Reviewer #4 (Public Review):

Summary:

In this work, the authors have used a mouse model of familial Amyotrophic lateral sclerosis (ALS) that carries a G93A mutation in the Sod1 gen to understand how the extraocular muscles (EOM) are preserved in ALS while other muscles undergo degeneration. Interestingly, the authors demonstrate that the integrity of neuromuscular junctions (NMJ) is affected by ALS in the limb and diaphragm muscles of G93A mice, while EOM is mostly preserved. The authors also further demonstrate that NaBu treatment partially restores the integrity of NMJ in the limb and diaphragm muscles of G93A mice. The results also indicate that chemokine Cxcl12 is expressed at higher levels in EOM myoblasts, and transduction with AAV encoding Cxcl12 improved the phenotypic characteristics of hindlimb-derived satellite cells.

Strengths:

The authors have used both in vivo and cell culture models. The findings have a translational potential.

Weaknesses:

The use of NaBu could be an issue as it has multiple effects and targets in ALS.

The sample size of animal experiments still needs to be improved.

The molecular mechanism of how Cxcl12 improved the phenotypic characteristics of hindlimb-derived satellite cells is still being determined.

eLife assessment

This useful study uses a chemoinformatics pipeline to identify a list of candidate mosquito repellants that may be pleasant to smell and safe for humans. The computational methodology is solid, but insufficiently benchmarked against other leading models. At the high concentrations tested, there may also be off-target effects of the repellents on the mosquitoes that are not considered. This paper may be of interest to specialists interested in the discovery of new mosquito repellents.

Reviewer #1 (Public Review):

Summary:

In this manuscript, the authors set up a pipeline to predict insect repellents that are pleasant and safe for humans. This is done by daisy-chaining a new classification model based on predicting repellents with a published model on predicting human perception. Models use a feature-engineered selection of chemical features to make their predictions. The predicted molecules are then validated against a proxy humanoid (heated brick) and its safety is tested by molecular assays of human cells. The humanistic approach to modeling these authors have taken (which considers cosmetic/aesthetic appeal and safety) is novel and a necessary step for consumer usage. However, the importance of pleasantness over effectiveness is still up for debate (DEET is unpleasant but still used often) and the generalization of safety tests is unknown and assumed. The effectiveness of the prediction models is also still warranted. They pass the authors' own behavioral tests, but their contribution to the field is unknown as both models (new and published) have not been rigorously benchmarked to previous models. Moreover, the author's breadth of literature in this field is sparse, ignoring directly related studies.

Strengths:

Humanistic approach to modeling considers pleasantness and safety. Chaining models can help limit the candidate odorants from the vastness of odor space.

Weaknesses:

The current models need to be bench-marked against leading models predicting similar outcomes. Similarly, many of these papers need to be addressed and discussed in the introduction. The authors might even consider their data sources for model training to increase performance and lexical categorization for interoperability. For instance, the Dravnikes data lexicon, currently used in the human perception lexicon, has been highly criticized for its overlapping and hard-to-interpret descriptive terms ("FRAGRANT", "AROMATIC").

Human Perception

Khan, R. M., Luk, C. H., Flinker, A., Aggarwal, A., Lapid, H., Haddad, R., & Sobel, N. (2007). Predicting odor pleasantness from odorant structure: pleasantness as a reflection of the physical world. Journal of Neuroscience, 27(37), 10015-10023.

Keller, A., Gerkin, R. C., Guan, Y., Dhurandhar, A., Turu, G., Szalai, B., ... & Meyer, P. (2017). Predicting human olfactory perception from chemical features of odor molecules. Science, 355(6327), 820-826.

Gutiérrez, E. D., Dhurandhar, A., Keller, A., Meyer, P., & Cecchi, G. A. (2018). Predicting natural language descriptions of mono-molecular odorants. Nature communications, 9(1), 4979.

Lee, B. K., Mayhew, E. J., Sanchez-Lengeling, B., Wei, J. N., Qian, W. W., Little, K. A., ... & Wiltschko, A. B. (2023). A principal odor map unifies diverse tasks in olfactory perception. Science, 381(6661), 999-1006.<br /> Related cleaned data: https://github.com/BioMachineLearning/openpom

Insect Repellents:

Wright, R. H. (1956). Physical basis of insect repellency. Nature, 178(4534), 638-638.

Katritzky, A. R., Wang, Z., Slavov, S., Tsikolia, M., Dobchev, D., Akhmedov, N. G., ... & Linthicum, K. J. (2008). Synthesis and bioassay of improved mosquito repellents predicted from chemical structure. Proceedings of the National Academy of Sciences, 105(21), 7359-7364.

Bernier, U. R., & Tsikolia, M. (2011). Development of Novel Repellents Using Structure− Activity Modeling of Compounds in the USDA Archival Database. In Recent Developments in Invertebrate Repellents (pp. 21-46). American Chemical Society.

Wei, J. N., Vlot, M., Sanchez-Lengeling, B., Lee, B. K., Berning, L., Vos, M. W., ... & Dechering, K. J. (2022). A deep learning and digital archaeology approach for mosquito repellent discovery. bioRxiv, 2022-09.

The current study assumes that insect repellents repel via their odor valence to the insect, but this is not accurate. Insect repellents also mask the body odor of humans making them hard to locate. The authors need to consult the literature to understand the localization and landing mechanisms of insects to their hosts. Here, they will understand that heat alone is not the attractant as their behavioral assay would have you believe. I suggest the authors test other behaviour assays to show more convincing evidence of effectiveness. See the following studies:

De Obaldia, M. E., Morita, T., Dedmon, L. C., Boehmler, D. J., Jiang, C. S., Zeledon, E. V., ... & Vosshall, L. B. (2022). Differential mosquito attraction to humans is associated with skin-derived carboxylic acid levels. Cell, 185(22), 4099-4116.

McBride, C. S., Baier, F., Omondi, A. B., Spitzer, S. A., Lutomiah, J., Sang, R., ... & Vosshall, L. B. (2014). Evolution of mosquito preference for humans linked to an odorant receptor. Nature, 515(7526), 222-227.

Wei, J. N., Vlot, M., Sanchez-Lengeling, B., Lee, B. K., Berning, L., Vos, M. W., ... & Dechering, K. J. (2022). A deep learning and digital archaeology approach for mosquito repellent discovery. bioRxiv, 2022-09.

Reviewer #2 (Public Review):

Summary:<br /> This is an interesting study that seeks to identify novel mosquito repellents that smell attractive to humans.

Strengths:<br /> The combination of standard machine learning methods with mosquito behavioral tests is a strength.

Weaknesses:<br /> The study would be strengthened by describing how other modern ML approaches (RF, decision trees) would classify and identify other potential repellents.

A comparison in the repellent activity between DEET and the top ten hits identified in this new study indicates little change in repellent activity (~3%), suggesting that DEET remains the gold standard. Without additional toxicity tests, the study is arguably incremental. The study's novelty should be better clarified.

The Methods in the repellency tests are sparse, and more information would be useful. Testing the top repellents at low doses (<<1%) and for long periods (2-12 h) would strengthen the manuscript. Without this information, the manuscript is lacking in depth.

Testing human subjects on their olfactory perceptions of the repellents would also increase the depth and utility of the manuscript. Without additional experiments, the authors' conclusions lack support and have limited impact on the state-of-the-art.

This manuscript is a mix of different approaches, which makes it lack cohesion. There is the ML method for classifying new repellents that smell good, but no testing of the repellents on human volunteers. The repellents are not tested at realistic concentrations and durations. And the calcium mobilization test is strange and makes little sense in the context of the other experiments and framing of the manuscript.

eLife assessment

This study reports a potentially important discovery that testosterone-induced metabolic changes in seminal vesicle epithelial cells lead to the production of oleic acids in seminal plasma to enhance sperm fertility. The evidence to support metabolic changes in seminal vesicles and the identification of oleic acid as a key factor in seminal plasma is solid. However, the evidence for how oleic acids support enhanced sperm fertility in vivo is not well supported, thus currently remains incomplete, and requires further study.

Reviewer #1 (Public Review):

Summary:

In this report, the authors investigated the effects of reproductive secretions on sperm function in mice. The authors attempt to weave together an interesting mechanism whereby a testosterone-dependent shift in metabolic flux patterns in the seminal vesicle epithelium supports fatty acid synthesis, which they suggest is an essential component of seminal plasma that modulates sperm function by supporting linear motility patterns.

Strengths:

The topic is interesting and of general interest to the field. The study employs an impressive array of approaches to explore the relationship between mouse endocrine physiology and sperm function mediated by seminal components from various glandular secretions of the male reproductive tract.

Weaknesses:

Unfortunately, support for the proposed mechanism is not convincingly supported by the data, and the experimental design and methodology need more rigor and details, and the presence of numerous (uncontrolled) confounding variables in almost every experimental group significantly reduce confidence in the overall conclusions of the study.

The methodological detail as described is insufficient to support replication of the work. Many of the statistical analyses are not appropriate for the apparent designs (e.g. t-tests without corrections for multiple comparisons). This is important because the notion that different seminal secretions will affect sperm function would likely have a different conclusion if the correct controls were selected for post hoc comparison. In addition, the HTF condition was not adjusted to match the protein concentrations of the secretion-containing media, likely resulting in viscosity differences as a major confounding factor on sperm motility patterns.

There is ambiguity in many of the measurements due to the lack of normalization (e.g. all Seahorse Analyzer measurements are unnormalized, making cell mass and uniformity a major confounder in these measurements). This would be less of a concern if basal respiration rates were consistently similar across conditions and there were sufficient independent samples, but this was not the case in most of the experiments.

The observation that oleic acid is physiologically relevant to sperm function is not strongly supported. The cellular uptake of 10-100uM labeled oleic acid is presumably due to the detergent effects of the oleic acid, and the authors only show functional data for nM concentrations of exogenous oleic acid. In addition, the effect sizes in the supporting data were not large enough to provide a high degree of confidence given the small sample sizes and ambiguity of the design regarding the number of biological and technical replicates in the extracellular flux analysis experiments.

Overall, the most confident conclusion of the study was that testosterone affects the distribution of metabolic fluxes in a cultured human seminal vesicle epithelial cell line, although the physiological relevance of this observation is not clear.

In the introduction, the authors suggest that their analyses "reveal the pathways by which seminal vesicles synthesize seminal plasma, ensure sperm fertility, and provide new therapeutic and preventive strategies for male infertility." These conclusions need stronger or more complete data to support them.

Reviewer #2 (Public Review):

Summary:

Using a combination of in vivo studies with testosterone-inhibited and aged mice with lower testosterone levels, as well as isolated mouse and human seminal vesicle epithelial cells, the authors show that testosterone induces an increase in glucose uptake. They find that testosterone induces differential gene expression with a focus on metabolic enzymes. Specifically, they identify increased expression of enzymes that regulate cholesterol and fatty acid synthesis, leading to increased production of 18:1 oleic acid.

Strength:

Oleic acid is secreted by seminal vesicle epithelial cells and taken up by sperm, inducing an increase in mitochondrial respiration. The difference in sperm motility and in vivo fertilization in the presence of 18:1 oleic acid and the absence of testosterone is small but significant, suggesting that the authors have identified one of the fertilization-supporting factors in seminal plasma.

Weaknesses:

Further studies are required to investigate the effect of other seminal vesicle components on sperm capacitation to support the author's conclusions. The author's experiments focused on potential testosterone-induced changes in the rate of seminal vesicle epithelial cell glycolysis and oxphos, however, provide conflicting results and a potential correlation with seminal vesicle epithelial cell proliferation should be confirmed by additional experiments.

Reviewer #3 (Public Review):

Summary:

Male fertility depends on both sperm and seminal plasma, but the functional effect of seminal plasma on sperm has been relatively understudied. The authors investigate the testosterone-dependent synthesis of seminal plasma and identify oleic acid as a key factor in enhancing sperm fertility.

Strengths:

The evidence for changes in cell proliferation and metabolism of seminal vesicle epithelial cells and the identification of oleic acid as a key factor in seminal plasma is solid.

Weaknesses:

The evidence that oleic acids enhance sperm fertility in vivo needs more experimental support, as the main phenotypic effect in vitro provided by the authors remains simply as an increase in the linearity of sperm motility, which does not necessarily correlate with enhanced sperm fertility.

eLife assessment

This work is potentially useful because it has generated a mineable yield of new candidate immune inhibitory receptors, which can serve both as drug targets and as subjects for further biological investigation. It is noted however that the work is rather incomplete, in that it does little to validate the putative new receptors, and instead makes a study of their putative distribution across cell types. Experimental follow-up to demonstrate the claimed properties for the proteins identified, or mining existing experimental data sources on gene expression across tissues to at least show that the pipeline correctly identified genes likely to be specific to immune cells, would make this work more complete.

Reviewer #1 (Public Review):

This manuscript proposes a new bioinformatics approach identifying several hundreds of previously unknown inhibitory immunoreceptors. When expressed in immune cells (such as neutrophils, monocytes, CD8+, CD4+, and T-cells), such receptors inhibit the functional activity of these cells. Blocking inhibitory receptors represents a promising therapeutic strategy for cancer treatment.

As such, this is a high-quality and important bioinformatics study. One general concern is the absence of direct experimental validation of the results. In addition to the fact that the authors bioinformatically identified 51 known receptors, providing such experimental evaluation (of at least one, or better few identified receptors) would, in my opinion, significantly strengthen the presented evidence.

I will now briefly summarize the results and give my comments.

First, using sequence comparison analysis, the authors identify a large set of putative receptors based on the presence of immunoreceptor tyrosine-based inhibitory motifs (ITIMs), or immunoreceptor tyrosine-based switch motifs (ITSMs). They further filter the identified set of receptors for the presence of the ITIMs or ITSMs in an intracellular domain of the protein. Second, using AlphaFold structure modeling, the authors select only receptors containing ITIMs/ITSMs in structurally disordered regions. Third, the evaluation of gene expression profiles of known and putative receptors in several immune cell types was performed. Fourth, the authors classified putative receptors into functional categories, such as negative feedback receptors, threshold receptors, threshold disinhibition, and threshold-negative feedback. The latter classification was based on the available data from Nat Rev Immunol 2020. Fifth, using publicly available single-cell RNA sequencing data of tumor-infiltrating CD4+ and CD8+ cells from nearly twenty types of cancer, the authors demonstrate that a significant fraction of putative receptors are indeed expressed in these datasets.

In summary, in my opinion, this is an interesting, important, high-quality bioinformatics work. The manuscript is clearly written and all technical details are carefully explained.

One comment/suggestion regarding the methodology of evaluating gene expression profiles of putative receptors: perhaps it might be important to look at clusters of genes that are co-expressed with putative inhibitory receptors.

Reviewer #2 (Public Review):

Summary:

The authors developed a bioinformatic pipeline to aid the screening and identification of inhibitory receptors suitable as drug targets. The challenge lies in the large search space and lack of tools for assessing the likelihood of their inhibitory function. To make progress, the authors used a consensus protein membrane topology and sequence motif prediction tool (TOPCOS) combined with both a statistical measure assessing their likelihood function and a machine learning protein structural prediction model (AlphaFold) to greatly cut down the search space. After obtaining a manageable set of 398 high-confidence known and putative inhibitory receptors through this pipeline, the authors then mapped these receptors to different functional categories across different cell types based on their expression both in the resting and activated state. Additionally, by using publicly available pan-cancer scRNA-seq for tumor-infiltrating T-cell data, they showed that these receptors are expressed across various cellular subsets.

Strengths:

The authors presented sound arguments motivating the need to efficiently screen inhibitory receptors and to identify those that are functional. Key components of the algorithm were presented along with solid justification for why they addressed challenges faced by existing approaches. To name a few:

• TOPCON algorithm was elected to optimize the prediction of membrane topology.<br /> • A statistical measure was used to remove potential false positives.<br /> • AlphaFold is used to filter out putative receptors that are low confidence (and likely intrinsically disordered).

To examine receptors screened through this pipeline through a functional lens, the authors proposed to look at their expression of various immune cell subsets to assign functional categories. This is a reasonable and appropriate first step for interpreting and understanding how potential drug targets are differentially expressed in some disease contexts.

Weaknesses:

The paper has strength in the pipeline they presented, but the weakness, in my opinion, lies in the lack of concrete demonstration on how this pipeline can be used to at least "rediscover" known targets in a disease-specific manner. For example, the result that both known and putative immune inhibitory receptors are expressed across a wide variety of tumor-infiltrating T-cell subsets is reassuring, but this would have been more informative and illustrative if the authors could demonstrate using a disease with known targets, as opposed to a pan-cancer context. Additionally, a discussion that contrasts the known and putative receptors in the context above would help readers better identify use cases suitable for their research using this pipeline. Particularly,<br /> • For known receptors, does the pipeline and the expression analysis above rediscover the known target in the disease of interest?<br /> • For putative receptors, what do the functional category mapping and the differential expression across various tumor-infiltrating T-cell subsets imply on a potential therapeutic target?

Author Response

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

eLife assessment

This manuscript presents a valuable approach to exploring CD4+ T-cell response in mice across stimuli and tissues through the analysis of their T-cell receptor repertoires. The authors use a transgenic mouse model, in which the possible diversity of the T-cell receptor repertoire is reduced, such that each of a diverse set of immune exposures elicits more detectably consistent T-cell responses across different individuals. However, whereas the proposed experimental system could be utilized to study convergent T-cell responses, the analyses done in this manuscript are incomplete and do not support the claims due to limitations in the statistical analyses and lack of data/code access.

We worked to address the reviewers' concerns below, point-by-point.

All data on immune repertoires are deposited here: https://figshare.com/articles/dataset/Convergence_plasticity_and_tissue_residence_of_regulatory_and_effector_T_cell_response/22226155

We added the Data availability statement to the manuscript.

Public Reviews:

Reviewer #1 (Public Review):

The authors investigate the alpha chain TCR landscape in conventional vs regulatory CD4 T cells. Overall I think it is a very well thought out and executed study with interesting conclusions. The authors have investigated CDR3 alpha repertoires coupled with a transgenic fixed CDR3beta in a mouse system.

Strengths:

• One of a kind evidence and dataset.

• State-of-the-art analyses using tools that are well-accepted in the literature.

• Interesting conclusions on the breadth of immune response to challenges across different types of challenges (tumor, viral and parasitic).

Thank you for the positive view.

Weaknesses:

• Some conclusions regarding the eCD4->eTreg transition are not so strong using only the data.

The overlaps between the top-nucleotide clones in both LLC and PYMT challenges are prominently above the average, and this result is reproducible in lungs and skin, so we have no doubts based on these data. Further experiments with different methods, including tracking the clonal fates, should clarify and confirm/correct/disprove our findings.

• Some formatting issues.

We are working on the manuscript to correct minor errors and formatting.

Reviewer #2 (Public Review):

This study investigates T-cell repertoire responses in a mouse model with a transgenic beta chain, such that all T-cells in all mice share a fixed beta chain, and repertoire diversity is determined solely by alpha chain rearrangements. Each mouse is exposed to one of a few distinct immune challenges, sacrificed, and T-cells are sampled from multiple tissues. FACS is used to sort CD4 and Treg cell populations from each sample, and TCR repertoire sequencing from UMI-tagged cDNA is done.

Various analyses using repertoire diversity, overlap, and clustering are presented to support several principal findings: 1) TCR repertoires in this fixed beta system have highly distinct clonal compositions for each immune challenge and each cell type, 2) these are highly consistent across mice, so that mice with shared challenges have shared clones, and 3) induction of CD4-to-Treg cell type transitions is challenge-specific.

The beta chain used for this mouse model was previously isolated based on specificity for Ovalbumin. Because the beta chain is essential for determining TCR antigen specificity, and is highly diverse in wildtype mice, I found it surprising that these mice are reported to have robust and consistently focused clonal responses to very diverse immune challenges, for which a fixed OVA-specific beta chain is unlikely to be useful. The authors don't comment on this aspect of their findings, but I would think it is not expected a priori that this would work. If this does work as reported, it is a valuable model system: due to massively reduced diversity, the TCR repertoire response is much more stereotyped across individual samples, and it is much easier to detect challenge-specific TCRs via the statistics of convergent responses.

This was to some extent expected, since these mice live almost normally and have productive adaptive immune responses and protection. In real life, there are frequent TCR-pMHC interactions where the TCR-alpha chain dominates (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5701794/; https://pubmed.ncbi.nlm.nih.gov/37047500/). On the fixed TCR-beta background this mechanics starts working full-fledged, essentially substituting TCR-beta diversity, at the extent of relatively simplified TCRab repertoire and probably higher cross-reactivity.

We agree that this is a valuable model, for sure, and indicated this in the last sentence of our Discussion. Now we are also adding this point to the abstract.

While the data and analyses present interesting signals, they are flawed in several ways that undermine the reported findings. I summarize below what I think are the most substantive data and analysis issues.

(1) There may be systematic inconsistencies in repertoire sampling depth that are not described in the manuscript. Looking at the supplementary tables (and making some plots), I found that the control samples (mice with mock challenge) have consistently much shallower sampling-in terms of both read count and UMI count-compared with the other challenge samples. There is also a strong pattern of lower counts for Treg vs CD4 cell samples within each challenge.

The immune response of control mice is less extensive, as it should be. Just like the fact that the number of Tregs in tissues is lower than CD4, this is normal. So this all follows the expectations. But please note that we were very accurate everywhere with appropriate data normalisation, using all our previous extensive experience (https://pubmed.ncbi.nlm.nih.gov/29080364/).

In particular (now adding more relevant details to Methods):

For diversity metrics calculations, we randomly sampled an equal number of 1000 UMI from each cloneset. Samples with UMI < 700 were excluded from analysis.

For amino acid overlap metrics calculations, we selected top-1000 largest clonotypes from each cloneset. Samples with clonotype counts < 700 were excluded from analysis.

For nucleotide overlaps metrics calculations (eCD4-eTreg), we selected top-100 clonotypes from each cloneset. Samples with clonotypes < 100 were excluded from analysis.

The top N clonotypes were selected as the top N clonotypes after randomly shuffling the sequences and aligning them in descending order. This was done in order to get rid of the alphabetical order for clonotypes with equal counts (e.g. count = 1 or 2).

Downsampling was carried out using software vdjtools v.1.2.1.

(2) FACS data are not reported. Although the graphical abstract shows a schematic FACS plot, there are no such plots in the manuscript. Related to the issue above, it would be important to know the FACS cell counts for each sample.

Yes, we agree that this is valuable information that should be provided. Unfortunately, this data has not been preserved.

(3) For diversity estimation, UMI-wise downsampling was performed to normalize samples to 1000 random UMIs, but this procedure is not validated (the optimal normalization would require downsampling cells). What is the influence of possible sampling depth discrepancies mentioned above on diversity estimation? All of the Treg control samples have fewer than 1000 total UMIs-doesn't that pose a problem for sampling 1000 random UMIs?

Indeed, I simulated this procedure and found systematic effects on diversity estimates when taking samples of different numbers of cells (each with a simulated UMI count) from the same underlying repertoire, even after normalizing to 1000 random UMIs. I don't think UMI downsampling corrects for cell sampling depth differences in diversity estimation, so it's not clear that the trends in Fig 1A are not artifactual-they would seem to show higher diversity for control samples, but these are the very same samples with an apparent systematic sampling depth bias.

We evaluated this approach through all our work, and summarised in the ref: https://pubmed.ncbi.nlm.nih.gov/29080364/. Altogether, normalising to the same count of randomly sampled UMI seems to be the best approach (although, preferably, the initial sequencing depth should be essentially higher for all samples than the sampling threshold used). Initial sorting of identical numbers of cells and ideally uniform library preparation and sequencing is generally not realistic and does not work in the real world, while UMI downsampling does the same work much better.

(4) The Figures may be inconsistent with the data. I downloaded the Supplementary Table corresponding to Fig 1 and made my own version of panels A-C. This looked quite different from the diversity estimations depicted in the manuscript. The data does not match the scale or trends shown in the manuscript figure.

There was a wrong column for Chao1, now correcting. Also, please note that we only used samples with > 700 UMI. Supplementary Table now corrected accordingly. Also, please note that Figure 1 shows the results for lung samples only.

(5) For the overlap analysis, a different kind of normalization was performed, but also not validated. Instead of sampling 1000 UMIs, the repertoires were reduced to their top 1000 most frequent clones. It is not made clear why a different normalization would be needed here. There are several samples (including all Treg control samples) with only a couple hundred clones. It's also likely that the noted systematic sampling depth differences may drive the separation seen in MDS1 between Treg and CD4 cell types. I also simulated this alternative downsampling procedure and found strong effects on MDS clustering due to sampling effects alone.

That’s right, for the overlap analysis (which values are mathematically proportional to the clonotype counts in both compared repertoires, so the difference in the counts causes major biases) the right way to do it is to choose the same number of clonotypes. See Ref. https://pubmed.ncbi.nlm.nih.gov/29080364/.

We kept only samples with > 700 for the overlap analyses. Some relatively poor samples are present in all challenges, while MDS1 localization has clear reproducible logic, so we are confident in these results.

It is not made clear how the overlap scores were converted to distances for MDS. It's hard to interpret this without seeing the overlap matrix.

This is a built-in feature in VDJtools software (https://pubmed.ncbi.nlm.nih.gov/26606115/). See also here: https://vdjtools-doc.readthedocs.io/en/master/overlap.html.

(6) The cluster analysis is superficial, and appears to have been cherry-picked. The clusters reported in the main text have illegibly small logo plots, and no information about V/J gene enrichments. More importantly, as the caption states they were chosen from the columns of a large (and messier-looking) cluster matrix in the supplementary figure based on association with each specific challenge. There's no detail about how this association was calculated, or how it controlled for multiple tests. I don't think it is legitimate to simply display a set of clusters that visually correlate; in a sufficiently wide random matrix you will find columns that seem to correlate with any given pattern across rows.

Particular CDR3 sequences and VJ segments do not mean much for the results of this manuscript. Logos are given just for visual explanation of how the consensus motifs of the clusters look like.

We now add two more Supplementary Tables and a Supplementary Figure with full information about clusters.

We disagree that the Supplementary Figure 1 (representing all the clusters) looks “messy”. Vice versa, it is surprisingly “digital”, showing the clear patterns of responses and homings. This becomes clear if you visually study it for a while. But yes, it is too big to let the reader focus on this or that aspect. That is why we need to select TCR clusters to illustrate this or that aspect discussed in the work, but they were selected from the overall already structured picture.

(7) The findings on differential plasticity and CD4 to Treg conversion are not supported. If CD4 cells are converting to Tregs, we expect more nucleotide-level overlap of clones. This intuition makes sense. But it seems that this section affirms the consequent: variation in nucleotide-level clone overlap is a readout of variation in CD4 to Treg conversion. It is claimed, based on elevated nucleotide-level overlap, that the LLC and PYMT challenges induce conversion more readily than the other challenges. It is not noted in the textual interpretations, but Fig 4 also shows that the control samples had a substantially elevated nucleotide-level overlap. There is no mention of a null hypothesis for what we'd expect if there was no induced conversion going on at all. This is a reduced-diversity mouse model, so convergent recombination is more likely than usual, and the challenges could be expected to differ in the parts of TCR sequence space they induce focus on. They use the top 100 clones for normalization in this case, but don't say why (this is the 3rd distinct normalization procedure).

Your point is absolutely correct: “This is a reduced-diversity mouse model, so convergent recombination is more likely than usual”. Distinct normalisation procedure was required to focus on the most expanded clonotypes to avoid the tail of (presumably cross-reactive) and identical TCRs present in all repertoires in these limited-repertoire mice. So we downsampled as strictly as possible to minimise this background signal of nucleotide overlap, and only this strict downsampling to the top-100 clonotypes allowed us to visualise the difference between the challenges. This is a sort of too complicated explanation that would overload the manuscript. But your comments and our answers will be available to the reader who wants to go into all the details.

The observed (at this strict downsampling) overlaps between the top-nucleotide clones in both LLC and PYMT challenges are prominently above the average, and this result is reproducible in lungs and skin, so we have no doubts in interpretations based on these data. Further experiments with different methods, including tracking the clonal fates, should clarify and confirm/correct/disprove our findings.

Although interpretations of the reported findings are limited due to the issues above, this is an interesting model system in which to explore convergent responses. Follow-up experimental work could validate some of the reported signals, and the data set may also be useful for other specific questions.

Yes, thank you for your really thorough analysis. We fully agree with your conclusion.

Reviewer #3 (Public Review):

Nakonechnaya et al present a valuable and comprehensive exploration of CD4+ T cell response in mice across stimuli and tissues through the analysis of their TCR-alpha repertoires.

The authors compare repertoires by looking at the relative overlap of shared clonotypes and observe that they sometimes cluster by tissue and sometimes by stimulus. They also compare different CD4+ subsets (conventional and Tregs) and find distinct yet convergent responses with occasional plasticity across subsets for some stimuli.

The observed lack of a general behaviour highlights the need for careful comparison of immune repertoires across cell subsets and tissues in order to better understand their role in the adaptive immune response.

In conclusion, this is an important paper to the community as it suggests several future directions of exploration.

Unfortunately, the lack of code and data availability does not allow the reproducibility of the results.

