm genes were still present. The gene loss may not only come from relaxation of natural selection, but also for better adaptation. For example, the grpr gene copies are absent or down-regulated in hadal snailfish, which could in turn increased their activity in the dark, allowing them to survive better in the dark environment (Wada et al., 1997). The loss of gpr27 may also increase the ability of lipid metabolism, which is essential for coping with short-term food deficiencies (Nath et al., 2020).”

Reviewer #2 (Recommendations For The Authors):

I have pointed out some of the examples that struck me as worthy of additional thought/writing/comments from the authors. Any changes/comments are relatively minor.

Reply: Thank you very much for your positive comments on this work.

For comparative transcriptome analyses, reads were mapped back to reference genomes and TPM values were obtained for gene-level count analyses. 1:1 orthologs were used for differential expression analyses. This is indeed the only way to normalize counts across species, by comparing the same gene set in each species. Differential expression statistics were run in DEseq2. This is a robust way to compare gene expression across species and where fold-change values are reported (e.g. Fig 3, creatively by coloring the gene name) the values are best-practice.

In other places, TPM values are reported (e.g. Fig 2D, Fig 4C, Fig 5A, Fig 4-Fig supp 4) to illustrate expression differences within a tissue across species. The comparisons look robust, although it is not made clear how the values were obtained in all cases. For example, in Fig 2D the TPM values appear to be from eyes of individual fish, but in Fig 4C and 5A they must be some kind of average? I think that information should be added to the figure legends.

Of note: TPM values are sensitive to the shape of the RNA abundance distribution from a given sample: A small number of very highly expressed genes might bias TPM values downward for other genes. From one individual to another or from one species to another, it is not obvious to me that we should expect the same TPM distribution from the same tissues, making it a challenging metric for comparison across samples, and especially across species. An alternative measure of RNA abundance is normalized counts that can be output from DEseq2. See:

Zhao, Y., Li, M.C., Konaté, M.M., Chen, L., Das, B., Karlovich, C., Williams, P.M., Evrard, Y.A., Doroshow, J.H. and McShane, L.M., 2021. TPM, FPKM, or normalized counts? A comparative study of quantification measures for the analysis of RNA-seq data from the NCI patient-derived models repository. Journal of translational medicine, 19(1), pp.1-15.

If the authors would like to keep the TPM values, I think it would be useful for them to visualize the TPM value distribution that the numbers were derived from. One way to do this would be to make a violin plot for species/tissue and plot the TPM values of interest on that. That would give a visualization of the ranked value of the gene within the context of all other TPM values. A more highly expressed gene would presumably have a higher rank in context of the specific tissue/species and be more towards the upper tail of the distribution. An example violin plot can be found in Fig 6 of:

Burns, J.A., Gruber, D.F., Gaffney, J.P., Sparks, J.S. and Brugler, M.R., 2022. Transcriptomics of a Greenlandic Snailfish Reveals Exceptionally High Expression of Antifreeze Protein Transcripts. Evolutionary Bioinformatics, 18, p.11769343221118347.

Alternatively, a comparison of TPM and normalized count data (heatmaps?) would be of use for at least some of the reported TPM values to show whether the different normalization methods give comparable outputs in terms of differential expression. One reason for these questions is that DEseq2 uses normalized counts for statistical analyses, but values are expressed as TPM in the noted figures (yes, TPM accounts for transcript length, but can still be subject to distribution biases).

Reply: Thank you for your suggestions. Following your suggestions, we modified Fig 2D, Fig 4C, Fig 4-Fig supp 4, and Fig 5-Fig supp 1, respectively. In the differential expression analyses, only one-to-one orthologues of hadal snailfish and Tanaka's snailfish can get the normalized counts output by DEseq2, so we showed the normalized counts by DEseq2 output for Fig 2D, Fig 4C, Fig 4-Fig supp 4, Fig 5-Fig supp 1, and for Fig 5A, since the copy number of fthl27 genes undergoes specific expansion in hadal snailfish, we visualized the ranking of all fthl27 genes across tissues by plotting violins in Fig 5-Fig supp 2.

Author response image 2.

(D) Log10-transformation normalized counts for DESeq2 (COUNTDESEQ2) of vision-related genes in the eyes of hadal snailfish and Tanka's snailfish. * represents genes significantly downregulated in hadal snailfish (corrected P < 0.05).

Author response image 3.

(C) The deletion of one copy of grpr and another copy of down-regulated expression in hadal snailfish. The relative positions of genes on chromosomes are indicated by arrows, with arrows to the right representing the forward strand and arrows to the left representing the reverse strand. The heatmap presented is the average of the normalized counts for DESeq2 (COUNTDESEQ2) in all replicate samples from each tissue. * represents tissue in which the grpr-1 was significantly down-regulated in hadal snailfish (corrected P < 0.05).

Author response image 4.

Expression of the vitamin D related genes in various tissues of hadal snailfish and Tanaka's snailfish. The heatmap presented is the average of the normalized counts for DESeq2 (COUNTDESEQ2) in all replicate samples from each tissue.

Author response image 5.

(B) Expression of the ROS-related genes in different tissues of hadal snailfish and Tanaka's snailfish. The heatmap presented is the average of the normalized counts for DESeq2 (COUNTDESEQ2) in all replicate samples from each tissue.

Author response image 6.

Ranking of the expression of individual copies of fthl27 gene in hadal snailfish and Tanaka's snailfish in various tissues showed that all copies of fthl27 in hadal snailfish have high expression. The gene expression presented is the average of TPM in all replicate samples from each tissue.

Line 96: Which BUSCOs? In the methods it is noted that the actinopterygii_odb10 BUSCO set was used. I think it should also be noted here so that it is clear which BUSCO set was used for completeness analysis. It could even be informally the ray-finned fish BUSCOs or Actinopterygii BUSCOs.

Reply: Thank you for this suggestion. We used Actinopterygii_odb10 database and we added the BUSCO set to the main text as follows (lines 92-95): “The new assembly filled 1.26 Mb of gaps that were present in our previous assembly and have a much higher level of genome continuity and completeness (with complete BUSCOs of 96.0 % [Actinopterygii_odb10 database]) than the two previous assemblies.”

Lines 102-105: The medaka genome paper proposes the notion that the ancestral chromosome number between medaka, tetraodon, and zebrafish is 24. There may be other evidence of that too. Some of that evidence should be cited here to support the notion that sticklebacks had chromosome fusions to get to 21 chromosomes rather than scorpionfish having chromosome fissions to get to 24. Here's the medaka genome paper:

Kasahara, M., Naruse, K., Sasaki, S., Nakatani, Y., Qu, W., Ahsan, B., Yamada, T., Nagayasu, Y., Doi, K., Kasai, Y. and Jindo, T., 2007. The medaka draft genome and insights into vertebrate genome evolution. Nature, 447(7145), pp.714-719.

Reply: Thank you for your great suggestion. Accordingly, we modified the sentence and added the citation as follows (lines 100-105): “We noticed that there is no major chromosomal rearrangement between hadal snailfish and Tanaka’s snailfish, and chromosome numbers are consistent with the previously reported MTZ-ancestor (the last common ancestor of medaka, Tetraodon, and zebrafish) (Kasahara et al., 2007), while the stickleback had undergone several independent chromosomal fusion events (Figure 1-figure supplements 4).”

Line 161-173: "Along with the expression data, we noticed that these genes exhibit a different level of relaxation of natural selection in hadal snailfish (Figure 2B; Figure 2-figure supplements 1)." With the above statment and evidence, the authors are presumably referring to gene losses and differences in expression levels. I think that since gene expression was not measured in a controlled way it may not be a good measure of selection throughout. The reported genes could be highly expressed under some other condition, selection intact. I find Fig2-Fig supp 1 difficult to interpret. I assume I am looking for regions where Tanaka’s snailfish reads map and Hadal snailfish reads do not, but it is not abundantly clear. Also, other measures of selection might be good to investigate: accumulation of mutations in the region could be evidence of relaxed selection, for example, where essential genes will accumulate fewer mutations than conditional genes or (presumably) genes that are not needed at all. The authors could complete a mutational/SNP analysis using their genome data on the discussed genes if they want to strengthen their case for relaxed selection. Here is a reference (from Arabidopsis) showing these kinds of effects:

Monroe, J.G., Srikant, T., Carbonell-Bejerano, P., Becker, C., Lensink, M., Exposito-Alonso, M., Klein, M., Hildebrandt, J., Neumann, M., Kliebenstein, D. and Weng, M.L., 2022. Mutation bias reflects natural selection in Arabidopsis thaliana. Nature, 602(7895), pp.101-105.

Reply: Thank you for pointing out this important issue. Following your suggestion, we have removed the mention of the down-regulation of some visual genes in the eyes of hadal snailfish and the results of the original Fig2-Fig supp 1 that were based on reads mapping to confirm whether the genes were lost or not. To investigate the potential relaxation of natural selection in the opn1sw2 gene in hadal snailfish, we conducted precise gene structure annotation. Our findings revealed that the opn1sw2 gene is pseudogenized in hadal snailfish, indicating a relaxation of natural selection. We have included this result in Figure 2-figure supplements 1.

Author response image 7.

Pseudogenization of opn1sw2 in hadal snailfish. The deletion changed the protein’s sequence, causing its premature termination.

Accordingly, we have toned down the related conclusions in the main text as follows (lines 164-173): “We noticed that the lws gene (long wavelength) has been completely lost in both hadal snailfish and Tanaka’s snailfish; rh2 (central wavelength) has been specifically lost in hadal snailfish (Figure 2B and 2C); sws2 (short wavelength) has undergone pseudogenization in hadal snailfish (Figure 2-figure supplements 1); while rh1 and gnat1 (perception of very dim light) is both still present and expressed in the eyes of hadal snailfish (Figure 2D). A previous study has also proven the existence of rhodopsin protein in the eyes of hadal snailfish using proteome data (Yan, Lian, Lan, Qian, & He, 2021). The preservation and expression of genes for the perception of very dim light suggests that they are still subject to natural selection, at least in the recent past.”

Line 161-170: What tissue were the transcripts derived from for looking at expression level of opsins? Eyes?

Reply: Thank you for your suggestions. The transcripts used to observe the expression levels of optic proteins were obtained from the eye.

Line 191: What does tmc1 do specifically?

Reply: Thank you for this suggestion. The tmc1 gene encodes transmembrane channel-like protein 1, involved in the mechanotransduction process in sensory hair cells of the inner ear that facilitates the conversion of mechanical stimuli into electrical signals used for hearing and homeostasis. We added functional annotations for the tmc1 in the main text (lines 190-196): “Of these, the most significant upregulated gene is tmc1, which encodes transmembrane channel-like protein 1, involved in the mechanotransduction process in sensory hair cells of the inner ear that facilitates the conversion of mechanical stimuli into electrical signals used for hearing and homeostasis (Maeda et al., 2014), and some mutations in this gene have been found to be associated with hearing loss (Kitajiri, Makishima, Friedman, & Griffith, 2007; Riahi et al., 2014).”

Line 208: "it is likely" is a bit proscriptive

Reply: Thank you for this suggestion. We rephrased the sentence as follows (lines 213-215): “Expansion of cldnj was observed in all resequenced individuals of the hadal snailfish (Supplementary file 10), which provides an explanation for the hadal snailfish breaks the depth limitation on calcium carbonate deposition and becomes one of the few species of teleost in hadal zone.”

Line 199: maybe give a little more info on exactly what cldnj does? e.g. "cldnj encodes a claudin protein that has a role in tight junctions through calcium independent cell-adhesion activity" or something like that.

Reply: Thank you for this suggestion. We have added functional annotations for the cldnj to the main text (lines 200-204): “Moreover, the gene involved in lifelong otolith mineralization, cldnj, has three copies in hadal snailfish, but only one copy in other teleost species, encodes a claudin protein that has a role in tight junctions through calcium independent cell-adhesion activity (Figure 3B, Figure 3C) (Hardison, Lichten, Banerjee-Basu, Becker, & Burgess, 2005).”

Lines 199-210: Paragraph on cldnj: there are extra cldnj genes in the hadal snailfish, but no apparent extra expression. Could the authors mention that in their analysis/discussion of the data?

Reply: Thank you for your suggestions. Despite not observing significant changes in cldnj expression in the brain tissue of hadal snailfish compared to Tanaka's snailfish, it is important to consider that the brain may not be the primary site of cldnj expression. Previous studies in zebrafish have consistently shown expression of cldnj in the otocyst during the critical early growth phase of the otolith, with a lower level of expression observed in the zebrafish brain. However, due to the unavailability of otocyst samples from hadal snailfish in our current study, our findings do not provide confirmation of any additional expression changes resulting from cldnj amplification. Consequently, it is crucial to conduct future comprehensive investigations to explore the expression patterns of cldnj specifically in the otocyst of hadal snailfish. Accordingly, we added a discussion of this result in the main text (lines 209-214): “In our investigation, we found that the expression of cldnj was not significantly up-regulated in the brain of the hadal snailfish than in Tanaka’s snailfish, which may be related to the fact that cldnj is mainly expressed in the otocyst, while the expression in the brain is lower. However, due to the immense challenge in obtaining samples of hadal snailfish, the expression of cldnj in the otocyst deserves more in-depth study in the future.”

Lines 225-231: I wonder whether low expression of a circadian gene might be a time of day effect rather than an evolutionary trait. Could the authors comment?

Reply: Thank you for your suggestions. Previous studies have shown that the grpr gene is expressed relatively consistently in mouse suprachiasmatic nucleus (SCN) throughout the day (Figure 4-figure supplements 1) and we hypothesize that the low expression of grpr-1 gene expression in hadal snailfish is an evolutionary trait. We have modified this result in the main text (lines 232-242): “In addition, in the teleosts closely related to hadal snailfish, there are usually two copies of grpr encoding the gastrin-releasing peptide receptor; we noticed that in hadal snailfish one of them is absent and the other is barely expressed in brain (Figure 4C), whereas a previous study found that the grpr gene in the mouse suprachiasmatic nucleus (SCN) did not fluctuate significantly during a 24-hour light/dark cycle and had a relatively stable expression (Pembroke, Babbs, Davies, Ponting, & Oliver, 2015) (Figure 4-figure supplements 1). It has been reported that grpr deficient mice, while exhibiting normal circadian rhythms, show significantly increased locomotor activity in dark conditions (Wada et al., 1997; Zhao et al., 2023). We might therefore speculate that the absence of that gene might in some way benefit the activity of hadal snailfish under complete darkness.”

Author response image 8.

(B) Expression of the grpr in a 24-hour light/dark cycle in the mouse suprachiasmatic nucleus (SCN). Data source with http://www.wgpembroke.com/shiny/SCNseq.

Line 253: What is gpr27? G protein coupled receptor?

Reply: We apologize for the ambiguous description. Gpr27 is a G protein-coupled receptor, belonging to the family of cell surface receptors. We introduced gpr27 in the main text as follows (lines 270-273): “Gpr27 is a G protein-coupled receptor, belonging to the family of cell surface receptors, involved in various physiological processes and expressed in multiple tissues including the brain, heart, kidney, and immune system.”

Line 253: Fig4 Fig supp 3 is a good example of pseudogenization!

Reply: Thank you very much for your recognition.

Line 279: What is bglap? It regulates bone mineralization, but what specifically does that gene do?

Reply: We apologize for the ambiguous description. The bglap gene encodes a highly abundant bone protein secreted by osteoblasts that binds calcium and hydroxyapatite and regulates bone remodeling and energy metabolism. We introduced bglap in the main text as follows (lines 300-304): “The gene bglap, which encodes a highly abundant bone protein secreted by osteoblasts that binds calcium and hydroxyapatite and regulates bone remodeling and energy metabolism, had been found to be a pseudogene in hadal fish (K. Wang et al., 2019), which may contribute to this phenotype.”

Line 299: Introduction of another gene without providing an exact function: acaa1.

Reply: We apologize for the ambiguous description. The acaa1 gene encodes acetyl-CoA acetyltransferase 1, a key regulator of fatty acid β-oxidation in the peroxisome, which plays a controlling role in fatty acid elongation and degradation. We introduced acaa1 in the main text as follows (lines 319-324): “In regard to the effect of cell membrane fluidity, relevant genetic alterations had been identified in previous studies, i.e., the amplification of acaa1 (encoding acetyl-CoA acetyltransferase 1, a key regulator of fatty acid β-oxidation in the peroxisome, which plays a controlling role in fatty acid elongation and degradation) may increase the ability to synthesize unsaturated fatty acids (Fang et al., 2000; K. Wang et al., 2019).”

Fig 5 legend: The DCFH-DA experiment is not an immunofluorescence assay. It is better described as a redox-sensitive fluorescent probe. Please take note throughout.

Reply: Thank you for pointing out our mistakes. We corrected the word. Line 1048 and 1151 as follows: “ROS levels were confirmed by redox-sensitive fluorescent probe using DCFH-DA molecular probe in 293T cell culture medium with or without fthl27-overexpression plasmid added with H2O2 or FAC for 4 hours.”

Line 326: Manuscript notes that ROS levels in transfected cells are "significantly lower" than the control group, but there is no quantification or statistical analysis of ROS levels. In the methods, I noticed the mention of flow cytometry, but do not see any data from that experiment. Proportion of cells with DCFH-DA fluorescence above a threshold would be a good statistic for the experiment... Another could be average fluorescence per cell. Figure 5B shows some images with green dots and it looks like more green in the "control" (which could better be labeled as "mock-transfection") than in the fthl27 overexpression, but this could certainly be quantified by flow cytometry. I recommend that data be added.

Reply: Thank you for your suggestions. We apologize for the error in the main text, we used a fluorescence microscope to observe fluorescence in our experiments, not a flow cytometer. We have corrected it in the methods section as follows (lines 651-653): “ROS levels were measured using a DCFH-DA molecular probe, and fluorescence was observed through a fluorescence microscope with an optional FITC filter, with the background removed to observe changes in fluorescence.” Meanwhile, we processed the images with ImageJ to obtain the respective mean fluorescence intensities (MFI) and found that the MFI of the fthl27-overexpression cells were lower than the control group, which indicated that the ROS levels of the fthl27-overexpression cells were significantly lower than the control group. MFI has been added to Figure 5B.

Author response image 9.

ROS levels were confirmed by redox-sensitive fluorescent probe using DCFH-DA molecular probe in 293T cell culture medium with or without fthl27-overexpression plasmid added with H2O2 or FAC for 4 hours. Images are merged from bright field images with fluorescent images using ImageJ, while the mean fluorescence intensity (MFI) is also calculated using ImageJ. Green, cellular ROS. Scale bars equal 100 μm.

Regarding the ROS experiment: Transfection of HEK293T cells should be reasonably straightforward, and the experiment was controlled appropriately with a mock transfection, but some additional parameters are still needed to help interpret the results. Those include: Direct evidence that the transfection worked, like qPCR, western blots (is the fthl27 tagged with an antigen?), coexpression of a fluorescent protein. Then transfection efficiency should be calculated and reported.