Thank you for your positive view.

All data on immune repertoires are deposited here: https://figshare.com/articles/dataset/Convergence_plasticity_and_tissue_residence_of_regulatory_and_effector_T_cell_response/22226155

We added the Data availability statement to the manuscript.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

• In the manuscript at "yielding 13,369 {plus minus} 1,255 UMI-labeled TCRα cDNA molecules and 3233 {plus minus} 310 TCRα CDR3 clonotypes per sample" I'm not sure how can there be fewer unique DNA molecules than clonotypes in each sample.

That was our mistake for sure, now corrected.

• In the manuscript at "This indicates that the amplitude and focused nature of the effector and regulatory T cell response in lungs is generally comparable."

I'm not sure it's possible to conclude that a drop in diversity in all conditions necessarily signals a focused nature. Since at this stage, the nature of the colotypes was not compared between conditions, it is not possible to claim a focused nature of the response.

We have softened the wording:

"This could indicate that the amplitude and focused nature of the effector and regulatory T cell response in lungs is generally comparable."

• What are your thoughts on why there is such a large overlap between Treg and Teff in the Lung in control? For some replicates it is almost as much as a post-LLC challenge!

There is some natural dispersion in the data, which is generally expectable. The overlaps between the top-nucleotide clones in both LLC and PYMT challenges are prominently above the average, and this result is reproducible in lungs and skin, so we have no doubts based on these data. Further experiments with different methods, including tracking the clonal fates, should clarify and confirm/correct/disprove our findings.

• In the manuscript at "These results indicate that distinct antigenic specificities are generally characteristic for eTreg cells that preferentially reside in particular lymphatic niches" I'm not sure we can conclude this from the Figure. Wouldn't you expect the samples to be grouped by color (the different challenges)? Maybe I'm not understanding the sentence!

This is a different story, about resident Tregs, irrespective of the challenge.

The whole explanation is here in the text:

“Global CDR3α cluster analysis revealed that characteristic eTreg TCR motifs were present in distinct lymphatic tissues, including spleen and thymus, irrespective of the applied challenge (Supplementary Fig. 1). To better illustrate this phenomenon, we performed MDS analysis of CDR3α repertoires for distinct lymphatic tissues, excluding the lungs due to their otherwise dominant response to the current challenge. This analysis demonstrated close proximity of eTreg repertoires obtained from the same lymphatic tissues upon all lung challenges and across all animals (Fig. 5a, b). These results indicate that distinct antigenic specificities are generally characteristic for eTreg cells that preferentially reside in particular lymphatic niches. Notably, the convergence of lymphatic tissue-resident TCR repertoires was less prominent for the eCD4 T cells (Fig. 5c, d).”

And in the abstract:

“Additionally, our TCRα repertoire analysis demonstrated that distinct antigenic specificities are characteristic for eTreg cells residing in particular lymphatic tissues, regardless of the challenge, revealing the homing-specific, antigen-specific resident Treg populations. ”

• In the manuscript at " Notably, the convergence of lymphatic tissue-resident TCR repertoires was less prominent for the eCD4 T cells ":

5b and 5d seem to have the same pattern: Spleen and MLN group together, AxLN and IgLN together and thymus is separate. Do you mean to say that the groups are more diffuse? I feel like the pattern really is the same and it's likely due to some noise in the data…

Yes, we just mean here that eTreg groups are less diffuse - means more convergent.

• I'm not sold on the eCD4 to eTreg conversion evidence. Why only limit to the top 100 clones? The top 1000 clones were used in previous analyses! Moreover, the authors claim that calculating relative overlap (via F2) of matching CDR3+V+J genes is evidence of a conversion between eCD4 and eTreg. I think to convince myself of a real conversion, I would track the cells between groups, unfortunately, I'm not sure how to track this.. Maybe looking at the thymus population? For example, what is the overlap in the thymus vs. after the challenge? I don't have an answer on how to verify but I feel that this conclusion is a bit on the weaker end.

Distinct normalisation procedure was required to focus on the most expanded clonotypes to avoid the tail of (presumably cross-reactive) and identical TCRs present in all repertoires in these limited-repertoire mice. So we downsampled as strictly as possible to minimise this background signal of nucleotide overlap, and only this strict downsampling to the top-100 clonotypes allowed us to visualise the difference between the challenges. This is a sort of too complicated explanation that would overload the manuscript. But your comments and our answers will be available to the reader who wants to go into all the details.

The observed (at this strict downsampling) overlaps between the top-nucleotide clones in both LLC and PYMT challenges are prominently above the average, and this result is reproducible in lungs and skin, so we have no doubts in interpretations based on these data. Further experiments with different methods, including tracking the clonal fates, should clarify and confirm/correct/disprove our findings.

• There is a nuance in the analysis between Figure 3 and Figure 5 which I think I am not grasping. Both Figures use the same method and the same data but what is different? I think the manuscript would benefit from making this crystal clear. The conclusions will likely be more evident as well!

As explained in the text and above, on Figure 5 “we performed MDS analysis of CDR3α repertoires for distinct lymphatic tissues, excluding the lungs due to their otherwise dominant response to the current challenge.”

The idea of this mini-chapter of the manuscript is to reveal tissue-resident Tregs, distinct for distinct tissues, resident there in all these mice, irrespectively of the challenge we applied. And they are really there (!).

• Do the authors plan to share their R scripts?

All calculations were performed in VDJtools. R was only used to build figures. Corrected this in Methods.

Minor typos and formatting issues to address:

• Typo in Figure 2a the category should read "worm" instead of "warm"

Corrected.

• Figure 2a heatmap is missing a color bar indicating the value ranges

The detailed information can be found in additional Supplementary materials.

• Figure 2f is never mentioned in the manuscript!

Corrected.

• "eTreg repertoire upon lung challenge is reflected in the draining lymph node" - the word upon is of a lower size

Corrected.

• The authors should make the spelling of eTreg uniform across the manuscript (reg in subscript vs just lower case letters. Same goes for CDR3a vs CDR3\alpha

Corrected.

• Figure 4a-d p-values annotations are not shown. Is it because they are not significant?

Corrected.

• The spelling of FACS buffer should be uniform (FACs vs FACS, see methods)

Corrected.

• In the gating strategy, I would make a uniform annotation for the cluster of differentiation, for example, "CD44 high" vs "CD44^{hi}", pos vs + etc.

Corrected.

• Citation for MIGEC software (if available) is missing from methods

Present in the text so probably sufficient.

Reviewer #2 (Recommendations For The Authors):

I noticed the data was made available via Figshare in the preprint, but there is no data availability statement in the current ms.

We provided Data availability statement.

The methods state that custom scripts were written to perform the various analyses. Those should be made available in a code repository, and linked in the ms.

All calculations were performed in VDJtools. R was only used to build figures. Corrected this in Methods.

The title mentioned "TCR repertoire prism", so I thought "prism" was the name of a new method or software. But then the word "prism" didn't appear anywhere in the ms.

We just mean viewing or understanding something from a different perspective or through a lens that reveals different aspects or nuances.

Figure 1D lacks an x-axis label.

Worked on the figures in general.

Reviewer #3 (Recommendations For The Authors):

• The paper is very concise, possibly a bit too much. It could use additional explanations to properly affirm its relevance, for example:

why the choice of fixing the CDR3beta background?

To make repertoire more similar across the mice, and to track all the features of repertoire using only one chain.

to what it is fixed?

As explained in Methods:

“C57BL/6J DO11.10 TCRβ transgenic mice (kindly provided by Philippa Marrack) and crossed to C57BL/6J Foxp3eGFP TCRa-/- mice.”

What do you expect to see and not to see in this specific system and why it is important?

As stated above: we expected repertoire to be more similar across the mice, and it is important to find antigen-specific TCR clusters across mice, and to be able to track all the features of the TCR repertoire using only one chain.

Does this system induce more convergent responses? If so, can we extrapolate the results from this system to the full alpha-beta response?

Such a model, compared to conventional mice, is much more powerful in terms of the ability of monitoring convergent TCR responses. At the same time, it behaves natural, mice live almost normally, so we believe it reflects natural behaviour of the full fledged alpha-beta T cell repertoire.

• Is the lack of similarity of other tissues to Lung/MLN due to a lack of a response?

As indicated in the title of the corresponding mini-chapter: “eTreg repertoire upon lung challenge is reflected in the draining lymph node”. And conclusion of this mini-chapter is that “these results demonstrate the selective tissue localization of the antigen-focused Treg response. ”

Can you do a dendrogram like 2a for the other tissues to better clarify what is going on there? There is space in the supplementary material.

We built lots of those, but in such single dimension mostly they are less informative compared to 2D MDS plots.

• Figure 5 seems a bit out of place as it looks more related to Figure 2. It could maybe be integrated there, sent to supplementary or become Figure 3?

This is a different story, about resident Tregs, irrespective of the challenge.

The whole explanation is here in the text:

“Global CDR3α cluster analysis revealed that characteristic eTreg TCR motifs were present in distinct lymphatic tissues, including spleen and thymus, irrespective of the applied challenge (Supplementary Fig. 1). To better illustrate this phenomenon, we performed MDS analysis of CDR3α repertoires for distinct lymphatic tissues, excluding the lungs due to their otherwise dominant response to the current challenge. This analysis demonstrated close proximity of eTreg repertoires obtained from the same lymphatic tissues upon all lung challenges and across all animals (Fig. 5a, b). These results indicate that distinct antigenic specificities are generally characteristic for eTreg cells that preferentially reside in particular lymphatic niches. Notably, the convergence of lymphatic tissue-resident TCR repertoires was less prominent for the eCD4 T cells (Fig. 5c, d).”

And in the abstract:

“Additionally, our TCRα repertoire analysis demonstrated that distinct antigenic specificities are characteristic for eTreg cells residing in particular lymphatic tissues, regardless of the challenge, revealing the homing-specific, antigen-specific resident Treg populations. ”

• Have you explored more systematically the role of individual variability? If you stratify by individual, do you observe any trend? If not this is also an interesting observation to highlight and discuss.

This is inside the calculations and figures/ one dot = 1 mice, so this natural variation is there inside.

• Regarding the MDS plots: why are 2 dimensions the right amount? Maybe with 3, you can see both tissue specificity and stimuli contributions. Can you do a stress vs # dimensions plot to check what should be the right amount of dimensions to more accurately reproduce the distance matrix?

Tissue specificity and stimuli contribution is hard to distinguish without focussing on appropriate samples, as we did on Fig. 3 and 5. The work is already not that simple as is, and attempting to analyse this in multidimensional space is far beyond our current abilities. But this is an interesting point for future work, thank you.

• Figure 2: A better resolution is needed in order to properly resolve the logo plots at the bottom.

Yes, we worked on Figures, and also provide new Supplementary Figure with all the logos.

• No code or data are made available. There is also a lack of supplementary figures that complement and expand the results presented in the main text.

We believe that the main text, although succinct, contains lots of information to analyse and conclusions (preliminary) to make. So we do not see it rational to overload it further.

eLife assessment

This manuscript presents a valuable approach to exploring CD4+ T-cell response in mice across stimuli and tissues through the analysis of their T-cell receptor repertoires. The authors use a transgenic mouse model with reduced diversity of the T-cell receptor repertoire to elicit more consistent T-cell responses across individuals, demonstrating challenge-specific and tissue-specific responses of regulatory T-cells. The evidence for the authors' conclusions is solid, and the work will be of interest to immunologists studying T cell responses.

Reviewer #1 (Public Review):

The authors investigate the alpha chain t cell receptor landscape in conventional vs regulatory CD4 T cells. Overall I think it is a very well thought out and executed study with interesting conclusions. Findings are valuable and are supported by convincing evidence. This work will be of interest for immunologists studying T cells.

Strengths:

- One of a kind evidence and dataset.

- State of the art analyses using well accepted in the literature tools.

- Interesting conclusions on the breadth of immune response to challenges across different types of challenges (tumor, viral and parasitic).

Reviewer #3 (Public Review):

This study presents a valuable exploration of CD4+ T cell response in a fixed TCRβ chain FoxP3-GFP mouse model across stimuli and tissues through the analysis of their TCRα repertoires. This is an insightful paper for the community as it suggests several future directions of exploration.

The authors compare Treg and conventional CD4+ repertoires by looking at diversity measures and the relative overlap of shared clonotypes to characterize similarity across different tissues and antigen challenges. They find distinct yet convergent responses with occasional plasticity across subsets for some stimuli. The observed lack of a general behavior highlights the need for careful comparison of immune repertoires across cell subsets and tissues. Such comparisons are crucial in order to better understand the heterogeneity of the adaptive immune response. This mouse model demonstrates its utility for this task due to the reduced diversity of the TCRα repertoire and the ability to track a single chain.

The revised manuscript has significantly improved in terms of clarity of explanations and presentations of the results.

Author Response

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

Overall Response

We thank the reviewers for reviewing our manuscript, recognizing the significance of our study, and offering valuable suggestions. Based on the reviewer’s comments and the updated eLife assessment, we would like to chose the current version of our manuscript as the Version of Record of our manuscript.

Public Reviews:

Reviewer #1 (Public Review):

Summary:

Given knowledge of the amino acid sequence and of some version of the 3D structure of two monomers that are expected to form a complex, the authors investigate whether it is possible to accurately predict which residues will be in contact in the 3D structure of the expected complex. To this effect, they train a deep learning model which takes as inputs the geometric structures of the individual monomers, per-residue features (PSSMs) extracted from MSAs for each monomer, and rich representations of the amino acid sequences computed with the pre-trained protein language models ESM-1b, MSA Transformer, and ESM-IF. Predicting inter-protein contacts in complexes is an important problem. Multimer variants of AlphaFold, such as AlphaFold-Multimer, are the current state of the art for full protein complex structure prediction, and if the three-dimensional structure of a complex can be accurately predicted then the inter-protein contacts can also be accurately determined. By contrast, the method presented here seeks state-of-the-art performance among models that have been trained end-to-end for inter-protein contact prediction.

Strengths:

The paper is carefully written and the method is very well detailed. The model works both for homodimers and heterodimers. The ablation studies convincingly demonstrate that the chosen model architecture is appropriate for the task. Various comparisons suggest that PLMGraph-Inter performs substantially better, given the same input, than DeepHomo, GLINTER, CDPred, DeepHomo2, and DRN-1D2D_Inter.

The authors control for some degree of redundancy between their training and test sets, both using sequence and structural similarity criteria. This is more careful than can be said of most works in the field of PPI prediction.

As a byproduct of the analysis, a potentially useful heuristic criterion for acceptable contact prediction quality is found by the authors: namely, to have at least 50% precision in the prediction of the top 50 contacts.

We thank the reviewer for recognizing the strengths of our work!

Weaknesses:

The authors check for performance drops when the test set is restricted to pairs of interacting proteins such that the chain pair is not similar as a pair (in sequence or structure) to a pair present in the training set. A more challenging test would be to restrict the test set to pairs of interacting proteins such that none of the chains are separately similar to monomers present in the training set. In the case of structural similarity (TM-scores), this would amount to replacing the two "min"s with "max"s in Eq. (4). In the case of sequence similarity, one would simply require that no monomer in the test set is in any MMSeqs2 cluster observed in the training set. This may be an important check to make, because a protein may interact with several partners, and/or may use the same sites for several distinct interactions, contributing to residual data leakage in the test set.

We thank the reviewer for the suggestion! In the case of protein-protein prediction (“0D prediction”) or protein-protein interfacial residue prediction(“1D prediction”), we think making none of the chains in the test set separately similar to monomers in the training set is necessary, as the reviewer pointed out that a protein may interact with several partners, and may even use the same sites for the interactions. Since the task of this study is predicting the inter-protein residue-residue contacts (“2D prediction”), even though a protein uses the same site to interact with different partners, as long as the interacting partners are different, the inter-protein contact maps would be different. Therefore, we don’t think that in our task, making this restriction to the test set is necessary.

The training set of AFM with v2 weights has a global cutoff of 30 April 2018, while that of PLMGraph-Inter has a cutoff of March 7 2022. So there may be structures in the test set for PLMGraph-Inter that are not in the training set of AFM with v2 weights (released between May 2018 and March 2022). The "Benchmark 2" dataset from the AFM paper may have a few additional structures not in the training or test set for PLMGraph-Inter. I realize there may be only few structures that are in neither training set, but still think that showing the comparison between PLMGraph-Inter and AFM there would be important, even if no statistically significant conclusions can be drawn.

We thank the reviewer for the suggestion! It is not enough to only use the date cutoff to remove the redundancy, since similar structures can be deposited in the PDB in different dates. Because AFM does not release the PDB codes of its training set, it is difficult for us to totally remove the redundancy. Therefore, we think no rigorous conclusion can be drawn by including these comparisons in the manuscript. Besides, the main point of this study is to demonstrate that the integration of multiple protein language models using protein geometric graphs can dramatically improve the model performance for inter-protein contact prediction, which can provide some important enlightenments for the future development of more powerful protein complex structure prediction methods beyond AFM, rather than providing a tool which can beat AFM at this moment. We think including too many stuffs in the comparison with AFM may distract the readers. Therefore, we choose to not include these comparisons in the manuscript.

Finally, the inclusion of AFM confidence scores is very good. A user would likely trust AFM predictions when the confidence score is high, but look for alternative predictions when it is low. The authors' analysis (Figure 6, panels c and d) seems to suggest that, in the case of heterodimers, when AFM has low confidence, PLMGraph-Inter improves precision by (only) about 3% on average. By comparison, the reported gains in the "DockQ-failed" and "precision-failed" bins are based on knowledge of the ground truth final structure, and thus are not actionable in a real use-case.

We agree with the reviewer that more studies are needed for providing a model which can well complement or even beat AFM. The main point of this study is to demonstrate that the integration of multiple protein language models using protein geometric graphs can dramatically improve the model performance for inter-protein contact prediction, which can provide some important enlightenments for the future development of more powerful protein complex structure prediction methods beyond AFM.

Reviewer #2 (Public Review):

This work introduces PLMGraph-Inter, a new deep learning approach for predicting inter-protein contacts, which is crucial for understanding proteinprotein interactions. Despite advancements in this field, especially driven by AlphaFold, prediction accuracy and efficiency in terms of computational cost still remains an area for improvement. PLMGraph-Inter utilizes invariant geometric graphs to integrate the features from multiple protein language models into the structural information of each subunit. When compared against other inter-protein contact prediction methods, PLMGraph-Inter shows better performance which indicates that utilizing both sequence embeddings and structural embeddings is important to achieve high-accuracy predictions with relatively smaller computational costs for the model training.

We thank the reviewer for recognizing the strengths of our work!

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

• I recommend renaming the section "Further potential redundancies removal between the training and the test" to "Further potential redundancies removal between the training and the test sets"

Changed.

• In lines 768-769, the sentence seems to end prematurely in "to use more stringent threshold in the redundancy removal"

Corrected.

• In Eq. (4), line 789, there are many instances of dashes that look like minus signs, creating some confusion.

Corrected.

• I think I may have mixed up figure references in my first review. When I said (Recommendations to the authors): "p. 22, line 2: from the figure, I would have guessed "greater than or equal to 0.7", not 0.8", I think I was referring to what is now lines 423-424, referring to what is now Figure 5c. The point stands there, I think.

Corrected.

• A couple of new grammatical mishaps have been introduced in the revision. These could be rectified.

We carefully rechecked our revisions, and corrected the grammatical issues we found.

Reviewer #2 (Recommendations For The Authors):

Most of my concerns were resolved through the revision. I have only one suggestion for the main figure.

The current scatter plots in Figure 2 are hard to understand as too many different methods are abstracted into a single plot with multiple colors. I would suggest comparing their performances using box plot or violin plot for the figure 2.

We thank the reviewer for the suggestion! In the revision, we tried violin plot, but it does not look good since too many different methods are included in the plot. Besides, we chose the scatter plot as it can provide much more details. We also provided the individual head-to-head scatter plots as supplementary figures, we think which can also be helpful for the readers to capture the information of the figures.

---

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

Overall Response

We would like to thank the reviewers for reviewing our manuscript, recognizing the significance of our study, and offering valuable suggestions. We have carefully revised the manuscript to address all the concerns and suggestions raised by the reviewers.

Public Reviews:

Reviewer #1 (Public Review):

Summary:

Given knowledge of the amino acid sequence and of some version of the 3D structure of two monomers that are expected to form a complex, the authors investigate whether it is possible to accurately predict which residues will be in contact in the 3D structure of the expected complex. To this effect, they train a deep learning model that takes as inputs the geometric structures of the individual monomers, per-residue features (PSSMs) extracted from MSAs for each monomer, and rich representations of the amino acid sequences computed with the pre-trained protein language models ESM-1b, MSA Transformer, and ESM-IF. Predicting inter-protein contacts in complexes is an important problem. Multimer variants of AlphaFold, such as AlphaFold-Multimer, are the current state of the art for full protein complex structure prediction, and if the three-dimensional structure of a complex can be accurately predicted then the inter-protein contacts can also be accurately determined. By contrast, the method presented here seeks state-of-the-art performance among models that have been trained end-to-end for inter-protein contact prediction.

Strengths:

The paper is carefully written and the method is very well detailed. The model works both for homodimers and heterodimers. The ablation studies convincingly demonstrate that the chosen model architecture is appropriate for the task. Various comparisons suggest that PLMGraph-Inter performs substantially better, given the same input than DeepHomo, GLINTER, CDPred, DeepHomo2, and DRN-1D2D_Inter. As a byproduct of the analysis, a potentially useful heuristic criterion for acceptable contact prediction quality is found by the authors: namely, to have at least 50% precision in the prediction of the top 50 contacts.

We thank the reviewer for recognizing the strengths of our work!

Weaknesses:

My biggest issue with this work is the evaluations made using bound monomer structures as inputs, coming from the very complexes to be predicted. Conformational changes in protein-protein association are the key element of the binding mechanism and are challenging to predict. While the GLINTER paper (Xie & Xu, 2022) is guilty of the same sin, the authors of CDPred (Guo et al., 2022) correctly only report test results obtained using predicted unbound tertiary structures as inputs to their model. Test results using experimental monomer structures in bound states can hide important limitations in the model, and thus say very little about the realistic use cases in which only the unbound structures (experimental or predicted) are available. I therefore strongly suggest reducing the importance given to the results obtained using bound structures and emphasizing instead those obtained using predicted monomer structures as inputs.

We thank the reviewer for the suggestion! In the revision, to emphasize the performance of PLMGraph-Inter using the predicted monomer structures, we moved the evaluation results based on the predicted monomer from the supplementary to the main text (see the new Table 1 and Figure 2 in the revised manuscript) and re-organized the two subsections “Evaluation of PLMGraph-Inter on HomoPDB and HeteroPDB test sets” and “Impact of the monomeric structure quality on contact prediction” in the main text.

In particular, the most relevant comparison with AlphaFold-Multimer (AFM) is given in Figure S2, not Figure 6. Unfortunately, it substantially shrinks the proportion of structures for which AFM fails while PLMGraph-Inter performs decently. Still, it would be interesting to investigate why this occurs. One possibility would be that the predicted monomer structures are of bad quality there, and PLMGraph-Inter may be able to rely on a signal from its language model features instead. Finally, AFM multimer confidence values ("iptm + ptm") should be provided, especially in the cases in which AFM struggles.

We thank the reviewer for the suggestion! It is worth noting that AFM automatically searches monomer templates in the prediction, and when we checked our AFM runs, we found that 99% of the targets in our study (including all the targets in the four datasets: HomoPDB, HeteroPDB, DHTest and DB5.5) at least 20 templates were identified (AFM employed the top 20 templates in the prediction), and 87.8% of the targets employed the native templates (line 455-462 in page 25 in the subsection of “Comparison of PLMGraph-Inter with AlphaFold-Multimer”). Therefore, we think Figure 6 not Figure S5 (the original Figure S2) shows a fairer comparison. Besides, it is also worth noting the targets used in this study would have a large overlap with the training set of AlphaFold-Multimer, since AFM used all protein complex structures in PDB deposited before 2018-04-30 in the model training, which would further cause the overestimation of the performance of AFM (line 450-455 in page 24-25 in the subsection of “Comparison of PLMGraph-Inter with AlphaFold-Multimer”).

To mimic the performance of AlphaFold2 in real practice and produce predicted monomeric structures with more diverse qualities, we only used the MSA searched from Uniref100 protein sequence database as the input to AlphaFold2 and set to not use the template (line 203~210 in page 12 in the subsection of “Evaluation of PLMGraph-Inter on HomoPDB and HeteroPDB test sets”). Since some of the predicted monomer structures are of bad quality, it is reasonable that the performance of PLMGraph-Inter drops when the predicted monomeric structures are used in the prediction. We provided a detailed analysis of the impact of the monomeric structure quality on the prediction performance in the subsection “Impact of the monomeric structure quality on contact prediction” in the main text.

We provided the analysis of the AFM multimer confidence values (“iptm + ptm”) in the revision (Figure 6, Figure S5 and line 495-501 in page 27 in the subsection of “Comparison of PLMGraph-Inter with AlphaFold-Multimer”).

Besides, in cases where any experimental structures - bound or unbound - are available and given to PLMGraph-Inter as inputs, they should also be provided to AlphaFold-Multimer (AFM) as templates. Withholding these from AFM only makes the comparison artificially unfair. Hence, a new test should be run using AFM templates, and a new version of Figure 6 should be produced. Additionally, AFM's mean precision, at least for top-50 contact prediction, should be reported so it can be compared with PLMGraph-Inter's.

We thank the reviewers for the suggestion, and we are sorry for the confusion! In the AFM runs to predict protein complex structures, we used the default setting of AFM which automatically searches monomer templates in the prediction. When we checked our AFM runs, we found that 99% of the targets in our study (including all the targets in the four datasets: HomoPDB, HeteroPDB, DHTest and DB5.5) employed at least 20 templates in their predictions (AFM only used the top 20 templates), and 87.8% of the targets employed the native template. We further clarified this in the revision (line 455462 in page 25 in the subsection of “Comparison of PLMGraph-Inter with AlphaFoldMultimer”). We also included the mean precisions of AFM (top-50 contact prediction) in the revision (Table S5 and line 483-484 in page 26 in the subsection of “Comparison of PLMGraph-Inter with AlphaFold-Multimer”).

It's a shame that many of the structures used in the comparison with AFM are actually in the AFM v2 training set. If there are any outside the AFM v2 training set and, ideally, not sequence- or structure-homologous to anything in the AFM v2 training set, they should be discussed and reported on separately. In addition, why not test on structures from the "Benchmark 2" or "Recent-PDB-Multimers" datasets used in the AFM paper?

We thank the reviewer for the suggestion! The biggest challenge to objectively evaluate AFM is that as far as we known, AFM does not release the PDB ids of its training set and the “Recent-PDB-Multimers” dataset. “Benchmark 2” only includes 17 heterodimer proteins, and the number would be further decreased after removing targets redundant to our training set. We think it is difficult to draw conclusions from such a small number of targets.

It is also worth noting that the AFM v2 weights have now been outdated for a while, and better v3 weights now exist, with a training cutoff of 2021-09-30.

Author response image 1.

The head-to-head comparison of qualities of complex predicted by AlphaFold-Multimer (2.2.0) and AlphaFold-Multimer (2.3.2) for each target PPI.

We thank the reviewer for reminding the new version of AFM. The only difference between AFM V3 and V2 is the cutoff date of the training set. During the revision, we also tested the new version of AFM on the datasets of HomoPDB and HeteroPDB, but we found the performance difference between the two versions of AFM is actually very little (see the figure above, not shown in the main text). One reason might be that some targets in HomoPDB and HeteroPDB are redundant with the training sets of the two version of AFM. Since our test sets would have more overlaps with the training set of AFM V3, we keep using the AFM V2 weights in this study.

Another weakness in the evaluation framework: because PLMGraph-Inter uses structural inputs, it is not sufficient to make its test set non-redundant in sequence to its training set. It must also be non-redundant in structure. The Benchmark 2 dataset mentioned above is an example of a test set constructed by removing structures with homologous templates in the AF2 training set. Something similar should be done here.

We thank the reviewer for the suggestion! In the revision, we explored the performance of PLMGraph-Inter when using different thresholds of fold similarity scores of interacting monomers to further remove potential redundancies between the training and test sets (i.e. redundancy in structure ) (line 353-386 in page 19-21 in the subsection “Ablation study”; line 762-797 in page 41-43 in the subsection “Further potential redundancies removal between the training and the test”). We found that for heteromeric PPIs (targets in HeteroPDB), the further removal of potential redundancy in structure has little impact on the model performance (~3%, when TM-score 0.5 is used as the threshold). However, for homomeric PPIs (targets in HomoPDB), the further removal of potential redundancy in structure significantly reduce the model performance (~18%, when TM-score 0.5 is used as the threshold) (see Table 2). One possible reason for this phenomenon is that the binding mode of the homomeric PPI is largely determined by the fold of its monomer, thus the does not generalize well on targets whose folds have never been seen during the training.

Whether the deep learning model can generalize well on targets with novel folds is a very interesting and important question. We thank the reviewer for pointing out this! However, to the best of our knowledge, this question has rarely been addressed by previous studies including AFM. For example, the Benchmark 2 dataset is prepared by ClusPro TBM (bioRxiv 2021.09.07.459290; Proteins 2020, 88:1082-1090) which uses a sequence-based approach (HHsearch) to identify templates not structure-based. Therefore, we don’t think this dataset is non-redundant in structure.

Finally, the performance of DRN-1D2D for top-50 precision reported in Table 1 suggests to me that, in an ablation study, language model features alone would yield better performance than geometric features alone. So, I am puzzled why model "a" in the ablation is a "geometry-only" model and not a "LM-only" one.

Using the protein geometric graph to integrate multiple protein language models is the main idea of PLMGraph-Inter. Comparing with our previous work (DRN-1D2D_Inter), we consider the building of the geometric graph as one major contribution of this work. To emphasize the efficacy of this geometric graph, we chose to use the “geometry-only” model as the base model.

Reviewer #1 (Recommendations For The Authors):

Some sections of the paper use technical terminology which limits accessibility to a broad audience. An obvious example is in the section "Results > Overview of PLMGraph-Inter > The residual network module": the average eLife reader is not a machine learning expert and might not be familiar with a "convolution with kernel size of 1 * 1". In general, the "Overview of PLMGraph-Inter" is a bit heavy with technical details, and I suggest moving many of these to Methods. This overview section can still be there but it should be shorter and written using less technical language.

We thank the reviewer for the suggestion! We moved some technical details to the Methods section in the revision (line 184-185 in page 11; line 729-735 in page 39).

List of typos and minor issues (page number according to merged PDF):

• p. 3. line -3: remove "to"

Corrected (line 36, page 3)

• p. 5, line 7: "GINTER" should be "GLINTER"

Corrected (line 64, page 5)

• p. 6, line -4: "Given structures" -> "Given the structures"

Corrected (line 95, page 6)

• p. 6, line -2: "with which encoded"... ?

We rephrased this sentence in revision. (line 97, page 6)

• p. 9, line 1: "principal" -> "principle"

Corrected (line 142, page 9)

• p. 13, line 1: "has" -> "but have"

Corrected (line 231, page 13)

• p. 14, lines 6-7: "As can be seen from the figure that the predicted" -> "As can be seen from the figure, the predicted"

We rephrased this paragraph, and the sentence was deleted in the revision (line 257-259 in page 15).

• p. 18, line 1: the "five models" are presumably models a-e? If so, say "of models a-e"

Corrected (line 310, page 17)

• p. 22, line 2: from the figure, I would have guessed "greater than or equal to 0.7", not 0.8

Based the Figure 3C, we think 0.8 is a more appropriate cutoff, since the precision drops significantly when the DTM-score is within 0.7~0.8.

• p. 23, lines 2-3: "worth to making" -> "worth making"

Corrected (line 443, page 24)

• p. 24, line -5: "predict" -> "predicted"

Corrected (line 484, page 26)

• p 28, line -5: Please clarify what you mean by "We doubt": are you saying that you don't think these rearrangements exist in nature? If not, then reword.

Corrected (line 566, page 30)

• Figure 2, panel c, "DCPred" in the legend should be "CDPred"

Corrected

• Figures 3 and 5: Please improve the y-axis title in panel C. "Percent" of what?

We changed the “Percent” to “% of targets” in the revision.

We thank the reviewer for carefully reading our manuscript!

Reviewer #2 (Public Review):

This work introduces PLMGraph-Inter, a new deep-learning approach for predicting inter-protein contacts, which is crucial for understanding proteinprotein interactions. Despite advancements in this field, especially driven by AlphaFold, prediction accuracy and efficiency in terms of computational cost) still remains an area for improvement. PLMGraph-Inter utilizes invariant geometric graphs to integrate the features from multiple protein language models into the structural information of each subunit. When compared against other inter-protein contact prediction methods, PLMGraph-Inter shows better performance which indicates that utilizing both sequence embeddings and structural embeddings is important to achieve high-accuracy predictions with relatively smaller computational costs for the model training.

The conclusions of this paper are mostly well supported by data, but test examples should be revisited with a more strict sequence identity cutoff to avoid any potential information leakage from the training data. The main figures should be improved to make them easier to understand.

We thank the reviewer for recognizing the significance of our work! We have carefully revised the manuscript to address the reviewer’s concerns.

(1) The sequence identity cutoff to remove redundancies between training and test set was set to 40%, which is a bit high to remove test examples having homology to training examples. For example, CDPred uses a sequence identity cutoff of 30% to strictly remove redundancies between training and test set examples. To make their results more solid, the authors should have curated test examples with lower sequence identity cutoffs, or have provided the performance changes against sequence identities to the closest training examples.

We thank the reviewer for the valuable suggestion! The “40 sequence identity” is a widely used threshold to remove redundancy when evaluating deep-learning based protein-protein interaction and protein complex structure prediction methods, thus we also chose this threshold in our study (bioRxiv 2021.10.04.463034, Cell Syst. 2021 Oct 20;12(10):969-982.e6). In the revision, we explored whether PLMGraph-inter can keep its performance when more stringent thresholds (30%,20%,10%) is applied (line 353386 in page 20-21 in the subsection of “Ablation study” and line 762-780 in page 40 in the subsection of “Further potential redundancies removal between the training and the test”). The result shows that even when using “10% sequence identity” as the threshold, mean precisions of the predicted contacts only decreases by ~3% (Table 2).

(2) Figures with head-to-head comparison scatter plots are hard to understand as scatter plots because too many different methods are abstracted into a single plot with multiple colors. It would be better to provide individual head-tohead scatter plots as supplementary figures, not in the main figure.

We thank the reviewer for the suggestion! We will include the individual head-to-head scatter plots as supplementary figures in the revision (Figure S1 and Figure S2 in the supplementary).

(3) The authors claim that PLMGraph-Inter is complementary to AlphaFoldmultimer as it shows better precision for the cases where AlphaFold-multimer fails. To strengthen the point, the qualities of predicted complex structures via protein-protein docking with predicted contacts as restraints should have been compared to those of AlphaFold-multimer structures.

We thank the reviewer for the suggestion! We included this comparison in the revision (Figure S7).

(4) It would be interesting to further analyze whether there is a difference in prediction performance depending on the depth of multiple sequence alignment or the type of complex (antigen-antibody, enzyme-substrates, single species PPI, multiple species PPI, etc).

We thank the reviewer for the suggestion! We analyzed the relationship between the prediction performance and the depth of MSA in the revision (Figure S4 and Line 253264 in page 15 in the subsection of “Evaluation of PLMGraph-Inter on HomoPDB and HeteroPDB test sets” and line 798-806 in page 42 in the subsection of “Calculating the normalized number of the effective sequences of paired MSA”).

Reviewer #2 (Recommendations For The Authors):

I have the following suggestions in addition to the public review.

(1) Overall, the manuscript is well-written; however, I recommend a careful review for minor grammar corrections to polish the final text.

We carefully checked the manuscript and corrected all the grammar issues and typos we found in the revision.

(2) It would be better to indicate that single sequence embeddings, MSA embeddings, and structure embeddings are ESM-1b, ESM-MSA & PSSM, and ESM-IF when they are first mentioned in the manuscript e.g. single sequence embeddings from ESM-1b, MSA embeddings from ESM-MSA and PSSM, and structural embeddings from ESM-IF.

We revised the manuscript according to the reviewer’s suggestion (line 86-88 in page 6; line 99-101 in page 7).

(3) I don't think "outer concatenation" is commonly used. Please specify whether it's outer sum, outer product, or horizontal & vertical tiling followed by concatenation.

It is horizontal & vertical tiling followed by concatenation. We clarified this in the revision (line 129-130 in page 8).

(4) 10th sentence on the page where the Results section starts, please briefly mention what are the other 2D pairwise features.

We clarified this in the revision (line 131-132 in page 8).

(5) In the result section, it states edges are defined based on Ca distances, but in the method section, it says edges are determined based on heavy atom distances. Please correct one of them.

It should be Ca distances. We are sorry for the carelessness, and we corrected this in the revision (line 646 in page 35).

(6) For the sentence, "Where ESM-1b and ESM-MSA-1b are pretrained PLMs learned from large datasets of sequences and MSAs respectively without label supervision,", I'd suggest replacing "without label supervision" with "with masked language modeling tasks" for clarity.

We revised the manuscript according to the reviewer’s suggestion (line 150-151 in page 9).

(7) It would be better to briefly explain what is the dimensional hybrid residual block when it first mentioned.

We explained the dimensional hybrid residue block when it first mentioned in the revision (line 107 in page 7).

(8) Please include error bars for the bar plots and standard deviations for the tables.

We thank the reviewer for the suggestion! Our understanding is the error bars and standard deviations are very informative for data which follow gaussian-like distributions, but our data (precisions of the predicted contacts) are obviously not this type. Most previous studies in protein contact prediction and inter-protein contact prediction also did not include these in their plots or tables. In our case, including these elements requires a dramatic change of the styles of our figures and tables, but we would like to not change our figures and tables too much in the revision.

(9) Please indicate whether the chain break is considered to generate attention map features from ESM-MSA-1b. If it's considered, please specify how.

The paired sequences were directly concatenated without using any letter to connect them, which means we did not consider chain break in generating the attention maps from ESM-MSA-1b.

Reviewer #1 (Public Review):

Summary:

Given knowledge of the amino acid sequence and of some version of the 3D structure of two monomers that are expected to form a complex, the authors investigate whether it is possible to accurately predict which residues will be in contact in the 3D structure of the expected complex. To this effect, they train a deep learning model which takes as inputs the geometric structures of the individual monomers, per-residue features (PSSMs) extracted from MSAs for each monomer, and rich representations of the amino acid sequences computed with the pre-trained protein language models ESM-1b, MSA Transformer, and ESM-IF. Predicting inter-protein contacts in complexes is an important problem. Multimer variants of AlphaFold, such as AlphaFold-Multimer, are the current state of the art for full protein complex structure prediction, and if the three-dimensional structure of a complex can be accurately predicted then the inter-protein contacts can also be accurately determined. By contrast, the method presented here seeks state-of-the-art performance among models that have been trained end-to-end for inter-protein contact prediction.

Strengths:

The paper is carefully written and the method is very well detailed. The model works both for homodimers and heterodimers. The ablation studies convincingly demonstrate that the chosen model architecture is appropriate for the task. Various comparisons suggest that PLMGraph-Inter performs substantially better, given the same input, than DeepHomo, GLINTER, CDPred, DeepHomo2, and DRN-1D2D_Inter.<br /> The authors control for some degree of redundancy between their training and test sets, both using sequence and structural similarity criteria. This is more careful than can be said of most works in the field of PPI prediction.<br /> As a byproduct of the analysis, a potentially useful heuristic criterion for acceptable contact prediction quality is found by the authors: namely, to have at least 50% precision in the prediction of the top 50 contacts.

Weaknesses:

The authors check for performance drops when the test set is restricted to pairs of interacting proteins such that the chain pair is not similar *as a pair* (in sequence or structure) to a pair present in the training set. A more challenging test would be to restrict the test set to pairs of interacting proteins such that *none* of the chains are separately similar to monomers present in the training set. In the case of structural similarity (TM-scores), this would amount to replacing the two "min"s with "max"s in Eq. (4). In the case of sequence similarity, one would simply require that no monomer in the test set is in any MMSeqs2 cluster observed in the training set. This may be an important check to make, because a protein may interact with several partners, and/or may use the same sites for several distinct interactions, contributing to residual data leakage in the test set.

The training set of AFM with v2 weights has a global cutoff of 30 April 2018, while that of PLMGraph-Inter has a cutoff of March 7 2022. So there may be structures in the test set for PLMGraph-Inter that are not in the training set of AFM with v2 weights (released between May 2018 and March 2022). The "Benchmark 2" dataset from the AFM paper may have a few additional structures not in the training or test set for PLMGraph-Inter. I realize there may be only few structures that are in neither training set, but still think that showing the comparison between PLMGraph-Inter and AFM there would be important, even if no statistically significant conclusions can be drawn.

Finally, the inclusion of AFM confidence scores is very good. A user would likely trust AFM predictions when the confidence score is high, but look for alternative predictions when it is low. The authors' analysis (Figure 6, panels c and d) seems to suggest that, in the case of heterodimers, when AFM has low confidence, PLMGraph-Inter improves precision by (only) about 3% on average. By comparison, the reported gains in the "DockQ-failed" and "precision-failed" bins are based on knowledge of the ground truth final structure, and thus are not actionable in a real use-case.

Reviewer #2 (Public Review):

This work introduces PLMGraph-Inter, a new deep learning approach for predicting inter-protein contacts, which is crucial for understanding protein-protein interactions. Despite advancements in this field, especially driven by AlphaFold, prediction accuracy and efficiency in terms of computational cost still remains an area for improvement. PLMGraph-Inter utilizes invariant geometric graphs to integrate the features from multiple protein language models into the structural information of each subunit. When compared against other inter-protein contact prediction methods, PLMGraph-Inter shows better performance which indicates that utilizing both sequence embeddings and structural embeddings is important to achieve high-accuracy predictions with relatively smaller computational costs for the model training.

Author Response

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

Reviewer #1 (Public Review):

The manuscript investigates the role of membrane contact sites (MCSs) and sphingolipid metabolism in regulating vacuolar morphology in the yeast Saccharomyces cerevisiae. The authors show that tricalbin (1-3) deletion leads to vacuolar fragmentation and the accumulation of the sphingolipid phytosphingosine (PHS). They propose that PHS triggers vacuole division through MCSs and the nuclear-vacuolar junction (NVJ). The study presents some solid data and proposes potential mechanisms underlying vacuolar fragmentation driven by this pathway. Although the manuscript is clear in what the data indicates and what is more hypothetical, the story would benefit from providing more conclusive evidence to support these hypothesis. Overall, the study provides valuable insights into the connection between MCSs, lipid metabolism, and vacuole dynamics.

We thank the positive review from the Reviewer #1. We hope that our hypotheses are supported by the "Author Response to Recommendations" and by further research in the future.

Reviewer #2 (Public Review):

This manuscript explores the mechanism underlying the accumulation of phytosphingosine (PHS) and its role in initiating vacuole fission. The study posits the involvement of membrane contact sites (MCSs) in two key stages of this process. Firstly, MCSs tethered by tricalbin between the endoplasmic reticulum (ER) and the plasma membrane (PM) or Golgi regulate the intracellular levels of PHS. Secondly, the amassed PHS triggers vacuole fission, most likely through the nuclear-vacuolar junction (NVJ). The authors propose that MCSs play a regulatory role in vacuole morphology via sphingolipid metabolism. While some results in the manuscript are intriguing, certain broad conclusions occasionally surpass the available data. Despite the authors' efforts to enhance the manuscript, certain aspects remain unclear. It is still uncertain whether subtle changes in PHS levels could induce such effects on vacuolar fission. Additionally, it is regrettable that the lipid measurements are not comparable with previous studies by the authors. Future advancements in methods for determining intracellular lipid transport and levels are anticipated to shed light on the remaining uncertainties in this study.

We thank the careful comment from Reviewer #2. As Reviewer #2 pointed out, the mechanism of how slight changes in PHS levels can induce the vacuolar fission event is still uncovered in this manuscript. We sincerely consider that this issue has to be resolved in further study.

Reviewer #3 (Public Review):

In this manuscript, the authors investigated the effects of deletion of the ER-plasma membrane/Golgi tethering proteins tricalbins (Tcb1-3) on vacuolar morphology to demonstrate the role of membrane contact sites (MCSs) in regulating vacuolar morphology in Saccharomyces cerevisiae. Their data show that tricalbin deletion causes vacuolar fragmentation possibly in parallel with TORC1 pathway. In addition, their data reveal that levels of various lipids including ceramides, long-chain base (LCB)-1P, and phytosphingosine (PHS) are increased in tricalbin-deleted cells. The authors find that exogenously added PHS can induce vacuole fragmentation and by performing analyses of genes involved in sphingolipid metabolism, they conclude that vacuolar fragmentation in tricalbin-deleted cells is due to the accumulated PHS in these cells. Importantly, exogenous PHS- or tricalbin deletion-induced vacuole fragmentation was suppressed by loss of the nucleus vacuole junction (NVJ), suggesting the possibility that PHS transported from the ER to vacuoles via the NVJ triggers vacuole fission. Of note, the authors find that hyperosmotic shock increases intracellular PHS levels, suggesting a general role of PHS in vacuole fission in response to physiological vacuolar division-inducing stimuli. This work provides valuable insights into the relationship between MCS-mediated sphingolipid metabolism and vacuole morphology. The conclusions of this paper are mostly supported by their results, but inclusion of direct evidence indicating increased transport of PHS from the ER to vacuoles via NVJ in response to vacuolar division-inducing stimuli would have strengthened this study. There is another weakness in their claim that the transmembrane domain of Tcb3 contributes to the formation of the tricalbin complex which is sufficient for tethering ER to the plasma membrane and the Golgi complex. Their claim is based only on the structural simulation, but not on by biochemical experiments such as co-immunoprecipitation and pull-down.

We appreciate the careful feedback from Reviewer #3. We have responded in the "Recommendations to Authors" section and hope it can partially support the weakness in our claim regarding the physical interaction between Tcb1, 2, and 3.

Reviewer #1 (Recommendations For The Authors):

I would suggest that the authors include some of the data (e.g., Tcb interactions) that they refer to in the response to the reviewers. I think that this could enhance the message in this manuscript. Also, maybe it's a typo and you were referring to some other image panel, but in the rebuttal letter a "Fig. S3B" is mentioned, but I could not find it.

Following the suggestions of reviewers #1 and #3, we have added the data of co-immunoprecipitation which confirmed that Tcb3 binds to both Tcb1 and Tcb2 as Supplemental Figure 2. With this change, the person (Ms. Saku Sasaki) who performed this analysis was also added as a co-author.

Also, we appreciate the careful remark and apologize for the mistake. In the previous Author's response, we mentioned the vacuole observation using SD medium, but this data was Fig 5C, not Fig S3B.

Reviewer #3 (Recommendations For The Authors):

I would recommend that the authors include the IP data mentioned in their rebuttal letter to show the interactions among Tcb1-3. Also, the authors should quantify all lipid species in Fig 5B, as shown in Fig 3A.

Following the suggestions of reviewers #1 and #3, we have added the co-immunoprecipitation data (Fig S2). In a further study, we would like to test if the transmembrane domain of Tcb3 is sufficient for the interaction among Tcb1-3. Also, we quantified all lipid species and replaced the data in Fig 5B.

Minor points:

(1) The function of vps4 is not mentioned in the manuscript.

(2) The function of Sur2p is not mentioned in the manuscript. It should be clearly mentioned that DHS is converted to PHS by Sur2p.

(1) We have added text sections which mention that VPS4 is needed for normal ESCRT function, and its deletion is an example for inhibition of GFP-Cps1p transport into the vacuole.

(2) We have added the text in the manuscript that states Sur2p is the hydroxylase that catalysis the conversion of DHS to PHS.

eLife assessment

This manuscript presents valuable findings that contribute to our understanding of how sphingolipids and membrane contact sites, formed by the tethering protein family tricalbins, are involved in regulating vacuolar morphology in S. cerevisiae. The evidence supporting the authors' claims is largely solid: while the reported correlation between sphingolipid levels and vacuole homeostasis is intriguing, the data do not completely substantiate the proposed mechanism. This study will be of interest to cell biologists focusing on intracellular organization and lipid metabolism.

eLife assessment

This manuscript presents valuable findings that contribute to our understanding of how sphingolipids and membrane contact sites, formed by the tethering protein family tricalbins, are involved in regulating vacuolar morphology in S. cerevisiae. The evidence supporting the authors' claims is largely solid. While the reported correlation between sphingolipid levels and vacuole homeostasis is interesting and intriguing, more work is needed to thoroughly substantiate the proposed mechanism. This study will be of interest to cell biologists focusing on intracellular organization and lipid metabolism.

Reviewer #1 (Public Review):

The manuscript investigates the role of membrane contact sites (MCSs) and sphingolipid metabolism in regulating vacuolar morphology in the yeast Saccharomyces cerevisiae. The authors show that tricalbin (1-3) deletion leads to vacuolar fragmentation and the accumulation of the sphingolipid phytosphingosine (PHS). They propose that PHS triggers vacuole division through MCSs and the nuclear-vacuolar junction (NVJ). The study presents some solid data and proposes potential mechanisms underlying vacuolar fragmentation driven by this pathway. Although the manuscript is clear in what the data indicates and what is more hypothetical, the story would benefit from providing more conclusive evidence to support these hypothesis. Overall, the study provides valuable insights into the connection between MCSs, lipid metabolism, and vacuole dynamics.

Reviewer #2 (Public Review):

This manuscript explores the mechanism underlying the accumulation of phytosphingosine (PHS) and its role in initiating vacuole fission. The study posits the involvement of membrane contact sites (MCSs) in two key stages of this process. Firstly, MCSs tethered by tricalbin between the endoplasmic reticulum (ER) and the plasma membrane (PM) or Golgi regulate the intracellular levels of PHS. Secondly, the amassed PHS triggers vacuole fission, most likely through the nuclear-vacuolar junction (NVJ). The authors propose that MCSs play a regulatory role in vacuole morphology via sphingolipid metabolism.

While some results in the manuscript are intriguing, certain broad conclusions occasionally surpass the available data. Despite the authors' efforts to enhance the manuscript, certain aspects remain unclear. It is still uncertain whether subtle changes in PHS levels could induce such effects on vacuolar fission. Additionally, it is regrettable that the lipid measurements are not comparable with previous studies by the authors. Future advancements in methods for determining intracellular lipid transport and levels are anticipated to shed light on the remaining uncertainties in this study.

Reviewer #3 (Public Review):

In this manuscript, the authors investigated the effects of deletion of the ER-plasma membrane/Golgi tethering proteins tricalbins (Tcb1-3) on vacuolar morphology to demonstrate the role of membrane contact sites (MCSs) in regulating vacuolar morphology in Saccharomyces cerevisiae. Their data show that tricalbin deletion causes vacuolar fragmentation possibly in parallel with TORC1 pathway. In addition, their data reveal that levels of various lipids including ceramides, long-chain base (LCB)-1P, and phytosphingosine (PHS) are increased in tricalbin-deleted cells. The authors find that exogenously added PHS can induce vacuole fragmentation and by performing analyses of genes involved in sphingolipid metabolism, they conclude that vacuolar fragmentation in tricalbin-deleted cells is due to the accumulated PHS in these cells. Importantly, exogenous PHS- or tricalbin deletion-induced vacuole fragmentation was suppressed by loss of the nucleus vacuole junction (NVJ), suggesting the possibility that PHS transported from the ER to vacuoles via the NVJ triggers vacuole fission. Of note, the authors find that hyperosmotic shock increases intracellular PHS levels, suggesting a general role of PHS in vacuole fission in response to physiological vacuolar division-inducing stimuli.

This work provides valuable insights into the relationship between MCS-mediated sphingolipid metabolism and vacuole morphology. The conclusions of this paper are mostly supported by their results, but inclusion of direct evidence indicating increased transport of PHS from the ER to vacuoles via NVJ in response to vacuolar division-inducing stimuli would have strengthened this study.

There is another weakness in their claim that the transmembrane domain of Tcb3 contributes to the formation of the tricalbin complex which is sufficient for tethering ER to the plasma membrane and the Golgi complex. Their claim is based only on the structural simulation, but not on by biochemical experiments such as co-immunoprecipitation and pull-down.

Author Response

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

Overall, the magnitude of the effect size due to FNDC5 deficiency in both male and female mice is rather modest. Looking at the data from a qualitative perspective, it is clear that knockout females still lose bone during lactation and on the low calcium diet (LCD). It is difficult to assess the physiologic consequence of the modest quantitative 'protection' seen in FNDC5 mutants since the mutants still show clear and robust effects of lactation and LCD on all parameters measured. Similarly, the magnitude of the 'increased' cortical bone loss in FNDC5 mutant males is also modest and perhaps could be related to the fact that these mice are starting with slightly more cortical bone. Since the authors do not provide a convincing molecular explanation for why FNDC5 deficiency causes these somewhat subtle changes, I would like to offer a suggestion for the authors to consider (below, point #2) which might de-emphasize the focus of the manuscript on FNDC5. If the authors chose not to follow this suggestion, the manuscript could be strengthened by addressing the consequences of the modest changes observed in WT versus FNDC5 KO mice.

Response: We agree that the magnitude of the effect size due to FNDC5 deficiency is modest with regards to the quantitative cortical bone parameters. However, if one examines the changes in osteocyte lacunar size and the mechanical properties of these bones, the differences are greater. As shown in Figure 3 E, the lacunar area of the WT females on a low calcium diet increases by over 30% and the KO by less than 20%, while in the males it is approximately 38% in WT compared to 46% in KO mice. According to Sims and Buenzli (PMID: 25708054) a potential total loss of ~16,000 mm3 (16 mL) of bone occurs through lactation in the human skeleton. This was based on our measurements in lactation-induced murine osteocytic osteolysis (Qing et al PMID: 22308018). They used our 2D section of tibiae from lactating mice showing an increase in lacunar size from 38 to 46 um2. In that paper we also showed that canalicular width is increased with lactation. Therefore, this would suggest a dramatic decrease in intracortical porosity due to the osteocyte lacunocanalicular system in female KO on a low calcium diet compared to WT females and a dramatic increase in KO males compared to WT males. Also, PTH was higher in the serum of female WT compared to female KO mice on a low calcium diet, the opposite for males in order to maintain normal calcium levels (See Table 1). Based on this data, using the FNDC5 null animals, we would speculate that the product of FNDC5, irisin, is having a highly significant effect on the ultrastructure of bone in both males and females challenged with a low calcium diet.

(2) The bone RNA-seq findings reported in Figures 4-6 are quite interesting. Although Youlten et al previously reported that the osteocyte transcriptome is sex-dependent, the work here certainly advances that notion to a considerable degree and likely will be of high interest to investigators studying skeletal biology and sexual dimorphism in general. To this end, one direction for the authors to consider might be to refocus their manuscript toward sexually-dimorphic gene expression patterns in osteocytes and the different effects of LCD on male versus female mice. This would allow the authors to better emphasize these major findings, and to then use FNDC5 deficiency as an illustrative example of how sexually-dimorphic osteocytic gene expression patterns might be affected by deletion of an osteocyte-acting endocrine factor. Ideally, the authors would confirm RNA-seq data comparing male versus female mice in osteocytes using in situ hybridization or immunostaining.

Response: Thank you for this suggestion. We have compared the different effects of LCD on male versus female mice in our revised version and have added a figure containing this information.

(3) Along the lines of point #2 (above), the presentation of the RNA-seq studies in Figures 4-6 is somewhat confusing in that the volcano plot titles seem to be reversed. For example, Figure 4A is titled "WT M: WT F", but the genes in the upper right quadrant appear to be up-regulated in female cortical bone RNA samples. Should this plot instead be titled "WT F: WT M"? If so, then all other volcano plots should be re-titled as well.

Response: We have now insured that the plots are appropriately labeled.

(4) Have the authors compared male versus female transcriptomes of LCD mice?

Response: We have now compared the male vs female transcriptomes of LCD mice and added an additional figure.

(5) It would be appreciated if the authors could provide additional serum parameters (if possible) to clarify incomplete data in both lactation and low-calcium diet models: RANKL/OPG ratio, Ctx, PTHrP, and 1,25-dihydroxyvitamin D levels.

Response: It is not possible to quantitate each of these as the serum has been exhausted. We have checked the RANKL/OPG ratio in the RNA seq and qPCR data using osteocyte enriched bone chips and found no difference.

(6) Lastly, the data that overexpressing irisin improved bone properties in Fig 2G was somewhat confusing. Based on Kim et al.'s (2018) work, irisin injection increased sclerostin gene expression and serum levels, thus reducing bone formation. Were sclerostin levels affected by irisin overexpression in this study? Was irisin's role in modulating sclerostin levels attenuated with additional calcium deficiency?

Response: We have not observed any differences in the osteocyte Sost mRNA expression between WT and KO normal and low-calcium-diet male and female mice in our RNAseq and qPCR data. As such, we did not check the Sost levels for the 2G experiment.

Reviewer #2 (Public Review):

Summary:

The goal of this study was to examine the role of FNDC5 in the response of the murine skeleton to either lactation or a calcium-deficient diet. The authors find that female FNDC5 KO mice are somewhat protected from bone loss and osteocyte lacunar enlargement caused by either lactation or a calcium-deficient diet. In contrast, male FNDC5 KO mice lose more bone and have a greater enlargement of osteocyte lacunae than their wild-type controls. Based on these results, the authors conclude that in males irisin protects bone from calcium deficiency but that in females it promotes calcium removal from bone for lactation.

While some of the conclusions of this study are supported by the results, it is not clear that the modest effects of FNDC5 deletion have an impact on calcium homeostasis or milk production.