Reply: Thank you for your suggestions. To assess the success of the transfection, we randomly selected a subset of fthl27-transfected HEK293T cells for transcriptome sequencing. This approach allowed us to examine the gene expression profiles and confirm the efficacy of the transfection process. As control samples, we obtained transcriptome data from two untreated HEK293T cells (SRR24835259 and SRR24835265) from NCBI. Subsequently, we extracted the fthl27 gene sequence of the hadal snailfish, along with 1,000 bp upstream and downstream regions, as a separate scaffold. This scaffold was then merged with the human genome to assess the expression levels of each gene in the three transcriptome datasets. The results demonstrated that the fthl27 gene exhibited the highest expression in fthl27-transfected HEK293T cells, while in the control group, the expression of the fthl27 gene was negligible (TPM = 0). Additionally, the expression patterns of other highly expressed genes were similar to those observed in the control group, confirming the successful fthl27 transfection. These findings have been incorporated into Figure 5-figure supplements 3.

Author response image 10.

(B) Reads depth of fthl27 gene in fthl27-transfected HEK293T cells and 2 untreated HEK293T cells (SRR24835259 and SRR24835265) transcriptome data. (C) Expression of each gene in the transcriptome data of fthl27-transfected HEK293T cells and 2 untreated HEK293T cells (SRR24835259 and SRR24835265), where the genes shown are the 4 most highly expressed genes in each sample.

Lines 383-386: expression of DNA repair genes is mentioned, but not shown anywhere in the results?

Reply: Thank you for your suggestions. Accordingly, we added a description of this finding in the results section (lines 337-343): “Next, we identified 34 genes that are significantly more highly expressed in all organs of hadal snailfish in comparison to Tanaka’s snailfish and zebrafish, while only seven genes were found to be significantly more highly expressed in Tanaka’s snailfish using the same criterion (Figure 5-figure supplements 1). The 34 genes are enriched in only one GO category, GO:0000077: DNA damage checkpoint (Adjusted P-value: 0.0177). Moreover, five of the 34 genes are associated with DNA repair.”. And we added the information in the Figure 5-figure supplements 1C.

Author response image 11.

(C) Genes were significantly more highly expressed in all tissues of the hadal snailfish compared to Tanaka's snailfish, and 5 genes (purple) were associated with DNA repair.

Author Response

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

Reviewer #3 comment

1) One suggestion for improvement is to consider incorporating the results from Figure S9 into in the main Figure 6, which would enhance readers' comprehension.

We appreciate your valuable feedback. Based on the reviewer’s suggestion, we have incorporated results from the Figure S9 into the main Figure 6, as shown below. Manuscripts and figure legends have also been modified accordingly.

Author response image 1.

Author Response

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

We thank the reviewers for truly valuable advice and comments. We have made multiple corrections and revisions to the original pre-print accordingly per the following comments:

1. Pro1153Leu is extremely common in the general population (allele frequency in gnomAD is 0.5). Further discussion is warranted to justify the possibility that this variant contributes to a phenotype documented in 1.5-3% of the population. Is it possible that this variant is tagging other rare SNPs in the COL11A1 locus, and could any of the existing exome sequencing data be mined for rare nonsynonymous variants?

One possible avenue for future work is to return to any existing exome sequencing data to query for rare variants at the COL11A1 locus. This should be possible for the USA MO case-control cohort. Any rare nonsynonymous variants identified should then be subjected to mutational burden testing, ideally after functional testing to diminish any noise introduced by rare benign variants in both cases and controls. If there is a significant association of rare variation in AIS cases, then they should consider returning to the other cohorts for targeted COL11A1 gene sequencing or whole exome sequencing (whichever approach is easier/less expensive) to demonstrate replication of the association.

Response: Regarding the genetic association of the common COL11A1 variant rs3753841 (p.(Pro1335Leu)), we do not propose that it is the sole risk variant contributing to the association signal we detected and have clarified this in the manuscript. We concluded that it was worthy of functional testing for reasons described here. Although there were several common variants in the discovery GWAS within and around COL11A1, none were significantly associated with AIS and none were in linkage disequilibrium (R2>0.6) with the top SNP rs3753841. We next reviewed rare (MAF<=0.01) coding variants within the COL11A1 LD region of the associated SNP (rs3753841) in 625 available exomes representing 46% of the 1,358 cases from the discovery cohort. The LD block was defined using Haploview based on the 1KG_CEU population. Within the ~41 KB LD region (chr1:103365089- 103406616, GRCh37) we found three rare missense mutations in 6 unrelated individuals, Table below. Two of them (NM_080629.2:c.G4093A:p.A1365T; NM_080629.2:c.G3394A:p.G1132S), from two individuals, are predicted to be deleterious based on CADD and GERP scores and are plausible AIS risk candidates. At this rate we could expect to find only 4-5 individuals with linked rare coding variants in the total cohort of 1,358 which collectively are unlikely to explain the overall association signal we detected. Of course, there also could be deep intronic variants contributing to the association that we would not detect by our methods. However, given this scenario, the relatively high predicted deleteriousness of rs3753841 (CADD= 25.7; GERP=5.75), and its occurrence in a GlyX-Y triplet repeat, we hypothesized that this variant itself could be a risk allele worthy of further investigation.

Author response table 1.

We also appreciate the reviewer’s suggestion to perform a rare variant burden analysis of COL11A1. We did conduct pilot gene-based analysis in 4534 European ancestry exomes including 797 of our own AIS cases and 3737 controls and tested the burden of rare variants in COL11A1. SKATO P value was not significant (COL11A1_P=0.18), but this could due to lack of power and/or background from rare benign variants that could be screened out using the functional testing we have developed.

1. COL11A1 p.Pro1335Leu is pursued as a direct candidate susceptibility locus, but the functional validation involves both: (a) a complementation assay in mouse GPCs, Figure 5; and (b) cultured rib cartilage cells from Col11a1-Ad5 Cre mice (Figure 4). Please address the following:

2A. Is Pro1335Leu a loss of function, gain of function, or dominant negative variant? Further rationale for modeling this change in a Col11a1 loss of function cell line would be helpful.

Response: Regarding functional testing, by knockdown/knockout cell culture experiments, we showed for the first time that Col11a1 negatively regulates Mmp3 expression in cartilage chondrocytes, an AIS-relevant tissue. We then tested the effect of overexpressing the human wt or variant COL11A1 by lentiviral transduction in SV40-transformed chondrocyte cultures. We deleted endogenous mouse Col11a1 by Cre recombination to remove the background of its strong suppressive effects on Mmp3 expression. We acknowledge that Col11a1 missense mutations could confer gain of function or dominant negative effects that would not be revealed in this assay. However as indicated in our original manuscript we have noted that spinal deformity is described in the cho/cho mouse, a Col11a1 loss of function mutant. We also note the recent publication by Rebello et al. showing that missense mutations in Col11a2 associated with congenital scoliosis fail to rescue a vertebral malformation phenotype in a zebrafish col11a2 KO line. Although the connection between AIS and vertebral malformations is not altogether clear, we surmise that loss of the components of collagen type XI disrupt spinal development. in vivo experiments in vertebrate model systems are needed to fully establish the consequences and genetic mechanisms by which COL11A1 variants contribute to an AIS phenotype.

2B. Expression appears to be augmented compared WT in Fig 5B, but there is no direct comparison of WT with variant.

Response: Expression of the mutant (from the lentiviral expression vector) is increased compared to mutant. We observed this effect in repeated experiments. Sequencing confirmed that the mutant and wildtype constructs differed only at the position of the rs3753841 SNP. At this time, we cannot explain the difference in expression levels. Nonetheless, even when the variant COL11A1 is relatively overexpressed it fails to suppress MMP3 expression as observed for the wildtype form.

2C. How do the authors know that their complementation data in Figure 5 are specific? Repetition of this experiment with an alternative common nonsynonymous variant in COL11A1 (such as rs1676486) would be helpful as a comparison with the expectation that it would be similar to WT.

Response: We agree that testing an allelic series throughout COL11A1 could be informative, but we have shifted our resources toward in vivo experiments that we believe will ultimately be more informative for deciphering the mechanistic role of COL11A1 in MMP3 regulation and spine deformity.

2D. The y-axes of histograms in panel A need attention and clarification. What is meant by power? Do you mean fold change?

Response: Power is directly comparable to fold change but allows comparison of absolute expression levels between different genes.

2E. Figure 5: how many technical and biological replicates? Confirm that these are stated throughout the figures.

Response: Thank you for pointing out this oversight. This information has been added throughout.

1. Figure 2: What does the gross anatomy of the IVD look like? Could the authors address this by showing an H&E of an adjacent section of the Fig. 2 A panels?

Response: Panel 2 shows H&E staining. Perhaps the reviewer is referring to the WT and Pax1 KO images in Figure 3? We have now added H&E staining of WT and Pax1 KO IVD as supplemental Figure 3E to clarify the IVD anatomy.

1. Page 9: "Cells within the IVD were negative for Pax1 staining ..." There seems to be specific PAX1 expression in many cells within the IVD, which is concerning if this is indeed a supposed null allele of Pax1. This data seems to support that the allele is not null.

Response: We have now added updated images for the COL11A1 and PAX1 staining to include negative controls in which we omitted primary antibodies. As can be seen, there is faint autofluorescence in the PAX1 negative control that appears to explain the “specific staining” referred to by the reviewer. These images confirm that the allele is truly a null.

1. There is currently a lack of evidence supporting the claim that "Col11a1 is positively regulated by Pax1 in mouse spine and tail". Therefore, it is necessary to conduct further research to determine the direct regulatory role of Pax1 on Col11a1.

Response: We agree with the reviewer and have clarified that Pax1 may have either a direct or indirect role in Col11a1 regulation.

1. There is no data linking loss of COL11A1 function and spine defects in the mouse model. Furthermore, due to the absence of P1335L point mutant mice, it cannot be confirmed whether P1335L can actually cause AIS, and the pathogenicity of this mutation cannot be directly verified. These limitations need to be clearly stated and discussed. A Col11a1 mouse mutant called chondroysplasia (cho), was shown to be perinatal lethal with severe endochondral defects (https://pubmed.ncbi.nlm.nih.gov/4100752/). This information may help contextualize this study.

Response: We partially agree with the reviewer. Spine defects are reported in the cho mouse (for example, please see reference 36 Hafez et al). We appreciate the suggestion to cite the original Seegmiller et al 1971 reference and have added it to the manuscript.

1. A recent article (PMID37462524) reported mutations in COL11A2 associated with AIS and functionally tested in zebrafish. That study should be cited and discussed as it is directly relevant for this manuscript.

Response: We agree with the reviewer that this study provides important information supporting loss of function I type XI collagen in spinal deformity. Language to this effect has been added to the manuscript and this study is now cited in the paper.

1. Please reconcile the following result on page 10 of the results: "Interestingly, the AISassociated gene Adgrg6 was amongst the most significantly dysregulated genes in the RNA-seq analysis (Figure 3c). By qRT-PCR analysis, expression of Col11a1, Adgrg6, and Sox6 were significantly reduced in female and male Pax1-/- mice compared to wild-type mice (Figure 3d-g)." In Figure 3f, the downregulation of Adgrg6 appears to be modest so how can it possibly be highlighted as one of the most significantly downregulated transcripts in the RNAseq data?

Response: By “significant” we were referring to the P-value significance in RNAseq analysis, not in absolute change in expression. This language was clearly confusing, and we have removed it from the manuscript.

1. It is incorrect to refer to the primary cell culture work as growth plate chondrocytes (GPCs), instead, these are primary costal chondrocyte cultures. These primary cultures have a mixture of chondrocytes at differing levels of differentiation, which may change differentiation status during the culturing on plastic. In sum, these cells are at best chondrocytes, and not specifically growth plate chondrocytes. This needs to be corrected in the abstract and throughout the manuscript. Moreover, on page 11 these cells are referred to as costal cartilage, which is confusing to the reader.

Response: Thank you for pointing out these inconsistencies. We have changed the manuscript to say “costal chondrocytes” throughout.

Minor points

• On 10 of the Results: "These data support a mechanistic link between Pax1 and Col11a1, and the AIS-associated genes Gpr126 and Sox6, in affected tissue of the developing tail." qRT-PCR validation of Sox6, although significant, appears to be very modestly downregulated in KO. Please soften this statement in the text.

Response: We have softened this statement.

• Have you got any information about how the immortalized (SV40) costal cartilage affected chondrogenic differentiation? The expression of SV40 seemed to stimulate Mmp13 expression. Do these cells still make cartilage nodules? Some feedback on this process and how it affects the nature of the culture what be appreciated.

Response: The “+ or –“ in Figure 5 refers to Ad5-cre. Each experiment was performed in SV40-immortalized costal chondrocytes. We have removed SV40 from the figure and have clarified the legend to say “qRT-PCR of human COL11A1 and endogenous mouse Mmp3 in SV40 immortalized mouse costal chondrocytes transduced with the lentiviral vector only (lanes 1,2), human WT COL11A1 (lane 3), or COL11A1P1335L. Otherwise we absolutely agree that understanding Mmp13 regulation during chondrocyte differentiation is important. We plan to study this using in vivo systems.

• Figure 1: is the average Odds ratio, can this be stated in the figure legend?

Response: We are not sure what is being asked here. The “combined odds ratio” is calculated as a weighted average of the log of the odds.

• A more consistent use of established nomenclature for mouse versus human genes and proteins is needed.

Human:GENE/PROTEIN

Mouse: Gene/PROTEIN

Response: Thank you for pointing this out. The nomenclature has been corrected throughtout the manuscript.

• There is no Figure 5c, but a reference to results in the main text. Please reconcile. -There is no Figure 5-figure supplement 5a, but there is a reference to it in the main text. Please reconcile.

Response: Figure references have been corrected.

• Please indicate dilutions of all antibodies used when listed in the methods.

Response: Antibody dilutions have been added where missing.

• On page 25, there is a partial sentence missing information in the Histologic methods; "#S36964 Invitrogen, CA, USA)). All images were taken..."

Response: We apologize for the error. It has been removed.

• Table 1: please define all acronyms, including cohort names.

Response: We apologize for the oversight. The legend to the Table has been updated with definitions of all acronyms.

• Figure 2: Indicate that blue staining is DAPI in panel B. Clarify that "-ab" as an abbreviation is primary antibody negative.

Response: A color code for DAPI and COL11A! staining has been added and “-ab” is now defined.

• Page 4: ADGRG6 (also known as GPR126)...the authors set this up for ADGRG6 but then use GPR126 in the manuscript, which is confusing. For clarity, please use the gene name Adgrg6 consistently, rather than alternating with Gpr126.

Response: Thank you for pointing this out. GPR126 has now been changed to ADGRG6 thoughout the manuscript.

• REF 4: Richards, B.S., Sucato, D.J., Johnston C.E. Scoliosis, (Elsevier, 2020). Is this a book, can you provide more clarity in the Reference listing?

Response: Thank you for pointing this out. This reference has been corrected.

• While isolation was addressed, the methods for culturing Rat cartilage endplate and costal chondrocytes are poorly described and should be given more text.

Response: Details about the cartilage endplate and costal chondrocyte isolation and culture have been added to the Methods.

• Page 11: 1st paragraph, last sentence "These results suggest that Mmp3 expression"... this sentence needs attention. As written, I am not clear what the authors are trying to say.

Response: This sentence has been clarified and now reads “These results suggest that Mmp3 expression is negatively regulated by Col11a1 in mouse costal chondrocytes.”

• Page 13: line 4 from the bottom, "ECM-clearing"? This is confusing do you mean ECM degrading?

Response: Yes and thank you. We have changed to “ECM-degrading”.

• Please use version numbers for RefSeq IDs: e.g. NM_080629.3 instead of NM_080629 Response: This change has been made in the revised manuscript.

• It would be helpful for readers if the ethnicity of the discovery case cohort was clearly stated as European ancestry in the Results main text.

Response: “European ancestry” has been added at first description of the discovery cohort in the manuscript.

• Avoid using the term "mutation" and use "variant" instead.

Response: Thank you for pointing this out. “Variant” is now used throughout the manuscript.

• Define error bars for all bar charts throughout and include individual data points overlaid onto bars.

Response: Thank you. Error bars are now clarified in the Figure legends.

Author Response

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

Positive comments:

We appreciate the positive comments of the editor and reviewers. The editor noted that the paper presents a “technological advance” that has enabled “important insights about the brain circuits through which the cerebellum could participate in social interactions.” Reviewer 1 thought this was a “timely and important study with solid evidence for correlative conclusions” and that the experiments were “technically challenging” and “well-performed”. Reviewer 2 stated that the finding of correlated activity between the regions is “interesting as non-motor functions of the cerebellum are relatively little explored.” They also thought “that the data are presented clearly, and the manuscript is well-written”. Reviewer 3 mentioned that “this approach can be useful for many neuroscientists”. We thank all the positive comments from the editors and all the reviewers.

Reviewer #1 (Public Review)

While the novelty of the device is strongly emphasized, I find that its value is somewhat diminished by the wire-free device developed by the same group as it should thus be possible to perform calcium imaging wire-free and electrophysiological recording via a single conventional cable (or also via wireless headstages).

While it would be potentially possible to use a wire-free Miniscope in parallel with a wired electrophysiology recording system, this would result in a larger footprint on the animal’s head, more than a gram in increased weight due to an added LiPo battery, a larger electrophysiology head-stage, and limited recording length due to a battery capacity of around 20 minutes. Our main goal for the development of the E-scope platform was to develop an expandable electrophysiology recording board that would work with all previously built UCLA Miniscopes while also streamlining the integration of power and data into the coaxial cable connection already familiar to hundreds of labs using Miniscopes. The vast majority of Miniscope experiments are done using wired systems and we aimed to support the expansion of those systems instead of requiring a more substantial switch to using wire-free Miniscopes.

The role of the identified network activations in social interactions is not touched upon.

We agree with the reviewer that we have not discovered a causal role for the co-modulated activity patterns we have observed. As these causal experiments will require the development of real-time techniques for blocking socially evoked changes in firing rate in cerebellum and ACC, we are currently planning experiments to address causality. These results will be described in a future publication.

Reviewer #1 (Recommendations for the Authors):

Please provide the number of recorded mice.

The number is now provided in the revised manuscript.

If the recorded areas (cerebellar cortex, DN, and ACC) are part of the same circuit regulating social interactions, it would be nice to get insights into the directionality of the circuit. The authors favor the possibility that during social behavior, cerebellar efferences indirectly influence ACC activities (as in Figure 4A), however, no evidence is presented to support this interpretation. ACC activities might also indirectly influence PC firing. It may be possible to get insights into this by comparing the timing of neuronal activity in the different areas with respect to social onset.

For this study, we mainly focused on the output of the cerebellar circuit to the cortex as previous work shows that dentate nucleus projects to the thalamus, which in turn projects to ACC and other cortical regions. (Badura et al.,eLife, 2018; Kelly et al., Nat. Neurosci., 2020) The temporal resolution of calcium imaging is limited (with the rise time of calcium events with genetically-encoded indicators taking hundreds of milliseconds) such that the resolution is insufficient to precisely assess the relative onset timing of the two regions. Our work certainly does not rule out cortical influences on PC firing.