Specific comments:

(1) The authors sometimes refer to FNDC5 and other times to irisin when describing causes for a particular outcome. Because irisin was not measured in any of the experiments, the authors should not conclude that lack of irisin is responsible. Along these lines, is there any evidence that either lactation or a calcium-deficient diet increases the production of irisin in mice?

therefore we have extrapolated that the observed effects are due to a lack of circulating irisin. However, this does not rule out that Fndc5 itself could have a function, but this would have to be most likely in muscle and not in the osteocyte as we do not detect significant levels of irisin in either primary osteoblasts nor primary osteocytes compared to muscle and C2C12 cells. As such, we concluded that the phenotypical differences we saw in our experiments are due to a lack of irisin. We now address the reviewer’s point in the discussion. The measurement of irisin in the circulation with lactation or with low calcium diet of normal mice has not been performed.

(2) The results of the irisin-rescue experiment shown in figure 2G cannot be appropriately interpreted without normal diet controls. In addition, some evidence that the AAV8-irisin virus actually increased irisin levels in the mice would strengthen the conclusion.

Response: We do not have the normal diet controls at this time. We have quantitate tagged irisin in other AAV experiments and found highly significant expression

(3) There is insufficient evidence to support the idea that the effect of FNDC5 on bone resorption and osteocytic osteolysis is important for the transfer of calcium from bone to milk. Previous studies by others have shown that bone resorption is not required to maintain milk or serum calcium when dietary calcium is sufficient but is critical if dietary calcium is low (Endo. 156:2762-73, 2015). To support the conclusions of the current study, it would be necessary to determine whether FNDC5 is required to maintain calcium levels when lactating mice lack sufficient dietary calcium.

Response: We agree that it would be important to measure calcium levels in the milk to test the hypothesis that FNDC5 is important to maintain calcium levels in milk. However, as the calcium levels are normal in the serum, we are assuming they are normal in milk. This would require future experiments.

(4) The amount of cortical bone loss due to lactation is very similar in both WT and FNDC5 KO mice. The results of the statistical analysis of the data presented in figure 1B are surprising given the very similar effect size of lactation. The key result from the 2-way ANOVA is whether there is an effect of genotype on the effect size of lactation (genotype-lactation interaction). The interaction terms were not provided. Similar concerns are noted for the results shown in figure 1G and H.

Response: We agree, thanks. We will now add the interaction terms in the figure legends.

(5) It is not clear what justifies the term 'primed' or 'activated' for resorption. Is there evidence that a certain level of TRAP expression lowers the threshold for osteocytic osteolysis in response to a stimulus?

Response: The number of TRAP positive osteocytes in female KO mice are lower than in female WT. The number of TRAP positive osteocytes are lower in WT males compared to WT females. We propose that irisin plays a role in the number of TRAP positive osteocytes in normal, WT females by readying or preparing these cells to rapidly respond to low calcium. We will use the term ‘primed’ and will not use the term ‘activated’. We are open to any terminology or description as to why this is observed and what irisin could be doing to the osteocyte.

Reviewer #3 (Public Review):

Summary:

Irisin has previously been demonstrated to be a muscle-secreted factor that affects skeletal homeostasis. Through the use of different experimental approaches, such as genetic knockout models, recombinant Irisin treatment, or different cell lines, the role of Irisin on skeletal homeostasis has been revealed to be more complex than previously thought and this warrants further examination of its role. Therefore, the current study sought to rigorously examine the effects of global Irisin knockout (KO) in male and female mouse bone. Authors demonstrated that in calcium-demanding settings, such as lactation or low-calcium diet, female Irisin KO mice lose less bone compared to wild-type (WT) female mice. Interestingly male Irisin KO mice exhibited worse skeletal deterioration compared to WT male mice when fed a low-calcium diet. When examined for transcriptomic profiles of osteocyte-enriched cortical bone, authors found that Irisin KO altered the expression of osteocytic osteolysis genes as well as steroid and fatty acid metabolism genes in males but not in females. These data support the authors' conclusion that Irisin regulates skeletal homeostasis in sex-dependent manner.

Strengths:

The major strength of the study is the rigorous examination of the effects of Irisin deletion in the settings of skeletal maturity and increased calcium demands in female and male mice. Since many of the common musculoskeletal disorders are dependent on sex, examining both sexes in the preclinical setting is crucial. Had the investigators only examined females or males in this study, the conclusions from each sex would have contradicted each other regarding the role of Irisin on bone. Also, the approaches are thorough and comprehensive that assess the functional (mechanical testing), morphological (microCT, BSEM, and histology), and cellular (RNA-seq) properties of bone.

Weaknesses: One of the weaknesses of this study is a lack of detailed mechanistic analysis of why Irisin has a sex-dependent role on skeletal homeostasis. This absence is particularly notable in the osteocyte transcriptomic results where such data could have been used to further probe potential candidate pathways between LC females vs. LC males.

Response: Our future studies will focus on understanding the molecular mechanism behind the sex-dependent effects of irisin. Our RNA seq data shows a significant difference in the lipid, steroid, and fat metabolism pathways between male and female mice, as well as between WT and KO mice. Future studies will focus on these pathways.

Another weakness is authors did not present data that convincingly demonstrate that Irisin secretion is altered in the skeletal muscle between female vs. male WT mice in response to calcium restriction. The supplement skeletal muscle data only present functional and electrophysiolgical outcomes. Since Itgav or Itgb5 were not different in any of the experimental groups, it is assumed that the changes in the level of Irisin is responsible for the phenotypes observed in WT mice. Assessing Irisin expression will further strengthen the conclusion based on observing skeletal changes that occur in Irisin KO male and female mice.

Response: The problem is that the commercial assays for irisin are not dependable, and results can differ widely across and beyond the physiologic range of 1-10 ng/ml. In part this is due to the nature of the polyclonal antibodies used and the resultant cross reactivity with other proteins. It was shown in Islam et al, 2021 (Nature Metabolism) that the commercial ELISAs were completely unreliable in mice and the only reliable method of measuring circulating irisin is mass spectrometry.

Reviewer #1 (Recommendations For The Authors):

Minor comments:

(1) Were there any low calcium diet food intake or body weight alterations between littermates and FDNC5 KO mice?

Response: Yes, and we can now include the body weight data and the food intake data in the supplement. We do not observe any significant difference between the groups.

(2) In Fig 1, ideally the authors would provide the osteocyte lacunar density along with the lacunar area.

Response: We do not observe any difference in osteocyte density in any of the groups. There is not sufficient time within 2 weeks to see a change in osteocyte density because there is no new bone formation.

(3) What is the author's comment on the involvement of irisin on TGF-B signaling since the authors observed peri lacunar remodeling in FDNC5 KO mice? Authors should also include this in the discussion section regarding the Irisin-TGF-B signaling in terms of observed increased matrix-related signals.

Response: Perilacunar modeling is the removal followed by the replacement of the perilacunar and pericanilucular matrix as occurs with lactation (Qing et al 2012). Osteocytic osteolysis is the first half of that process where the matrix is removed. Alliston and colleagues generated transgenic mice with reduced expression of the TGFb Type II receptor in mice by using the Dmp1-Cre (PMID: 32282961). They clearly found a significant difference in bone parameters, the appearance of the osteocyte lacunocanalicular network, and markers of the osteocyte perilacunar remodeling between the sexes, however they did not compare the lacunar remodeling process in males as compared to females. The females were subjected to lactation and were found to be resistant to osteocytic osteolysis. To compare males and females, they would have had to challenge both sexes to a high calcium demanding condition such as low calcium diet as performed in the current study. Their study does suggest that TGF is involved in the osteocytic osteolysis that occurs with lactation. However, as the null males showed an abnormal lacunocanlicular network compared to wildtype males, this does not necessarily indicate a defect in perilacunar remodeling. It is more likely that the defect occurred during bone formation when osteoblasts were differentiating into osteocytes. Therefore, we will reference this paper regarding the role of TGF in osteocytic osteolysis in females with lactation but not in the comparison of males to females. We have examined the normalized expression of TGF1, 2, and 3 in the present study and found no significant differences in TGF1 or 2 in any of the groups, but did find significantly higher expression of TGF3 in females compared to males for WT (fdr < 0.05), LCD WT (fdr < 0.05), and Control KO (p value < 0.01). Perhaps this isoform is playing a major role in osteocytic osteolysis that occurs with lactation.

(4) Did the authors compare the transcriptomic dataset between lactated female WT vs. KO groups? Or were the RNA-seq studies only performed on LCD study samples?

Response: We have examined RNA sequence on the LCD study samples, and not in the lactating females.

Reviewer #2 (Recommendations For The Authors):

Line 401 on page 14 states that the sexes respond differently to calcium deficiency. Lacunar area increases in both sexes, so the response is very similar. What appears to be different between the sexes is the role of FNDC5 in this process.

Response: Female WT mice have higher osteocyte lacunar area at baseline with normal diet compared to WT males. With the low calcium diet, lacunar area increases in both sexes, with female WTs having a greater increase. We agree that what appears to be different between the sexes is the role of FNDC5 when challenged with high calcium demand.

Reviewer #3 (Recommendations For The Authors):

• The authors state in the abstract and discussion that 'We propose Irisin ensures the survival of offspring by targeting the osteocytes...'. However, this appears to be over interpretation of their findings as they have not assessed the number of offspring surviving to weaning or their growth rate between WT and KO breeders.

Response: That was a proposal and we agree that it could be an over interpretation. However we would like to keep this as a speculation that could be tested in future studies.

• Figures 1 and 2 should include cortical Total Area (and maybe Marrow Cavity data from Supp as well). These data will help readers to assess whether the thinning of the cortex is driven by impaired periosteal expansion or accelerated endosteal resorption (or both). Marrow cavity area data seem to suggest increased endosteal resorption (Supp. Table 2), but unclear if periosteal expansion is altered.

Response: The data are included in the supplementary tables. We do not observe any difference in the periosteal area between the groups.

• To further support the author's statement that male KO mice exhibit different material properties of bone compared to WT mice, estimated elastic modulus should be calculated from the stiffness data (see https://doi.org/10.1002/jbmr.2539).

Response: We looked at the elastic modulus and it requires a stress strain curve instead of the force displacement we used in our calculations, therefore we were not able to get the estimated elastic modulus from the raw data we have.

• In Figure 3 there is no legend indicating females or males. Based on the data and results texts it is assumed that red is Female and blue is Male. However, please confirm in the figure legend.

Response: This is now added in the figure legends.

• Transcriptomic data should be deposited to NCBI GEO data repository. Also, please indicate whether cutoff p-value for DEG analysis was adjusted or not.

Response: We have submitted our data to the GEO data repository: GSE242445. Significant genes were defined as genes with p-value less than 0.01 and absolute log2 fold change larger than 1. The p-value is not adjusted. This information is now added.

• The statistical analysis section indicates that a two-way repeated-measure ANOVA was used. However, the data presented in the study are from independent groups, in which case repeated-measure statistical approaches should not be used. Please clarify the statistical tests that were used.

Response: We now use regular ANOVA instead of repeated-measure ANOVA. Repeated-measure ANOVA is used for paired tests. The data remain significant.

In summary, we thank the reviewers for their very useful and thoughtful suggestions for improving our manuscript.

eLife assessment

The study presents valuable findings on sexually dimorphic patterns of osteocytic transcriptomes and low calcium diet-induced osteocytic osteolysis in FNDC5-deficient mice. The authors present solid evidence for sex-specific changes in osteocyte morphology and gene expression under a calcium-demanding setting in this particular strain of mice, although the protective role of FNDC5-deficiency in lactation and low-calcium diet in female mice remains unclear due to lack of mechanistic studies. The study also lacks evidence that irisin, a proteolytically cleaved product of FNDC5, is responsible for the observed phenotypes, as irisin was not directly measured.

Reviewer #1 (Public Review):

In this manuscript, Shimonty and colleagues study the effects of FNDC5/irisin deletion on osteocytes in a sex-specific manner using models of lactation induced bone loss and bone loss due to low calcium diet (LCD). Consistent with the previous findings of Kim et al. (2018), the authors report 'protective' effects of irisin deficiency in lactating female FNDC5-null mice due to reduced osteocytic osteolysis. Interestingly, FNDC5 null mice show distinct changes when placed on LCD, with mutant females showing some protection from hyperparathyroidism-induced bone loss, while mutant males (which have more cortical bone at baseline) show increased LCD-induced bone loss. Furthermore, new insights into irisin's role in osteocytes regarding cellular energetic metabolism were provided by sex and gene-dependent transcriptomic datasets. Strengths of the well-written manuscript include clear description of sex-dependent effects, strong transcriptomic datasets, and focus on cortical bone changes using microCT, histomorphometry, BSEM, and serum analysis. Despite these strengths, important weaknesses are noted (below) which could be addressed to improve the impact of the work for a broad audience.

Major comments:

(1) Overall, the magnitude of the effect size due to FNDC5 deficiency in both male and female mice is rather modest at the level of bone mass. Looking at the data from a qualitative perspective, it is clear that knockout females still lose bone during lactation and on the low calcium diet (LCD). It is difficult to assess the physiologic consequence of the modest quantitative 'protection' seen in FNDC5 mutants since the mutants still show clear and robust effects of lactation and LCD on all parameters measured. Similarly, the magnitude of the 'increased' cortical bone loss in FNDC5 mutant males is also modest, and perhaps could be related to the fact that these mice are starting with slightly more cortical bone. Since the authors do not provide a convincing molecular explanation for why FNDC5 deficiency causes these somewhat subtle changes, I would like to offer a suggestion for the authors to consider (below, point #2) which might de-emphasize the focus of the manuscript on FNDC5. If the authors chose not to follow this suggestion, the manuscript could be strengthened by addressing the consequences of the modest changes observed in WT versus FNDC5 KO mice. I understand that the effects of FNDC5 are more obvious at the level of osteocyte morphology, and it is reasonable to emphasize these findings here.

(2) The bone RNA-seq findings reported in Figures 4-6 are quite interesting. Although Youlten et al previously reported that the osteocyte transcriptome is sex-dependent, the work here certainly advances that notion to a considerable degree, and likely will be of high interest to investigators studying skeletal biology and sexual dimorphism in general. To this end, one direction for the authors to consider might be to refocus their manuscript towards sexually-dimorphic gene expression patterns in osteocytes and the different effects of LCD on male versus female mice. This would allow the authors to better emphasize these major findings, and then to use FNDC5 deficiency as an illustrative example of how sexually-dimorphic osteocytic gene expression patterns might be affected by deletion of an osteocyte-acting endocrine factor. Ideally, the authors would confirm RNA-seq data comparing male versus female mice in osteocytes using in situ hybridization or immunostaining. Of course, this point is only a suggestion for the authors to consider.

(3) It would be appreciated if the authors could provide additional serum parameters (if possible) to clarify incomplete data in both lactation and low-calcium diet models: RANKL/OPG ratio, Ctx, PTHrP, and 1,25-dihydroxyvitamin D levels. I understand that this may not be possible due to lack of available material.

Reviewer #2 (Public Review):

Summary:

The goal of this study was to examine the role of FNDC5 in the response of the murine skeleton to either lactation or a calcium-deficient diet. The authors find that female FNDC5 KO mice are somewhat protected from the bone loss and osteocyte lacunar enlargement caused by either lactation or a calcium-deficient diet. In contrast, male FNDC5 KO mice lose more bone and have a greater enlargement of osteocyte lacunae than their wild type controls. Based on these results, the authors conclude that in males irisin protects bone from calcium deficiency but that in females it promotes calcium removal from bone for lactation.

While some of the conclusions of this study are supported by the results, it is not clear that the modest effects of FNDC5 deletion have an impact on calcium homeostasis or milk production.

Specific comments.

(1) The authors sometimes refer to FNDC5 and other times to irisin when describing causes for a particular outcome. Because irisin was not measured in any of the experiments, the authors should not conclude that lack of irisin is responsible. Along these lines, is there any evidence that either lactation or a calcium-deficient diet increases production of irisin in mice?

(2) The results of the irisin-rescue experiment shown in figure 2G cannot be appropriately interpreted without normal diet controls. In addition, some evidence that the AAV8-irisin virus actually increased irisin levels in the mice would strengthen the conclusion.

(3) There is insufficient evidence to support the idea that the effect of FNDC5 on bone resorption and osteocytic osteolysis is important for the transfer of calcium from bone to milk. Previous studies by others have shown that bone resorption is not required to maintain milk or serum calcium when dietary calcium is sufficient but is critical if dietary calcium is low (Endo. 156:2762-73, 2015). To support the conclusions of the current study, it would be necessary to determine whether FNDC5 is required to maintain calcium levels when lactating mice lack sufficient dietary calcium.

(4) The amount of cortical bone loss due to lactation is very similar in both WT and FNDC5 KO mice. The results of the statistical analysis of the data presented in figure 1B are surprising given the very similar effect size of lactation. The key result from the 2-way ANOVA is whether there is an effect of genotype on the effect size of lactation (genotype-lactation interaction). The interaction terms were not provided. Similar concerns are noted for the results shown in figure 1G and H.

(5) It is not clear what justifies the term 'primed' or 'activated' for resorption. Is there evidence that a certain level of TRAP expression lowers the threshold for osteocytic osteolysis in response to a stimulus?

Reviewer #3 (Public Review):

Summary: Irisin has previously been demonstrated to be a muscle-secreted factor that affects skeletal homeostasis. Through the use of different experimental approaches, such as genetic knockout models, recombinant Irisin treatment, or different cell lines, the role of Irisin on skeletal homeostasis has been revealed to be more complex than previously thought and this warrants further examination of its role. Therefore, the current study sought to rigorously examine the effects of global Irisin knockout (KO) in male and female mouse bone. Authors demonstrated that in calcium-demanding settings, such as lactation or low-calcium diet, female Irisin KO mice lose less bone compared to wildtype (WT) female mice. Interestingly male Irisin KO mice exhibited worse skeletal deterioration compared to WT male mice when fed low-calcium diet. When examined for transcriptomic profiles of osteocyte-enriched cortical bone, authors found that Irisin KO altered the expression of osteocytic osteolysis genes as well as steroid and fatty acid metabolism genes in males but not in females. These data support authors' conclusion that Irisin regulates skeletal homeostasis in a sex-dependent manner.

Strengths:

The major strength of the study is rigorous examination of the effects of Irisin deletion in the settings of skeletal maturity and increased calcium demands in female and male mice. Since many of the common musculoskeletal disorders are dependent on sex, examining both sexes in the preclinical setting is crucial. Had the investigators only examined females or males in this study, the conclusion from each sex would have contradicted each other regarding the role of Irisin on bone. Also, the approaches are thorough and comprehensive that assess the functional (mechanical testing), morphological (microCT, BSEM, and histology), and cellular (RNA-seq) properties of bone. Transcriptomic data deposited to NCBI GEO data repository will be a valuable resource to musculoskeletal researchers who aim to further assess the affects of Irisin on skeleton.

Weaknesses:

One of the weaknesses of this study is a lack of detailed mechanistic analysis of why Irisin has sex-dependent role on skeletal homeostasis. However, the osteocyte transcriptome comparisons between LC females vs. LC males lay a foundation for such future mechanistic studies.

Another weakness is authors did not present data that convincingly demonstrate that Irisin secretion is altered in the skeletal muscle between female vs. male WT mice in response to calcium restriction. The supplement skeletal muscle data only present functional and electrophysiological outcomes. Since Itgav or Itgb5 were not different in any of the experimental groups, it is assumed that the changes in the level of Irisin is responsible for the phenotypes observed in WT mice. Assessing Irisin expression will further strengthen the conclusion based on observing skeletal changes that occur in Irisin KO male and female mice.

Author Response

Reviewer #1 (Public Review):

Response to reviewer 1 comments on “weaknesses”:

“A weakness in the approach is the use of genetic models that do not offer complete deletion of the prolactin receptor from targeted neuronal populations...”

We acknowledge that neither model used provided a complete deletion of the prolactin receptor (Prlr) from the targeted neuronal populations. We suspect that incomplete deletion of targeted genes is not uncommon in these sort of studies, but this remains the best approach to addressing our question, and we believe we have been thorough and transparent in reporting the degree of deletion observed. We thought we had appropriately discussed the implications of the low proportion of Kiss1 cells still expressing Prlr, but will certainly revisit to ensure it is discussed thoroughly. This does not detract, however, from the key conclusion that prolactin action is necessary for full suppression of fertility in lactation in the mouse.

“Results showing no impact of progesterone on LH secretion during lactation are surprising, given the effectiveness of progesterone-containing birth control in lactating women...”

We think that this comment misrepresents what has been done in our study. We did not report a lack of impact of progesterone, as exogenous progesterone was never administered to mice. We did, however, give mifepristone as a progesterone receptor antagonist to determine whether endogenous progesterone contributed to the suppression of kisspeptin neuronal activity. We found that mifepristone, at levels sufficient to terminate pregnancy, had no effect on pulsatile LH secretion in lactating mice. This is consistent with our prior observation that progesterone levels are low in mouse lactation, suggesting that progesterone does not contribute significantly to the suppression of kisspeptin neuronal activity during lactation in the mouse. We agree with the reviewer that if we had given exogenous progesterone, it likely would result in suppression of pulsatile LH secretion (as it does in women). Indeed, in other work, we have found that progesterone administration profoundly suppresses activity of the kisspeptin neurons in mice (https://doi.org/10.1210/en.2019-00193). But this was not the point of the present experiment. We will review how we have described this experiment to ensure that this is absolutely clear.

“While the authors assert their findings may reflect an important role for prolactin in lactational infertility in other mammalian species, that remains to be seen….”

We acknowledge that our study cannot address whether prolactin is necessary for the suppression of lactation in other mammalian species. We hope our data may stimulate a re-examination of this question in other species, however, as some of the prior methodology (such as using pharmacological suppression of prolactin) may have had off target effects that confound interpretation. We thought that this point was discussed appropriately in the manuscript but we will certainly check and make sure this is addressed suitably.

eLife assessment

This study addresses an important long-standing controversy in the field of reproductive biology. Using cutting-edge techniques the authors provide compelling evidence supporting a pivotal role of prolactin as the mediator of lactational infertility. Some methodological, technical, interpretive, stylistic, and typographical aspects of the paper need to be strengthened.

Reviewer #1 (Public Review):

Summary:

In this paper, Hackwell and colleagues performed technically impressive, long-term, GCaMP fiber photometry recordings from Kiss1 neurons in the arcuate nucleus of mice during multiple reproductive states. The data show an immediate suppression of activity of arc Kiss1 neuronal activity during pregnancy that is maintained during lactation. In the absence of any apparent change in suckling stimulus or milk production, mice lacking prolactin receptors in arcuate Kiss1 neurons regained Kiss1 episodic activity and estrous cyclicity faster than control mice, demonstrating that direct prolactin action on Kiss1 neurons is at least partially responsible for suppressing fertility in this species. The effect of loss of prolactin receptors from CamK2a expressing neurons was even greater, indicating either that prolactin sensitivity in Kiss1 neurons of the RP3V contributes to lactational infertility or that other prolactin-sensitive neurons are involved. These data demonstrate the important role of prolactin in suppressing Kiss1 neuron activity and thereby fertility during the lactational period in the mouse.

Strengths:

This is the first study to monitor the activity of the GnRH pulse-generating system across different reproductive states in the same animal. Another strength in the study design is that it isolated the effects of prolactin by maintaining normal lactation and suckling (assessed indirectly using pup growth curves). The study also offers insight into the phenomenon of postpartum ovulation in mice. The results showed a brief reactivation of arcuate Kiss1 activity immediately prior to parturition, attributed to falling progesterone levels at the end of pregnancy. This hypothesis will be of interest to the field and is likely to inspire testing in future studies. With the exceptions mentioned below, the conclusions of the paper are well supported by the data, and the aims of the study were achieved. This paper is likely to raise the standard for technical expectations in the field and spark new interest in the direct impact of prolactin on Kiss1 neurons during lactation in other species.

Weaknesses:

A weakness in the approach is the use of genetic models that do not offer complete deletion of the prolactin receptor from targeted neuronal populations. A substantial proportion of Kiss1 neurons in both models retain the receptor. As a result, it is not clear whether the partial maintenance of cyclicity during lactation in the genetic models is due to incomplete deletion or to the involvement of other factors. This weakness should be more fully discussed in the text. In addition, results showing no impact of progesterone on LH secretion during lactation are surprising, given the effectiveness of progesterone-containing birth control in lactating women. The progesterone-related experiments were not well justified or discussed in the text. While the authors assert their findings may reflect an important role for prolactin in lactational infertility in other mammalian species, that remains to be seen. Hyperprolactinemia is known to suppress GnRH release, but its importance in the suppression of cyclicity during lactation is controversial. Indeed, in several species, the stimulus of suckling is considered to be the main driver of lactational fertility suppression. Data from rats shows that exogenous prolactin was unable to suppress LH release in dams deprived of their pups shortly after birth; both suckling and prolactin were necessary to suppress a post ovariectomy rise in LH levels. The duration of amenorrhea does not correlate with average prolactin levels in humans, and suckling but not prolactin was required to suppress the postpartum rise in LH in the rhesus monkey. The authors should discuss more thoroughly whether the protocol of this or other studies might result in discordant results and whether mice are likely to be an outlier in their mechanism of cycle suppression.

Reviewer #2 (Public Review):

Summary:

The overall goal of Eleni et al. is to determine if the suppression of LH pulses during lactation is mediated by prolactin signaling at kisspeptin neurons. To address this, the authors used GCaMP fiber photometry and serial blood sampling to reveal that in vivo episodic arcuate kisspeptin neuron activity and LH pulses are suppressed throughout pregnancy and lactation. The authors further utilized knockout models to demonstrate that the loss of prolactin receptor signaling at kisspeptin cells prevents the suppression of kisspeptin function and results in early reestablishment of fertility during lactation. The work demonstrates exemplary design and technique, and the outcomes of these experiments are sophistically discussed.

Strengths:

This manuscript demonstrates exemplary skill with powerful techniques and reveals a key role for arcuate kisspeptin neurons in maintaining lactation-induced infertility in mice. In a difficult feat, the authors used fiber photometry to map the activity of arcuate kisspeptin cells into lactation and weaning without disrupting parturition, lactation, or maternal behavior. The authors used a knockout approach to identify if prolactin inhibition of fertility is mediated by direct signaling at arcuate kisspeptin cells. Although the model does not perfectly eliminate prolactin receptor expression in all kisspeptin neurons, results from the achieved knockdown support the conclusion that prolactin signaling at kisspeptin neurons is required to maintain lactational infertility. The methods were advanced and appropriate for the aims, the studies were rigorously conducted, and the conclusions were thoughtfully discussed. Overall, the aims of this study were achieved.

Reviewer #3 (Public Review):

Summary:

Grattan and colleagues were trying to establish the neural mechanism underlying lactational infertility, in particular trying to establish the relative importance of the neurogenic effects of the suckling stimulus versus prolactin per se. They have shown that in the mouse it is rather prolactin and more specifically its action on the hypothalamic arcuate kisspeptin neuronal system, which is the key neural construct underlying gonadotrophin-releasing hormone (GnRH) pulse generation and central to the neuroendocrine control of reproduction, that mediates lactational infertility. The authors have taken a measured tone to emphasise the data pertaining to the mouse without extravagant extrapolation to humans. Nevertheless, the key findings provide a substantial foundation to facilitate interpretation of studies in other species.

Strengths:

The major strength of this study is the use of a combination of cutting-edge technologies, which of course underlie the majority of scientific advances rather than intellectual prowess favoured by the majority of scientists. Their approach avoided the major confounding effects of using pharmacological strategies to suppress prolactin action that has complicated the vast majority of previous studies. The study also provides an elegant and comprehensive contiguous description of GnRH pulse generator frequency across the ovarian cycle, through pregnancy and lactation, and into weaning in individual animals.

Weaknesses:<br /> There are no significant weaknesses.

Author Response

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

Reviewer #1 (Public Review):

The authors present a number of deep-learning models to analyse the dynamics of epithelia. In this way, they want to overcome the time-consuming manual analysis of such data and also remove a potential operator bias. Specifically, they set up models for identifying cell division events and cell division orientation. They apply these tools to the epithelium of the developing Drosophila pupal wing. They confirm a linear decrease of the division density with time and identify a burst of cell division after the healing of a wound that they had induced earlier. These division events happen a characteristic time after and a characteristic distance away from the wound. These characteristic quantities depend on the size of the wound.

Strength:

The methods developed in this work achieve the goals set by the authors and are a very helpful addition to the toolbox of developmental biologists. They could potentially be used on various developing epithelia. The evidence for the impact of wounds on cell division is solid.

Weakness:

Some aspects of the deep-learning models remained unclear, and the authors might want to think about adding details. First of all, for readers not being familiar with deep-learning models, I would like to see more information about ResNet and U-Net, which are at the base of the new deep-learning models developed here. What is the structure of these networks?

We agree with the Reviewer and have included additional information on page 8 of the manuscript, outlining some background information about the architecture of ResNet and U-Net models.

How many parameters do you use?

We apologise for this omission and have now included the number of parameters and layers in each model in the methods section on page 25.

What is the difference between validating and testing the model? Do the corresponding data sets differ fundamentally?

The difference between ‘validating’ and ‘testing’ the model is validating data is used during training to determine whether the model is overfitting. If the model is performing well on the training data but not on the validating data, this a key signal the model is overfitting and changes will need to be made to the network/training method to prevent this. The testing data is used after all the training has been completed and is used to test the performance of the model on fresh data it has not been trained on. We have removed refence to the validating data in the main text to make it simpler and add this explanation to the methods. There is no fundamental (or experimental) difference between each of the labelled data sets; rather, they are collected from different biological samples. We have now included this information in the Methods text on page 24.