Reviewer #2 (Public Review)

However, the causal relationship is far from established with the methods used, leaving it unclear if these two brain regions are similarly engaged by the behavior or if they form a pathway/loop.

As indicated in our response to Reviewer #1’s similar critique, the goal of the presented study is to demonstrate the feasibility and capabilities of this novel device. This new tool will allow us to conduct a comprehensive and rigorous study to assess the causal role of the interactions between the cerebellum and ACC in social behavior (as well as other behaviors). These experiments are being designed now.

Reviewer #2 (Recommendations for the Authors):

It is unclear what is entirely unique about the E-scope. It seems that its advance is simply a common cable that allows interfacing with both devices (lighter weight than two cables is stated in the Discussion). Is this really an advance? What are its limitations? E.g., how close can the recording sites be to one another? How can it be configured for any other extracellular recording approach (tetrodes, 64-channel arrays, or Neuropixels)?

In our experience, multiple lines of wires tethered to different head-mounted devices on an animal significantly impacts their behavior. Therefore, one of the major advantages of the UCLA Miniscope Platform is the use of a single, flexible coaxial cable to minimize the impact on tethering on behavior. The E-Scope platform builds on top of this work by incorporating electrophysiology recording capabilities into this single, flexible coaxial cable. Additionally, the electrophysiology recording hardware is backwards compatible with all previously built UCLA Miniscopes and can run through open-source and commercial commutators already used in Miniscope experiments.

The available bandwidth within the shared single coaxial cable can handle megapixel Miniscope imaging along with the maximum data output of a 32 channel Intan Ephys IC. The E-Scope platform presented here does run the Intan Ephys IC at 20KSps for all 32 channels instead of the maximum 30KSps due to microcontroller speed limitations, but this could be overcome by using a fast microcontroller or clock, or slightly reducing the total number of electrodes samples. Finally, the E-Scope was designed to support any electrode types supported by the Intan Ephys IC. This includes up to 32 channels of passive probes such as single electrodes, tetrodes, silicon probes, and flexible multi-channel arrays but does not include Neuropixels as Neuropixels use custom active electronics on the probe to multiplex, sample, and serialize electrophysiology data.

The authors only analyzed simple spikes in PCs for social-related activity. What about complex spikes? Is this correlated with ACC activity?

Complex spikes were detectable to the extent that we were able to define that the recorded cell was a PC, but because these cells were recorded in freely behaving mice, accurate complex spike detection was not reliable enough to be used for further correlational analyses.

The data is sampled in the two regions (cerebellum and ACC) at very different rates (imaging is much slower than electrophysiology; ephys data was binned). How does this affect the correlation plots?

We generated firing rate maps for the cerebellar neural activity using a binning size that matched the sampling frequency of calcium imaging (see Methods). As mentioned in our methods, to study the relationship between the electrophysiology and calcium imaging data we binned the spike trains using 33 ms bins to match the calcium imaging sampling rate (~30 Hz). This limits the temporal resolution to calculate fine-scale correlations, but the correlations that we report are on a behaviorally relevant temporal scale. The fine temporal resolution of the electrophysiology data however can still be used to further examine at a higher temporal resolution the relationship between cerebellar output and specific social behavior epochs.

For the correlation analysis, over what time frame was the activity relationship examined? How was this duration determined?

Author response image 1.

The main criteria for the time frame used to study the correlation analysis was the behavioral timescale of social interaction [see figure above for the number of social (red) and object (blue) interaction bouts (a), their duration (b) and coefficient of variation (CV) (c)]. Overall, the activity relationship time frame was based on the average duration of the social interactions (~3 sec). Periods of 3.8 before and 5.8 sec after interaction onset were used to study. Accordingly, the cross-correlograms were constructed using a maximum lag length of 5 sec. In the article we reported correlation at lag 0.

The relationship between the cerebellum and ACC seems unconvincing. If two brain regions are similarly engaged by the behavior, wouldn't they have a high correlation? Is the activity in one region driving the other?

We reference studies showing an anatomical and functional indirect connection between the cerebellum and the ACC or prefrontal cortex (Badura et al., eLife, 2018). Also, as stated in the introduction, the ACC is a recognized brain area for social behavioral studies. In the results, we stated that correlations increase in groups of neurons that are similarly engaged during a specific epoch in the social interaction was an expected finding. What was not expected was that there would be no difference in the distribution's correlation when the social epochs were removed, suggesting that intrinsic connectivity does not drive a difference in correlations.

Although, since there is a cerebello-cortical loop, further study will be needed to understand which area initiates this type of activity during social behavior,

• In the figures, the color-coded scale bars should be labeled as z-scores (confusing without them).

• In Figure 4, the color differences for Soc-ACC, Soc+ACC and SocNS ACC should be more striking as it is hard to tell them apart because they are all similar shades of blue-gray.

We thank the reviewer for their suggestions for improving the figures. We have incorporated these changes in Figures 2, 3 and along with their figure supplements. Graphs in Figure 4D-G have been edited to make the lines more visible to the reader.

Reviewer #3 (Public Review)

However, a mouse weighs between 20 and 40 g, so that an implant of 4.5 g is still quite considerable. It can be expected that this has an impact on the behavior and, possibly, the well-being of the animals. Whether this is the case or not, is not really addressed in this study.

The weight of the E-Scope (4.5 g) is near the maximum that is tolerated by animals in our experience. We therefore acclimated the mouse to the weight with dummy scopes of increasing weights over a 7-10 day period. During this period, we observed the animal to have normal exploratory behavior. Specifically, there is no change in the sociability of the animals (Figure 2A) and animals cover the large arena (48x 48 cm, Figure 2H).

Overall, the description of animal behavior is rather sparse. The methods state only that stranger age-matched mice were used, but do not state their gender. The nature of the social interactions was not described? Was their aggressive behavior, sexual approach and/or intercourse? Did the stranger mice attack/damage the E-Scope? Were the interactions comparable (using which parameters?) with and without E-Scope attached? It is not even described what the authors define as an "interaction bout" (Figure 2A). The number of interaction bouts is counted per 7 minutes, I presume? This is not specified explicitly.

As mentioned in the methods section of the original version of our manuscript, all the target mice were age-matched “male” mice. As per the reviewer’s suggestion, we now have added in the manuscript that before any of our social interaction behavioral experiments, aggressive or agitated mice were removed after assessing their behavior in the arena during habituation. For all trials, all mice were introduced for the first time.

We also mention in the methods section of our manuscript, that social behaviors were evaluated by proximity between the subject mouse and novel target mouse (2 cm from the body, head, or base of tail). From our recordings, we did not observe any aggressive, mounting, nor any other dominance behavior over the E-Scope subject mouse during the 7 minutes of social interaction assessment. Social interaction bouts in Figure 2A show the average number of social interaction bouts during the recording time. This has now been expanded upon in our revised manuscript.

It would be very insightful if the authors would describe which events they considered to be action potentials, and which not. Similarly, the raw traces of Figure 1E are declared to be single-unit recordings of Purkinje cells. Partially due to the small size of the traces (invisible in print and pixelated in the digital version), I have a hard time recognizing complex spikes and simple spikes in these traces. This is a bit worrisome, as the authors declare the typical duration of the pause in simple spike firing after a complex spike to be 20-100 ms. In my experience, such long pauses are rare in this region, and definitely not typical. In the right panel of Figure 1A, an example of a complex spike-induced pause is shown. This pause is around 15 ms, so not typical according to the text, and starts only around 4 ms after the complex spike, which should not be the case and suggests either a misalignment of the figure or the detection of complex spike spikelets as simple spikes, while the abnormally long pause suggests that the authors fail to detect a lot of simple spikes. The authors could provide more confidence in their data by including more raw data, making explicit how they analyzed the signals, and by reporting basic statistics of firing properties (like rate, cv or cv2, pause duration). In this respect, Figure 2 - figure supplement 3 shows quite a large percentage of cells to have either a very low or a very high firing rate.

We now provide a better example of simple spikes and complex spikes in Fig 1E and corrected our comment in the body of the manuscript. Previous version of the SS x CS cross-correlation histogram in Figure 1G as the reviewer mentions, was not the best example, because of the detected CS spikelets. However, the detection of CS spikelets has little impact on the interpretation of the results. We have replaced this figure with a better example of the SS x CS cross-correlation histogram.

The number of Purkinje cells recorded during social interactions is quite low: only 11 cells showed a modulation in their spiking activity (unclear whether in complex spikes, simple spikes or both. During object interaction, only 4 cells showed a significant modulation. Unclear is whether the latter 4 are a subset of the former 11, or whether "social cells" and "object cells" are different categories. Having so few cells, and with these having different types of modulation, the group of cells for each type of modulation is really small, going down to 2 cells/group. It is doubtful whether meaningful interpretation is possible here.

While the number of neurons is not as high as those reported for other regions, the number presented depicts the full range of responses to social behavior. It is extremely difficult to obtain stable neurons in freely behaving socially interacting animals and only a handful of neurons could be recorded in each animal. Among these recorded neurons only a subset responds to social interactions further reducing the numbers. The results however are consistent among cell types and the direction of modulation fits with the inhibitory connectivity between PCs and DN neurons. To our knowledge, we are the first group to publish neuronal activity of PC and DN neurons from freely behaving mice during social behavior.

Neural activity patterns observed during social interaction do not necessarily relate specifically to social interaction, but can also occur in a non-social context. The authors control this by comparing social interactions with object interactions, but I miss a direct comparison between the two conditions, both in terms of behavior (now only the number of interactions is counted, not their duration or intensity), and in terms of neural activity. There is some analysis done on the interaction between movement and cerebellar activity (Figure 2 - figure supplement 4), but it is unclear to what extent social interactions and movements are separated here. It would already help to indicate in the plots with trajectories (e.g., Fig. 2H) indicate the social interactions (e.g., social interaction-related movements in red, the rest of the trajectories in black).

We have updated the social interaction plots in Figure 2H in the revised version of the manuscript.

Reviewer #3 (Recommendations for the Authors):

Increase the number of cerebellar neurons that are recorded.

Due to the difficulty of the experiment and the low yield which we get for cerebellar recordings, substantially increasing the number of neurons will require many more experiments which are not feasible at this time.

Include more raw data and make the analysis procedure more insightful with illustrations of intermediate steps.

We have included a more thorough description of the analysis in the methods section of the revised manuscript.

Provide a better description of the behavior.

We have increased the level of detail regarding the mouse behavior in the Results and Methods sections. This includes a more detailed description of the parameters we used to analyze the social interaction.

Author Response

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

eLife assessment

This valuable paper examines gene expression differences between male and female individuals over the course of flower development in the dioecious angiosperm Trichosantes pilosa. The authors show that male-biased genes evolve faster than female-biased and unbiased genes. This is frequently observed in animals, but this is the first report of such a pattern in plants. In spite of the limited sample size, the evidence is mostly solid and the methods appropriate for a non-model organism. The resources produced will be used by researchers working in the Cucurbitaceae, and the results obtained advance our understanding of the mechanisms of plant sexual reproduction and its evolutionary implications: as such they will broadly appeal to evolutionary biologists and plant biologists.

Public Reviews:

Reviewer #1 (Public Review):

The evolution of dioecy in angiosperms has significant implications for plant reproductive efficiency, adaptation, evolutionary potential, and resilience to environmental changes. Dioecy allows for the specialization and division of labor between male and female plants, where each sex can focus on specific aspects of reproduction and allocate resources accordingly. This division of labor creates an opportunity for sexual selection to act and can drive the evolution of sexual dimorphism.

In the present study, the authors investigate sex-biased gene expression patterns in juvenile and mature dioecious flowers to gain insights into the molecular basis of sexual dimorphism. They find that a large proportion of the plant transcriptome is differentially regulated between males and females with the number of sex-biased genes in floral buds being approximately 15 times higher than in mature flowers. The functional analysis of sex-biased genes reveals that chemical defense pathways against herbivores are up-regulated in the female buds along with genes involved in the acquisition of resources such as carbon for fruit and seed production, whereas male buds are enriched in genes related to signaling, inflorescence development and senescence of male flowers. Furthermore, the authors implement sophisticated maximum likelihood methods to understand the forces driving the evolution of sex-biased genes. They highlight the influence of positive and relaxed purifying selection on the evolution of male-biased genes, which show significantly higher rates of non-synonymous to synonymous substitutions than female or unbiased genes. This is the first report (to my knowledge) highlighting the occurrence of this pattern in plants. Overall, this study provides important insights into the genetic basis of sexual dimorphism and the evolution of reproductive genes in Cucurbitaceae.

Reviewer #2 (Public Review):

Summary:

This study uses transcriptome sequence from a dioecious plant to compare evolutionary rates between genes with male- and female-biased expression and distinguish between relaxed selection and positive selection as causes for more rapid evolution. These questions have been explored in animals and algae, but few studies have investigated this in dioecious angiosperms, and none have so far identified faster rates of evolution in male-biased genes (though see Hough et al. 2014 https://doi.org/10.1073/pnas.1319227111).

Strengths:

The methods are appropriate to the questions asked. Both the sample size and the depth of sequencing are sufficient, and the methods used to estimate evolutionary rates and the strength of selection are appropriate. The data presented are consistent with faster evolution of genes with male-biased expression, due to both positive and relaxed selection.

This is a useful contribution to understanding the effect of sex-biased expression in genetic evolution in plants. It demonstrates the range of variation in evolutionary rates and selective mechanisms, and provides further context to connect these patterns to potential explanatory factors in plant diversity such as the age of sex chromosomes and the developmental trajectories of male and female flowers.

Weaknesses:

The presence of sex chromosomes is a potential confounding factor, since there are different evolutionary expectations for X-linked, Y-linked, and autosomal genes. Attempting to distinguish transcripts on the sex chromosomes from autosomal transcripts could provide additional insight into the relative contributions of positive and relaxed selection.

Reviewer #3 (Public Review):

The potential for sexual selection and the extent of sexual dimorphism in gene expression have been studied in great detail in animals, but hardly examined in plants so far. In this context, the study by Zhao, Zhou et al. al represents a welcome addition to the literature.

Relative to the previous studies in Angiosperms, the dataset is interesting in that it focuses on reproductive rather than somatic tissues (which makes sense to investigate sexual selection), and includes more than a single developmental stage (buds + mature flowers).<br /> Some aspects of the presentation have been improved in this new version of the manuscript.

Specifically:

• the link between sex-biased and tissue-biased genes is now slightly clearer,

• the limitation related to the de novo assembled transcriptome is now formally acknowledged,

• the interpretation of functional categories of the genes identified is more precise,

• the legends of supplementary figures have been improved - a large number of typos have been fixed.

in response to this first round of reviews. As I detail below, many of the relevant and constructive suggestions by the previous reviewers were not taken into account in this revision.

For instance:

• Reviewer 2 made precise suggestions for trying to take into account the potential confounding factor of sex-chromosomes. This suggestion was not followed.

For the question of reviewer 2:

The presence of sex chromosomes is a potential confounding factor, since there are different evolutionary expectations for X-linked, Y-linked, and autosomal genes. Attempting to distinguish transcripts on the sex chromosomes from autosomal transcripts could provide additional insight into the relative contributions of positive and relaxed selection.

Empirically, the analyses could be expanded by an attempt to distinguish between genes on the autosomes and the sex chromosomes. Genotypic patterns can be used to provisionally assign transcripts to XY or XX-like behavior when all males are heterozygous and all females are homozygous (fixed X-Y SNPs) and when all females are heterozygous and males are homozygous (lost or silenced Y genes). Comparing such genes to autosomal genes with sex-biased expression would sharpen the results because there are different expectations for the efficacy of selection on sex chromosomes. See this paper (Hough et al. 2014; https://www.pnas.org/doi/abs/10.1073/pnas.1319227111), which should be cited and does in fact identify faster substitution rates in Y-linked genes.

Authors’ response: We have cited Hough et al. (2014) and Sandler et al. (2018) in the revised manuscript. We agree that the presence of sex chromosomes is potentially a confounding factor. By adopting methods in Hough et al. (2014) and Sandler et al. (2018), we tried to distinguish transcripts on sex chromosomes from autosomal chromosomes. For a total of 2,378 unbiased genes, we found that 36 genes were putatively sex chromosomal genes, 20 of which were exclusively heterozygous and homozygous for males and females, respectively; while the other 16 genes showing an opposite genotyping patterns between males and females. For 343 male-biased genes, only three ones exhibit a pattern of potentially sex-linked. For the 1,145 female-biased genes, we identified 19 genes which might located on the sex chromosomes. Among the 19 genes, five genes were exclusively heterozygous for males and exclusively homozygous for females, while reversed genotyping patterns presented in the other 14 genes. So, sex-linked genes may contribute relatively little to rapid evolution of male-biased genes. An alternative explanation is that the results could be unreliable due to small sample sizes. Thus, we did not describe them in the Results section. We will investigate the issue when whole genome sequences and population datasets become available in the near future.

• Reviewer 1 & 3 indicated that results were mentioned in the discussion section without having been described before. This was not fixed in this new version.

For the question of reviewer 1:

2) Paragraph (407-416) describes the analysis of duplicated genes under relaxed selection but there is no mention of this in the results.

Authors’ response: Following this suggestion, in the Results section, we have added a sentence, “We also found that most of them were members of different gene families generated by gene duplication (Table S13)” on line 310-311 in the revised manuscript (Rapid_evolution_of_malebiased_genes_Trichosanthes_pilosa_Tracked_change_2023_11_06.docx).

For the question of reviewer 1:

38- line 417-424. The discussion should not contain new results.

Authors’ response: Thank you for pointing out this. In the Results section, we have added a few sentences as following: “Similarly, given that dN/dS values of sex-biased genes were higher due to codon usage bias, lower dS rates would be expected in sex-biased genes relative to unbiased genes (Ellegren & Parsch, 2007; Parvathy et al., 2022). However, in our results, the median of dS values in male-biased genes were much higher than those in female-biased and unbiased genes in the results of ‘free-ratio’ (Fig. S4A, female-biased versus male-biased genes, P = 6.444e-12 and malebiased versus unbiased genes, P = 4.564e-13) and ‘two-ratio’ branch model (Fig. S4B, femalebiased versus male-biased genes, P = 2.2e-16 and male-biased versus unbiased genes, P = 9.421e08, respectively). ” on line 323-331, and consequently, removed the following sentence, “femalebiased vs male-biased genes, P = 6.444e-12 and male-biased vs unbiased genes, P = 4.564e-13” and “female-biased versus male-biased genes, P = 2.2e-16 and male-biased versus unbiased genes, P = 9.421e-08, respectively” in the Discussion section.

• Reviewer 1 asked for a comparison between the number of de novo assembled unigenes in this transcriptome and the number of genes in other Cucurbitaceae species. I could not see this comparison reported.

Authors’ response: In the first revision, we described only percentages. We have now added the number of genes. We modify this part as follows: “The majority of unigenes were annotated by homologs in species of Cucurbitaceae (61.6%, 36,375), including Momordica charantia (16.3%, 9,625), Cucumis melo (11.9%, 7,027), Cucurbita pepo (11.9%, 7,027), Cucurbita moschata (11.5%, 6,791), Cucurbita maxima (10.1%, 5,964) and other species (38.4%, 22,676) (Fig. S1C).”.

• Reviewer 1 pointed out that permutation tests were more appropriate, but no change was made to the manuscript.

Authors’ response: Thank you for your suggestion. In the first revision, we have indirectly responded to the issues. Wilcoxon rank sum test is more commonly used for all comparisons between sex-biased and unbiased genes in many papers. Additionally, we tested datasets using permutation t-tests, which is consistent with the results of Wilcoxon rank sum test. For example, we found that only in floral buds, there are significant differences in ω values in the results of ‘free-ratio’ (female-biased versus male-biased genes, P = 0.04282 and male-biased versus unbiased genes, P = 0.01114) and ‘two-ratio’ model (female-biased versus male-biased genes, P = 0.01992 and male-biased versus unbiased genes, P = 0.02127, respectively). We also described these results in the Results section accordingly (line 278-284).

• Reviewer 3 pointed out the small sample size (both for the RNA-seq and the phylogenetic analysis), but again this limitation is not acknowledged very clearly.

Authors’ response: Sorry, we acknowledged that our sample size was relatively small. In the revised version, we have added a sentence as follows, “Additionally, our sample size is relatively small, and may provide low power to detect differential expression.” in the Discussion section.

• Reviewer 1 & 3 pointed out that Fig 3 was hard to understand and asked for clarifications that I did not see in the text and the figure in unchanged.

Authors’ response: Thank you for your suggestions. We have revised the manuscript to clarify the meaning of the acronym (F1TGs, F2TGs, M1TGs, M2TGs, F1BGs, F2BGs, M1BGs and M2BGs) and presented the number of genes. We have added two labels, indicating that panels A and B correspond to males and C and D to females in Fig. 3.

• Reviewer 3 suggested to combine all genes with sex-bias expression when evaluating the evolutionary rate, in addition to the analyses already done. This suggestion was not followed.

For the question of reviewer 3：line 196 and following: In these analyses, I could not understand the rationale for keeping buds vs mature flowers as separate analyses throughout. Why not combine both and use the full set of genes showing sex-bias in any tissue? This would increase the power and make the presentation of the results a lot more straightforward.

Authors’ response: Thank you for your suggestions. In the first revision, we tried to respond to the issues. First, we observed strong sexual dimorphism in floral buds, such as racemose versus solitary, early-flowering versus late-flowering. Second, as you pointed out earlier, “the dataset is interesting in that it focuses on reproductive rather than somatic tissues (which makes sense to investigate sexual selection), and includes more than a single developmental stage (buds + mature flowers)”, we totally agree with you on this point. Third, according to your suggestions, we combined all genes with sex-bias expression to evaluate the evolutionary rates. We found significant differences (please see a Figure below) in ω values in the results of ‘free-ratio’ (female-biased versus male-biased genes, P =0.005622 and male-biased versus unbiased genes, P = 0.001961) and ‘two-ratio’ model (female-biased versus male-biased genes, P = 0.008546 and male-biased versus unbiased genes, P = 0.009831, respectively) using Wilcoxon rank sum test. However, the significance is lower than previous results in floral buds due to sex-biased genes of mature flower joined, especially compared to the results of “free-ratio model”. Additionally, we also test all combined genes with sex-bias expression using permutation t-test. Unfortunately, there are no significant differences in ω values expect for male-biased versus unbiased genes in the results of ‘free-ratio’ model (P = 0.03034) and ‘two-ratio’ model (P = 0.0376), respectively. To a certain extent, the combination of all genes with sex-bias expression may cover the signals of rapid evolution of sex-biased genes in floral buds. Therefore, these results are not described in our manuscript. In the near future, we would like to make further investigations through more development stages of flowers and new technologies (e.g. Single-Cell method, See Murat et al., 2023) in each sex to consolidate the conclusion, and it is hoped that we could find more meaningful results.

Author response image 1.

• Reviewer 3 pointed out that hand-picking specific categories of genes was not statistically valid, and in fact not necessary in the present context. This was not changed.

For the question of reviewer3: removing genes on a post-hoc basis seems statistically suspicious to me. I don't think your analysis has enough power to hand-pick specific categories of genes, and it is not clear what this brings here. I suggest simply removing these analyses and paragraphs.

Authors’ response: Thank you for your suggestions. We have changed them accordingly. We removed a part of the following paragraph, “To confirm the contributions of positive selection and relaxed selection to rapid rates of male-biased genes in floral buds, we generated three datasets of OGs by excluding different sets of genes. Specifically, we excluded 18 relaxed selective male-biased genes (5.23%), 98 positively selected male-biased genes (28.57%), and 112 male-biased genes (32.65%) under positive and relaxed selection from 343 OGs (Fig. S4). We observed that after excluding male-biased genes under relaxed purifying selection, the median (0.264) decreased by 0.34% compared to the median (0.265) of all OGs (Fig. S4A-B). However, after excluding positively selected male-biased genes, the median (0.236) was reduced by 11% (Fig. S4A, C) in the results of ‘free-ratio’ branch model. This pattern was consistent with the results of ‘two-ratio’ branch model as well (Fig. S4E-G).” on line 290 to 300.

However, we kept the following paragraph, “We also analyzed female-biased and unbiased genes that underwent positive and relaxed selection in floral buds (Tables S6-S10). We identified 216 (18.86%) positively selected, and 69 (6.03%) relaxed selective female-biased genes from 1,145 OGs, respectively. Similarly, we found 436 (18.33%) positively selected, and 43 (1.81%) unbiased genes under relaxed selection from 2,378 OGs, respectively. Notably, male-biased genes have a higher proportion (10%) of positively selected genes compared to female-biased and unbiased genes. However, relaxed selective male-biased genes have a higher proportion (3.24%) than unbiased genes, but about 0.8% lower than that of female-biased genes.”. In this way, we can compare the proportion of sex-biased genes that have undergone positive selection and release selection among female-biased genes, unbiased genes and male-biased genes in floral buds in the Discussion section.

• Reviewer 1 asked for all data to be public, but I could not find in the manuscript where the link to the data on ResearchGate was provided.

Authors’ response: We have added a link in the Data Availability section.

• Reviewers 1 & 3 pointed out that since only two tissues were compared, the claims on pleiotropy should have been toned down, but no change was made to the text.

Authors’ response: Thank you for your suggestions. We revised “due to low pleiotropic constraints” to “due to low evolutionary constraints” and revised “low pleiotropy” to “low constraints”.

• Reviewer 1 asked for a clarification on which genes are plotted on the heatmap of Fig3C and an explanation of the color scale. No change was made.

Authors’ response: Sorry for the confusion. Actually, Reviewer 1 asked that “Fig. 2C, which genes are plotted on the heatmap and what is the color scale corresponding to?” In the previous revision, we have revised them (See Fig. 2 Sex-biased gene expression for floral buds and flowers at anthesis in males and females of Trichosanthes pilosa). Sex-biased genes (the union of sex-biased genes in F1, M1, F2 and M2) are plotted on the heatmap. The color gradient represents from high to low (from red to green) gene expression.

• Reviewer 1 asked for panel B in Fig S5 and S6 to be removed. They are still there. They asked for abbreviations to be explained in the legend of Fig S8. This was not done. They asked for details about columns headers. Such detailed were not added. They asked for more recent references on line 53-56: this was not done.

Authors’ response: We have removed panel B in Fig. S5 and S6. We explained abbreviations in text and Fig. S8. We added more details about the column headers in Supplementary Table S4, S5, S6, S7, S8, S9 and S10. We also added more recent references on line 53-56.

Recommendations for the authors:

Reviewer #3 (Recommendations For The Authors):

Authors’ response: Thank you for your suggestions. We have revised/fixed these issues following your concerns and suggestions.

Line 46-48 would be clearer as « Sexual dimorphism is the condition where sexes of the same species exhibit different morphological, ecological and physiological traits in gonochoristic animals and dioecious plants, despite male and female individuals sharing the same genome except for sex chromosomes or sex-determining loci »

Authors’ response: Thanks. We have revised it accordingly.

Line 50: replace «in both » by «between the two »

Authors’ response: We have revised it.

Line 51: « genes exclusively » -> « genes expressed exclusively »

Authors’ response: We have revised it.

Line 58: « in many animals » -> « in several animal species »

Authors’ response: We have revised it to “in some animal species”.

Line 58: « to which » -> « of this bias »

Authors’ response: We have revised it.

Line 64: « Most dioecious plants possess homomorphic sex-chromosomes that are roughly similar in size when viewed by light microscopy. » : a reference is missing

Authors’ response: We have added the reference.

Line 67: remove « that »

Authors’ response: We have revised it.

line 96: change to: « only the five above-mentioned studies »

Authors’ response: We have revised it.

Line 97: remove « the »

Authors’ response: We have revised it.

Line 111: « Drosophia » -> Drosophila

Authors’ response: We have revised it.

Line 114: exhibiting -> « exhibited »

Authors’ response: We have revised it.

Line 115: suggest -> « suggesting »

Authors’ response: We have revised it.

Line 117: « studies in plants have rarely reported elevated rates of sex-biased genes » : is it « rarely » or « never » ?

Authors’ response: We have revised to “never”.

Line 143: « It’s » -> « Its »

Authors’ response: We have revised it.

Line 143-146: say whether the male parts (e.g. anthers) are still present in females flowers, and the female parts (pistil+ ovaries) in the male flowers, or whether these respective organs are fully aborted.

Authors’ response: We have added the following sentence, “The male parts (e. g., anthers) of female flowers, and the female parts (e. g., pistil and ovaries) of male flowers are fully aborted” in line 148150 of the Introduction section.

Line 158: this is now clearer, but please specify whether you are talking about 12 floral buds in total, or 12 per individual (i.e. 72 buds in total).

Authors’ response: We have revised it to “Using whole transcriptome shotgun sequencing, we sequenced floral buds and flowers at anthesis from female and male of dioecious T. pilosa. We set up three biological replicates from three female and three male plants, including 12 samples in total (six floral buds and six flowers at anthesis)”.

Line 194-198: These sentences are unclear and hard to link to the figure. Consider changing for « In male plants, the number of tissue-biased genes in flowers at anthesis (M2TGs: n = 2795) was higher than that in floral buds (M1TGs: n = 1755, Fig. 3A and 3B). Figure 3 is also very hard to read. Adding a label on the side to indicate that panels A and B correspond to male-biased genes and C and D to female-biased genes could be useful.

Authors’ response: Thank you for your suggestions. We have revised the text to clarify the meaning of the acronym (F1TGs, F2TGs, M1TGs, M2TGs, F1BGs, F2BGs, M1BGs and M2BGs) and presented the number of genes. We have added two labels, indicating that panels A and B correspond to males and C and D to females in Figure 3.

Line 208: explain the approach: e.g. « We then compared rates of protein evolution among malebiased, female-biased and unbiased genes. To do this, we sequenced floral bud transcriptomes from the closely related T. anguina, as well as two more distant outgroups, T. kirilowii and Luffa cylindrica. T. kirilowii is a dioecious species like T. pilosa, and the other two are monoecious. We identified one-to-one orthologous groups (OGs) for 1,145 female-biased, 343 male-biased, and 2,378 unbiased genes. »

Authors’ response: We have revised this paragraph to the following, “We compared rates of protein evolution among male-biased, female-biased and unbiased genes in four species with phylogenetic relationships (((T. anguina, T. pilosa), T. kirilowii), Luffa cylindrica), including dioecious T. pilosa, dioecious T. kirilowii, monoecious T. anguina in Trichosanthes, together with monoecious Luffa cylindrica. To do this, we sequenced transcriptomes of T. pilosa. We also collected transcriptomes of T. kirilowii, as well as genomes of T. anguina and Luffa cylindrica.”

Line 220: « the same ω value was in all branches » -> « all branches are constrained to have the same ω value ».

Authors’ response: We have revised it.

Line 221: « results of the 'two-ratio' branch model ... »

Authors’ response: We have revised it.

Line 235: add a few words to explain why the effect size is bigger than for buds, but still is not significant: e.g. «possibly because of limited statistical power due to the low number of sex-biased genes in flowers at anthesis »

Authors’ response: We have revised this to “However, there is no statistically significant difference in the distribution of ω values using Wilcoxon rank sum tests for female-biased versus male-biased genes (P = 0.0556), female-biased versus unbiased genes (P = 0.0796), and male-biased versus unbiased genes (P = 0.3296) possibly because of limited statistical power due to the low number of sex-biased genes in flowers at anthesis.” in line 260-261.

Line 255: explain in plain English what the « A model » is. This was already requested in the previous version.

Authors’ response: We have revised “A model” to “classical branch-site model A”.

Line 258: explain in plain English what the « foreground 2b ω value » corresponds to

Authors’ response: We have revised to as follows, “foreground 2b ω value” to “foreground ω >1”. Additionally, we also added the sentence “The classical branch-site model assumes four site classes (0, 1, 2a, 2b), with different ω values for the foreground and background branches. In site classes 2a and 2b, the foreground branch undergoes positive selection when there is ω > 1.” in line 624-627.

Line 259: explain how these different approaches complement each other rather than being redundant. This was also already requested in the previous version.

Authors’ response: Sorry. We have now revised it as follows, “As a complementary approach, we utilized the aBSREL and BUSTED methods that are implemented in HyPhy v.2.5 software, which avoids false positive results by classical branch-site models due to the presence of rate variation in background branches, and detected significant evidence of positive selection.” in line 292-295.

Line 270: remove « dramatically », and also remove « or eliminated at both gene-wide and genomewide levels », as well as « relative to positive selection »

Authors’ response: Thank you for your suggestions. We have revised it.

Line 290-309: remove this section - this was already pointed out in the previous reviews as a « ad hoc » procedure, and this point has already been made clear with the RELAX analysis.

Authors’ response: Thank you for your suggestions. We revised this section accordingly. We remove the following paragraph, “To confirm the contributions of positive selection and relaxed selection to rapid rates of male-biased genes in floral buds, we generated three datasets of OGs by excluding different sets of genes. Specifically, we excluded 18 relaxed selective male-biased genes (5.23%), 98 positively selected male-biased genes (28.57%), and 112 male-biased genes (32.65%) under positive and relaxed selection from 343 OGs (Fig. S4). We observed that after excluding malebiased genes under relaxed purifying selection, the median (0.264) decreased by 0.34% compared to the median (0.265) of all OGs (Fig. S4A-B). However, after excluding positively selected malebiased genes, the median (0.236) was reduced by 11% (Fig. S4A, C) in the results of ‘free-ratio’ branch model. This pattern was consistent with the results of ‘two-ratio’ branch model as well (Fig. S4E-G).” on line 334-344.

However, we kept the other parts “We also analyzed female-biased and unbiased genes that underwent positive and relaxed selection in floral buds (Tables S6-S10). We identified 216 (18.86%) positively selected, and 69 (6.03%) relaxed selective female-biased genes from 1,145 OGs, respectively. Similarly, we found 436 (18.33%) positively selected, and 43 (1.81%) unbiased genes under relaxed selection from 2,378 OGs, respectively. Notably, male-biased genes have a higher proportion (10%) of positively selected genes compared to female-biased and unbiased genes. However, relaxed selective male-biased genes have a higher proportion (3.24%) than unbiased genes, but about 0.8% lower than that of female-biased genes.”. In this way, we can compare the proportion of sex-biased genes that have undergone positive selection and release selection among female-biased genes, unbiased genes and male-biased genes in floral buds in the Discussion sections.

Line 348: Here you talk about « Numerous studies », but then only report three studies. Please clarify.

Authors’ response: Thank you for your suggestions. We have revised it to “Several studies”.

Line 352: Cut the sentence: « In contrast, the wind-pollinated dioecious plant Populus balsamifera ... »

Authors’ response: Thank you for your suggestions. We have revised it.

Line 357: « In contrast to the above studies... »: If I understand correctly, this is not in contrast to the observation in Populus balsamifera. Please clarify.

Authors’ response: Thank you for your suggestions. We have revised to “Similar to the above study of Populus balsamifera.”.

Line 420: « our results » -> « we »; « that underwent » -> « undergoing »

Authors’ response: Thank you for your suggestions. We have revised it.

Figure 3 is very hard to read and poorly labeled (see my comments on line 194 above). It is also hard to link to the text, since the numbers reported in the text are actually not present in the figure unless the readers makes some calculations themselves. This should be improved. Also, the use of acronyms (e.g. M1BG, F2TG etc.) contributes to making the text very difficult to read. The acronyms should at least be explained very clearly in the text when they are used.

Authors’ response: Thank you for your suggestions. We have revised the text to clarify the meaning of the acronym (F1TGs, F2TGs, M1TGs, M2TGs, F1BGs, F2BGs, M1BGs and M2BGs) and give the number of genes. We have added two labels, indicating that panels A and B correspond to males and C and D to females in Figure 3.

Author Response

We are grateful to the reviewers for their positive feedback with their comments and suggestions on the manuscript. Reviewer 1 has indicated two weaknesses and Reviewer 2 has none. With this provisional reply, we address the two concerns of the Reviewer 1:

1) Data obtained from a single aminoacyl-tRNA (D-Tyr-tRNATyr) have been generalized to imply that what is relevant to this model substrate is true for all other D-aa-tRNAs. This is not a risk-free extrapolation. Why do the authors believe that the length of the amino acid side chain will not matter in the activity of DTD2?

We thank the reviewer for bringing up this important point. We wish to clarify that only a few of the aminoacyl-tRNA synthetases are known to charge D-amino acids and only D-Leu (Yeast), D-Asp (Bacteria, Yeast), D-Tyr (Bacteria, Cyanobacteria, Yeast) and D-Trp (Bacteria) show toxicity in vivo in the absence of known DTD (Soutourina J. et al., JBC, 2000; Soutourina O. et al., JBC, 2004; Wydau S. et al., JBC, 2009). D-Tyr-tRNATyr is used as a model substrate to test the DTD activity in the field because of the conserved toxicity of D-Tyr in various organisms. DTD2 has been shown to recycle D-Asp-tRNAAsp and D-Tyr-tRNATyr with the same efficiency both in vitro and in vivo (Wydau S. et al., NAR, 2007). Moreover, we have previously shown that it recycles acetaldehyde-modified D-Phe-tRNAPhe and D-Tyr-tRNATyr in vitro (Mazeed M. et al., Science Advances, 2021). We have earlier shown that DTD1, another conserved chiral proofreader across bacteria and eukaryotes, acts via a side chain independent mechanism (Ahmad S. et al., eLife, 2013). Considering the action on multiple side chains with different chemistry and size, it can be proposed with reasonable confidence that DTD2 also operates based on a side chain independent manner.