How did you assess the quality of the training data classification?

These data were generated and hand-labelled by an expert with many years of experience in identifying cell divisions in imaging data, to give the ground truth for the deep learning model.

Reviewer #1 (Recommendations For The Authors):

You repeatedly use 'new', 'novel' as well as 'surprising' and 'unexpected'. The latter are rather subjective and it is not clear based on what prior knowledge you make these statements. Unless indicated otherwise, it is understood that the results and methods are new, so you can delete these terms.

We have deleted these words, as suggested, for almost all cases.

p.4 "as expected" add a reference or explain why it is expected.

A reference has now been included in this section, as suggested.

p.4 "cell divisions decrease linearly with time" Only later (p.10) it turns out that you think about the density of cell divisions.

This has been changed to "cell division density decreases linearly with time".

p.5 "imagine is largely in one plane" while below "we generated a 3D z-stack" and above "our in vivo 3D image data" (p.4). Although these statements are not strictly contradictory, I still find them confusing. Eventually, you analyse a 2D image, so I would suggest that you refer to your in vivo data as being 2D.

We apologise for the confusion here; the imaging data was initially generated using 3D z-stacks but this 3D data is later converted to a 2D focused image, on which the deep learning analysis is performed. We are now more careful with the language in the text.

p.7 "We have overcome (...) the standard U-Net model" This paragraph remains rather cryptic to me. Maybe you can explain in two sentences what a U-Net is or state its main characteristics. Is it important to state which class you have used at this point? Similarly, what is the exact role of the ResNet model? What are its characteristics?

We have included more details on both the ResNet and U-Net models and how our model incorporates properties from them on Page 8.

p.8 Table 1 Where do I find it? Similarly, I could not find Table 2.

These were originally located in the supplemental information document, but have been moved to the main manuscript.

p.9 "developing tissue in normal homeostatic conditions" Aren't homeostatic and developing contradictory? In one case you maintain a state, in the other, it changes.

We agree with the Reviewer and have removed the word ‘homeostatic’.

p.9 "Develop additional models" I think 'models' refers to deep learning models, not to physical models of epithelial tissue development. Maybe you can clarify this?

Yes, this is correct; we have phrased this better in the text.

p.12 "median error" median difference to the manually acquired data?

Yes, and we have made this clearer in the text, too.

p.12 "we expected to observe a bias of division orientation along this axis" Can you justify the expectation? Elongated cells are not necessarily aligned with the direction of a uniaxially applied stress.

Although this is not always the case, we have now included additional references to previous work from other groups which demonstrated that wing epithelial cells do become elongated along the P/D axis in response to tension.

p.14 "a rather random orientation" Please, quantify.

The division orientations are quantified in Fig. 4F,G; we have now changed our description from ‘random’ to ‘unbiased’.

p.17 "The theories that must be developed will be statistical mechanical (stochastic) in nature" I do not understand. Statistical mechanics refers to systems at thermodynamic equilibrium, stochastic to processes that depend on, well, stochastic input.

We have clarified that we are referring to non-equilibrium statistical mechanics (the study of macroscopic systems far from equilibrium, a rich field of research with many open problems and applications in biology).

Reviewer #2 (Public Review):

In this manuscript, the authors propose a computational method based on deep convolutional neural networks (CNNs) to automatically detect cell divisions in two-dimensional fluorescence microscopy timelapse images. Three deep learning models are proposed to detect the timing of division, predict the division axis, and enhance cell boundary images to segment cells before and after division. Using this computational pipeline, the authors analyze the dynamics of cell divisions in the epithelium of the Drosophila pupal wing and find that a wound first induces a reduction in the frequency of division followed by a synchronised burst of cell divisions about 100 minutes after its induction.

In general, novelty over previous work does not seem particularly important. From a methodological point of view, the models are based on generic architectures of convolutional neural networks, with minimal changes, and on ideas already explored in general. The authors seem to have missed much (most?) of the literature on the specific topic of detecting mitotic events in 2D timelapse images, which has been published in more specialized journals or Proceedings. (TPMAI, CCVPR etc., see references below). Even though the image modality or biological structure may be different (non-fluorescent images sometimes), I don't believe it makes a big difference. How the authors' approach compares to this previously published work is not discussed, which prevents me from objectively assessing the true contribution of this article from a methodological perspective.

On the contrary, some competing works have proposed methods based on newer - and generally more efficient - architectures specifically designed to model temporal sequences (Phan 2018, Kitrungrotsakul 2019, 2021, Mao 2019, Shi 2020). These natural candidates (recurrent networks, long-short-term memory (LSTM) gated recurrent units (GRU), or even more recently transformers), coupled to CNNs are not even mentioned in the manuscript, although they have proved their generic superiority for inference tasks involving time series (Major point 2). Even though the original idea/trick of exploiting the different channels of RGB images to address the temporal aspect might seem smart in the first place - as it reduces the task of changing/testing a new architecture to a minimum - I guess that CNNs trained this way may not generalize very well to videos where the temporal resolution is changed slightly (Major point 1). This could be quite problematic as each new dataset acquired with a different temporal resolution or temperature may require manual relabeling and retraining of the network. In this perspective, recent alternatives (Phan 2018, Gilad 2019) have proposed unsupervised approaches, which could largely reduce the need for manual labeling of datasets.

We thank the reviewer for their constructive comments. Our goal is to develop a cell detection method that has a very high accuracy, which is critical for practical and effective application to biological problems. The algorithms need to be robust enough to cope with the difficult experimental systems we are interested in studying, which involve densely packed epithelial cells within in vivo tissues that are continuously developing, as well as repairing. In response to the above comments of the reviewer, we apologise for not including these important papers from the division detection and deep learning literature, which are now discussed in the Introduction (on page 4).

A key novelty of our approach is the use of multiple fluorescent channels to increase information for the model. As the referee points out, our method benefits from using and adapting existing highly effective architectures. Hence, we have been able to incorporate deeper models than some others have previously used. An additional novelty is using this same model architecture (retrained) to detect cell division orientation. For future practical use by us and other biologists, the models can easily be adapted and retrained to suit experimental conditions, including different multiple fluorescent channels or number of time points. Unsupervised approaches are very appealing due to the potential time saved compared to manual hand labelling of data. However, the accuracy of unsupervised models are currently much lower than that of supervised (as shown in Phan 2018) and most importantly well below the levels needed for practical use analysing inherently variable (and challenging) in vivo experimental data.

Regarding the other convolutional neural networks described in the manuscript:

(1) The one proposed to predict the orientation of mitosis performs a regression task, predicting a probability for the division angle. The architecture, which must be different from a simple Unet, is not detailed anywhere, so the way it was designed is difficult to assess. It is unclear if it also performs mitosis detection, or if it is instead used to infer orientation once the timing and location of the division have been inferred by the previous network.

The neural network used for U-NetOrientation has the same architecture as U-NetCellDivision10 but has been retrained to complete a different task: finding division orientation. Our workflow is as follows: firstly, U-NetCellDivision10 is used to find cell divisions; secondly, U-NetOrientation is applied locally to determine the division orientation. These points have now been clarified in the main text on Page 14.

(2) The one proposed to improve the quality of cell boundary images before segmentation is nothing new, it has now become a classic step in segmentation, see for example Wolny et al. eLife 2020.

We have cited similar segmentation models in our paper and thank the referee for this additional one. We had made an improvement to the segmentation models, using GFP-tagged E-cadherin, a protein localised in a thin layer at the apical boundary of cells. So, while this is primarily a 2D segmentation problem, some additional information is available in the z-axis as the protein is visible in 2-3 separate z-slices. Hence, we supplied this 3-focal plane input to take advantage of the 3D nature of this signal. This approach has been made more explicit in the text (Pages 14, 15) and Figure (Fig. 2D).

As a side note, I found it a bit frustrating to realise that all the analysis was done in 2D while the original images are 3D z-stacks, so a lot of the 3D information had to be compressed and has not been used. A novelty, in my opinion, could have resided in the generalisation to 3D of the deep-learning approaches previously proposed in that context, which are exclusively 2D, in particular, to predict the orientation of the division.

Our experimental system is a relatively flat 2D tissue with the orientation of the cell divisions consistently in the xy-plane. Hence, a 2D analysis is most appropriate for this system. With the successful application of the 2D methods already achieving high accuracy, we envision that extension to 3D would only offer a slight increase in effectiveness as these measurements have little room for improvement. Therefore, we did not extend the method to 3D here. However, of course, this is the next natural step in our research as 3D models would be essential for studying 3D tissues; such 3D models will be computationally more expensive to analyse and more challenging to hand label.

Concerning the biological application of the proposed methods, I found the results interesting, showing the potential of such a method to automatise mitosis quantification for a particular biological question of interest, here wound healing. However, the deep learning methods/applications that are put forward as the central point of the manuscript are not particularly original.

We thank the referee for their constructive comments. Our aim was not only to show the accuracy of our models but also to show how they might be useful to biologists for automated analysis of large datasets, which is a—if not the—bottleneck for many imaging experiments. The ability to process large datasets will improve robustness of results, as well as allow additional hypotheses to be tested. Our study also demonstrated that these models can cope with real in vivo experiments where additional complications such as progressive development, tissue wounding and inflammation must be accounted for.

Major point 1: generalisation potential of the proposed method.

The neural network model proposed for mitosis detection relies on a 2D convolutional neural network (CNN), more specifically on the Unet architecture, which has become widespread for the analysis of biology and medical images. The strategy proposed here exploits the fact that the input of such an architecture is natively composed of several channels (originally 3 to handle the 3 RGB channels, which is actually a holdover from computer vision, since most medical/biological images are gray images with a single channel), to directly feed the network with 3 successive images of a timelapse at a time. This idea is, in itself, interesting because no modification of the original architecture had to be carried out. The latest 10-channel model (U-NetCellDivision10), which includes more channels for better performance, required minimal modification to the original U-Net architecture but also simultaneous imaging of cadherin in addition to histone markers, which may not be a generic solution.

We believe we have provided a general approach for practical use by biologists that can be applied to a range of experimental data, whether that is based on varying numbers of fluorescent channels and/or timepoints. We envisioned that experimental biologists are likely to have several different parameters permissible for measurement based on their specific experimental conditions e.g., different fluorescently labelled proteins (e.g. tubulin) and/or time frames. To accommodate this, we have made it easy and clear in the code on GitHub how these changes can be made. While the model may need some alterations and retraining, the method itself is a generic solution as the same principles apply to very widely used fluorescent imaging techniques.

Since CNN-based methods accept only fixed-size vectors (fixed image size and fixed channel number) as input (and output), the length or time resolution of the extracted sequences should not vary from one experience to another. As such, the method proposed here may lack generalization capabilities, as it would have to be retrained for each experiment with a slightly different temporal resolution. The paper should have compared results with slightly different temporal resolutions to assess its inference robustness toward fluctuations in division speed.

If multiple temporal resolutions are required for a set of experiments, we envision that the model could be trained over a range of these different temporal resolutions. Of course, the temporal resolution, which requires the largest vector would be chosen as the model's fixed number of input channels. Given the depth of the models used and the potential to easily increase this by replacing resnet34 with resnet50 or resnet101 the model would likely be able to cope with this, although we have not specifically tested this. (page 27)

Another approach (not discussed) consists in directly convolving several temporal frames using a 3D CNN (2D+time) instead of a 2D, in order to detect a temporal event. Such an idea shares some similarities with the proposed approach, although in this previous work (Ji et al. TPAMI 2012 and for split detection Nie et al. CCVPR 2016) convolution is performed spatio-temporally, which may present advantages. How does the authors' method compare to such an (also very simple) approach?

We thank the Reviewer for this insightful comment. The text now discusses this (on Pages 8 and 17). Key differences between the models include our incorporation of multiple light channels and the use of much deeper models. We suggest that our method allows for an easy and natural extension to use deeper models for even more demanding tasks e.g. distinguishing between healthy and defective divisions. We also tested our method with ‘difficult conditions’ such as when a wound is present; despite the challenges imposed by the wound (including the discussed reduction in fluorescent intensities near the wound edge), we achieved higher accuracy compared to Nie et al. (accuracy of 78.5% compared to our F1 score of 0.964) using a low-density in vitro system.

Major point 2: innovatory nature of the proposed method.

The authors' idea of exploiting existing channels in the input vector to feed successive frames is interesting, but the natural choice in deep learning for manipulating time series is to use recurrent networks or their newer and more stable variants (LSTM, GRU, attention networks, or transformers). Several papers exploiting such approaches have been proposed for the mitotic division detection task, but they are not mentioned or discussed in this manuscript: Phan et al. 2018, Mao et al. 2019, Kitrungrotaskul et al. 2019, She et al 2020.

An obvious advantage of an LSTM architecture combined with CNN is that it is able to address variable length inputs, therefore time sequences of different lengths, whereas a CNN alone can only be fed with an input of fixed size.

LSTM architectures may produce similar accuracy to the models we employ in our study, however due to the high degree of accuracy we already achieve with our methods, it is hard to see how they would improve the understanding of the biology of wound healing that we have uncovered. Hence, they may provide an alternative way to achieve similar results from analyses of our data. It would also be interesting to see how LTSM architectures would cope with the noisy and difficult wounded data that we have analysed. We agree with the referee that these alternate models could allow an easier inclusion of difference temporal differences in division time (see discussion on Page 20). Nevertheless, we imagine that after selecting a sufficiently large input time/ fluorescent channel input, biologists could likely train our model to cope with a range of division lengths.

Another advantage of some of these approaches is that they rely on unsupervised learning, which can avoid the tedious relabeling of data (Phan et al. 2018, Gilad et al. 2019).

While these are very interesting ideas, we believe these unsupervised methods would struggle under the challenging conditions within ours and others experimental imaging data. The epithelial tissue examined in the present study possesses a particularly high density of cells with overlapping nuclei compared to the other experimental systems these unsupervised methods have been tested on. Another potential problem with these unsupervised methods is the difficulty in distinguishing dynamic debris and immune cells from mitotic cells. Once again despite our experimental data being more complex and difficult, our methods perform better than other methods designed for simpler systems as in Phan et al. 2018 and Gilad et al. 2019; for example, analysis performed on lower density in vitro and unwounded tissues gave best F1 scores for a single video was 0.768 and 0.829 for unsupervised and supervised respectively (Phan et al. 2018). We envision that having an F1 score above 0.9 (and preferably above 0.95), would be crucial for practical use by biologists, hence we believe supervision is currently still required. We expect that retraining our models for use in other experimental contexts will require smaller hand labelled datasets, as they will be able to take advantage of transfer learning (see discussion on Page 4).

References :

We have included these additional references in the revised version of our Manuscript.

Ji, S., Xu, W., Yang, M., & Yu, K. (2012). 3D convolutional neural networks for human action recognition. IEEE transactions on pattern analysis and machine intelligence, 35(1), 221-231. >6000 citations

Nie, W. Z., Li, W. H., Liu, A. A., Hao, T., & Su, Y. T. (2016). 3D convolutional networks-based mitotic event detection in time-lapse phase contrast microscopy image sequences of stem cell populations. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops (pp. 55-62).

Phan, H. T. H., Kumar, A., Feng, D., Fulham, M., & Kim, J. (2018). Unsupervised two-path neural network for cell event detection and classification using spatiotemporal patterns. IEEE Transactions on Medical Imaging, 38(6), 1477-1487.

Gilad, T., Reyes, J., Chen, J. Y., Lahav, G., & Riklin Raviv, T. (2019). Fully unsupervised symmetry-based mitosis detection in time-lapse cell microscopy. Bioinformatics, 35(15), 2644-2653.

Mao, Y., Han, L., & Yin, Z. (2019). Cell mitosis event analysis in phase contrast microscopy images using deep learning. Medical image analysis, 57, 32-43.

Kitrungrotsakul, T., Han, X. H., Iwamoto, Y., Takemoto, S., Yokota, H., Ipponjima, S., ... & Chen, Y. W. (2019). A cascade of 2.5 D CNN and bidirectional CLSTM network for mitotic cell detection in 4D microscopy image. IEEE/ACM transactions on computational biology and bioinformatics, 18(2), 396-404.

Shi, J., Xin, Y., Xu, B., Lu, M., & Cong, J. (2020, November). A Deep Framework for Cell Mitosis Detection in Microscopy Images. In 2020 16th International Conference on Computational Intelligence and Security (CIS) (pp. 100-103). IEEE.

Wolny, A., Cerrone, L., Vijayan, A., Tofanelli, R., Barro, A. V., Louveaux, M., ... & Kreshuk, A. (2020). Accurate and versatile 3D segmentation of plant tissues at cellular resolution. Elife, 9, e57613.

eLife assessment

In this potentially useful study, the authors use deep learning models to provide solid evidence that epithelial wounding triggers bursts of cell division at a characteristic distance away from the wound. The usefulness of the methods to the community will depend on documenting their robustness toward variability in temporal resolution and/or mitotic event duration, demonstrating their overall superiority over existing approaches and making the code possible to use by others.

Reviewer #1 (Public Review):

The authors present a number of deep learning models to analyse the dynamics of epithelia. In this way they want to overcome the time-consuming manual analysis of such data and also remove a potential operator bias. Specifically, they set up models for identifying cell division events and cell division orientation. They apply these tools to the epithelium of the developing Drosophila pupal wing. They confirm a linear decrease of the division density with time and identify a burst of cell division after healing of a wound that they had induced earlier. These division events happen a characteristic time after and a characteristic distance away from the wound. These characteristic quantities depend on the size of the wound.

Strengths:

The methods developed in this work achieve the goals set by the authors and are a very helpful addition to the toolbox of developmental biologists. They could potentially be used on various developing epithelia. The evidence for the impact of wounds on cell division is compelling.

The methods presented in this work should prove to be very helpful for quantifying cell proliferation in epithelial tissues.

Reviewer #2 (Public Review):

In this manuscript, the authors propose a computational method based on deep convolutional neural networks (CNNs) to automatically detect cell divisions in two-dimensional fluorescence microscopy timelapse images. Three deep learning models are proposed to detect the timing of division, predict the division axis, and enhance cell boundary images to segment cells before and after division. Using this computational pipeline, the authors analyze the dynamics of cell divisions in the epithelium of the Drosophila pupal wing and find that a wound first induces a reduction in the frequency of division followed by a synchronised burst of cell divisions about 100 minutes after its induction.

Comments on revised version:

Regarding the Reviewer's 1 comment on the architecture details, I have now understood that the precise architecture (number/type of layers, activation functions, pooling operations, skip connections, upsampling choice...) might have remained relatively hidden to the authors themselves, as the U-net is built automatically by the fast.ai library from a given classical choice of encoder architecture (ResNet34 and ResNet101 here) to generate the decoder part and skip connections.

Regarding the Major point 1, I raised the question of the generalisation potential of the method. I do not think, for instance, that the optimal number of frames to use, nor the optimal choice of their time-shift with respect to the division time (t-n, t+m) (not systematically studied here) may be generic hyperparameters that can be directly transferred to another setting. This implies that the method proposed will necessarily require re-labeling, re-training and re-optimizing the hyperparameters which directly influence the network architecture for each new dataset imaged differently. This limits the generalisation of the method to other datasets, and this may be seen as in contrast to other tools developed in the field for other tasks such as cellpose for segmentation, which has proven a true potential for generalisation on various data modalities. I was hoping that the authors would try themselves testing the robustness of their method by re-imaging the same tissue with slightly different acquisition rate for instance, to give more weight to their work.

In this regard, and because the authors claimed to provide clear instructions on how to reuse their method or adapt it to a different context, I delved deeper into the code and, to my surprise, felt that we are far from the coding practice of what a well-documented and accessible tool should be.

To start with, one has to be relatively accustomed with Napari to understand how the plugin must be installed, as the only thing given is a pip install command (that could be typed in any terminal without installing the plugin for Napari, but has to be typed inside the Napari terminal, which is mentioned nowhere). Surprisingly, the plugin was not uploaded on Napari hub, nor on PyPI by the authors, so it is not searchable/findable directly, one has to go to the Github repository and install it manually. In that regard, no description was provided in the copy-pasted templated files associated to the napari hub, so exporting it to the hub would actually leave it undocumented.

Regarding now the python notebooks, one can fairly say that the "clear instructions" that were supposed to enlighten the code are really minimal. Only one notebook "trainingUNetCellDivision10.ipynb" has actually some comments, the other have (almost) none nor title to help the unskilled programmer delving into the script to guess what it should do. I doubt that a biologist who does not have a strong computational background will manage adapting the method to its own dataset (which seems to me unavoidable for the reasons mentioned above).

Finally regarding the data, none is shared publicly along with this manuscript/code, such that if one doesn't have a similar type of dataset - that must be first annotated in a similar manner - one cannot even test the networks/plugin for its own information. A common and necessary practice in the field - and possibly a longer lasting contribution of this work - could have been to provide the complete and annotated dataset that was used to train and test the artificial neural network. The basic reason is that a more performant, or more generalisable deep-learning model may be developed very soon after this one and for its performance to be fairly compared, it requires to be compared on the same dataset. Benchmarking and comparison of methods performance is at the core of computer vision and deep-learning.

Author Response

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

eLife assessment

This important study used Voltage Sensitive Dye Imaging (VSDI) to measure neural activity in the primary visual cortex of monkeys trained to detect an oriented grating target that was presented either alone or against an oriented mask. The authors show convincingly that the initial effect of the mask ran counter to the behavioral effects of the mask, a pattern that reversed in the latter phase of the response. They interpret these results in terms of influences from the receptive field center, and although an alternative view that emphasizes the role of the receptive field surround also seems reasonable, this study stands as an interesting and important contribution to our understanding of mechanisms of visual perception.

Public Reviews:

Reviewer #1 (Public Review):

This is a clear account of some interesting work. The experiments and analyses seem well done and the data are useful. It is nice to see that VSDI results square well with those from prior extracellular recordings.

The authors have done a good job responding to the main points of my previous review. One important question remains, as stated in that review:

"My reading is that this is primarily a study of surround suppression with results that follow pretty directly from what we already know from that literature, and although they engage with some of the literature they do not directly mention surround suppression in the text. Their major effect - what they repeatedly describe as a "paradoxical" result in which the responses initially show a stronger response to matched targets and backgrounds and then reverse - seems to pretty clearly match the expected outcome of a stimulus that initially evokes additional excitation due to increased center contrast followed by slightly delayed surround suppression tuned to the same peak orientation. Their dynamics result seems entirely consistent with previous work, e.g. Henry at al 2020, particularly their Fig. 3 https://elifesciences.org/articles/54264, so it seems like a major oversight to not engage with that work at all, and to explain what exactly is new here."

Their rebuttal of my first review is not convincing -- I still believe that surround influences are important and perhaps predominant in determining the outcome of the experiments. This is particularly clear for the "paradoxical" dynamics that they observe, which seem exactly to reflect the behavior of the surround.

The authors' arguments to the contrary are based on three main points. First, their stimuli cover the center and surround, unlike those of many previous experiments, so they argue that this somehow diminishes the impact of the surround. But the argument is not accompanied by data showing the effects of center stimuli alone or surround stimuli alone. Second, their model -- a normalization model -- does not need surround influences to account for the masking effect. Third, they cite human psychophysical masking results from their collaborators (Sebastian et al 2017), but do not cite an equally convincing demonstration that surround contrast creates potent orientation selective masking when presented alone (Petrov et al 2005, https://doi.org/10.1523/JNEUROSCI.2871-05.2005).

At the end of the day, these issues will be resolved by further experiments, not argumentation. The paper stands as an excellent contribution, but it might be wise for the authors to be less doctrinaire in their interpretations.

We thank the reviewer for their positive comments and constructive criticism. In general, we agree with the reviewer’s comments. Importantly, we do not claim that there is no effect from the surround. What we say in the discussion is:

“Because our targets are added to the background rather than occluding it, it is likely that a significant portion of the behavioral and neural masking effects that we observe come from target-mask interactions at the target location rather than from the effect of the mask in the surround.”

We still stand by this assessment. We also make the point that, at least within the framework of our delayed normalization model, there is no need for the normalization mechanism to extend beyond the center mechanism to account for our results, and even if the normalization mechanism is somewhat larger than the center, the overlap region at the center would still have a large contribution to the modulations. Overall, we agree that these issues will be need to be resolved by future experiments.

For the reasons discussed in our previous reply, we disagree with the reviewers’ statement “…this is primarily a study of surround suppression with results that follow pretty directly from what we already know from that literature”. For similar reasons we disagree with the statement “It is nice to see that VSDI results square well with those from prior extracellular recordings”.

Reviewer #2 (Public Review):

Summary

In this experiment, Voltage Sensitive Dye Imaging (VSDI) was used to measure neural activity in macaque primary visual cortex in monkeys trained to detect an oriented grating target that was presented either alone or against an oriented mask. Monkeys' ability to detect the target (indicated by a saccade to its location) was impaired by the mask, with the greatest impairment observed when the mask was matched in orientation to the target, as is also the case in human observers. VSDI signals were examined to test the hypothesis that the target-evoked response would be maximally suppressed by the mask when it matched the orientation of the target. In each recording session, fixation trials were used to map out the spatial response profile and orientation domains that would then be used to decode the responses on detection trials. VSDI signals were analyzed at two different scales: a coarse scale of the retinotopic response to the target and a finer scale of orientation domains within the stimulus-evoked response. Responses were recorded in three conditions: target alone, mask alone, and target presented with mask. Analyses were focused on the target evoked response in the presence of the mask, defined to be the difference in response evoked by the mask with target (target present) versus the mask alone (target absent). These were computed across five 50 msec bins (total, 250 msec, which was the duration of the mask (target present trials, 50% of trials) / mask + target (target present trials, 50% of trials). Analyses revealed that in an initial (transient) phase the target evoked response increased with similarity between target and mask orientation. As the authors note, this is surprising given that this was the condition where the mask maximally impaired detection of the target in behavior. Target evoked responses in a later ('sustained') phase fell off with orientation similarity, consistent with the behavioral effect. When analyzed at the coarser scale the target evoked response, integrated over the full 250 msec period showed a very modest dependence on mask orientation. The same pattern held when the data were analyzed on the finer orientation domain scale, with the effect of the mask in the transient phase running counter to the perceptual effect of the mask and the sustained response correlating the perceptual effect. The effect of the mask was more pronounced when analyzed at the scale.

Strengths

The work is on the whole very strong. The experiments are thoughtfully designed, the data collection methods are good, and the results are interesting. The separate analyses of data at a coarse scale that aggregates across orientation domains and a more local scale of orientation domains is a strength and it is reassuring that the effects at the more localized scale are more clearly related to behavior, as one would hope and expect. The results are strengthened by modeling work shown in Figure 8, which provides a sensible account of the population dynamics. The analyses of the relationship between VSDI data and behavior are well thought out and the apparent paradox of the anti-correlation between VSDI and behavior in the initial period of response, followed by a positive correlation in the sustained response period is intriguing.

We thank the reviewer for their positive comments.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

None, except perhaps for a more balanced representation of the "surround" possibility in the Discussion. The Petrov et al paper (https://doi.org/10.1523/JNEUROSCI.2871-05.2005) should be considered and cited.

As discussed above, we believe that our discussion of possible contribution from the surround is balanced. While the paper by Petrov et al is interesting, the stimuli used to study the surround effects are quite different (e.g., gap between center and surround, and the sharp edge of the surround inner boundary) so direct comparison with our results is not possible.

Reviewer #2 (Recommendations For The Authors):

The authors have addressed the questions/suggestions I raised in my review.

---

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

We thank the reviewers for their helpful comments and suggestions.

eLife assessment

This is an important contribution that extends earlier single-unit work on orientation-specific center-surround interactions to the domain of population responses measured with Voltage Sensitive Dye (VSD) imaging and the first to relate these interactions to orientation-specific perceptual effects of masking. The authors provide convincing evidence of a pattern of results in which the initial effect of the mask seems to run counter to the behavioral effects of the mask, a pattern that reversed in the latter phase of the response. It seems likely that the physiological effects of masking reported here can be attributed to previously described signals from the receptive field surround.

We thank the reviewers for bringing up the relation of our results to findings from previous orientation-specific center-surround interactions studies. In our final manuscript, we added a paragraph discussing this important issue. Briefly, for multiple reasons, we believe that the orientation-dependent behavioral and neural masking effects that we observe are unlikely to depend on previously described center-surround interactions in V1. First, in human subjects, perceptual similarity masking effects are almost entirely accounted for by target-mask interactions at the target location and are recapitulated when the mask has the same size and location as the target (Sebastian et al 2017). Second, in our computational model, the effect of mask orientation on the dynamics of the response are qualitatively the same if the mask is restricted to the size and location of the target while mask contrast is increased (Fig. 8 – figure supplement 3). Third, in our model, the results are qualitatively the same when the spatial pooling region for the normalization signal is the same as that for the excitation signal (Fig. 8 – figure supplement figure 1). These considerations suggest that center-surround interactions may not be necessary for neural and behavioral similarity masking effects with additive targets.