2) While the use of EFTu supports that the ternary complex formation by the elongation factor can resist modifications of L-Tyr-tRNATyr by the aldehydes or other agents, in the context of the present work on the role of DTD2 in plants, one would want to see the data using eEF1alpha. This is particularly relevant because there are likely to be differences in the way EFTu and eEF1alpha may protect aminoacyl-tRNAs (for example see description in the latter half of the article by Wolfson and Knight 2005, FEBS Letters 579, 3467-3472).

We thank the reviewer for bringing another important point. We analysed the aa-tRNA bound elongation factor structures from both bacteria (PDB id: 1TTT) and mammal (PDB id: 5LZS) and found that the amino acid binding site is highly conserved where side chain of amino acid is projected outside. Modelling of D-amino acid in the same site shows serious clashes, indicating D-chiral rejection during aa-tRNA binding by elongation factor. In addition, the amino group of amino acid is tightly selected by the main chain atoms of elongation factor thereby lacking a space for aldehydes to enter and then modify the L-aa-tRNAs and Gly-tRNAs. Minor differences near the amino acid side chain binding site (as indicated in Wolfson and Knight, FEBS Letters, 2005) might induce the amino acid specific binding differences. However, those changes will have no influence when the D-chiral amino acid enters the pocket, as the whole side chain would clash with the active site. We will present a sequence and structural conservation analysis to clarify this important point in our revised manuscript. Overall, our structural analysis suggests a conserved mode of aa-tRNA selection by elongation factor across life forms and therefore, our biochemical results with bacterial elongation factor Tu (EF-Tu) reflect the protective role of elongation factor in general across species.

In our revised manuscript, we will provide a thorough point-by-point response to the above as well as all the specific reviewer comments. We also intend to include new analysis with updated data that would address the key questions raised by the reviewers.

Author Response

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

Reviewer #1 (Public Review):

The biogenesis of outer membrane proteins (OMPs) into the outer membranes of Gram-negative bacteria is still not fully understood, particularly substrate recognition and insertion by beta-assembly machinery (BAM). In the studies, the authors present their studies that in addition to recognition by the last strand of an OMP, sometimes referred to as the beta-signal, an additional signal upstream of the last strand is also important for OMP biogenesis.

Strengths:

1. Overall the manuscript is well organized and written, and addresses an important question in the field. The idea that BAM recognizes multiple signals on OMPs has been presented previously, however, it was not fully tested.

2. The authors here re-address this idea and propose that it is a more general mechanism used by BAM for OMP biogenesis.

3. The notion that additional signals assist in biogenesis is an important concept that indeed needs fully tested in OMP biogenesis.

4. A significant study was performed with extensive experiments reported in an attempt to address this important question in the field.

5. The identification of important crosslinks and regions of substrates and Bam proteins that interact during biogenesis is an important contribution that gives clues to the path substrates take en route to the membrane.

Weaknesses:

Major critiques (in no particular order):

1. The title indicates 'simultaneous recognition', however no experiments were presented that test the order of interactions during OMP biogenesis.

We have replaced the word “Simultaneous” with “Dual” so as not to reflect on the timing of the recognition events for the distinct C-terminal signal and -5 signal.

1. Aspects of the study focus on the peptides that appear to inhibit OmpC assembly, but should also include an analysis of the peptides that do not to determine this the motif(s) present still or not.

We thank the reviewer for this comment. Our study focuses on the peptides which exhibited an inhibitory effect in order to elucidate further interactions between the BAM complex and substrate proteins, especially in early stage of the assembly process. In the case of peptide 9, which contains all of our proposed elements but did not have an inhibitory effect, there is the presence of an arginine residue at the polar residue next to hydrophobic residue in position 0 (0 Φ). As seen in Fig S5, S6, and S7, there are no positively charged amino acids in the polar residue positions in the -5 or last strands. This might be the reason why peptide 9, as well as peptide 24, the β-signal derived from the mitochondrial OMP Tom40 and contains a lysine at the polar position, did not display an inhibitory effect. Incorporating the reviewer's suggestions might elucidate conditions that should not be added to the elements, but this is not the focus of this paper and was not discussed to avoid complicating the paper.

1. The β-signal is known to form a β-strand, therefore it is unclear why the authors did not choose to chop OmpC up according to its strands, rather than by a fixed peptide size. What was the rationale for how the peptide lengths were chosen since many of them partially overlap known strands, and only partially (2 residues) overlap each other? It may not be too surprising that most of the inhibitory peptides consist of full strands (#4, 10, 21, 23).

A simple scan of known β-strands would have been an alternative approach, however this comes with the bias of limiting the experiments to predicted substrate (strand) sequences, and it presupposes that the secondary structure element would be formed by this tightly truncated peptide.

Instead, we allowed for the possibility that OMPs meet the BAM complex in an unfolded or partially folded state, and that the secondary structure (β-strand) might only form via β-argumentation after the substrate is placed in the context of the lateral gate. We therefore used peptides that mapped right across the entirety of OmpC, with a two amino acid overlap.

To clarify this important point regarding the unbiased nature of our screen, we have revised the text:

(Lines 147-151) "We used peptides that mapped the entirety of OmpC, with a two amino acid overlap. This we considered preferable to peptides that were restricted by structural features, such as β-strands, in consideration that β-strand formation may or may not have occurred in early-stage interactions at the BAM complex."

1. It would be good to have an idea of the propensity of the chosen peptides to form β-stands and participate in β-augmentation. We know from previous studies with darobactin and other peptides that they can inhibit OMP assembly by competing with substrates.

We appreciate the reviewer's suggestion. However, we have not conducted biophysical characterizations of the peptides to calculate the propensity of each peptide to form β-stands and participate in β-augmentation. The sort of detailed biophysical analysis done for Darobactin (by the Maier and Hiller groups, The antibiotic darobactin mimics a β-strand to inhibit outer membrane insertase Nature 593:125-129) was a Nature publication based on this single peptide. A further biophysical analysis of all of the peptides presented here goes well beyond the scope of our study.

1. The recognition motifs that the authors present span up to 9 residues which would suggest a relatively large binding surface, however, the structures of these regions are not large enough to accommodate these large peptides.

The β-signal motif (ζxGxx[Ω/Φ]x[Ω/Φ]) is an 8-residue consensus, some of the inhibitory peptides include additional residues before and after the defined motif of 8 residues, and the lateral gate of BamA has been shown interact with a 7-residue span (eg. Doyle et al, 2022). Cross-linking presented in our study showed BamD residues R49 and G65 cross-linked to the positions 0 and 6 of the internal signal in OmpC (Fig. 6D).

We appreciate this point of clarification and have modified the text to acknowledge that in the final registering of the peptide with its binding protein, some parts of the peptide might sit beyond the bounds of the BamD receptor’s binding pocket and the BamA lateral gate:

(Lines 458-471) "The β-signal motif (ζxGxx[Ω/Φ]x[Ω/Φ]) is an eight-residue consensus, and internal signal motif is composed of a nine-residue consensus. Recent structures have shown the lateral gate of BamA interacts with a 7-residue span of substrate OMPs. Interestingly, inhibitory compounds, such as darobactin, mimic only three resides of the C-terminal side of β-signal motif. Cross-linking presented here in our study showed that BamD residues R49 and G65 cross-linked to the positions 0 and 6 of the internal signal in OmpC (Fig. 6D). Both signals are larger than the assembly machineries signal binding pocket, implying that the signal might sit beyond the bounds of the signal binding pocket in BamD and the lateral gate in BamA. These finding are consistent with similar observations in other signal sequence recognition events, such as the mitochondrial targeting presequence signal that is longer than the receptor groove formed by the Tom20, the subunit of the translocator of outer membrane (TOM) complex (Yamamoto et al., 2011). The presequence has been shown to bind to Tom20 in several different conformations within the receptor groove (Nyirenda et al., 2013)."

Moreover, the distance between amino acids of BamD which cross-linked to the internal signal, R49 and Y62, is approximately 25 Å (pdbID used 7TT3). The distance of the maximum amino acid length of the internal signal of OmpC, from F280 to Y288, is approximately 22 Å (pdbID used 2J1N). This would allow for the signal to fit within the confines of the TRP motif of BamD.

Author response image 1.

1. The authors highlight that the sequence motifs are common among the inhibiting peptides, but do not test if this is a necessary motif to mediate the interactions. It would have been good to see if a library of non-OMP related peptides that match this motif could also inhibit or not.

With respect, this additional work would not address any biological question relevant to the function of BamD. To randomize sequences and then classify those that do or don’t fit the motif would help in refining the parameters of the β-signal motif, but that was not our intent.

We have identified the peptides from within the total sequence of an OMP, shown which peptides inhibit in an assembly assay, and then observed that the inhibitory peptides conform to a previously published (β-signal) motif.

1. In the studies that disrupt the motifs by mutagenesis, an effect was observed and attributed to disruption of the interaction of the 'internal signal'. However, the literature is filled with point mutations in OMPs that disrupt biogenesis, particular those within the membrane region. F280, Y286, V359, and Y365 are all residues that are in the membrane region that point into the membrane. Therefore, more work is needed to confirm that these mutations are in parts of a recognition motif rather than on the residues that are disrupting stability/assembly into the membrane.

As the reviewer pointed out, the side chains of the amino acids constituting the signal elements we determined were all facing the lipid side, of which Y286 and Y365 were important for folding as well as to be recognized. However, F280A and V359A had no effect on folding, but only on assembly through the BAM complex. The fact that position 0 functions as a signal has been demonstrated by peptidomimetics (Fig. 1) and point mutant analysis (Fig. 2). We appreciate this clarification and have modified the text to acknowledge that the all of the signal element faces the lipid side, which contributes to their stability in the membrane finally, and before that the BAM complex actively recognizes them and determines their orientation:

(Lines 519-526) After OMP assembly, all elements of the internal signal are positioned such that they face into the lipid-phase of the membrane. This observation may be a coincidence, or may be utilized by the BAM complex to register and orientate the lipid facing amino acids in the assembling OMP away from the formative lumen of the OMP. Amino acids at position 6, such as Y286 in OmpC, are not only component of the internal signal for binding by the BAM complex, but also act in structural capacity to register the aromatic girdle for optimal stability of the OMP in the membrane.

1. The title of Figure 3 indicates that disrupting the internal signal motif disrupts OMP assembly, however, the point mutations did not seem to have any effect. Only when both 280 and 286 were mutated was an effect observed. And even then, the trimer appeared to form just fine, albeit at reduced levels, indicating assembly is just fine, rather the rate of biogenesis is being affected.

We appreciate this point and have revised the title of Figure 3 to be:

(Lines 1070-1071) "Modifications in the putative internal signal slow the rate of OMP assembly in vivo."

1. In Figure 4, the authors attempt to quantify their blots. However, this seems to be a difficult task given the lack of quality of the blots and the spread of the intended signals, particularly of the 'int' bands. However, the more disturbing trend is the obvious reduction in signal from the post-urea treatment, even for the WT samples. The authors are using urea washes to indicate removal of only stalled substrates. However a reduction of signal is also observed for the WT. The authors should quantify this blot as well, but it is clear visually that both WT and the mutant have obvious reductions in the observable signals. Further, this data seems to conflict with Fig 3D where no noticeable difference in OmpC assembly was observed between WT and Y286A, why is this the case?

We have addressed this point by adding a statistical analysis on Fig. 4A. As the reviewer points out, BN-PAGE band quantification is a difficult task given the broad spread of the bands on these gels. Statistical analysis showed that the increase in intermediates (int) was statistically significant for Y286A at all times until 80 min, when the intermediate form signals decrease.

(Lines 1093-1096) "Statistical significance was indicated by the following: N.S. (not significant), p<0.05; , p<0.005; *. Exact p values of intermediate formed by Wt vs Y286A at each timepoint were as follows; 20 minutes: p = 0.03077, 40 minutes: p = 0.02402, 60 minutes: p = 0.00181, 80 minutes: p = 0.0545."

Further regarding the Int. band, we correct the statement as follows.

(Lines 253-254) "Consistent with this, the assembly intermediate which was prominently observed at the OmpC(Y286A) can be extracted from the membranes with urea;"

OMP assembly in vivo has additional periplasmic chaperones and factors present in order to support the assembly process. Therefore, it is likely that some proteins were assembled properly in vivo compared to their in vitro counterparts. Such a decrease has been observed not only in E. coli but also in mitochondrial OMP import (Yamano et al., 2010).

1. The pull-down assays with BamA and BamD should include a no protein control at the least to confirm there is no non-specific binding to the resin. Also, no detergent was mentioned as part of the pull downs that contained BamA or OmpC, nor was it detailed if OmpC was urea solubilized.

We have performed pull down experiments with a no-protein (Ni-NTA only) control as noted (Author response image 1). The results showed that the amount of OmpC carrying through on beads only was significantly lower than the amount of OmpC bound in the presence of BamD or BamA. The added OmpC was not treated with urea, but was synthesized by in vitro translation; the in vitro translated OmpC is the standard substrate in the EMM assembly assay (Supp Fig. S1) where it is recognized by the BAM complex. Thus, we used it for pull-down as well and, to make this clearer, we have revised as follows:

Author response image 2.

Pull down assay of radio-labelled OmpC with indicated protein or Ni-NTA alone (Ni-NTA) . T; total, FT; Flow throw, W; wash, E; Elute.

(Lines 252-265) "Three subunits of the BAM complex have been previously shown to interact with the substrates: BamA, BamB, and BamD (Hagan et al., 2013; Harrison, 1996; Ieva et al., 2011). In vitro pull-down assay showed that while BamA and BamD can independently bind to the in vitro translated OmpC polypeptide (Fig .S9A), BamB did not (Fig. S9B)."

11.

• The neutron reflectometry experiments are not convincing primarily due to the lack controls to confirm a consistent uniform bilayer is being formed and even if so, uniform orientations of the BamA molecules across the surface.

• Further, no controls were performed with BamD alone, or with OmpC alone, and it is hard to understand how the method can discriminate between an actual BamA/BamD complex versus BamA and BamD individually being located at the membrane surface without forming an actual complex.

• Previous studies have reported difficulty in preparing a complex with BamA and BamD from purified components.

• Additionally, little signal differences were observed for the addition of OmpC. However, an elongated unfolded polypeptide that is nearly 400 residues long would be expected to produce a large distinct signal given that only the C-terminal portion is supposedly anchored to BAM, while the rest would be extended out above the surface.

• The depiction in Figure 5D is quite misleading when viewing the full structures on the same scales with one another.

We have addressed these five points individually as follows.

i. The uniform orientation of BamA on the surface is guaranteed by the fixation through a His-tag engineered into extracellular loop 6 of BamA and has been validated in previous studies as cited in the text. Moreover, to explain this, we reconstructed another theoretical model for BamA not oriented well in the system as below. However, we found that the solid lines (after fitting) didn’t align well with the experimental data. We therefore assumed that BamA has oriented well in the membrane bilayer.

Author response image 3.

Experimental (symbols) and fitted (curves) NR profiles of BamA not oriented well in the POPC bilayer in D2O (black), GMW (blue) and H2O (red) buffer.

ii. There would be no means by which to do a control with OmpC alone or BamD alone as neither protein binds to the lipid layer chip. OmpC is diluted from urea and then the unbound OmpC is washed from the chip before NR measurements. BamD does not have an acyl group to anchor it to the lipid layer, without BamA to anchor to, it too is washed from the chip before NR measurements. We have reconstructed another theoretical model for both of BamA + BamD embedding in the membrane bilayer, and the fits were shown below. Apparently, the fits didn’t align well with the experimental data, which discriminate the BamA/BamD individually being located at the membrane surface without forming an actual complex.

Author response image 4.

Experimental (symbols) and fitted (curves) NR profiles of BamA+D embedding together in the POPC bilayer in D2O (black), GMW (blue) and H2O (red) buffer.

iii. The previous studies that reported difficulty in preparing a complex with BamA and BamD from purified components were assays done in aqueous solution including detergent solubilized BamA, or with BamA POTRA domains only. Our assay is superior in that it reports the binding of BamD to a purified BamA that has been reconstituted in a lipid bilayer.

iv. The relatively small signal differences observed for the addition of OmpC are expected, since OmpC is an elongated, unfolded polypeptide of nearly 400 residues long which, in the context of this assay, can occupy a huge variation in the positions at which it will sit with only the C-terminal portion anchored to BAM, and the rest moving randomly about and extended from the surface.

v. We appreciate the point raised and have now added a note in the Figure legend that these are depictions of the results and not a scale drawing of the structures.

1. In the crosslinking studies, the authors show 17 crosslinking sites (43% of all tested) on BamD crosslinked with OmpC. Given that the authors are presenting specific interactions between the two proteins, this is worrisome as the crosslinks were found across the entire surface of BamD. How do the authors explain this? Are all these specific or non-specific?

The crosslinking experiment using purified BamD was an effective assay for comprehensive analysis of the interaction sites between BamD and the substrate. However, as the reviewer pointed out, cross-linking was observed even at the sites that, in the context of the BAM complex, interact with BamC as a protein-protein interaction and would not be available for substrate protein-protein interactions. To complement this, analysis and to address this issue, we also performed the experiment in Fig. 6C.

In Fig. 6C, the interaction of BamD with the substrate is examined in vivo, and the results demonstrate that if BPA is introduced into the site, we designated as the substrate recognition site, it is cross-linked to the substrate. On the other hand, position 114 was found to crosslink with the substrate in vitro crosslinking, but not in vivo. It should be noted that position 114 has also been confirmed to form cross-link products with BamC, we believe that BamD-substrate interactions in the native state have been investigated. To explain the above, we have added the following description to the Results section.

(Lines 319-321) "Structurally, these amino acids locate both the lumen side of funnel-like structure (e.g. 49 or 62) and outside of funnel-like structure such as BamC binding site (e.g. 114) (fig. S12C). (Lines 350-357) Positions 49, 53, 65, and 196 of BamD face the interior of the funnel-like structure of the periplasmic domain of the BAM complex, while position 114 is located outside of the funnel-like structure (Bakelar et al., 2016; Gu et al., 2016; Iadanza et al., 2016). We note that while position 114 was cross-linked with OmpC in vitro using purified BamD, that this was not seen with in vivo cross-linking. Instead, in the context of the BAM complex, position 114 of BamD binds to the BamC subunit and would not be available for substrate binding in vivo (Bakelar et al., 2016; Gu et al., 2016; Iadanza et al., 2016)."

1. The study in Figure 6 focuses on defined regions within the OmpC sequence, but a more broad range is necessary to demonstrate specificity to these regions vs binding to other regions of the sequence as well. If the authors wish to demonstrate a specific interaction to this motif, they need to show no binding to other regions.