We would also like to point out some key differences between the stimuli that we use and the ones used in most previous center-surround studies. First, in our experiments, the target and the mask were additive, while in most previous center-surround studies the target occludes the background. Such studies therefore restrict the mask effect to the surround, while in our study we allow target-mask interactions at the center. Second, most center-surround studies have a sharp-edged target/surround, while in our experiments no sharp edges were present. Unpublished results from our lab suggest that such sharp edges have a large impact on V1 population responses. A third key difference is that our stimuli were flashed for a short interval of 250 ms corresponding to a typical duration of a fixation in natural vision, while most previous center-surround studies used either longer-duration drifting stimuli or very short-duration random-order stimuli for reverse-correlation analysis.

In addition, we would like to emphasize that our results go beyond previous studies in two important ways. First, we study the effect of similarity masking in behaving animals and quantitatively compare the effect of similarity masking on behavior and physiology in the same subjects and at the same time. Second, VSD imaging allows us to capture the dynamics of superficial V1 population responses over the entire population of millions of neurons activated by the target at two important spatial scales. Such results therefore complement electrophysiological studies that examine the activity of a very small subset of the active neurons.

Public Reviews:

Reviewer #1 (Public Review):

This is a clear account of some interesting work. The experiments and analyses seem well done and the data are useful. It is nice to see that VSDI results square well with those from prior extracellular recordings. But the work may be less original than the authors propose, and their overall framing strikes me as odd. Some additional clarifications could make the contribution more clear.

Please see our reply above regarding the agreement with previous studies and framing.

My reading is that this is primarily a study of surround suppression with results that follow pretty directly from what we already know from that literature, and although they engage with some of the literature they do not directly mention surround suppression in the text. Their major effect - what they repeatedly describe as a "paradoxical" result in which the responses initially show a stronger response to matched targets and backgrounds and then reverse - seems to pretty clearly match the expected outcome of a stimulus that initially evokes additional excitation due to increased center contrast followed by slightly delayed surround suppression tuned to the same peak orientation. Their dynamics result seems entirely consistent with previous work, e.g. Henry et al 2020, particularly their Fig. 3 https://elifesciences.org/articles/54264, so it seems like a major oversight to not engage with that work at all, and to explain what exactly is new here.

We thank the reviewer for the pointing out this previous work which we now cite in the final version of the manuscript. For the reasons discussed above, while this study is interesting and related to our work, we believe that our results are quite distinct.

• In the discussion (lines 315-316), they state "in order to account for the reduced neural sensitivity with target-background similarity in the second phase of the response, the divisive normalization signal has to be orientation selective." I wonder whether they observed this in their modeling. That is, how robust were the normalization model results to the values of sigma_e and sigma_n? It would be useful to know how critical their various model parameters were for replicating the experimental effects, rather than just showing that a good account is possible.

Thank you for this suggestion. In the final manuscript we include a supplementary figure that shows how the model’s predictions are affected by the orientation tuning and spatial extent of the normalization signal, and by the size and contrast of the mask (Fig. 8 – figure supplement 1-4).

• The majority of their target/background contrast conditions were collected only in one animal. This is a minor limitation for work of this kind, but it might be an issue for some.

We agree that this is a limitation of the current study. These are challenging experiments and we were unable to collect all target/background contrast combinations from both monkeys. However, in the common conditions, the results appear similar in the two animals, and the key results seem to be robust to the contrast combination in the animal in which a wider range of contrast combinations was tested. We added these points to the discussion in the final manuscript.

• The authors point out (line 193-195) that "Because the first phase of the response is shorter than the second phase, when V1 response is integrated over both phases, the overall response is positively correlated with the behavioral masking effect." I wonder if this could be explored a bit more at the behavioral level - i.e. does the "similarity masking" they are trying to explain show sensitivity to presentation time?

We agree that testing the effect of stimulus duration on similarity masking is interesting, but unfortunately, it is beyond the scope of the current study. We would also like to point out that the duration of the presentation was selected to match the typical time of fixation during natural behaviors, so much shorter or much longer stimulus durations would be less relevant for natural vision.

• From Fig. 3 it looks like the imaging ROI may include some opercular V2. If so, it's plausible that something about the retinotopic or columnar windowing they used in analysis may remove V2 signals, but they don't comment. Maybe they could tell us how they ensured they only included V1?

We thank the reviewer for this comment. As part of our experiments, we extract a detailed retinotopic map for each chamber, so we were able to ensure that the area used for the decoding analysis lays entirely within V1. We now incorporate this information in the final manuscript (Fig. 3 – figure supplement 1).

• In the discussion (lines 278-283) they say "The positive correlation between the neural and behavioral masking effects occurred earlier and was more robust at the columnar scale than at the retinotopic scale, suggesting that behavioral performance in our task is dominated by columnar scale signals in the second phase of the response. To the best of our knowledge, this is the first demonstration of such decoupling between V1 responses at the retinotopic and columnar scales, and the first demonstration that columnar scale signals are a better predictor of behavioral performance in a detection task." I am having trouble finding where exactly they demonstrate this in the results. Is this just by comparison of Figs. 4E,K and 5E,K? I may just be missing something here, but the argument needs to be made more clearly since much of their claim to originality rests on it.

We thank the reviewer for this comment. In the final manuscript we are more explicit when we discuss this point and refer to the relevant panels in Figs. 4, 5 and their figure supplements. To substantiate this key claim, we also report the timing of the transition between the two phases in all temporal correlation panels and report the neural-behavioral correlation for the integration period.

Reviewer #2 (Public Review):

Summary

In this experiment, Voltage Sensitive Dye Imaging (VSDI) was used to measure neural activity in macaque primary visual cortex in monkeys trained to detect an oriented grating target that was presented either alone or against an oriented mask. Monkeys' ability to detect the target (indicated by a saccade to its location) was impaired by the mask, with the greatest impairment observed when the mask was matched in orientation to the target, as is also the case in human observers. VSDI signals were examined to test the hypothesis that the target-evoked response would be maximally suppressed by the mask when it matched the orientation of the target. In each recording session, fixation trials were used to map out the spatial response profile and orientation domains that would then be used to decode the responses on detection trials. VSDI signals were analyzed at two different scales: a coarse scale of the retinotopic response to the target and a finer scale of orientation domains within the stimulus-evoked response. Responses were recorded in three conditions: target alone, mask alone, and target presented with mask. Analyses were focused on the target evoked response in the presence of the mask, defined to be the difference in response evoked by the mask with target (target present) versus the mask alone (target absent). These were computed across five 50 msec bins (total, 250 msec, which was the duration of the mask (target present trials, 50% of trials) / mask + target (target present trials, 50% of trials). Analyses revealed that in an initial (transient) phase the target evoked response increased with similarity between target and mask orientation. As the authors note, this is surprising given that this was the condition where the mask maximally impaired detection of the target in behavior. Target evoked responses in a later ('sustained') phase fell off with orientation similarity, consistent with the behavioral effect. When analyzed at the coarser scale the target evoked response, integrated over the full 250 msec period showed a very modest dependence on mask orientation. The same pattern held when the data were analyzed on the finer orientation domain scale, with the effect of the mask in the transient phase running counter to the perceptual effect of the mask and the sustained response correlating the perceptual effect. The effect of the mask was more pronounced when analyzed at the scale.

Strengths

The work is on the whole very strong. The experiments are thoughtfully designed, the data collection methods are good, and the results are interesting. The separate analyses of data at a coarse scale that aggregates across orientation domains and a more local scale of orientation domains is a strength and it is reassuring that the effects at the more localized scale are more clearly related to behavior, as one would hope and expect. The results are strengthened by modeling work shown in Figure 8, which provides a sensible account of the population dynamics. The analyses of the relationship between VSDI data and behavior are well thought out and the apparent paradox of the anti-correlation between VSDI and behavior in the initial period of response, followed by a positive correlation in the sustained response period is intriguing.

Points to Consider / Possible Improvements

The biphasic nature of the relationship between neural and behavioral modulation by the mask and the surprising finding that the two are anticorrelated in the initial phase are left as a mystery. The paper would be more impactful if this mystery could be resolved.

We thank the reviewer for the positive comments. In our view, while our results are surprising, there may not be a remaining mystery that needs to be resolved. As our model shows, the biphasic nature of V1’s response can be explained by a delayed orientation-tuned gain control. Our results are consistent with the hypothesis that perception is based on columnar-scale V1 signals that are integrated over an approximately 200 ms long period that incorporates both the early and the late phase of the response, since such decoded V1 signals are positively correlated with the behavioral similarity masking effect (Fig. 5D, J; Fig. 5 – figure supplement 1). We now explain this more clearly in the discussion of our final manuscript.

The finding is based on analyses of the correlation between behavior and neural responses. This appears in the main body of the manuscript and is detailed in Figures S1 and S2, which show the correlation over time between behavior and target response for the retinotopic and columnar scale.

One possible way of thinking of this transition from anti- to positive correlation with behavior is that it might reflect the dynamics of a competitive interaction between mask and target, with the initial phase reflecting predominantly the mask response, with the target emerging, on some trials, in the latter phase. On trials when the mask response is stronger, the probability of the target emerging in the latter phase, and triggering a hit, might be lower, potentially explaining the anticorrelation in the initial phase. The sustained response may be a mixture of trials on which the target response is or is not strong enough to overcome the effect of the mask sufficiently to trigger target detection.

It would, I think, be worth examining this by testing whether target dynamics may vary, depending on whether the monkey detected the target (hit trials) or failed to detect the target (miss trials). Unless I missed it I do not think this analysis was done. Consistent with this possibility, the authors do note (lines 226-229) that "The trajectories in the target plus mask conditions are more complex. For example, when mask orientation is at +/- 45 deg to the target, the population response is initially dominated by the mask, but then in mid-flight, the population response changes direction and turns toward the direction of the target orientation." This suggests (to this reviewer, at least) that the emergence of a positive correlation between behavioral and neural effects in the latter phase of the response could reflect either a perceptual decision that the target is present or perhaps deployment of attention to the location of the target.

It may be that this transition reflected detection, in which it might be more likely on hit trials than miss trials. Given the SNR it would presumably be difficult to do this analysis on a trial-by-trial basis, but the hit and miss trials (which make each make up about 1/2 of all trials) could be averaged separately to see if the mid-flight transition is more prominent on hit trials. If this is so for the +/- 45 degree case it would be good to see the same analysis for other combinations of target and mask. It would also be interesting to separate correct reject trials from false alarms, to determine whether the mid-flight transition tends to occur on false alarm trials.

If these analyses do not reveal the predicted pattern, they might still merit a supplemental figure, for the sake of completeness.

We thank the reviewer for suggesting this interesting possibility. The original analysis in the manuscript was based on both correct and incorrect trials, raising the possibility that our results reflect some contribution from decision- and/or attention-related signals rather than from low-level nonlinear encoding mechanisms in V1 that we postulate in our model (Fig. 8). To explore this possibility, we re-examined our results while excluding error trials. We found that our key results from Figs 4 and 5 – namely that there is an early transient phase in which the neural and behavioral similarity effects are anti-correlated, and a later sustained phase in which they are positively correlated – hold even for the subset of correct trials, reducing the possibility that decision/attention-related signals play a major role in explaning our results. We now include the results of this analysis as a supplementary figure in the final manuscript (Fig. 4 – figure supplement 2). While there may be some interesting differences in the response dynamics between correct and incorrect trials, the current study was not designed to address this question and the large number of conditions and small number of repeats that it necessitated make this data set suboptimal for examining these phenomena.

References

Sebastian S, Abrams J, Geisler WS. 2017. Constrained sampling experiments reveal principles of detection in natural scenes. Proc Natl Acad Sci U S A 114: E5731-e40

Reviewer #1 (Public Review):

This is a clear account of some interesting work. The experiments and analyses seem well done and the data are useful. It is nice to see that VSDI results square well with those from prior extracellular recordings.

The authors have done a good job responding to the main points of my previous review. One important question remains, as stated in that review:

"My reading is that this is primarily a study of surround suppression with results that follow pretty directly from what we already know from that literature, and although they engage with some of the literature they do not directly mention surround suppression in the text. Their major effect - what they repeatedly describe as a "paradoxical" result in which the responses initially show a stronger response to matched targets and backgrounds and then reverse - seems to pretty clearly match the expected outcome of a stimulus that initially evokes additional excitation due to increased center contrast followed by slightly delayed surround suppression tuned to the same peak orientation. Their dynamics result seems entirely consistent with previous work, e.g. Henry at al 2020, particularly their Fig. 3 https://elifesciences.org/articles/54264, so it seems like a major oversight to not engage with that work at all, and to explain what exactly is new here."

Their rebuttal of my first review is not convincing -- I still believe that surround influences are important and perhaps predominant in determining the outcome of the experiments. This is particularly clear for the "paradoxical" dynamics that they observe, which seem exactly to reflect the behavior of the surround.

The authors' arguments to the contrary are based on three main points. First, their stimuli cover the center and surround, unlike those of many previous experiments, so they argue that this somehow diminishes the impact of the surround. But the argument is not accompanied by data showing the effects of center stimuli alone or surround stimuli alone. Second, their model -- a normalization model -- does not need surround influences to account for the masking effect. Third, they cite human psychophysical masking results from their collaborators (Sebastian et al 2017), but do not cite an equally convincing demonstration that surround contrast creates potent orientation selective masking when presented alone (Petrov et al 2005, https://doi.org/10.1523/JNEUROSCI.2871-05.2005).

At the end of the day, these issues will be resolved by further experiments, not argumentation. The paper stands as an excellent contribution, but it might be wise for the authors to be less doctrinaire in their interpretations.

Reviewer #2 (Public Review):

Summary

In this experiment, Voltage Sensitive Dye Imaging (VSDI) was used to measure neural activity in macaque primary visual cortex in monkeys trained to detect an oriented grating target that was presented either alone or against an oriented mask. Monkeys' ability to detect the target (indicated by a saccade to its location) was impaired by the mask, with the greatest impairment observed when the mask was matched in orientation to the target, as is also the case in human observers. VSDI signals were examined to test the hypothesis that the target-evoked response would be maximally suppressed by the mask when it matched the orientation of the target. In each recording session, fixation trials were used to map out the spatial response profile and orientation domains that would then be used to decode the responses on detection trials. VSDI signals were analyzed at two different scales: a coarse scale of the retinotopic response to the target and a finer scale of orientation domains within the stimulus-evoked response. Responses were recorded in three conditions: target alone, mask alone, and target presented with mask. Analyses were focused on the target evoked response in the presence of the mask, defined to be the difference in response evoked by the mask with target (target present) versus the mask alone (target absent). These were computed across five 50 msec bins (total, 250 msec, which was the duration of the mask (target present trials, 50% of trials) / mask + target (target present trials, 50% of trials). Analyses revealed that in an initial (transient) phase the target evoked response increased with similarity between target and mask orientation. As the authors note, this is surprising given that this was the condition where the mask maximally impaired detection of the target in behavior. Target evoked responses in a later ('sustained') phase fell off with orientation similarity, consistent with the behavioral effect. When analyzed at the coarser scale the target evoked response, integrated over the full 250 msec period showed a very modest dependence on mask orientation. The same pattern held when the data were analyzed on the finer orientation domain scale, with the effect of the mask in the transient phase running counter to the perceptual effect of the mask and the sustained response correlating the perceptual effect. The effect of the mask was more pronounced when analyzed at the scale.

Strengths

The work is on the whole very strong. The experiments are thoughtfully designed, the data collection methods are good, and the results are interesting. The separate analyses of data at a coarse scale that aggregates across orientation domains and a more local scale of orientation domains is a strength and it is reassuring that the effects at the more localized scale are more clearly related to behavior, as one would hope and expect. The results are strengthened by modeling work shown in Figure 8, which provides a sensible account of the population dynamics. The analyses of the relationship between VSDI data and behavior are well thought out and the apparent paradox of the anti-correlation between VSDI and behavior in the initial period of response, followed by a positive correlation in the sustained response period is intriguing.

eLife assessment

This study provides valuable new insights as to how two evolutionary conserved motifs in CD4 contribute to the CD4-mediated enhancement of TCR signaling independently of the CD4-LCK interaction. The data at hand are convincing, even if confined to a cell line model and not substantiated in vivo and with little new mechanistic insight provided regarding the domains of CD4 shown to have significant roles in the signaling process. Without the data from primary cells it is difficult to make statements about the quantitative contribution of LCK-dependent and independent functions of CD4 in TCR signaling.

Author Response

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

Reviewer #1 (Public Review):

Summary:

This study by Lee et al. is a direct follow-up on their previous study that described an evoluBonary conservancy among placental mammals of two moBfs (a transmembrane moBf and a juxtamembrane palmitoylaBon site) in CD4, an anBgen co-receptor, and showed their relevance for T-cell anBgen signaling. In this study, they describe the contribuBon of these two moBfs to the CD4-mediated anBgen signaling in the absence of CD4-LCK binding. Their approach was the comparison of anBgen-induced proximal TCR signaling and distal IL-2 producBon in 58-/- T-cell hybridoma expressing exogenous truncated version of CD4 (without the interacBon with LCK), called T1 with T1 version with the mutaBons in either or both of the conserved moBfs. They show that the T1 CD4 can support signaling to the extend similar to WT CD4, but the mutaBon of the conserved moBfs substanBally reduced the signaling. The authors conclude that the role of these moBfs is independent of the LCK-binding.

Strengths:

The authors convincingly show that T1 CD4, lacking the interacBon with LCK supports the TCR signaling and also that the two studied moBfs have a significant contribuBon to it.

Weaknesses:

The study has several weaknesses.

(1) The whole study is based on a single experimental system, geneBcally modified 58-/- hybridoma. It is unclear at this moment, how the molecular moBfs studied here contribute to the signaling in a real T cell. The evoluBonary conservancy suggests that these moBfs are important for T cell biology. However, the LCK-binding moBf is conserved as well (perhaps even more) and it plays a very minor role in their model. Without verifying their results in primary cells, the quanBtaBve, but even qualitaBve, importance of these moBfs for T-cell signaling and biology is unclear. Although the authors discuss this issue in the Discussion, it should be noted in all important parts of the manuscript, where conclusions are made (abstract, end of introducBon, perhaps also in the Btle) that the results are coming from the hybridoma cells.

We appreciate the Reviewer’s thoughWul comments and suggesBon. We now state in the abstract and introducBon that wet-lab experiments were performed with T cell hybridomas. We have also beXer highlighted work from Killeen and LiXman (PMID: 8355789) wherein they showed that C-terminally truncated CD4, which lacked the moBfs that mediate CD4-Lck interacBons, can drive CD4+ T cell development, proliferaBon, and T-helper funcBon because we now provide mechanisBc data to help explain those in vivo results. Also, as noted by the reviewer, we discuss how the sum of our data provides jusBficaBon for the investment in and use of mouse models to interrogate how the funcBonally important residues/moBfs idenBfied and studied here influence T cell biology.

We will take the opportunity to reiterate here that, while the study is based on a well characterized, albeit single, wet-lab experimental system, the whole study is based on two lines of invesBgaBon. The other approach was a systems biology computaBonal approach that analyzes data from real-world experiments in a variety of jawed vertebrate species over evoluBon. Specifically, we used a computaBonal reconstrucBon of the evoluBonary history of CD4 by performing mulBple analyses of CD4 from 99 jawed vertebrates spanning ~435 million years of evoluBon. This analysis allowed us to idenBfy residues, and networks of evoluBonarily coupled residues, that are predicted to be funcBonally important in vivo. Like other systems biology approaches, this allowed us to look at the larger picture by evaluaBng data points that have emerged from constant tesBng and adjustments of CD4 funcBon in vivo through selecBon on an evoluBonary Bmescale in more jawed vertebrate species, and under more real-world condiBons, than can be tested in the laboratory. Our structure-funcBon analysis provided a second, wet-lab reducBonist experimental system to cross-validate that the residues idenBfied by our evoluBonary analysis are funcBonally significant. This experimental validaBon is criBcal and elevates the relevance of our studies above ad hoc observaBons. Our work also provides mechanisBc insights for why the residues studied here are funcBonally significant (i.e., key determinants of pMHCII-specific signaling iniBaBon). In short, using both systems allowed us to cross-validate the funcBonal significance of the residues within the GGXXG and (C/F)CV+C moBfs studied here by two independent methods.

(2) Many of the experiments lack the negaBve control. I believe that two types of negaBve controls should be included in all experiments. First, hybridoma cells without CD4 (or with CD4 mutant unable to bind MHCII). Second, no pepBde control, i.e., acBvaBon of the hybridoma cells with the APC not loaded with the cognate pepBde. These controls are required to disBnguish the basal levels of phoshorylaBon and CD4-independent anBgen-induced phosphorylaBon to quanBfy, what is the contribuBon of the parBcular moBfs to the CD4-mediated support. Although these controls are included in some of the experiments, they are missing in other ones. The binding mutant appears in some FC results as a horizontal bar (without any error bar/variability), showing that CD4 does not give a huge advantage in these readouts. Why don't the authors show no pepBde controls here as well? Why the primary FC data (histograms) are not shown? Why neither of these two controls is shown for the % of responders plots? Although the IL-2 producBon is a very robust and convincing readout, the phosphoflow is much less sensiBve. It seems that the signaling is elevated only marginally. Without the menBoned controls and showing the raw data, the precise interpretaBon is not possible.

These comments, and those in point #3, concern our flow cytometry-based analysis of early intracellular signaling events where we asked: how do the moBfs under invesBgaBon impact phosphorylaBon of CD3z, ZAP-70, and PLCg1 in response to agonist pMHCII? Thank you for poinBng out areas of confusion regarding these analyses. We will try to clarify here and have worked to clarify the text.

Our approach was to mutate consBtuent residues within the moBfs that our evoluBonary analysis predicted to be funcBonally significant, compare the performance of the mutants to that of controls bearing WT moBfs, and then infer the funcBon of the moBfs based on the differenBal phenotype of the mutants relaBve to their controls. In most cases, the C-terminally truncated CD4-T1 mutant served as the appropriate CD4 control backbone against which to evaluate the phenotypes of the GGXXG and (C/F)CV+C moBf mutants. This is a convenBonal structure-funcBon strategy.

All experiments included APCs expressing null pMHCII (Hb:I-Ek) as negaBve controls. These were a necessary component of the data analysis, explained further below, which involved background subtracBon of the signal from control or mutant T cell hybridomas bound to these negaBve control APCs from those bound to the agonist pMHCII (MCC:I-Ek). Doing so allowed us to establish a true signal over background for calculaBng percent responders and signaling intensity. These negaBve controls served the same purpose of APCs expressing I-Ek not loaded with cognate pepBde requested by the reviewer. It is important to note that we previously published that TCR-CD3-pMHCII interacBons reciprocally increase CD4-pMHCII dwell Bme, and vice versa, such that dwell Bmes of the 5c.c7 TCR and CD4 to the null Hb:I-Ek are both basal in this system relaBve to antagonist, weak agonist, and agonist pMHCII (PMID 29386113). A recent study using different techniques also concluded that TCR-CD3 and CD4 cooperaBvely enhance signaling to pMHCII (PMID 36396644). The use of the null pMHCII, Hb:I-Ek, in each experiment thus serves as a well-characterized negaBve control for both TCR and CD4 engagement in this experimental system with regards to assembly of the TCR-CD3 and CD4 around pMHCII to drive signaling. In our view, it is the most important negaBve control for interpreBng our results, and it is present in each experiment. In Fig 1B and related supplemental figures we compare the Cterminally truncated CD4-T1 mutant to the full-length WT CD4 to evaluate the contribuBons of the intracellular domains to early signaling events. We found no significant differences for pCD3z, pZAP-70, and pPLCg1 levels demonstraBng that, in our system, CD4 WT and T1 are staBsBcally indisBnguishable.

In Fig 1C we asked: what is the contribuBon of CD4-pMHCII interacBons made by CD4 T1, which lacks the intracellular domain, using our CD4 T1Dbind mutant. Fig 2C and Table 3 show that pCD3z levels for T1Dbind were ~54% of T1, meaning that CD4 binding to pMHCII roughly doubles pCD3z levels (even without the intracellular domain). We also showed that the percent of responders were not different between the CD4 T1 and T1Dbind mutant in Fig 2C. The impact on ZAP-70 and PLCg1 are shown in Figure 2—figure supplement 4. These differences, including the magnitude of the decrease, were observed reproducibly (p<0.001) in three independently generated sets of lines. We believe that this analysis saBsfies the request by the reviewer for an analysis of the contribuBons of CD4 binding to pMHCII. We did not include this as a negaBve control in experiments evaluaBng the contribuBons of the GGXXG and (C/F)CV+C moBfs to CD4 T1 signaling because the quesBon being asked in those experiments was how do the moBfs impact signaling in the absence of the intracellular domain (i.e., within the CD4 T1 backbone, making CD4 T1 the proper comparator for the quesBon we were asking). We showed the average normalized intensity for the T1Dbind mutant, relaBve to T1, for this lower bound of signaling mediated by TCR-CD3-only as a doXed line in those figures to provide a reference point for the readers to evaluate and put into perspecBve how the mutants we generated impacted the overall contribuBon of CD4 to these early signaling events. The T1Dbind mutants were not always measured in the same experiment at the same Bme with other mutants, because the cell lines used were not always made at the same Bme, so we did not think it appropriate to graph the results together.

We do not know how to interpret the comment “Although the IL-2 producBon is a very robust and convincing readout, the phosphoflow is much less sensiBve. It seems that the signaling is elevated only marginally.” We will offer our perspecBve that we do not know how to equate the sensiBvity of the phos-flow to the IL-2. Because the IL-2 is a signaling output, it results from signaling amplificaBon from the membrane to the nucleus. If CD3z phosphorylaBon is the iniBaBng event for a signaling cascade that leads to IL-2 gene transcripBon and transducBon, as is widely believed, our data strongly suggests that the ~2-fold difference in pCD3z levels between CD4 T1 and T1Dbind (Fig 2C/Table 3 data) contributes to the difference between no IL-2 output for T1Dbind and IL-2 output by T1 in this experimental system. Because CD4 WT and T1 have significantly different levels of IL-2 output, but show no significant differences in pCD3z, pZAP-70, or pPLCg1 levels, there are likely to be other differences we did not measure via other pathways that intersect at the nucleus. At many levels, biology works on gradients such that small differences can Bp a system in one direcBon or another. The kineBc discriminaBon model (PMID 8643643), which is thought to be a reasonable descripBon of the relaBonship between pMHC engagement and signaling outcomes, suggests that very small differences in molecular interacBons at the earliest stages of a response can lead to big differences in signaling outcome. We therefore have no basis at this juncture to think that ~2-fold differences in pCD3z levels could not account for bigger differences in signaling output such as IL-2.

(3) The processing of the data is not clear. Some of the figures seem to be overprocessed. For instance, I am not sure what "Normalized % responders of pCD3zeta" means (e.g., Fig. 1C and elsewhere)? Why do not the authors show the actual % of pCD3zeta+ cells including the gaBng strategy? Why do the authors subtract the two histograms in Fig. 2- Fig.S3? It is very unusual.

We did develop and implement a novel strategy for measuring the impact of our mutaBons on CD3z, ZAP-70, and PLCg1 phosphorylaBon. This was explained in more detail in our prior study. The instrucBons to authors indicated that we should not repeat methods in the current manuscript. However, we will go through the approach here, and address why we did not show primary FC histograms for all experiments from above. First, we think that a brief explanaBon as to what moBvated us to develop our approach will add to a beXer understanding:

(1) For experimental and staBsBcal rigor, our goal was to perform both experimental and biological replicates by measuring and comparing the average of at least three independently generated sets of paired WT/T1 control Vs. mutant cells lines generated at different Bmes to determine the staBsBcal significance of the difference, if any, between averages of the control and mutant lines.

(2) Our quesBons necessitated that we measure signals generated naturally by the cooperaBve engagement of cognate pMHCII by TCR-CD3 and CD4 on APCs, rather than through aCD3/aCD4 crosslinking.

(3) We chose to use flow cytometry rather than bulk cell analysis by Western Bloung to analyze signaling occurring in cells that were engaged to the agonist APC in order to avoid diluBon of that signal by cells that are not engaged to APCs and not signaling. 4. For each experiment, we wanted to subtract background signals from cells bound to APCs expressing a null pMHCII (Hb:I-Ek) from signals generated by cells bound to APCs expressing agonist pMHCII (MCC:I-Ek). Doing so allowed us to idenBfy cells that are signaling (responders) to agonist over null pMHCII. The goal here was to quanBtate the level of signaling in an objecBve manner with a method that can be applied to all samples uniformly rather than seung a flow cytometry gate on posiBve cells (e.g. pCD3z) because gaBng is subjecBve and can vary from experiment to experiment. To put that another way, as detailed below, we used our subtracBon method to idenBfy signaling responders rather than seung a signaling gate on the posiBve populaBon.

Regarding gaBng schemes, controls, and data processing:

Figure 2—figure supplement 3 of the current study and Figure 6—figure supplement 1 of our prior study are designed to walk the reader through our experimental design, gaBng, data processing and thinking. Here we will provide a detailed explanaBon to complement the figure legend as well as the methods provided in our prior manuscript (see pt #4 below).