The region of affinity for the BAM complex was determined by peptidomimetic analysis, and the signal region was further identified by mutational analysis of OmpC. Subsequently, the subunit that recognizes the signal region was identified as BamD. In other words, in the process leading up to Fig. 6, we were able to analyze in detail that other regions were not the target of the study. We have revised the text to make clear that we focus on the signal region including the internal signal, and have not also analyzed other parts of the signal region:

(Lines 329-332) "As our peptidomimetic screen identified conserved features in the internal signal, and cross-linking highlighted the N-terminal and C-terminal TPR motifs of BamD as regions of interaction with OmpC, we focused on amino acids specifically within the β-signals of OmpC and regions of BamD which interact with β-signal."

1. The levels of the crosslinks are barely detectable via western blot analysis. If the interactions between the two surfaces are required, why are the levels for most of the blots so low?

These are western blots of cross-linked products – the efficiency of cross-linking is far less than 100% of the interacting protein species present in a binding assay and this explains why the levels for the blots are ‘so low’. We have added a sentence to the revised manuscript to make this clear for readers who are not molecular biologists:

(Lines 345-348) "These western blots reveal cross-linked products representing the interacting protein species. Photo cross-linking of unnatural amino acid is not a 100% efficient process, so the level of cross-linked products is only a small proportion of the molecules interacting in the assays."

15.

• Figure 7 indicates that two regions of BamD promote OMP orientation and assembly, however, none of the experiments appears to measure OMP orientation?

• Also, one common observation from panel F was that not only was the trimer reduced, but also the monomer. But even then, still a percentage of the trimer is formed, not a complete loss.

(i) We appreciate this point and have revised the title of Figure 7 to be:

(Lines 1137-1138) "Key residues in two structurally distinct regions of BamD promote β-strand formation and OMP assembly."

(ii) In our description of Fig. 7F (Lines 356-360) we do not distinguish between the amount of monomer and trimer forms, since both are reflective of the overall assembly rate i.e. assembly efficiency. Rather, we state that:

"The EMM assembly assay showed that the internal signal binding site was as important as the β-signal binding site to the overall assembly rates observed for OmpC (Fig. 7F), OmpF (fig. S15D), and LamB (fig. S15E). These results suggest that recognition of both the C-terminal β-signal and the internal signal by BamD is important for efficient protein assembly."

16.

• The experiment in Fig 7B would be more conclusive if it was repeated with both the Y62A and R197A mutants and a double mutant. These controls would also help resolve any effect from crowding that may also promote the crosslinks.

• Further, the mutation of R197 is an odd choice given that this residue has been studied previously and was found to mediate a salt bridge with BamA. How was this resolved by the authors in choosing this site since it was not one of the original crosslinking sites?

As stated in the text, the purpose of the experiment in Figure 7B is to measure the impact of pre-forming a β-strand in the substrate (OmpC) before providing it to the receptor (BamD). We thank the reviewer for the comment on the R197 position of BamD. The C-terminal domain of BamD has been suggested to mediate the BamA-BamD interface, specifically BamD R197 amino acid creates a salt-bridge with BamA E373 (Ricci et al., 2012). It had been postulated that the formation of this salt-bridge is not strictly structural, with R197 highlighted as a key amino acid in BamD activity and this salt-bridge acts as a “check-point” in BAM complex activity (Ricci et al., 2012, Storek et al., 2023). Our results agree with this, showing that the C-terminus of BamD acts in substrate recognition and alignment of the β-signal (Fig. 6, Fig S12). We show that amino acids in the vicinity of R197 (N196, G200, D204) cross-linked well to substrate and mutations to the β-signal prevent this interaction (Fig S12B, D). For mutational analysis of BamD, we looked then at the conservation of the C-terminus of BamD and determined R197 was the most highly conserved amino acid (Fig 6C). In order to account for this, we have adjusted the manuscript:

(Lines 376-377) "R197 has previously been isolated as a suppressor mutation of a BamA temperature sensitive strain (Ricci et al., 2012)."

(Lines 495-496) "This adds an additional role of the C-terminus of BamD beyond a complex stability role (Ricci et al., 2012; Storek et al., 2023)."

1. As demonstrated by the authors in Fig 8, the mutations in BamD lead to reduction in OMP levels for more than just OmpC and issues with the membrane are clearly observable with Y62A, although not with R197A in the presence of VCN. The authors should also test with rifampicin which is smaller and would monitor even more subtle issues with the membrane. Oddly, no growth was observed for the Vec control in the lower concentration of VCN, but was near WT levels for 3 times VCN, how is this explained?

While it would be interesting to correlate the extent of differences to the molecular size of different antibiotics such as rifampicin, such correlations are not the intended aim of our study. Vancomycin (VCN) is a standard measure of outer membrane integrity in our field, hence its use in our tests for membrane integrity.

We apologize to the reviewer as Figure 8 D-G may have been misleading. Figure 8D,E are using bamD shut-down cells expressing plasmid-borne BamD mutants. Whereas Figure 8F, G are the same strain as used in Figure 3. We have adjusted the figure as well as the figure legend: (Lines 1165-1169) D, E E coli bamD depletion cells expressing mutations at residues, Y62A and R197A, in the β-signal recognition regions of BamD were grown with of VCN. F, G, E coli cells expressing mutations to OmpC internal signal, as shown in Fig 3, grown in the presence of VCN. Mutations to two key residues of the internal signal were sensitive to the presence of VCN.

1. While Fig 8I indeed shows diminished levels for FY as stated, little difference was observed for the trimer for the other mutants compared to WT, although differences were observed for the dimer. Interestingly, the VY mutant has nearly WT levels of dimer. What do the authors postulate is going on here with the dimer to trimer transition? How do the levels of monomer compare, which is not shown?

The BN-PAGE gel system cannot resolve protein species that migrate below ~50kDa and the monomer species of the OMPs is below this size. We can’t comment on effects on the monomer because it is not visualized. The non-cropped gel image is shown here. Recently, Hussain et al., has shown that in vitro proteo-liposome system OmpC assembly progresses from a “short-lived dimeric” form before the final process of trimerization (Hussain et al., 2021). However, their findings suggest that LPS plays the final role in stimulation of dimer-to-trimer, a step well past the recognition step of the β-signals. Mutations to the internal signal of OmpC results in the formation of an intermediate, the substrate stalled on the BAM complex. This stalling, presumably, causes a hinderance to the BAM complex resulting in reduced timer and loss of dimer OmpF signal in the EMM of cells expressing OmpC double mutant strain, FY. cannot resolve protein species that migrate below ~50kDa and the monomer species of the OMPs is below this size. We can’t comment on effects on the monomer because it is not visualized. The non-cropped gel image is shown here. We have noted this in the revised text:

Author response image 5.

Non-cropped gel of Fig. 8I. the asterisk indicates a band observed in the sample loading wells at the top of the gel.

(Lines 417-418) "The dimeric form of endogenous OmpF was prominently observed in both the OmpC(WT) as well as the OmpC(VY) double mutant cells."

1. In the discussion, the authors indicate they have '...defined an internal signal for OMP assembly', however, their study is limited and only investigates a specific region of OmpC. More is needed to definitively say this for even OmpC, and even more so to indicate this is a general feature for all OMPs.

We acknowledge the reviewer's comment on this point and have expanded the statement to make sure that the conclusion is justified with the specific evidence that is shown in the paper and the supplementary data. We now state:

(Lines 444-447) "This internal signal corresponds to the -5 strand in OmpC and is recognized by BamD. Sequence analysis shows that similar sequence signatures are present in other OMPs (Figs. S5, S6 and S7). These sequences were investigated in two further OMPs: OmpF and LamB (Fig. 2C and D)."

Note, we did not state that this is a general feature for all OMPs. That would not be a reasonable proposition.

20.

• In the proposed model in Fig 9, it is hard to conceive how 5 strands will form along BamD given the limited surface area and tight space beneath BAM.

• More concerning is that the two proposal interaction sites on BamD, Y62 and R197, are on opposite sides of the BamD structure, not along the same interface, which makes this model even more unlikely.

• As evidence against this model, in Figure 9E, the two indicates sites of BamD are not even in close proximity of the modeled substrate strands.

We can address the reviewer’s three concerns here:

i. The first point is that the region (formed by BamD engaged with POTRA domains 1-2 and 5 of BamA) is not sufficient to accommodate five β-strands. Structural analysis reveals that the interaction between the N-terminal side of BamD and POTRA1-2 is substantially changed the conformation by substrate binding, and that this surface is greatly extended. This surface does have enough space to accommodate five beta-strands, as now documented in Fig. 9D, 9E using the latest structures (7TT5 and 7TT2) as illustrations of this. The text now reads:

(Lines 506-515) "Spatially, this indicates the BamD can serve to organize two distinct parts of the nascent OMP substrate at the periplasmic face of the BAM complex, either prior to or in concert with, engagement to the lateral gate of BamA. Assessing this structurally showed the N-terminal region of BamD (interacting with the POTRA1-2 region of BamA) and the C-terminal region of BamD (interacting with POTRA5 proximal to the lateral gate of BamA) (Bakelar et al., 2016; Gu et al., 2016; Tomasek et al., 2020) has the N-terminal region of BamD changing conformation depending on the folding states of the last four β-strands of the substrate OMP, EspP (Doyle et al., 2022). The overall effect of this being a change in the dimensions of this cavity change, a change which is dependent on the folded state of the substrate engaged in it (Fig 9 B-E)."

ii. The second point raised regards the orientation of the substrate recognition residues of BamD. Both Y62A and R197 were located on the lumen side of the funnel in the EspP-BAM transport intermediate structure (PDBID;7TTC); Y62A is relatively located on the edge of BamD, but given that POTRA1-2 undergoes a conformational change and opens this region, as described above, both are located in locations where they could bind to substrates. This was explained in the following text in the results section of revised manuscript.

(Lines 377-379) "Each residue was located on the lumen side of the funnel-like structure in the EspP-BAM assembly intermediate structure (PDBID; 7TTC) (Doyle et al., 2022)."

**Reviewer #2 (Public Review):"

Previously, using bioinformatics study, authors have identified potential sequence motifs that are common to a large subset of beta-barrel outer membrane proteins in gram negative bacteria. Interestingly, in that study, some of those motifs are located in the internal strands of barrels (not near the termini), in addition to the well-known "beta-signal" motif in the C-terminal region.

Here, the authors carried out rigorous biochemical, biophysical, and genetic studies to prove that the newly identified internal motifs are critical to the assembly of outer membrane proteins and the interaction with the BAM complex. The author's approaches are rigorous and comprehensive, whose results reasonably well support the conclusions. While overall enthusiastic, I have some scientific concerns with the rationale of the neutron refractory study, and the distinction between "the intrinsic impairment of the barrel" vs "the impairment of interaction with BAM" that the internal signal may play a role in. I hope that the authors will be able to address this.

Strengths:

1. It is impressive that the authors took multi-faceted approaches using the assays on reconstituted, cell-based, and population-level (growth) systems.

2. Assessing the role of the internal motifs in the assembly of model OMPs in the absence and presence of BAM machinery was a nice approach for a precise definition of the role.

Weaknesses:

1. The result section employing the neutron refractory (NR) needs to be clarified and strengthened in the main text (from line 226). In the current form, the NR result seems not so convincing.

What is the rationale of the approach using NR?

We have now modified the text to make clear that:

(Lines 276-280) "The rationale to these experiments is that NR provides: (i) information on the distance of specified subunits of a protein complex away from the atomically flat gold surface to which the complex is attached, and (ii) allows the addition of samples between measurements, so that multi-step changes can be made to, for example, detect changes in domain conformation in response to the addition of a substrate."

What is the molecular event (readout) that the method detects?

We have now modified the text to make clear that:

(Lines 270-274) "While the biochemical assay demonstrated that the OmpC(Y286A) mutant forms a stalled intermediate with the BAM complex, in a state in which membrane insertion was not completed, biochemical assays such as this cannot elucidate where on BamA-BamD this OmpC(Y286A) substrate is stalled."

What are "R"-y axis and "Q"-x axis and their physical meanings (Fig. 5b)?

The neutron reflectivity, R, refers to the ratio of the incoming and exiting neutron beams and it is measured as a function of Momentum transfer Q, which is defined as Q=4π sinθ/λ, where θ is the angle of incident and λ is the neutron wavelength. R(Q)is approximately given byR(Q)=16π2/ Q2 |ρ(Q)|2, where R(Q) is the one-dimensional Fourier transform of ρ(z), the scattering length density (SLD) distribution normal to the surface. SLD is the sum of the coherent neutron scattering lengths of all atoms in the sample layer divided by the volume of the layer. Therefore, the intensity of the reflected beams is highly dependent on the thickness, densities and interface roughness of the samples. This was explained in the following text in the method section of revised manuscript.

(Lines 669-678) "Neutron reflectivity, denoted as R, is the ratio of the incoming to the exiting neutron beams. It’s calculated based on the Momentum transfer Q, which is defined by the formula Q=4π sinθ/λ, where θ represents the angle of incidence and λ stands for the neutron wavelength. The approximate value of R(Q) can be expressed as R(Q)=16π2/ Q2 |ρ(Q)|2, where R(Q) is the one-dimensional Fourier transform of ρ(z), which is the scattering length density (SLD) distribution perpendicular to the surface. SLD is calculated by dividing the sum of the coherent neutron scattering lengths of all atoms in a sample layer by the volume of that layer. Consequently, factors such as thickness, volume fraction, and interface roughness of the samples significantly influence the intensity of the reflected beams."

How are the "layers" defined from the plot (Fig. 5b)?

The “layers” in the plot (Fig. 5b) represent different regions of the sample being studied. In this study, we used a seven-layer model to fit the experimental data (chromium - gold - NTA - HIS8 - β-barrel - P3-5 - P1-2. This was explained in the following text in the figure legend of revised manuscript. (Lines 1115-1116) The experimental data was fitted using a seven-layer model: chromium - gold - NTA - His8 - β-barrel - P3-5 - P1-2.

What are the meanings of "thickness" and "roughness" (Fig. 5c)?

We used neutron reflectometry to determine the relative positions of BAM subunits in a membrane environment. The binding of certain subunits induced conformational changes in other parts of the complex. When a substrate membrane protein is added, the periplasmic POTRA domain of BamA extends further away from the membrane surface. This could result in an increase in thickness as observed in neutron reflectometry measurements.

As for roughness, it is related to the interface properties of the sample. In neutron reflectometry, the intensity of the reflected beams is highly dependent on the thickness, densities, and interface roughness of the samples. An increase in roughness could suggest changes in these properties, possibly due to protein-membrane interactions or structural changes within the membrane.

(Lines 1116-1120) "Table summarizes of the thickness, roughness and volume fraction data of each layer from the NR analysis. The thickness refers to the depth of layered structures being studied as measured in Å. The roughness refers to the irregularities in the surface of the layered structures being studied as measured in Å."

What does "SLD" stand for?

We apologize for not explaining abbreviation when the SLD first came out. We explained it in revised manuscript. (Line 298)

1. In the result section, "The internal signal is necessary for insertion step of assembly into OM" This section presents an important result that the internal beta-signal is critical to the intrinsic propensity of barrel formation, distinct from the recognition by BAM complex. However, this point is not elaborated in this section. For example, what is the role of these critical residues in the barrel structure formation? That is, are they involved in any special tertiary contacts in the structure or in membrane anchoring of the nascent polypeptide chains?

We appreciate the reviewer's comment on this point. Both position 0 and position 6 appear to be important amino acids for recognition by the BAM complex, since mutations introduced at these positions in peptide 18 prevent competitive inhibition activity.

In terms of the tertiary structure of OmpC, position 6 is an amino acid that contributes to the aromatic girdle, and since Y286A and Y365A affected OMP folding as measured in folding experiments, it is perhaps their position in the aromatic girdle that contributes to the efficiency of β-barrel folding in addition to its function as a recognition signal. We have added a sentence in the revised manuscript:

(Lines 233-236) "Position 6 is an amino acid that contributes to the aromatic girdle. Since Y286A and Y365A affected OMP folding as measured in folding experiments, their positioning into the aromatic girdle may contributes to the efficiency of β-barrel folding, in addition to contributing to the internal signal."

The mutations made at position 0 had no effect on folding, so this residue may function solely in the signal. Given the register of each β-strand in the final barrel, the position 0 residues have side-chains that face out into the lipid environment. From examination of the OmpC crystal structure, the residue at position 0 makes no special tertiary contacts with other, neighbouring residues.  

Reviewer #1 (Recommendations For The Authors):

Minor critiques (in no particular order):

1. Peptide 18 was identified based on its strong inhibition for EspP assembly but another peptide, peptide 23, also shows inhibition and has no particular consensus.

We would correct this point. Peptide 23 has a strong consensus to the canonical β-signal. We had explained the sequence consensus of β-signal in the Results section of the text. In the third paragraph, we have added a sentence indicating the relationship between peptide 18 and peptide 23.

(Lines 152-168) "Six peptides (4, 10, 17, 18, 21, and 23) were found to inhibit EspP assembly (Fig. 1A). Of these, peptide 23 corresponds to the canonical β-signal of OMPs: it is the final β-strand of OmpC and it contains the consensus motif of the β-signal (ζxGxx[Ω/Φ]x[Ω/Φ]). The inhibition seen with peptide 23 indicated that our peptidomimetics screening system using EspP can detect signals recognized by the BAM complex. In addition to inhibiting EspP assembly, five of the most potent peptides (4, 17, 18, 21, and 23) inhibited additional model OMPs; the porins OmpC and OmpF, the peptidoglycan-binding OmpA, and the maltoporin LamB (fig. S3). Comparing the sequences of these inhibitory peptides suggested the presence of a sub-motif from within the β-signal, namely [Ω/Φ]x[Ω/Φ] (Fig. 1B). The sequence codes refer to conserved residues such that: ζ, is any polar residue; G is a glycine residue; Ω is any aromatic residue; Φ is any hydrophobic residue and x is any residue (Hagan et al., 2015; Kutik et al., 2008). The non-inhibitory peptide 9 contained some elements of the β-signal but did not show inhibition of EspP assembly (Fig. 1A).

Peptide 18 also showed a strong sequence similarity to the consensus motif of the β-signal (Fig. 1B) and, like peptide 23, had a strong inhibitory action on EspP assembly (Fig. 1A). Variant peptides based on the peptide 18 sequence were constructed and tested in the EMM assembly assay (Fig. 1C)."

1. It is unclear why the authors immediately focused on BamD rather than BamB, given that both were mentioned to mediate interaction with substrate. Was BamB also tested?

We thank the reviewer for this comment. Following the reviewer's suggestion, we have now performed a pull-down experiment on BamB and added it to Fig. S9. We also modified the text of the results as follows.

(Lines 262-265) "Three subunits of the BAM complex have been previously shown to interact with the substrates: BamA, BamB, and BamD (Hagan et al., 2013; Harrison, 1996; Ieva et al., 2011). In vitro pull-down assay showed that while BamA and BamD can independently bind to the in vitro translated OmpC polypeptide (Fig .S9A), BamB did not (Fig. S9B)."

1. For the in vitro folding assays of the OmpC substrates, labeled and unlabeled, no mention of adding SurA or any other chaperone which is known to be important for mediating OMP biogenesis in vitro.

We appreciate the reviewer’s concerns on this point, however chaperones such as SurA are non-essential factors in the OMP assembly reaction mediated by the BAM complex: the surA gene is not essential and the assembly of OMPs can be measured in the absence of exogenously added SurA. It remains possible that addition of SurA to some of these assays could be useful in detailing aspects of chaperone function in the context of the BAM complex, but that was not the intent of this study.

1. For the supplementary document, it would be much easier for the reader to have the legends groups with the figures.

Following the reviewer's suggestion, we have placed the legends of Supplemental Figures together with each Figure.

1. Some of the figures and their captions are not grouped properly and are separated which makes it hard to interpret the figures efficiently.

We thank the reviewer for this comment, we have revised the manuscript and figures to properly group the figures and captions together on a single page.

1. The authors begin their 'Discussion' with a question (line 454), however, they don't appear to answer or even attempt to address it; suggest removing rhetorical questions.

As per the reviewers’ suggestion, we removed this question.

1. Line 464, 'unbiased' should be removed. This would imply that if not stated, experiments are 'negatively' biased.

We removed this word and revised the sentence as follows:

(Lines 431-433) "In our experimental approach to assess for inhibitory peptides, specific segments of the major porin substrate OmpC were shown to interact with the BAM complex as peptidomimetic inhibitors."

1. Lines 466-467; '...go well beyond expected outcomes.' What does this statement mean?

Our peptidomimetics led to unexpected results in elucidating the additional essential signal elements. The manuscript was revised as follows:

(Lines 433-435) "Results for this experimental approach went beyond expected outcomes by identifying the essential elements of the signal Φxxxxxx[Ω/Φ]x[Ω/Φ] in β-strands other than the C-terminal strand."

1. Line 478; '...rich information that must be oversimplified...'?

We appreciate the reviewer’s pointed out. For more clarity, the manuscript was revised as follows:

(Lines 450-453) "The abundance of information which arises from modeling approaches and from the multitude of candidate OMPs, is generally oversimplified when written as a primary structure description typical of the β-signal for bacterial OMPs (i.e. ζxGxx[Ω/Φ]x[Ω/Φ]) (Kutik et al., 2008)."

1. There are typos in the supplementary figures.

We have revised and corrected the Supplemental Figure legends.  

Reviewer #2 (Recommendations For The Authors):

1. In Supplementary Information, I recommend adding the figure legends directly to the corresponding figures. Currently, it is very inconvenient to go back and forth between legends and figures.

Following the reviewer's suggestion, we have placed the legends of Supplemental Figures together with each Figure.

1. Line 94 (p.3): "later"

Lateral?

Yes. We have corrected this.

1. Line 113 (p.3): The result section, "Peptidomimetics derived from E. coli OmpC inhibit OMP assembly" Rationale of the peptide inhibition assay is not clear. How can the peptide sequence that effectively inhibit the assembly interpreted as the b-assembly signal? By competitive binding to BAM or by something else? What is the authors' hypothesis in doing this assay?

In revision, we have added following sentence to explain the aim and design of the peptidomimetics:

(Lines 140-145) "The addition of peptides with BAM complex affinity, such as the OMP β-signal, are capable of exerting an inhibitory effect by competing for binding of substrate OMPs to the BAM complex (Hagan et al., 2015). Thus, the addition of peptides derived from the entirety of OMPs to the EMM assembly assay, which can evaluate assembly efficiency with high accuracy, expects to identify novel regions that have affinity for the BAM complex."

1. Line 113- (p.3) and Fig. S1: The result section, "Peptidomimetics derived from E. coli OmpC inhibit OMP assembly"

Some explanation seems to be needed why b-barrel domain of EspP appears even without ProK?

We appreciate the reviewer’s pointed out. We added following sentence to explain:

(Lines 128-137) "EspP, a model OMP substrate, belongs to autotransporter family of proteins. Autotransporters have two domains; (1) a β-barrel domain, assembled into the outer membrane via the BAM complex, and (2) a passenger domain, which traverses the outer membrane via the lumen of the β-barrel domain itself and is subsequently cleaved by the correctly assembled β-barrel domain (Celik et al., 2012). When EspP is correctly assembled into outer membrane, a visible decrease in the molecular mass of the protein is observed due to the self-proteolysis. Once the barrel domain is assembled into the membrane it becomes protease-resistant, with residual unassembled and passenger domains degraded (Leyton et al., 2014; Roman-Hernandez et al., 2014)."

1. Line 186 (p.6): "Y285"

Y285A?

We have corrected the error, it was Y285A.

1. Lines 245- (p. 7)/ Lines 330- (p. 10)

It needs to be clarified that the results described in these paragraphs were obtained from the assays with EMM.

We appreciate the reviewer’s concerns on these points. For the first half, the following text was added at the beginning of the applicable paragraph to indicate that all of Fig. 4 is the result of the EMM assembly assay.

(Line 241) "We further analyzed the role of internal β-signal by the EMM assembly assay. At the second half, we used purified BamD but not EMM. We described clearly with following sentence."

(Lines 316-318) "We purified 40 different BPA variants of BamD, and then irradiated UV after incubating with 35S-labelled OmpC."

Author Response

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

eLife assessment

The bacterial neurotransmitter:sodium symporter homoglogue LeuT is an well-established model system for understanding the fundamental basis for how human monoamine transporters, such as the dopamine and serotonin, couple ions with neurotransmitter uptake. Here the authors provide convincing data to show that the K+ catalyses the return step of the transport cycle in LeuT by binding to one of the two sodium sites. The paper is an important contribution, but it's still unclear exactly where K+ binds in LeuT, and how to incorporate K+ binding into a transport cycle mechanism.

Public Reviews:

Reviewer #1 (Public Review):

This manuscript tackles an important question, namely how K+ affects substrate transport in the SLC6 family. K+ effects have previously been reported for DAT and SERT, but the prototypical SLC6fold transporter LeuT was not known to be sensitive to the K+ concentration. In this manuscript, the authors demonstrate convincingly that K+ inhibits Na+ binding, and Na+-dependent amino acid binding at high concentrations, and that K+ inside of vesicles containing LeuT increases the transport rate. However, outside K+ apparently had very little effect. Uptake data are supplemented with binding data, using the scintillation proximity assay, and transition metal FRET, allowing the observation of the distribution of distinct conformational states of the transporter.<br /> Overall, the data are of high quality. I was initially concerned about the use of solutions of very high ionic strength (the Km for K+ is in the 200 mM range), however, the authors performed good controls with lower ionic strength solutions, suggesting that the K+ effect is specific and not caused by artifacts from the high salt concentrations.

The major issue I have with this manuscript is with the interpretation of the experimental data. Granted that the K+ effect seems to be complex. However, it seems counterintuitive that K+ competes with Na+ for the same binding site, while at the same time accelerating the transport rate. Even if K+ prevents rebinding of Na+ on the inside of vesicles, it would be expected that K+ then stabilizes this Na+-free conformation, resulting in a slowing of the transport rate. However, the opposite is found. I feel that it would be useful to perform some kinetic modeling of the transport cycle to identify a mechanism that would allow K+ to act as a competitive inhibitor of Na+ binding and rate-accelerator at the same time.

This ties into the second point: It is not mentioned in the manuscript what the configuration of the vesicles is after LeuT reconstitution. Are they right-side out? Is LeuT distributed evenly in inside-out and right-side out orientation? Is the distribution known? If yes, how does it affect the interpretation of the uptake data with and without K+ gradient?

Finally, mutations were only made to the Na1 cation binding site. These mutations have an effect mostly to be expected, if K+ would bind to this site. However, indirect effects of mutations can never be excluded, and the authors acknowledge this in the discussion section. It would be interesting to see the effect of K+ on a couple of mutants that are far away from Na+/substrate binding sites. This could be another piece of evidence to exclude indirect effects, if the K+ affinity is less affected.

Reviewer #2(Public Review):

To characterize the relationship between Na+ and K+ binding to LeuT, the effect of K+ on Na+- dependent [3 H] leucine binding was studied using a scintillation proximity assay. In the presence of K+ the apparent affinity for sodium was reduced but the maximal binding capacity for this ion was unchanged, consistent with a competitive mechanism of inhibition between Na+ and K+.

To obtain a more direct readout of K+ binding to LeuT, tmFRET was used. This method relies on the distance-dependent quenching of a cysteine-conjugated fluorophore (FRET donor) by a transition metal (FRET acceptor). This method is a conformational readout for both ion- and ligand-binding. Along with the effect of K+ on Na+-dependent [3 H] leucine binding, the findings support the existence of a specific K+ binding site in LeuT and that K+ binding to this site induces an outward closed conformation.

It was previously shown that in liposomes inlaid with LeuT by reconstitution, intra-vesicular K+ increases the concentrative capacity of [ 3 H] alanine. To obtain insights into the mechanistic basis of this phenomenon, purified LeuT was reconstituted into liposomes containing a variety of cations, including Na+ and K+ followed by measurements of [ 3 H] alanine uptake driven by a Na+ gradient.

The ionic composition of the external medium was manipulated to determine if the stimulation of [3 H] alanine uptake by K+ was due to an outward directed potassium gradient serving as a driving force for sodium-dependent substrate transport by moving in the direction opposite to that of sodium and the substrate. Remarkably it was found that it is the intra-liposomal K+ per se that increases the transport rate of alanine and not a K+ gradient, suggesting that binding of K+ to the intra-cellular face of the transporter could prevent the rebinding of sodium and the substrate thereby reducing their efflux from the cell. These conclusions assume that the measured radioactive transport is via right-side-out liposomes rather than from their inverted counterparts (in case of a random orientation of the transporters in the proteoliposomes). Even though this assumption is likely to be correct, it should be tested.

Since K+- and Na+-binding are competitive and K+ excludes substrate binding, the Authors chose to focus on the Na1 site where the carboxyl group of the substrate serves as one of the groups which coordinate the sodium ion. This was done by the introduction of conservative mutations of the amino acid residues forming the Na1 site. The potassium interaction in these mutants was monitored by sodium dependent radioactive leucine binding. Moreover, the effect the effect of Na+ with and without substrate as well as that of potassium on the conformational equilibria was measured by tmFRET measurements on the mutants introduced in the construct enabling the measurements. The results suggest that K+-binding to LeuT modulates substrate transport and that the K+ affinity and selectivity for LeuT is sensitive to mutations in the Na1 site, pointing toward the Na1 site as a candidate site for facilitating the interaction between K+ in some NSS members.

The data presented in this manuscript are of very high quality. They are a detailed extension of results by the same group (Billesbolle et. al, Ref. 16 from the list) providing more detailed information on the importance of the Na1 site for potassium interaction. Clearly this begs for the identification of the binding site in a potassium bound LeuT structure in the future. Presumably LeuT was studied here because it appears that it is relatively easy to determine structures of many conformational states. Furthermore, convincing evidence showed that the stimulatory effect of K+ on transport is not because of energization of substrate accumulation but is rather due to the binding of this cation to a specific site.

Reviewer #1 (Recommendations For The Authors):

• Include a transport mechanism that can account for the K+ effects.

We appreciate the opportunity to elaborate further regarding how we envision this complex mechanism. It is generally known that, within the LeuT-fold transporters, the return step is ratelimiting for the transport process. Our data suggests that K+ binds to the inward-facing apo form.

Accordingly, we propose that the role of K+ binding is to facilitate LeuT to overcome the rate-limiting step. We propose the following mechanistic model: When Na+ and substrate is released to the intracellular environment the transporter must return to the outward-facing conformation. This can happen in (at least) two ways: 1) The transporter in its apo-form closes the inner gate and opens to the extracellular side, now ready to perform a new transport cycle. 2) The transporter rebinds Na+, which allows for the rebinding of substrate. It can now go in reverse (efflux) or it once again release its content. The transporter can naturally also only rebind Na+ and release it again to the cytosol.

The purpose of K+ binding is to prevent Na+ rebinding and to promote a conformational state of the transporter, which does not allow Na+ binding. Even though Na+ has a higher affinity for the site, K+ is much more abundant.

This model is supported by our previous experiment, showing that intravesicular K+ prevents [3H]alanine efflux while LeuT performs Na+-dependent alanine transport. Thus, the increase in Vmax could be due to a decreased efflux (exchange mode), or a facilitation of the rate-limiting step, or a combination of the two.

Note that the model does not require that K+ is counter-transported. It just has to prevent Na+ rebinding. However, even though we failed to show K+ counter-transport, it does not mean that it does not happen. Further experiments must clarify this issue.

To be more explicit about our proposed mechanistic model, we have expanded the last paragraph in the Discussion section. It now reads:

“We propose that K+ binding either facilitates LeuT transition from inward- to outward-facing (the rate limiting step of the transport cycle), or solely prevents the rebinding and possible efflux of Na+ and substrate. It could also be a combination of both. Either way, intracellular K+ will lead to an increase in Vmax and concentrative capacity. Note that our previous experiment showed an increased [3H]alanine efflux when LeuT transports alanine in the absence of intra-vesicular K+16. Specifically, the mechanistic impact of K+ could be to catalyze LeuT away from the state that allows the rebinding of Na+ and substrate. This way, K+ binding would decrease the possible rebinding of intracellularly released Na+ and substrate, thereby rectifying the transport process and increase the concentrative capacity and Vmax (Figure 6). Our results suggest that K+ is not counter-transported but rather promotes LeuT to overcome an internal rate limiting energy barrier. However, further investigations must be performed before any conclusive statement can be made here.”

• Describe the orientation of the transporter in the vesicles.

When working with reconstituted NSS, the transport activity is determined by the Na+ gradient. This is also evident in the experiments where we dissipate the Na+ gradient. Here we find transport activity compatible to background. We can also see in the literature, that directionality is rarely determined for transport proteins in reconstituted systems. When that is said, it is difficult to know how the inside-out LeuT contribute to the transport process. Will they work in reverse and contribute to the accumulation of intravesicular [3H]alanine? If so, to what extent? They will likely not be affected by the intravesicular K+. Therefore, their possible contribution will ‘work against’ our results and decrease the apparent K+ effects reported herein. Taken together, unless the vast majority of LeuT molecules are inside-out, knowing the actual proportion will not, in our perspective, affect our interpretations and conclusions of the data.

When that is said, we have also been curious about this issue and with the question raised by the reviewer, we performed the suggested experiment. We have inserted the results in Figure 3 – Figure supplement 1D. The figure shows that a fraction of the reconstituted LeuT are susceptible to thrombin cleavage of the accessible C-terminal. We have quantified the cleaved fraction to around 40% of the total (see Author response image 1 below). It is, however, a crude estimate since it is difficult to perform reliable dosimetry with fractions that close together. Thus, we are reluctant to add a quantitative measure in the article text.

Author response image 1.

We have inserted the following in the main text:

“It is difficult to control the directionality of proteins when they are reconstituted into lipid vesicles. They will be inserted in both orientations. Outside-out and inside-out. In the case of LeuT it is the imposed Na+-gradient which is determines the directionality of transport. Uptake through the insideout transporters will probably also happen. Note that the inside-out LeuT will not have the K+ binding site exposed to the intra-vesicular environment. Accordingly, a propensity of transporters will likely not be influenced by the added K+ and will tend to mask the contribution of K+ to the transport mode from the right-side out LeuT. To investigate LeuT directionality in our reconstituted samples, we performed thrombin cleavage of accessible C-terminals on intact and perforated vesicles, respectively. The result suggests that the proportion of LeuT inserted as outside-out is larger than the proportion with an inside-out directionality (Figure 3 – Figure supplement 1D).”

For the inserted Figure 3 – Figure supplement 1D, we have added the following legend:<br /> “(D) SDS-PAGE analysis of LeuT proteoliposomes following time-dependent thrombin digestion of accessible C-terminals (reducing the mass of LeuT by ~1.3 kDa). The reaction was terminated by the addition of PMSF at the specified time points. The lanes corresponding to the time-dependent proteolysis are flanked by lanes containing proteoliposomes without thrombin (left, 0 min) or digested in the presence of DDM (right, 180 min+DDM). Arrows indicate bands of full-length (top) and cleaved (bottom) LeuT.”

• Check the effects of mutations away from the Na1 cation binding site.

We have included the LeuT K398C in the study as a negative control for unspecific effects on Na+ and K+ binding. The mutant exhibit Na+ dependent [3H]leucine binding and K+-dependency similar to LeuT WT – see Table 2 and Table 2 - Figure Supplement 1G.

As a minor point, the authors use the term "affinity" liberally. However, unless these are direct binding experiments, the term "apparent affinity" may be more appropriate, since Km values are affected by the transport cycle (in uptake), as well as binding of cations/substrate.

We thank the reviewer for emphasizing this important point. We have revised the manuscript accordingly. We use ‘affinity’ when it has been determined under equilibrium conditions, either as a SPA binding experiment or based on tmFRET. We use the term ‘Km’ when the apparent affinity has been determined during non-equilibrium conditions such as during substrate transport.

Reviewer #2 (Recommendations For The Authors):

As mentioned in part 2, it is important to show the effect of internal potassium on transport in-sided liposomes. This could be done using the methodology developed by Tsai et. al. Biochemistry 51 (2012) 1557-1585.

We appreciate this important point and have performed the suggested experiment. See reviewer 1 comment #2

In the Abstract and throughout it is mentioned that K+ is not counter transported, yet on the bottom of p. 16 it is mentioned that this is possible.

We have tried to be very cautious with any interpretation about whether K+ is only binding or whether it is also counter-transported. Either way, it must facilitate a transition towards a non-Na+ binding state. We tried to differentiate between the two possibilities by investigating if an outwarddirected K+ gradient alone could drive transport (Figure 3E). We do not observe any significant difference from background (no gradient). However, the gained information is rather weak: It is still possible that K+ is counter-transported, but the K+ gradient does not impose any driving force. Instead, it ensures a rectification of the Na+-dependent substrate transport. If so, this experiment would come up negative even if K+ is counter-transported.

To be more explicit, we have changed the wording on page 16.

Our results suggests that K+ is not counter-transported, but rather promote LeuT to overcome an internal rate limiting energy barrier. However, further investigations must be performed before any conclusive statement can be made here.

Fig.2-Fig. Supplement 1: it is important to show that the effect of leucine is sodium-dependent by adding the control K+ and leucine.

We thank the reviewer for suggesting this important control. We have added the experiment to Figure 2 – Figure supplement 1 as suggested. The effect is not different from K+ alone supporting the SPA-binding data that K+-binding does not promote substrate binding.

Point for discussion: Whereas potassium is counter transported in SERT, there are conflicting interpretations on this in DAT (Ref. 15 from the list and Bhat et. al eLife (2021) 10:e67996). The situation in LeuT seems like the scenario described by Bhat et. al.

We appreciate the suggestion for a proposed link between LeuT and hDAT. Although, as mentioned above, we find it early days to be too certain on this option. We have now mentioned the mechanistic similarity in the Discussion following our description of the proposed mechanistic model (see first request from reviewer #1):

“If K+ is not counter-transported, LeuT might comply with the mechanism previously suggested for the human DAT31.”