We will refer to Figure 2—figure supplement 3 here:

Panel A. The dot plots show our approach to idenBfying 5c.c7+ CD4+ 58a-b- T cell hybridomas (yaxis, GFP posiBve) coupled to M12 cells (x-axis, TagIt Violet) expressing the null pMHCII Hb:I-Ek (lev) or agonist pMHCII MCC:I-Ek (right). The gaBng shows the frequency of GFP+ T cell hybridomas that are bound to TagIt violet posiBve APCs (i.e., cell couples). The histogram on the right then shows the staining intensity for pCD3z on the x-axis for the 10,000 coupled events collected wherein the APCs express the null pMHCII (filled cyan) or the agonist pMHCII (black line).

Panel B. The data presented here is the same as in Panel A, but for CD4 T1 cells.

Panel C. The data presented here walks through how we idenBfy 5c.c7+ CD4+ 58a-b- T cell hybridomas responding (i.e., signaling) to agonist pMHCII, as well as the mean signaling intensity of the responding populaBon, in a gaBng-independent manner aver background subtracBon. For the lev graph, we exported the data for the histograms shown in Panel A from FlowJo 10 sovware and ploXed them here using Prism 9 as smoothed lines (500 nearest neighbors). The cyan line is therefore a replicate of the flow cytometry histogram shown in Panel A for pCD3z intensity from 5c.c7+ CD4+ 58a-b- T cell hybridomas coupled to M12 cells expressing the null pMHCII (Hb:I-Ek), while the black histogram is a replicate of the pCD3z intensity for 5c.c7+ CD4+ 58a-b- T cell hybridomas coupled to M12 cells expressing the agonist pMHCII (MCC:I-Ek). Next, to idenBfy the responding populaBon in a gaBng-independent manner, we used Excel to subtract the pCD3z intensity for the null pMHCII (cyan) negaBve control populaBon on a bin-by-bin bases from the pCD3z intensity for the agonist pMHCII (black) responding populaBon. We then transferred the background subtracted values to Prism 9 for smoothing and ploung (grey line: MCC:I-Ek minus Hb:I-Ek). The middle graph shows the same data processing for the data from Panel B for the CD4 T1 cells. Please note that the background subtracted grey line has negaBve values and posiBve values. The negaBve values represent intensity bins where signaling in response to agonist pMHCII leads to fewer cells per bin than in the null pMHCII populaBon that is not signaling, while the posiBve values represent bins of intensity where signaling cells outnumber non-signaling cells. The right graph in this panel shows the populaBons aver background subtracBon for intensity bins that had more cells with pCD3z signal in the agonist pMHCII populaBon than the null pMHCII populaBon (grey = WT full length CD4 and blue = T1). In short, the right graph shows idenBficaBon of those cells that are signaling in response to agonist pMHCII. This approach miBgated the need for subjecBve gaBng in FlowJo to idenBfy signaling cells (i.e., pCD3z posiBve) and allowed for background subtracBon which could not be done in FlowJo. We used this approach for all analyses of pCD3z, pZAP-70, and pPLCg1 in this study.

The number of cells in these background-subtracted populaBons were divided by 10,000 (the number of events collected and analyzed) to calculate the percent of responding 5c.c7+ CD4+ 58a-b- T cell hybridomas, while the mean fluorescent intensity for the cells within these populaBon represent the signaling intensity.

Panel D. The graph on the lev shows the mean fluorescence intensity (MFI) ± SEM for the posiBve signaling populaBon from the right graph of panel C. We see in this example comparing a WT and T1 cell line, generated at the same Bme from the same parental 58a-b- T cell hybridoma populaBon, that the T1 MFI is significantly greater than the WT. These intensity values represent one of the paired intensity values used in the main Fig 2B (Lev graph), where we show the paired MFI analysis of responding populaBons from 5 independently generated sets of cell lines. Please note that these single MFI values are directly derived from the flow cytometry histograms aver background subtracBon. Figure 2B, and similar figures, therefore equate to a disBllaBon of all of the histograms for the populaBons tested in a manner that we consider easier to digest than either overlaying all histograms or showing mulBple panels individually. It also conserves more space. This is why we only showed representaBve flow cytometry histograms, rather than all histograms.

The graph on the right shows the % responders for the posiBve signaling populaBon from the right graph of panel C. Specifically, the total number of cells that were determined to be signaling in response to agonist pMHCII was divided by 10,000 (the number of coupled cells collected by flow cytometry) to determine the percent responders. These values represent one of five sets of values used to determine the average normalized percent responders (all normalized to WT). There was no significant difference between these two populaBons in terms of percent responders.

Regarding graphing normalized values for the mean MFI for signaling intensity or the percent responders: in our first manuscript, we presented the individual MFI intensity values for matched pairs of cells as well as the actual percent responders per group. The feedback we received from colleagues on this presentaBon was that it was confusing, distracBng, and otherwise hard to digest. It was suggested to us by mulBple individuals that the normalized values would be preferable because it is easier and faster to understand. Upon reflecBon, we agreed with this feedback because the normalized presentaBon with staBsBcs allows for the two key relevant quesBons to be quickly evaluated: 1. Are the mutants different than the control? 2. By how much? We have lev the raw intensity values and well as the normalized intensity values in the version of record. Given the Reviewer’s comments, we have now graphed the average % responders instead of normalized values in the figures, and lev the normalized values in Table 3.

(4) The manuscript lacks Materials and Methods. It only refers to the previous paper, which is very unusual. Although most of the methods are the same, they sBll should be menBoned here. Moreover, some of the mutants presented here were not generated in the previous study, as far as I understand. Perhaps the authors plan to include Materials and Methods during the revision...

Because we submiXed this as a Research Advances arBcle we followed the journal instrucBons to reference the Materials and Methods in our prior publicaBon, upon which this work builds, as the methods used are the same. They are detailed in that study. We have now included a copy of the Materials and Methods for the eLife staff to determine how best to link with this manuscript. We have also included the gene sequences for the novel constructs used in this study. Thank you for poinBng out the omission.

(5) Membrane rafts are a very controversial topic. I recommend the authors stick to the more consensual term "detergent resistant microdomains" in all cases/occurances.

We agree this is a controversial topic with a variety of viewpoints. Because we are not experts in the field of membrane composition, we turned to the literature to inform our view of how best to refer to these membrane subdomains. In our reading, we found a 2006 meeting report from a Keystone symposium on lipid rafts and cell function authored by Linda Pike (PMID 16645198). At this meeting, a central focus was reaching a consensus on how best to refer to these domains. The consensus term agreed upon by this group was “membrane rafts”. Specifically, we will quote from this report published in the Journal of Lipid Research, ‘Together, the discussions permitted the generation of a definition for “lipid rafts” in an ad hoc session on the final day of the meeting. All participants were invited to contribute to this effort, and the work product reflects the consensus of this broad-based group…… First and foremost, the term “lipid raft” was discarded in favor of the term “membrane raft.”’ We chose to use the term “membrane raft” based on this consensus opinion.

(6) Last, but not least, the mechanistic explanation (beyond the independence of LCK binding) of the role of these motifs is very unclear at the moment.

We agree with this comment. One goal in making these results, and those in our prior study, available to the field at large is to provide evidence in support of our view that the dominant paradigm that is thought to explain the earliest events in T cell signaling needs re-evaluating. How T cell signaling is initiated in response to pMHCII is clearly more complex than is currently thought. However, out data is inconsistent with the dominant paradigm in which CD4 recruits Lck to TCR-CD3 to phosphorylate ITAMs to initiate signaling.

Reviewer #2 (Public Review):

Summary:

The paper by Kuhn and colleagues follows upon a 2022 eLife paper in which they identified residues in CD4 constrained by evolutionary purifying selection in placental mammals and then performed functional analyses of these conserved sequences. They showed that sequences distinct from the CXC "clamp" involved in recruitment of Lck have critical roles in TCR signaling, and these include a glycine-rich motif in the transmembrane (TM) domain and the cyscontaining juxtamembrane (JM) motif that undergoes palmitoylation, both of which promote TCR signaling, and a cytoplasmic domain helical motif, also involved in Lck binding, that constrains signaling. Mutations in the transmembrane and juxtamembrane sequences led to reduced proximal signaling and IL-2 production in a hybridoma's response to antigen presentation, despite retention of abundant CD4 association with Lck in the detergent-soluble membrane fraction, presumably mislocalized outside of lipid rafts and distal to the TCR. A major conclusion of that study was that CD4 sequences required for Lck association, including the CXC "clasp" motif, are not as consequential for CD4 co-receptor function in TCR signaling as the conserved TM and JM motifs. However, the experiments did not determine whether the functions of the TM and JM motifs are dependent on the Lck-binding properties of CD4 - the mutations in those motifs could result in free Lck redistributing to associate with CD4 in signaling-incompetent membrane domains or could function independently of CD4-Lck association. The current study addresses this specific question.

Using the same model system as in the earlier eLife paper (the entire methods section is a citation to the earlier paper), the authors show that truncation of the Lck-binding intracellular domain resulted in a moderate reduction in IL-2 response, as previously shown, but there was no apparent effect on proximal phosphorylation events (CD3z, Lck, ZAP70, PLCg1). They then evaluated a series of TM and JM motif mutations in the context of the truncated Lck-nonbinding molecule, and showed that these had substantially impaired co-receptor function in the IL-2 assay and reduced proximal signaling. The proximal signaling could be observed at high ligand density even with a MHC non-binding mutation in CD4, although there was still impaired IL-2 production. This result additionally illustrates that phosphorylation of the proximal signaling molecules is not sufficient to activate IL-2 expression in the context of antigen presentation.

Strengths:

The strength of the paper is the further clear demonstration that the classical model of CD4 coreceptor function (MHCII-binding CD4 bringing Lck to the TCR complex, for phosphorylation of the CD3 chain ITAMs and of the ZAP70 kinase) is not sufficient to explain TCR activation. The data, combined with the earlier eLife paper, further implicate the gly-rich TM sequence and the palmitoylation targets in the JM region as having critical roles in productive co-receptordependent TCR activation.

Weaknesses:

The major weakness of the paper is the lack of mechanistic insight into how the TM and JM motifs function. The new results are largely incremental in light of the earlier paper from this group as well as other literature, cited by the authors, that implicates "free" Lck, not associated with co-receptors, as having the major role in TCR activation. It is clear that the two motifs are important for CD4 function at low pMHCII ligand density. The proposal that they modulate interactions of TCR complex with cholesterol or other membrane lipids is an interesting one, and it would be worth further exploring by employing approaches that alter membrane lipid composition. The JM sequence presumably dictates localization within the membrane, by way of palmitoylation, which may be critical to regulate avidity of the TCR:CD4 complex for pMHCII or TCR complex allosteric effects that influence the activation threshold. Experiments that explore the basis of the mutant phenotype could substantially enhance the impact of this study.

We appreciate these thoughtful comments and suggestions. We will restate what we wrote in our preliminary response to the reviews to explain the scope of the current study:

To address comments about the limited scope of this study and referencing of the Methods secBon to our prior study, we would like to note that we submiXed the current study via the Research Advance mechanism. Our goal was to build upon the conclusions of our 2022 eLife publicaBon (PMID: 35861317) and address an unresolved quesBon from that study (as nicely summarized by Reviewer #2). In the current manuscript we present data from reducBonist experiments that were designed specifically for this purpose and, as noted by the reviewers, we provide answers to the quesBon being asked. We think that the Research Advance mechanism is an ideal opportunity to make these results available to the field given the stated purpose of such arBcles (for reference: “A Research Advance might use a new technique or a different experimental design to generate results that build upon the conclusions of the original research by, for example, providing new mechanis=c insights or extend the pathway under inves=ga=on…”). Now that we have provided evidence that CD4 does not recruit Lck to phosphorylate TCR-CD3 ITAMs in our system, nor do the GGXXG and (C/F)CV+C motifs play a role in enabling CD4 to regulate Lck proximity to TCR-CD3, we agree that it is important to form and test alternative hypotheses for how TCR-CD3 signaling is initiated.

Reviewer #1 (Public Review):

Summary:

This study by Lee et al. is a direct follow-up on their previous study that described an evolutionary conservancy among placental mammals of two motifs (a transmembrane motif and a juxtamembrane palmitoylation site) in CD4, an antigen co-receptor, and showed their relevance for T-cell antigen signaling. In this study, they describe the contribution of these two motifs to the CD4-mediated antigen signaling in the absence of CD4-LCK binding. Their approach was the comparison of antigen-induced proximal TCR signaling and distal IL-2 production in 58-/- T-cell hybridoma expressing exogenous truncated version of CD4 (without the interaction with LCK), called T1 and T1 version with the mutations in either or both of the conserved motifs. They show that the T1 CD4 can support signaling to extend similar to WT CD4, but the mutation of the conserved motifs substantially reduced the signaling. The authors conclude that the role of these motifs is independent of the LCK-binding.

Strengths:

The authors convincingly show that CD4 is capable of contributing to TCR signaling in a manner independent of LCK, but dependent on the two studied motifs in CD4.

Weaknesses:

(1) Experiments in primary T cells are required to estimate the relative contribution of LCK-dependent and LCK-independent mechanisms of CD4 signaling.

(2) The mechanistic explanation (beyond the independence of LCK binding) of the role of these motifs is unclear at the moment.

Reviewer #2 (Public Review):

Summary:

The paper by Kuhn and colleagues follows upon a 2022 eLife paper in which they identified residues in CD4 constrained by evolutionary purifying selection in placental mammals, and then performed functional analyses of these conserved sequences. They showed that sequences distinct from the CXC "clamp" involved in recruitment of Lck have critical roles in TCR signaling, and these include a glycine-rich motif in the transmembrane (TM) domain and the cys-containing juxtamembrane (JM) motif that undergoes palmitylation, both of which promote TCR signaling, and a cytoplasmic domain helical motif, also involved in Lck binding, that constrains signaling. Mutations in the transmembrane and juxtamembrane sequences led to reduced proximal signaling and IL-2 production in a hybridoma's response to antigen presentation, despite retention of abundant CD4 association with Lck in the detergent-soluble membrane fraction, presumably mislocalized outside of lipid rafts and distal to the TCR. A major conclusion of that study was that CD4 sequences required for Lck association, including the CXC "clasp" motif, are not as consequential for CD4 co-receptor function in TCR signaling as the conserved TM and JM motifs. However, the experiments did not determine whether the functions of the TM and JM motifs are dependent on the Lck-binding properties of CD4 - the mutations in those motifs could result in free Lck redistributing to associate with CD4 in signaling-incompetent membrane domains or could function independently of CD4-Lck association. The current study addresses this specific question.

Using the same model system as in the earlier eLife paper (the entire methods section is a citation to the earlier paper), the authors show that truncation of the Lck-binding intracellular domain resulted in a moderate reduction in IL-2 response, as previously shown, but there was no apparent effect on proximal phosphorylation events (CD3z, Lck, ZAP70, PLCg1). They then evaluated a series of TM and JM motif mutations in the context of the truncated Lck-nonbinding molecule and showed that these had substantially impaired co-receptor function in the IL-2 assay and reduced proximal signaling. The proximal signaling could be observed at high ligand density even with a MHC non-binding mutation in CD4, although there was still impaired IL-2 production. This result additionally illustrates that phosphorylation of the proximal signaling molecules is not sufficient to activate IL-2 expression in the context of antigen presentation.

Strengths:

The strength of the paper is the further clear demonstration that the classical model of CD4 co-receptor function (MHCII-binding CD4 bringing Lck to the TCR complex, for phosphorylation of the CD3 chain ITAMs and of the ZAP70 kinase) is not sufficient to explain TCR activation. The data, combined with the earlier eLife paper, further implicate the gly-rich TM sequence and the palmitylation targets in the JM region as having critical roles in productive co-receptor-dependent TCR activation.

Weaknesses:

The major weakness of the paper is the lack of mechanistic insight into how the TM and JM motifs function. The new results are largely incremental in light of the earlier paper from this group as well as other literature, cited by the authors, that implicates "free" Lck, not associated with co-receptors, as having the major role in TCR activation. It is clear that the two motifs are important for CD4 function at low pMHCII ligand density. The proposal that they modulate interactions of TCR complex with cholesterol or other membrane lipids is an interesting one, and it would be worth further exploring by employing approaches that alter membrane lipid composition. The JM sequence presumably dictates localization within the membrane, by way of palmitylation, which may be critical to regulate avidity of the TCR:CD4 complex for pMHCII or TCR complex allosteric effects that influence the activation threshold. Experiments that explore the basis of the mutant phenotype could substantially enhance the impact of this study.

Additional comments:

- Is the "IL-2 sensitivity" measurement for the T1-TP (3C) meaningful (Table 3)? It is showing only a moderate reduction compared to T1 control, while TP (2C) or just the 3C palmitylation mutations essentially eliminate response.

- It is unclear how the pairs of control and mutant cells connected by lines in the figures are related. They are presumably cells from distinct biological experiments, with technical replicates for each, but are they paired because they were derived at the same time with different constructs? This should be explained in this paper, not in a reference.

Author Response

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

We would like to express our sincere appreciation for the invaluable comments provided by the reviewers and their constructive suggestions to enhance the quality of our manuscript. In response to their feedback, we have diligently revised and resubmitted our paper as an article, introducing five primary figures, seven supplementary figures, and two supplementary data files. Importantly, this work represents a noteworthy contribution to the field, presenting novel findings for the first time without any prior publication.

Within the enclosed document, we have provided a comprehensive response to the reviewer comments, addressing each point in a meticulous and specific manner. We extend our sincere gratitude to the reviewers for their diligent examination of our manuscript and for offering insightful recommendations.

In our latest revision, we have taken great care to respond to every reviewer's comment, ensuring that we clarify the manuscript and provide robust evidence where required. The primary focus of these revisions was to provide additional context regarding the cooperative role between PR-Set-7 and PARP-1 in the repression of metabolic genes, accompanied by a thorough description of the current state of the field. Substantial modifications and new analyses, presented in the supplemental figures, have been included to comprehensively address this concern.

Another concern raised was regarding the interaction between PARP-1 and mono-methylated active histone marks, which was not adequately described in the previous version of our manuscript. In this revised version, we have updated our Fig. 1 and Supplemental Fig. S1 and introduced Supplemental Fig. S2 to properly demonstrate that PARP-1 binds to all mono-methylated active histone marks tested. Furthermore, we extensively revised the Discussion section of our manuscript to discuss the implications of this discovery and how it fits into the broader context of PARP-1 research.

Addressing another reviewer's concern about the potential indirect regulation of transcription by PARP1 and PR-SET7, we revised the discussion section and incorporated findings from our recent study. These findings clearly demonstrate PARP1's binding to the loci of misregulated genes, suggesting a direct involvement in their regulation.

Furthermore, we have improved the description of the reagents and Drosophila lines used in this study to provide a more comprehensive understanding for readers. Finally, we conducted a comprehensive revision of the entire manuscript to rectify the identified typos and grammatical errors.

Enclosed, you will find a detailed, point-by-point response to each of the reviewer's comments, showcasing our commitment to addressing their concerns with precision.

We firmly believe that our revisions successfully resolve all the concerns raised by the reviewers, and we are confident that this improved version of our manuscript contributes significantly to the scientific discourse.

Reviewer #1:

The study investigates the role of PARP-1 in transcriptional regulation. Biochemical and ChIP-seq analyses demonstrate specific binding of PARP-1 to active histone marks, particularly H4K20me, in polytene chromosomes of Drosophila third instar larvae. Under heat stress conditions, PARP-1's dynamic repositioning from the Hsp70 promoter to its gene body is observed, facilitating gene activation. PARP-1, in conjunction with PR-Set7, plays a crucial role in the activation of Hsp70 and a subset of heat shock genes, coinciding with an increase in H4K20me1 levels at these gene loci. This study proposes that H4K20me1 is a key facilitator of PARP-1 binding and gene regulation. However, there are several critical concerns that are yet to be addressed. The experimental validation and demonstration of results in the main manuscript are scant. Recent developments in the area are omitted, as an important publication hasn't been discussed anywhere in the work (PMID: 36434141). The proposed mechanism operates quite selectively, and any extrapolations require intensive scientific evidence.

Major Comments:

(1) PARP1 hypomorphic mutant validation data must be provided at RNA levels as the authors have mentioned about its global reduction in RNA levels.

We sincerely appreciate Reviewer 1 for their meticulous review of our manuscript and for providing valuable insights. In response to the raised concern, we would like to highlight that the validation data for the PARP1 hypomorphic mutant at the RNA level has been previously documented in our study (PMID: 20371698), where we found that PARP1 RNA level was deeply impacted in parp1C03256. To enhance clarity, we have made corresponding modifications to the Materials and Methods section to explicitly articulate this aspect: parp-1C03256 significantly lowers the level of PARP-1 RNA and protein level (14) but also significantly diminishes the level of pADPr (11).

We hope these revisions effectively address the reviewer's suggestion and contribute to a more comprehensive understanding of our findings.

(2) The authors should provide immunoblot data for global Poly (ADP) ribosylation levels in PARP1 hypomorphic mutant condition as compared to the control. They must also provide the complete details of the mouse anti-pADPr antibody used in their immunoblot in Figure 5B.

We extend our gratitude to Reviewer 1 for drawing attention to aspects requiring further clarification. In response to the inquiry about global Poly (ADP) ribosylation levels in the PARP1 hypomorphic mutant condition, we want to emphasize that our study extensively reported on the diminished levels of pADPr in comparison to the wildtype, as documented in our previous work (PMID: 21444826). To address this, we have incorporated pertinent details in the Materials and Methods section, providing a comprehensive account of our findings. parp-1C03256 significantly lowers the level of PARP-1 RNA and protein level (14) but also significantly diminishes the level of pADPr (11).

Furthermore, in addressing the request for complete details of the mouse anti-pADPr antibody (10H) used in Figure 5B, we have taken steps to enhance transparency. The Materials and Methods section has been revised to incorporate more comprehensive information about the antibody, ensuring a clearer understanding of our experimental procedures. anti-pADPr (Mouse monoclonal, 1:500, 10H - sc-56198, Santa Cruz).

We appreciate the reviewer's diligence in ensuring the robustness of our methodology, and we believe these modifications strengthen the overall quality and transparency of our study.

(3) PR-Set7 mutant validation results should be provided in the main manuscript, as done by the authors using qRT-PCR. Also, immunoblot data for the PR-set7 null condition should be supplemented in the main manuscript as the authors have already mentioned their anti-PR-Set7 (Rabbit, 1:1000, Novus Biologicals, 44710002) antibody in the materials and methods section.

We appreciate Reviewer 1's thorough examination of our manuscript and their constructive feedback. The pr-set7 null mutant has been rigorously characterized in a study conducted by Dr. Ruth Steward's laboratory (PMID: 15681608). Additionally, we employed our PR-SET7 antibody to validate the mutant, and the corresponding data can be found in Supplemental Figure 3. To enhance clarity, we have made necessary modifications to both the results and Materials and Methods sections, providing explicit details on the validation process. Result section: To validate our hypothesis, we initially confirmed that the pr-set720 mutant not only eliminated PR-SET7 RNA and protein but also abrogated H4K20me1 modification (Supplemental Fig.S3).

Material and methods section: The pr-set720 null mutant was validated in (15) and we confirmed that this mutant abolishes PR-SET7 RNA and protein level but also leads to the absence of H4K20me1 (Supplemental Fig. S3).

We believe these revisions address the reviewer's concerns and contribute to a more comprehensive presentation of our study.

(4) The authors have probably missed out on a very important recent report (PMID: 36434141), suggesting the antagonistic nature of the PARP1 and PR-SET7 association. In light of these important observations, the authors must check for the levels of PR-SET7 in PARP1 hypomorphic conditions.

We appreciate the insightful comment from Reviewer 1, drawing our attention to the recent study by Estève et al. (PMID: 36434141) highlighting the potential antagonistic relationship between PARP1 and PR-SET7. To address this important point, we have carefully examined the levels of PR-SET7 in PARP1 hypomorphic conditions.

In response to this concern, we have added two new supplemental figures, Supplemental Fig. S4 and S5, which specifically address the impact of PARP1 deficiency on PR-Set7 expression. These figures clearly demonstrate that there were no significant changes observed in PR-SET7 RNA (Fig. S4) or protein levels (Fig. S5) in the absence of Parp1. This finding supports the conclusion that Parp1 is not directly involved in the regulation of PR-SET7 in Drosophila.

Furthermore, we have updated the Results section to explicitly mention this observation:

Interestingly, in the absence of PARP-1, neither PR-SET7 RNA nor protein levels were affected (Supplemental Fig. S4-5), indicating that PARP-1 is not directly implicated in the regulation of PR-SET7.

Additionally, we have included information about the anti-H3 antibody used in Supplemental Fig. S4 in the Materials and Methods section: anti-H3 (Rabbit polyclonal, 1/1000, FL-136 sc-10809 Santa Cruz).

We believe that these modifications effectively address the raised concern and provide a more comprehensive understanding of the relationship between PARP1 and PR-SET7 in our study. We hope these clarifications enhance the overall robustness and clarity of our findings.

(5) Also, the results of the aforementioned study should be adequately discussed in the present study along with its implications in the same.

We appreciate Reviewer 1's valuable suggestion to discuss the implications of the study by Estève et al. (PMID: 36434141) within the context of our own findings. Estève et al. reported a potential antagonistic relationship between PARP1 and PR-SET7, showing that a decrease in PARP1 proteins leads to an increase in PR-SET7 protein levels. In our investigation, however, we did not observe significant changes in PR-SET7 RNA and protein levels in the parp1C03256 mutant, as demonstrated in the newly added Supplemental Fig. S3 and S4.

We acknowledge the discrepancy between our results and those of Estève et al., and we propose that this difference may be due to distinct experimental approach: Estève et al.'s study focused on mammalian cell populations and in vitro experiments, whereas our investigation employed Drosophila third-instar larvae as the whole organism model. It is plausible that regulatory mechanisms governing PR-SET7 differ between mammals and Drosophila. Another possibility is that PARP-1 may cooperate with PR-SET7 in the context of Drosophila development but could exhibit antagonistic roles against PR-SET7 in specific cell lines and under certain biological or developmental conditions.

In the Discussion section, we have incorporated this information, stating: A recent study demonstrated that in human cells overexpressing PARP-1, PR-SET7/SET8 is degraded (33). This implies that the absence of PARP-1 might lead to increased levels of PR-SET7. However, in our study involving parp-1 mutant in Drosophila third-instar larvae, we observed a slightly different scenario: we detected a minor but not significant reduction in both PR-SET7 RNA and protein levels (Supplemental Fig.S4 and S5). This outcome stands in stark contrast to the previous study's findings. The discrepancy could be due to the distinct experimental approaches used: the previous research focused on mammalian cells and in vitro experiments, whereas our study examined the functions of PARP-1 in whole Drosophila third-instar larvae during development. Consequently, while PARP-1 may cooperate with PR-SET7 in the context of Drosophila development, it could exhibit antagonistic roles against PR-SET7 in specific cell lines and under certain biological or developmental conditions.

We believe these modifications provide a comprehensive discussion of the observed discrepancies and enhance the overall interpretation of our findings. We hope that these clarifications satisfactorily address the concerns raised by Reviewer 1.

(6) Gene transcriptional activation requires open chromatin and RNA polymerase II binding to the promoter. Since, differentially expressed genes in both PR-Set7 null and PARP1 hypomorph mutants, co-enriched with PARP-1 and H4K20me1 were mainly upregulated, the authors should provide RNA polymerase II occupancy data of these genes via RNA-Pol II ChIP-seq to further attest their claims.

We appreciate the insightful comment from Reviewer 1 regarding the necessity for RNA-polymerase II (PolII) occupancy data to further support our claims on gene transcriptional activation. To address this concern, we conducted an analysis of PolII occupancy around genes co-enriched with PARP-1 and H4K20me1 that are upregulated in both pr-set720 and parp-1C03256 mutants during the third instar larvae stage. The results of this analysis have been included in the newly added supplemental Fig. S5.

Our findings reveal that these upregulated genes exhibit higher PolII occupancy compared to other genes, both at their promoter regions and gene bodies, suggesting heightened activity during third instar larval stage in wild type animals (Supplemental Fig. S6). To further validate these results, we cross-referenced publicly available RNA-seq data at the same developmental stage, confirming that, on average, these upregulated genes display a 40% higher expression compared to other genes (supplemental Fig. S6B).

Moreover, we would like to highlight the consistency of our current findings with our previous study (PMID: 38012002), where we reported the critical involvement of PARP-1 in tempering the expression of active metabolic genes at the end of the third instar larvae. The current data, suggesting a role for PR-SET7 in this regulatory process, adds another layer to our understanding of the nuanced control exerted by PARP-1 on the expression of active metabolic genes during this critical developmental transition.

In light of these results, we have modified the Results section to emphasize these findings: Intriguingly, under wild-type conditions, these genes displayed expression levels approximately 40% higher than the average and demonstrated increased RNA-Polymerase II occupancy both at their promoter regions and gene bodies compared to other genes (supplemental Fig.S6), indicating their high activity in wild type context.