Fig. 5-Fig. Supplement 1: Why are no data on N27Q and N286Q given? If these mutants have no transport activity this should be stated. Moreover, alanine uptake by A22V is almost sodium independent and is also very fast, suggesting binding, not transport. Are the counts sensitive to ionophores like nigericin?

We appreciate this important point. Indeed, the LeuT N27Q and N286Q are transport inactive. This information is now inserted in the main text when describing the conformational dynamics of N27QtmFRET and N286QtmFRET.

We agree with the reviewer that the [3H]alanine uptake for A22V is not very conclusive. The vesicles with Na+ on both sides (open diamonds) do allow [3H]alanine binding. Vesicles with added gramicidin are similar in activity. The fast rate could indeed suggest a binding event. This we also do not rule out in the main text. However, the contribution in activity from LeuT A22V in vesicles with a Na+ gradient cannot be explained by a binding event alone. Then it should bind more [3H]alanine in the presence of a Na+ gradient, which is possible, but hard to imagine. Also, the alanine affinity for LeuT A22V is ~1 µM (Table 1). At this affinity it should be literally impossible to detect any binding because the off-rate is so fast that it would all dissociate during the washing procedure.

We have described the data and left out any interpretation (e.g. changed ‘[3H]alanine transport’ to ‘[3H]alanine activity’). In addition, we have replaced: “This correlates with the lack of changes in conformational equilibrium observed in the tmFRET data between the NMDG+, Na+ and K+ states.” with: “Further investigations must clarify whether the changes in observed [3H]alanine activity constitutes a transport- or a binding event.”

Lower part of p. 16. The Authors speculate "that the mechanistic impact of K+ binding could be to accelerate a transition away from the conformation where Na+ and substrate are released, to a state where they can no longer rebind and thus revert the transport process (efflux)". This could be easily tested by measuring exchange, which should not be influenced by potassium.

We performed this experiment in Billesbolle et al. 2016. Nat Commun (Fig. 1f). We show that the exchange is decreased in the presence of K+. We hypothesize that this is because K+ binding forces LeuT away from the exchange mode.

Author Response

Response to the Reviews

We are grateful for these balanced, nuanced evaluations of our work concerning the observed epistatic trends and our interpretations of their mechanistic origins. Overall, we think the reviewers have done an excellent job at recognizing the novel aspects of our findings while also discussing the caveats associated with our interpretations of the biophysical effects of these mutations. We believe it is important to consider both of these aspects of our work in order to appreciate these advances and what sorts of pertinent questions remain.

Notably, both reviewers suggest that a lack of experimental approaches to compare the conformational properties of GnRHR variants weakens our claims. We would first humbly suggest that this constitutes a more general caveat that applies to nearly all investigations of the cellular misfolding of α-helical membrane proteins. Whether or not any current in vitro folding measurements report on conformational transitions that are relevant to cellular protein misfolding reactions remains an active area of debate (discussed further below). Nevertheless, while we concede that our structural and/ or computational evaluations of various mutagenic effects remain speculative, prevailing knowledge on the mechanisms of membrane protein folding suggest our mutations of interest (V276T and W107A) are highly unlikely to promote misfolding in precisely the same way. Thus, regardless of whether or not we were able experimentally compare the relevant folding energetics of GnRHR variants, we are confident that the distinct epistatic interactions formed by these mutations reflect variations in the misfolding mechanism and that they are distinct from the interactions that are observed in the context of stable proteins. In the following, we provide detailed considerations concerning these caveats in relation to the reviewers’ specific comments.

Reviewer #1 (Public Review):

The paper carries out an impressive and exhaustive non-sense mutagenesis using deep mutational scanning (DMS) of the gonadotropin-releasing hormone receptor for the WT protein and two single point mutations that I) influence TM insertion (V267T) and ii) influence protein stability (W107A), and then measures the effect of these mutants on correct plasma membrane expression (PME).

Overall, most mutations decreased mGnRHR PME levels in all three backgrounds, indicating poor mutational tolerance under these conditions. The W107A variant wasn't really recoverable with low levels of plasma membrane localisation. For the V267T variant, most additional mutations were more deleterious than WT based on correct trafficking, indicating a synergistic effect. As one might expect, there was a higher degree of positive correlation between V267T/W107A mutants and other mutants located in TM regions, confirming that improper trafficking was a likely consequence of membrane protein co-translational folding. Nevertheless, context is important, as positive synergistic mutants in the V27T could be negative in the W107A background and vice versa. Taken together, this important study highlights the complexity of membrane protein folding in dissecting the mechanism-dependent impact of disease-causing mutations related to improper trafficking.

Strengths

This is a novel and exhaustive approach to dissecting how receptor mutations under different mutational backgrounds related to co-translational folding, could influence membrane protein trafficking.

Weaknesses

The premise for the study requires an in-depth understanding of how the single-point mutations analysed affect membrane protein folding, but the single-point mutants used seem to lack proper validation.

Given our limited understanding of the structural properties of misfolded membrane proteins, it is unclear whether the relevant conformational effects of these mutations can be unambiguously validated using current biochemical and/ or biophysical folding assays. X-ray crystallography, cryo-EM, and NMR spectroscopy measurements have demonstrated that many purified GPCRs retain native-like structural ensembles within certain detergent micelles, bicelles, and/ or nanodiscs. However, helical membrane protein folding measurements typically require titration with denaturing detergents to promote the formation of a denatured state ensemble (DSE), which will invariably retain considerable secondary structure. Given that the solvation provided by mixed micelles is clearly distinct from that of native membranes, it remains unclear whether these DSEs represent a reasonable proxy for the misfolded conformations recognized by cellular quality control (QC, see https://doi.org/10.1021/acs.chemrev.8b00532). Thus, the use and interpretation of these systems for such purposes remains contentious in the membrane protein folding community. In addition to this theoretical issue, we are unaware of any instances in which GPCRs have been found to undergo reversible denaturation in vitro- a practical requirement for equilibrium folding measurements (https://doi.org/10.1146/annurev-biophys-051013-022926). We note that, while the resistance of GPCRs to aggregation, proteolysis, and/ or mechanical unfolding have also been probed in micelles, it is again unclear whether the associated thermal, kinetic, and/ or mechanical stability should necessarily correspond to their resistance to cotranslational and/ or posttranslational misfolding. Thus, even if we had attempted to validate the computational folding predictions employed herein, we suspect that any resulting correlations with cellular expression may have justifiably been viewed by many as circumstantial. Simply put, we know very little about the non-native conformations are generally involved in the cellular misfolding of α-helical membrane proteins, much less how to measure their relative abundance. From a philosophical standpoint, we prefer to let cells tell us what sorts of broken protein variants are degraded by their QC systems, then do our best to surmise what this tells us about the relevant properties of cellular DSEs.

Despite this fundamental caveat, we believe that the chosen mutations and our interpretation of their relevant conformational effects are reasonably well-informed by current modeling tools and by prevailing knowledge on the physicochemical drivers of membrane protein folding and misfolding. Specifically, the mechanistic constraints of translocon-mediated membrane integration provide an understanding of the types of mutations that are likely to disrupt cotranslational folding. Though we are still learning about the protein complexes that mediate membrane translocation (https://doi.org/10.1038/s41586-022-05336-2), it is known that this underlying process is fundamentally driven by the membrane depth-dependent amino acid transfer free energies (https://doi.org/10.1146/annurev.biophys.37.032807.125904). This energetic consideration suggests introducing polar side chains near the center of a nascent TMDs should almost invariably reduce the efficiency of topogenesis. To confirm this in the context of TMD6 specifically, we utilized a well-established biochemical reporter system to confirm that V276T attenuates its translocon-mediated membrane integration (Fig. S1)- at least in the context of a chimeric protein. We also constructed a glycosylation-based topology reporter for full-length GnRHR, but ultimately found its’ in vitro expression to be insufficient to detect changes in the nascent topological ensemble. In contrast to V276T, the W107A mutation is predicted to preserve the native topological energetics of GnRHR due to its position within a soluble loop region. W107A is also unlike V276T in that it clearly disrupts tertiary interactions that stabilize the native structure. This mutation should preclude the formation of a structurally conserved hydrogen bonding network that has been observed in the context of at least 25 native GPCR structures (https://doi.org/10.7554/eLife.5489). However, without a relevant folding assay, the extent to which this network stabilizes the native GnRHR fold in cellular membranes remains unclear. Overall, we admit that these limitations have prevented us from measuring how much V276T alters the efficiency of GnRHR topogenesis, how much the W107A destabilizes the native fold, or vice versa. Nevertheless, given these design principles and the fact that both reduce the plasma membrane expression of GnRHR, as expected, we are highly confident that the structural defects generated by these mutations do, in fact, promote misfolding in their own ways. We also concede that the degree to which these mutagenic perturbations are indeed selective for specific folding processes is somewhat uncertain. However, it seems exceedingly unlikely that these mutations should disrupt topogenesis and/ or the folding of the native topomer to the exact same extent. From our perspective, this is the most important consideration with respect to the validity of the conclusions we have made in this manuscript.

Furthermore, plasma membrane expression has been used as a proxy for incorrect membrane protein folding, but this not necessarily be the case, as even correctly folded membrane proteins may not be trafficked correctly, at least, under heterologous expression conditions. In addition, mutations can affect trafficking and potential post-translational modifications, like glycosylation.

While the reviewer is correct that the sorting of folded proteins within the secretory pathway is generally inefficient, it is also true that the maturation of nascent proteins within the ER generally bottlenecks the plasma membrane expression of most α-helical membrane proteins. Our group and several others have demonstrated that the efficiency of ER export generally appears to scale with the propensity of membrane proteins to achieve their correct topology and/ or to achieve their native fold (see https://doi.org/10.1021/jacs.5b03743 and https://doi.org/10.1021/jacs.8b08243). Notably, these investigations all involved proteins that contain native glycosylation and various other post-translational modification sites. While we cannot rule out that certain specific combinations of mutations may alter expression through their perturbation of post-translational GnRHR modifications, we feel confident that the general trends we have observed across hundreds of variants predominantly reflect changes in folding and cellular QC. This interpretation is supported by the relationship between observed trends in variant expression and Rosetta-based stability calculations, which we identified using unbiased unsupervised machine learning approaches (compare Figs. 6B & 6D).

Reviewer #2 (Public Review):

Summary:

In this paper, Chamness and colleagues make a pioneering effort to map epistatic interactions among mutations in a membrane protein. They introduce thousands of mutations to the mouse GnRH Receptor (GnRHR), either under wild-type background or two mutant backgrounds, representing mutations that destabilize GnRHR by distinct mechanisms. The first mutant background is W107A, destabilizing the tertiary fold, and the second, V276T, perturbing the efficiency of cotranslational insertion of TM6 to the membrane, which is essential for proper folding. They then measure the surface expression of these three mutant libraries, using it as a proxy for protein stability, since misfolded proteins do not typically make it to the plasma membrane. The resulting dataset is then used to shed light on how diverse mutations interact epistatically with the two genetic background mutations. Their main conclusion is that epistatic interactions vary depending on the degree of destabilization and the mechanism through which they perturb the protein. The mutation V276T forms primarily negative (aggravating) epistatic interactions with many mutations, as is common to destabilizing mutations in soluble proteins. Surprisingly, W107A forms many positive (alleviating) epistatic interactions with other mutations. They further show that the locations of secondary mutations correlate with the types of epistatic interactions they form with the above two mutants.

Strengths:

Such a high throughput study for epistasis in membrane proteins is pioneering, and the results are indeed illuminating. Examples of interesting findings are that: (1) No single mutation can dramatically rescue the destabilization introduced by W107A. (2) Epistasis with a secondary mutation is strongly influenced by the degree of destabilization introduced by the primary mutation. (3) Misfolding caused by mis-insertion tends to be aggravated by further mutations. The discussion of how protein folding energetics affects epistasis (Fig. 7) makes a lot of sense and lays out an interesting biophysical framework for the findings.

Weaknesses:

The major weakness comes from the potential limitations in the measurements of surface expression of severely misfolded mutants. This point is discussed quite fairly in the paper, in statements like "the W107A variant already exhibits marginal surface immunostaining" and many others. It seems that only about 5% of the W107A makes it to the plasma membrane compared to wild-type (Figures 2 and 3). This might be a low starting point from which to accurately measure the effects of secondary mutations.

The reviewer raises an excellent point that we considered at length during the analysis of these data and the preparation of the manuscript. Though we remain confident in the integrity of these measurements and the corresponding analyses, we now realize this aspect of the data merits further discussion and documentation in our forthcoming revision, in which we will outline the following specific lines of reasoning.

Still, the authors claim that measurements of W107A double mutants "still contain cellular subpopulations with surface immunostaining intensities that are well above or below that of the W107A single mutant, which suggests that this fluorescence signal is sensitive enough to detect subtle differences in the PME of these variants". I was not entirely convinced that this was true.

We made this statement based on the simple observation that the surface immunostaining intensities across the population of recombinant cells expressing the library of W107A double mutants was consistently broader than that of recombinant cells expressing W107A GnRHR alone (see Author response image 1 for reference). Given that the recombinant cellular library represents a mix of cells expressing ~1600 individual variants that are each present at low abundance, the pronounced tails within this distribution presumably represent the composite staining of many small cellular subpopulations that express collections of variants that deviate from the expression of W107A to an extent that is significant enough to be visible on a log intensity plot.

Author response image 1.

Firstly, I think it would be important to test how much noise these measurements have and how much surface immunostaining the W107A mutant displays above the background of cells that do not express the protein at all.

For reference, the average surface immunostaining intensity of HEK293T cells transiently expressing W107A GnRHR was 2.2-fold higher than that of the IRES-eGFP negative, untransfected cells within the same sample- the WT immunostaining intensity was 9.5-fold over background by comparison. Similarly, recombinant HEK293T cells expressing the W107A double mutant library had an average surface immunostaining intensity that was 2.6-fold over background across the two DMS trials. Thus, while the surface immunostaining of this variant is certainly diminished, we were still able to reliably detect W107A at the plasma membrane even under distinct expression regimes. We will include these and other signal-to-noise metrics for each experiment in a new table in the revised version of this manuscript.

Beyond considerations related to intensity, we also previously noticed the relative intensity values for W107A double mutants exhibited considerable precision across our two biological replicates. If signal were too poor to detect changes in variant expression, we would have expected a plot of the intensity values across these two replicates to form a scatter. Instead, we found DMS intensity values for individual variants to be highly correlated from one replicate to the next (Pearson’s R= 0.97, see Author response image 2 for reference). This observation empirically demonstrates that this assay consistently differentiated between variants that exhibit slightly enhanced immunostaining from those that have even lower immunostaining than W107A GnRHR.

Author response image 2.

But more importantly, it is not clear if under this regimen surface expression still reports on stability/protein fitness. It is unknown if the W107A retains any function or folding at all. For example, it is possible that the low amount of surface protein represents misfolded receptors that escaped the ER quality control.

While we believe that such questions are outside the scope of this work, we certainly agree that it is entirely possible that some of these variants bypass QC without achieving their native fold. This topic is quite interesting to us but is quite challenging to assess in the context of GPCRs, which have complex fitness landscapes that involve their propensity to distinguish between different ligands, engage specific components associated with divergent downstream signaling pathways, and navigate between endocytic recycling/ degradation pathways following activation. In light of the inherent complexity of GPCR function, we humbly suggest our choice of a relatively simple property of an otherwise complex protein may be viewed as a virtue rather than a shortcoming. Protein fitness is typically cast as the product of abundance and activity. Rather than measuring an oversimplified, composite fitness metric, we focused on one variable (plasma membrane expression) and its dominant effector (folding). We believe restraining the scope in this manner was key for the elucidation of clear mechanistic insights.

The differential clustering of epistatic mutations (Fig. 6) provides some interesting insights as to the rules that dictate epistasis, but these too are dominated by the magnitude of destabilization caused by one of the mutations. In this case, the secondary mutations that had the most interesting epistasis were exceedingly destabilizing. With this in mind, it is hard to interpret the results that emerge regarding the epistatic interactions of W107A. Furthermore, the most significant positive epistasis is observed when W107A is combined with additional mutations that almost completely abolish surface expression. It is likely that either mutation destabilizes the protein beyond repair. Therefore, what we can learn from the fact that such mutations have positive epistasis is not clear to me. Based on this, I am not sure that another mutation that disrupts the tertiary folding more mildly would not yield different results. With that said, I believe that the results regarding the epistasis of V276T with other mutations are strong and very interesting on their own.

We agree with the reviewer. In light of our results we believe it is virtually certain that the secondary mutations characterized herein would be likely to form distinct epistatic interactions with mutations that are only mildly destabilizing. Indeed, this insight reflects one of the key takeaway messages from this work- stability-mediated epistasis is difficult to generalize because it should depend on the extent to which each mutation changes the stability (ΔΔG) as well as initial stability of the WT/ reference sequence (ΔG, see Figure 7). Frankly, we are not so sure we would have pieced this together as clearly had we not had the fortune (or misfortune?) of including such a destructive mutation like W107A as a point of reference.

Additionally, the study draws general conclusions from the characterization of only two mutations, W107A and V276T. At this point, it is hard to know if other mutations that perturb insertion or tertiary folding would behave similarly. This should be emphasized in the text.

We agree and will be sure to emphasize this point in the revised manuscript.

Some statistical aspects of the study could be improved:

1. It would be nice to see the level of reproducibility of the biological replicates in a plot, such as scatter or similar, with correlation values that give a sense of the noise level of the measurements. This should be done before filtering out the inconsistent data.

We thank the reviewer for this suggestion and will include scatters for each genetic background like the one shown above in the supplement of the revised version of the manuscript.

1. The statements "Variants bearing mutations within the C- terminal region (ICL3-TMD6-ECL3-TMD7) fare consistently worse in the V276T background relative to WT (Fig. 4 B & E)." and "In contrast, mutations that are 210 better tolerated in the context of W107A mGnRHR are located 211 throughout the structure but are particularly abundant among residues 212 in the middle of the primary structure that form TMD4, ICL2, and ECL2 213 (Fig. 4 C & F)." are both hard to judge. Inspecting Figures 4B and C does not immediately show these trends, and importantly, a solid statistical test is missing here. In Figures 4E and F the locations of the different loops and TMs are not indicated on the structure, making these statements hard to judge.

We apologize for this oversight and thank the reviewer for pointing this out. We will include additional statistical tests to reinforce these conclusions in the revised version of the manuscript.

1. The following statement lacks a statistical test: "Notably, these 98 variants are enriched with TMD variants (65% TMD) relative to the overall set of 251 variants (45% TMD)." Is this enrichment significant? Further in the same paragraph, the claim that "In contrast to the sparse epistasis that is generally observed between mutations within soluble proteins, these findings suggest a relatively large proportion of random mutations form epistatic interactions in the context of unstable mGnRHR variants". Needs to be backed by relevant data and statistics, or at least a reference.

We will include additional statistical tests for this in the revised manuscript and will ensure the language we use is consistent with the strength of the indicated statistical enrichment.