Additionally, we have incorporated this information into the Discussion section to underscore the cooperative role of PARP-1 and PR-SET7 in repressing the expression of active metabolic genes: Notably, genes co-enriched with PARP-1 and H4K20me1, and are upregulated in both parp-1C03256 and pr-set720 mutants, are predominantly metabolic genes exhibiting high expression levels under wild-type conditions and a high occupancy of polymerase II both at their promoter region and gene body (Supplemental Fig. S6). In our previous study, we discovered that PARP-1 plays a crucial role in repressing highly active metabolic genes during the development of Drosophila by binding directly to their loci (34). Also, PARP-1 is required for maintaining optimum glucose and ATP levels at the third-instar larval stage (34). During Drosophila development, repression of metabolic genes is crucial for larval to pupal transition (35, 36). This repression is linked to the reduced energy requirements as the organism prepares for its sedentary pupal stage (35, 37). Notably, we observed that PARP-1 shows a high affinity for binding to the gene bodies of these metabolic genes (34).

Our data indicates that in both parp-1 and pr-set7 mutant animals, there was a preferential repression of metabolic genes at sites where PARP-1 and H4K20me1 are co-bound (Fig.3E), while these metabolic genes are highly active during third-instar larval stage (Supplemental Fig.S6). Thus, we propose that the presence of H4K20me1 may be essential for the binding of PARP-1 at these gene bodies, contributing to their repression. Importantly, this mechanism of gene repression has broader developmental implications. As earlier stated, mutant animals lacking functional PARP-1 and PR-SET7 undergo developmental arrest during larval to pupal transition. This arrest could be directly linked to the disruption of the normal metabolic gene repression during development. Without the repressive action of PARP-1 and PR-SET7, key metabolic processes might remain unchecked, leading to metabolic imbalances that are incompatible with the normal progression to the pupal stage.

Finaly, we have updated the Materials and Methods section to include information about the RNA-seq and PolII ChIP-seq datasets used: GSE15292 (RNA-polymerase II). In addition, we used the Developmental time-course RNA-seq dataset (54), SRP001065.

We believe that these modifications comprehensively address Reviewer 1's concern and provide a more robust foundation for our claims regarding the role of PARP-1 and PR-SET7 in the transcriptional regulation of co-enriched genes during the critical developmental transition.

(7) As discussed in Figure 4, the authors found transcriptional activation of group B genes even after a significant reduction of H3K20me1 in their gene body after heat shock. Given the dynamic equilibrium shift in epigenetic marks that regulate gene expression and their locus-specific transcriptional regulation, the authors should further look for the enrichment of other epigenetic marks and even H4K20me1 specific demethylases such as PHF8 (PMID: 20622854), and their cross-talk with PARP1 to further bridge the missing links of this tale. This will add more depth to this work.

We appreciate the thoughtful input provided by Reviewer 1 and acknowledge the importance of exploring additional epigenetic marks and potential cross-talk association with PARP1 to enhance the depth of our study. Our investigation has primarily focused on the interplay between PR-SET7/H4K20me1 and PARP-1, as evidenced by the colocalization and robust binding affinity observed between PARP-1 and H4K20me1 (Fig 1C, 2B, and 3A). This interaction is particularly noteworthy in the context of regulating specific heat shock genes, as highlighted in Figure 4A. While we recognize the potential significance of examining a broader spectrum of epigenetic marks and considering the involvement of specific demethylases, such as PHF8 (PMID: 20622854), in this regulatory network, our research strategy is intentionally tailored to leverage the unique characteristics of the PR-SET7/H4K20me1 and PARP-1 interplay in Drosophila. A key consideration is the technical advantage afforded by the fact that PR-SET7 is the exclusive methylase responsible for H4K20 in Drosophila (PMID: 15681608), allowing for specific depletion of H4K20me1 without the confounding influence of other methyltransferases.

This specificity is pivotal, especially given the similar developmental arrest patterns observed in both PR-SET7 and PARP-1 mutants. Such parallel phenotypes provide a distinct opportunity to delve deeply into the intricacies of their interaction during organismal development and in response to heat stress. Additionally, the identity of the demethylase for H4K20me1 in Drosophila remains unknown, further underscoring the rationale for our focused approach.

While we acknowledge the broader implications of exploring additional epigenetic marks, we believe that our deliberate focus on the PR-SET7/H4K20me1 and PARP-1 pathway provides a unique and valuable perspective on the regulation of gene expression in Drosophila. We hope that this clarification addresses the concerns raised by Reviewer 1 and conveys the rationale behind our chosen research strategy.

Reviewer #2:

Summary:

This study from Bamgbose et al. identifies a new and important interaction between H4K20me and Parp1 that regulates inducible genes during development and heat stress. The authors present convincing experiments that form a mostly complete manuscript that significantly contributes to our understanding of how Parp1 associates with target genes to regulate their expression.

Strengths:

The authors present 3 compelling experiments to support the interaction between Parp1 and H4K20me, including:

(1) PR-Set7 mutants remove all K4K20me and phenocopy Parp mutant developmental arrest and defective heat shock protein induction.

(2) PR-Set7 mutants have dramatically reduced Parp1 association with chromatin and reduced poly-ADP ribosylation.

(3) Parp1 directly binds H4K20me in vitro.

Weaknesses:

(1) The histone array experiment in Fig1 strongly suggests that PARP binds to all mono-methylated histone residues (including H3K27, which is not discussed). Phosphorylation of nearby residues sometimes blocks this binding (S10 and T11 modifications block binding to K9me1, and S28P blocks binding to K27me1). However, H3S3P did not block H3K4me1, which may be worth highlighting. The H3K9me2/3 "blocking effect" is not nearly as strong as some of these other modifications, yet the authors chose to focus on it. Rather than focusing on subtle effects and the possibility that PARP "reads" a "histone code," the authors should consider focusing on the simple but dramatic observation that PARP binds pretty much all mono-methylated histone residues. This result is interesting because nucleosome mono-methylation is normally found on nucleosomes with high turnover rates (Chory et al. Mol Cell 2019)- which mostly occurs at promoters and highly transcribed genes. The author's binding experiments could help to partially explain this correlation because PARP could both bind mono-methylated nucleosomes and then further promote their turnover and lower methylation state.

We appreciate the comprehensive review and valuable insights provided. In response to the comments, we have made substantial revisions to address the concerns and enhance the clarity of our findings. In Figure 1B, C, D, F, and G, we have expanded our data presentation to demonstrate PARP-1's binding affinity for H3K27me1. This addition is now incorporated into the revised results section. Additionally, we have updated Supplemental Fig.S1 and introduced new supplemental data (Supplemental Fig.S2) to illustrate the inhibition of PARP-1 binding by H3S10P, H3S28P, and H3T11P. The comprehensive exploration of PARP-1's interaction with mono-methylated histones, as suggested by the reviewer, is now more robustly documented in our revised figures and supplementary materials.

Our Discussion section has been refined to articulate more clearly how PARP-1 may be selectively recruited to active chromatin domains through its interaction with mono-methylated histone marks. We have proposed a model where PARP-1 actively participates in the turnover process, contributing to the maintenance of an active chromatin environment. This proposed mechanism involves PARP-1 selectively binding to mono-methylated active histone marks associated with highly transcribed genes. Upon activation, PARP-1 undergoes automodification, leading to its release from chromatin and facilitating the reassembly of nucleosomes carrying the mono-methylated marks. The enzymatic action of Poly(ADP)-ribose glycohydrolase (PARG) subsequently cleaves pADPr, allowing for the restoration of PARP-1's binding affinity to mono-methylated active histone marks. This proposed hypothesis is consistent with existing research across various model organisms and aligns with the known association of PARP-1 with highly expressed genes, as well as its role in mediating nucleosome dynamics and assembly.

Our Discussion section is modified a followed: Finaly, highly transcribed genes have been reported to present a high turnover of mono-methylated modifications, maintaining a state of low methylation (50). Then, our findings suggest that PARP-1 might actively participate in the turnover process to uphold an active chromatin environment. The proposed mechanism unfolds as follows: 1) PARP-1 selectively binds to mono-methylated active histone marks associated with highly transcribed genes. 2) Upon activation, PARP-1 undergoes automodification and is subsequently released from chromatin, facilitating the reassembly of nucleosomes carrying the mono-methylated marks. 3) The enzymatic action of Poly(ADP)-ribose glycohydrolase (PARG) cleaves pADPr, allowing for the restoration of PARP-1's binding affinity to mono-methylated active histone marks. This proposed hypothesis aligns cohesively with existing research conducted across various model organisms, including mice, Drosophila, and Humans (7, 23, 29, 51-53). Notably, previous studies have consistently demonstrated that PARP-1 predominantly associates with highly expressed genes and plays a crucial role in mediating nucleosome dynamics and assembly. Thus, our proposed model provides a molecular framework that may contribute to understanding the relationship between PARP-1 and the epigenetic regulation of gene expression. Further experimental validation is warranted to elucidate the precise details of this proposed mechanism and its implications in the broader context of chromatin dynamics and transcriptional control.

We hope that these revisions address the reviewer's concerns and contribute to the overall strength and clarity of our manuscript.

(2) The RNAseq analysis of Parp1/PR-Set7 mutants is reasonable, but there is a caveat to the author's conclusion (Line 251): "our results indicate H4K20me1 may be required for PARP-1 binding to preferentially repress metabolic genes and activate genes involved in neuron development at co-enriched genes." An alternative possibility is that many of the gene expression changes are indirect consequences of altered development induced by Parp1 or PR-Set7 mutants. For example, Parp1 could activate a transcription factor that represses the metabolic genes that they mention. The authors should consider discussing this possibility.

We hope that these revisions address the reviewer's concerns and contribute to the overall strength and clarity of our manuscript.

We extend our gratitude to Reviewer 2 for their thoughtful consideration of our manuscript and the insightful suggestion. In response to the raised concern regarding the conclusion on Line 251, where we proposed that "our results indicate H4K20me1 may be required for PARP-1 binding to preferentially repress metabolic genes and activate genes involved in neuron development at co-enriched genes," we acknowledge the alternative possibility suggested by the reviewer. It is plausible that many of the observed gene expression changes are indirect consequences of altered development induced in parp-1 or pr-set7 mutants. For example, PARP-1 could activates a transcription factor that represses the mentioned metabolic genes.

To address this concern, we have revisited our data and incorporated relevant findings from one of our recent studies that utilized a ChIP-seq approach. The results from this study suggest a direct binding of PARP-1 to the loci of metabolic genes, providing support for the notion that PARP-1 may indeed directly regulate their expression (PMID: 37347109). We have updated the Discussion section to reflect this information, aiming to provide a more comprehensive perspective on the potential mechanisms underlying the observed gene expression changes: In our previous study, we discovered that PARP-1 plays a crucial role in repressing highly active metabolic genes during the development of Drosophila by binding directly to their loci (34). Also, PARP-1 is required for maintaining optimum glucose and ATP levels at the third-instar larval stage (34). During Drosophila development, repression of metabolic genes is crucial for larval to pupal transition (35, 36). This repression is linked to the reduced energy requirements as the organism prepares for its sedentary pupal stage (35, 37). Notably, we observed that PARP-1 shows a high affinity for binding to the gene bodies of these metabolic genes (34).

We believe these modifications contribute to a more informed interpretation of our findings.

(3) The section on the inducibility of heat shock genes is interesting but missing an important control that might significantly alter the author's conclusions. Hsp23 and Hsp83 (group B genes) are transcribed without heat shock, which likely explains why they have H4K20me without heat shock. The authors made the reasonable hypothesis that this H4K20me would recruit Parp-1 upon heat shock (line 270). However, they observed a decrease of H4K20me upon heat shock, which led them to conclude that "H4K20me may not be necessary for Parp1 binding/activation" (line 275). However, their RNA expression data (Fig4A) argues that both Parp1 and H40K20me are important for activation. An alternative possibility is that group B genes indeed recruit Parp1 (through H4K20me) upon heat shock, but then Parp1 promotes H3/H4 dissociation from group B genes. If Parp1 depletes H4, it will also deplete H4K20me1. To address this possibility, the authors should also do a ChIP for total H4 and plot both the raw signal of H4K20me1 and total H4 as well as the ratio of these signals. The authors could also note that Group A genes may similarly recruit Parp1 and deplete H3/H4 but with different kinetics than Group B genes because their basal state lacks H4K20me/Parp1. To test this possibility, the authors could measure Parp association, H4K20methylation, and H4 depletion at more time points after heat shock at both classes of genes.

We thank Reviewer 2 for their valuable comment on our manuscript. We acknowledge your hypothesis suggesting that PARP-1 may induce H3/H4 dissociation from group B genes, potentially leading to a reduction in H4K20me1. However, our findings support a different interpretation.

Our data indicate that while H4K20me1 is present under normal conditions at group B genes, its reduction following heat shock does not appear to hinder PARP-1's role in transcriptional activation (Fig 4A, C and E). We propose that the observed decrease in H4K20me1 might reflect a regulatory shift in chromatin structure that is conducive to transcriptional activation during heat shock, facilitated by PARP-1 independently of sustained H4K20me1 levels at group B genes. Additionally, the literature suggests a dual role for H4K20me1 in gene regulation, from facilitating transcriptional elongation in certain contexts to acting as a repressor in others.

Unlike in group A genes which had low enrichment of H4K20me1 before heat shock (Fig 4B and D), the high enrichment of H4K20me1 in group B genes (Fig 4C and E) could imply a repressive role for this mark prior to heat stress. Thus, in the context of group B genes, it's conceivable that the removal of H4K20me1 might be necessary for their activation during heat stress. Thus, PR-SET7 may possess functions beyond its role as a histone methylase, which are crucial for activating group B genes under heat stress conditions. These functions could include methylation of non-histone substrates and non-catalytic activities.

Furthermore, our analysis of gene expression in pr-set720 and parp-1C03256 mutants indicates that while PARP-1 and H4K20me1 interaction may have overlapping roles in gene regulation, they also possess distinct functions in the modulation of gene expression (Fig 3E). Thus, we propose that the relationship between PR-SET7 and PARP-1 in transcriptional regulation involves a complex regulatory mechanism that extends beyond the presence of H4K20me1.

We modified the discussion section to address this point: Another plausible explanation could be that the recruitment of PARP-1 to group B genes loci promotes H4 dissociation and then leads to a reduction of H4K20me1. However, our findings suggest an alternative interpretation: the decrease in H4K20me1 at group B genes during heat shock does not seem to impede PARP-1's role in transcriptional activation, (Fig.4A, C and E). Rather than disrupting PARP-1 function, we propose that this reduction in H4K20me1 may signify a regulatory shift in chromatin structure, priming these genes for transcriptional activation during heat shock, with PARP-1 playing an independent facilitating role. Moreover, existing studies have highlighted the dual role of H4K20me1, acting as a promoter of transcription elongation in certain contexts and as a repressor in others (13, 25, 38, 39, 41-45). The elevated enrichment of H4K20me1 in group B genes under normal conditions may indicate a repressive state that requires alleviation for transcriptional activation. Additionally, we cannot discount the possibility of unique regulatory functions associated with PR-SET7, extending beyond its recognized role as a histone methylase. Non-catalytic activities and potential interactions with non-histone substrates might contribute to the nuanced control exerted by PR-SET7 on group B genes during heat stress (46, 47). Furthermore, our exploration of pr-set720 and ParpC03256 mutants reveals distinct roles for PARP-1 and H4K20me1 in modulating gene expression (Fig 3E). This reinforces the notion that the interplay between PR-SET7 and PARP-1 involves a multifaceted regulatory mechanism. Understanding the intricate relationship between these molecular players is crucial for elucidating the complexities of gene expression modulation under heat stress conditions.

We hope that this modification will adequately address Reviewer 2 concerns and enhance the clarity of our conclusions.

Reviewer #1 (Recommendations For The Authors):

(1) Please check the entire manuscript for grammatical errors and typos. PR-set7 has been wrongly written as PR-ste7 in quite a few places in the manuscript. Poly (ADP)-ribosylation has been written as poly(ADP-ribosyl)ation in the last result heading. There are more such errors. Please rectify them.

We express our sincere appreciation to Reviewer 1 for their meticulous review of our manuscript, and we acknowledge the importance of ensuring grammatical accuracy and clarity. We have taken your feedback seriously and conducted a comprehensive revision of the entire manuscript to rectify the identified typos and grammatical errors. We hope that these revisions contribute to an improved overall presentation of our research, and we appreciate the reviewer's diligence in ensuring the accuracy of the manuscript.

(2) The authors can also look up publicly available mammalian ChIP-seq data for H4K20me1 and PARP1, in order to further ossify their findings and increase the breadth of their work.

We appreciate the suggestion from Reviewer 1 and have taken steps to further validate and broaden the scope of our findings. Specifically, we compared the distribution of PARP1 and H4K20me1 in Human K562 cells. The results of this analysis revealed a correlation in their distribution, supporting the idea that the observed correlation between PARP-1 and H4K20me1 is not limited to fruit flies. We have incorporated these findings into the Results section and added a new Supplemental Fig. S6 to visually highlight this correlation: Finally, to extend the generalizability of our observations beyond Drosophila, we compared the distribution of PARP1 and H4K20me1 in Human K562 cells. Strikingly, we observed a correlation in their distribution, suggesting that the interplay between PARP-1 and H4K20me1 is not limited to fruit flies (Supplemental Fig. S6).

We believe that this modification addresses Reviewer 1's suggestion by providing additional evidence that supports the broader relevance of our findings beyond the Drosophila model system.

(3) Please discuss in greater detail how the PARP1-H4K20me1 axis orchestrates the repression program (metabolic pathways in this case) with proper references.

We appreciate Reviewer 1's continued engagement with our manuscript and have adjusted the discussion section to provide a more detailed insight into how the PARP1-H4K20me1 axis orchestrates the repression program, particularly focusing on metabolic pathways. The modified discussion section now reads: In our previous study, we discovered that PARP-1 plays a crucial role in repressing highly active metabolic genes during the development of Drosophila by binding directly to their loci (34). Also, PARP-1 is required for maintaining optimum glucose and ATP levels at the third-instar larval stage (34). During Drosophila development, repression of metabolic genes is crucial for larval to pupal transition (35, 36). This repression is linked to the reduced energy requirements as the organism prepares for its sedentary pupal stage (35, 37). Notably, we observed that PARP-1 shows a high affinity for binding to the gene bodies of these metabolic genes (34). Our data indicates that in both parp-1 and pr-set7 mutant animals, there was a preferential repression of metabolic genes at sites where PARP-1 and H4K20me1 are co-bound (Fig.3E), while these metabolic genes are highly active during third-instar larval stage (Supplemental Fig.S6). Thus, we propose that the presence of H4K20me1 may be essential for the binding of PARP-1 at these gene bodies, contributing to their repression. Importantly, this mechanism of gene repression has broader developmental implications. As earlier stated, mutant animals lacking functional PARP-1 and PR-SET7 undergo developmental arrest during larval to pupal transition. This arrest could be directly linked to the disruption of the normal metabolic gene repression during development. Without the repressive action of PARP-1 and PR-SET7, key metabolic processes might remain unchecked, leading to metabolic imbalances that are incompatible with the normal progression to the pupal stage.

We hope these modifications provide a more comprehensive discussion on how the PARP1-H4K20me1 axis influences the repression program, particularly within metabolic pathways, and how this mechanism contributes to the broader context of Drosophila development.

eLife assessment

This valuable study presents convincing evidence for an association between PARP-1 and H4K20me1 in transcriptional regulation, supported by biochemical and ChIP-seq analyses. The work contributes significantly to our understanding of how Parp1 associates with target genes to regulate their expression.

Reviewer #2 (Public Review):

Summary:

This study from Bamgbose et al. identifies a new and important interaction between H4K20me and Parp1 that regulates inducible genes during development and heat stress. The authors present convincing experiments that form a mostly complete manuscript that significantly contributes to our understanding of how Parp1 associates with target genes to regulate their expression.

Strengths:

The authors present 3 compelling experiments to support the interaction between Parp1 and H4K20me, including:

(1) PR-Set7 mutants remove all K4K20me and phenocopy Parp mutant developmental arrest and defective heat shock protein induction.

(2) PR-Set7 mutants have dramatically reduced Parp1 association with chromatin and reduced poly-ADP ribosylation.

(3) Parp1 directly binds H4K20me in vitro.

Weaknesses:

(1) The RNAseq analysis of Parp1/PR-Set7 mutants is reasonable, but there is a caveat to the author's conclusion (Line 251): "our results indicate H4K20me1 may be required for PARP-1 binding to preferentially repress metabolic genes and activate genes involved in neuron development at co-enriched genes." An alternative possibility is that many of the gene expression changes are indirect consequences of altered development induced by Parp1 or PR-Set7 mutants. For example, Parp1 could activate a transcription factor that represses metabolic genes. The authors counter this model by stating that Parp1 directly binds to "repressed" metabolic genes. While this argument supports their model, it does not rule out the competing indirect transcription factor model. Therefore, they should still mention the competing model as a possibility.

(2) The section on inducibility of heat shock genes is interesting but missing an important control that might significantly alter the author's conclusions. Hsp23 and Hsp83 (group B genes) are transcribed without heat shock, which likely explains why they have H4K20me without heat shock. The authors made the reasonable hypothesis that this H4K20me would recruit Parp-1 upon heat shock (line 270). However, they observed a decrease of H4K20me upon heat shock, which led them to conclude that "H4K20me may not be necessary for Parp1 binding/activation" (line 275). However, their RNA expression data (Fig4A) argues that both Parp1 and H40K20me are important for activation. An alternative possibility is that group B genes indeed recruit Parp1 (through H4K20me) upon heat shock, but then Parp1 promotes H3/H4 dissociation from group B genes. If Parp1 depletes H4, it will also deplete H4K20me1. To address this possibility, the authors should also do a ChIP for total H4 and plot both the raw signal of H4K20me1 and total H4 as well as the ratio of these signals. The authors could also note that Group A genes may similarly recruit Parp1 and deplete H3/H4 but with different kinetics than Group B genes because their basal state lacks H4K20me/Parp1. To test this possibility, the authors could measure Parp association, H4K20methylation, and H4 depletion at more time points after heat shock at both classes of genes.

eLife assessment

This solid study addresses the unresolved question of why many thousands of small-effect loci contribute more to the heritability of a trait than the large-effect lead variants. The authors explore resource competition within the transcriptional machinery as one possible explanation with a simple theoretical model, concluding that the effects of resource competition would be too small to explain the heritability effects. The topic and approximation of the problem are important and offer an intuitive way to think about polygenic variation, but there are concerns on the derivation of the equations with respect to dropping vs. including certain terms that deal inherently with small numbers.

Author Response

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

eLife assessment

This important study addresses the fundamentally unresolved question of why many thousands of small-effect loci contribute more to the heritability of a trait than the large-effect lead variants. The authors explore resource competition within the transcriptional machinery as one possible explanation with a simple theoretical model, concluding that the effects of resource competition would be too small to explain the heritability effects. The topic and approximation of the problem are very timely and offer an intuitive way to think about polygenic variation, but the analysis of the simple model appears to be incomplete, leaving the main claims only partially supported.

We thank eLife for recognizing the importance of our work. We hope the revised manuscript addresses the reviewers’ reservations.

Public Reviews:

Reviewer #1 (Public Review):

This study explores whether the extreme polygenicity of common traits can be explained in part by competition among genes for limiting molecular resources (such as RNA polymerases) involved in gene regulation. The authors hypothesise that such competition would cause the expression levels of all genes that utilise the same molecular resource to be correlated and could thus, in principle, partly explain weak trans-regulatory effects and the observation of highly polygenic architectures of gene expression. They study this hypothesis under a very simple model where the same molecule binds to regulatory elements of a large number m of genes, and conclude that this gives rise to trans-regulatory effects that scale as 1/m, and which may thus be negligible for large m.

We thank the reviewer for their thorough and thoughtful review of our manuscript.

The main limitation of this study lies in the details of the mathematical analysis, which does not adequately account for various small effects, whose magnitude scales inversely with the number m of genes that compete for the limiting molecular resource. In particular, the fraction of "free" molecule (which is unbound to any of the genes) also scales as 1/m, but is not accounted for in the analysis, making it difficult to assess whether the quantitative conclusions are indeed correct.

It is explicitly accounted for in the supplement.

Second, the questions raised in this study are better analysed in the framework of a sensitivity or perturbation analysis, i.e., by asking how changes in expression level or binding affinity at one gene (rather than the total expression level or total binding affinity) affect expression level at other genes. In the context of complex traits, where an increase in gene expression can either increase or decrease the trait, we believe the most important quantity of interest is variation in expression and, therefore, trait variation. Nevertheless, our results do show that the relative change in expression due to competition is also small.

Thus, while the qualitative conclusion that resource competition in itself is unlikely to mediate trans-regulatory effects and explain highly polygenic architectures of gene expression traits probably holds, the mathematical reasoning used to arrive at this conclusion requires more care.

In my opinion, the potential impact of this kind of analysis rests at least partly on the plausibility of the initial hypothesis- namely whether most molecular resources involved in gene regulation are indeed "limiting resources". This is not obvious, and may require a careful assessment of existing evidence, e..g., what is the concentration of bound vs. unbound molecular species (such as RNA polymerases) in various cell types?

We intentionally looked at the most extreme case of extreme resource limitation, and we conclude that since extreme resource limitation is a small effect, the same would be true of weak resource limitation, when unbound molecules play an important role. We put more emphasis on this point in our revised text.

Reviewer #1 (Recommendations For The Authors):

While the main conclusion that resource competition in itself is unlikely to mediate trans effects and explain high levels of polygenicity may well be correct, I am not convinced that the mathematical reasoning presented in support of this conclusion is entirely correct. I will attempt to outline my concerns mainly in the context of section 2, since the arguments in sections 3 and 4 build upon this.

(a) The key assumption underlying the approximations in equations 3, 4, and 5 is that there is very little free polymerase, in other words /_0 is a small quantity. However, the second and third terms that emerge in equation 7 are also small quantities and (as far as I can see) of the same order as /_0. Thus, one cannot simply use equation 4 or 5 as a starting point to derive eq. 7 and should instead use the exact x_i = (g_i [G])/ (1+g_tot [G]), in order to make sure that all (and not just some) terms that are similar in order of magnitude are accounted for in the analysis.

The concentration of free polymerase is marked as [P], and we explicitly assume (just before eq. 2) that [P]<<[P]0 with [P]0 being the overall concentration of polymerase. This is a conservative assumption – we consider extreme resource competition with little free polymerase and since we since only a small effect in this extreme scenario we assume it would be a small effect also for less extreme scenarios. We put more emphasis on this point in our revised text.

More concretely, the difference between the exact x_i = (g_i [G])/ (1+g_tot [G]) and the approximate x_i = (g_i / g_tot) is precisely 1/m (for large m) in the example considered line 246 onwards. Thus, I suspect that the conclusion that Var[x_i] = (1-1/m)Var[g_i] in that example is just an artefact of starting with eqs. 4 and 5. As a sanity check, it may be useful to actually simulate resource competition explicitly (maybe using a deterministic simulation) under the explicit model [PG_i] = g_i [G] and _0 = + Sum[[PG]_i , i=1,m] without making any further approximations to see if perturbations in g_i actually produce Order [1/m] effects in the variance of x_i for the example considered line 246 onwards (this would require simulating with a few different m and plotting Var[x_i] vs. m for example).

The exact equation the reviewer is alluding to describes a scenario of non-extreme resource competition. If g_tot [G]>>1, i.e. if most polymerase is bound to a gene then x_i is equal to g_i/g_tot and this is the scenario we are considering of extreme competition. If g_tot [G]<<1, then x_i=g_i [G] and competition has no effect. While the intermediate case is interesting, we see no reason for the effects to be larger than in the extreme competition case. We have added the results of simulations in the supplement to validate our arguments.

Lines 231-239: Because of the concerns highlighted above and questions about the validity of equation 7, I am not convinced that the interpretations given here and also in section 4 are correct.

(b) Lines 219-230 (including equations 6 and 7): I think to address the question of whether genetic changes in cis-regulatory elements for a given gene have an effect on other genes (under this model of resource competition), it is better to spell out the argument in terms of Var[ dx_i ] rather than Var[x_i], where dx_i is the change in expression level at gene i due to changes at all m genes, dg_i is the change in gene activity due to (genetic) changes in the relevant regulatory elements associated with gene i etc. Var[ dx_i ] can then be expressed as a sum of Var[dg_i], Var[dg_tot] and Cov[d g_i, dg_tot]. However, I suspect that to do this correctly, one should not start with the approximate x_i=g_i/g_tot : see previous comment.

The variance of the deviation from the mean is mathematically identical to the overall variance, Var[ dx_i ]= Var[ x_i ]. Our analysis is therefore equivalent to the suggested analysis.

Somewhere in all of this, there is also an implicit assumption that E[dg_i] is zero, i.e, mutations are as likely to increase as to decrease binding affinities so that one needs to only consider Var[dx_i] and not E[dx_i]; this assumption should be spelled out.

Our results concern the variation around trait means and therefore we have not included a possible mean effect of mutation, which would not affect the results but just shift the mean.

Some minor comments (mostly related to the introduction and general context):

• I think it would be worth connecting more with the literature on molecular competition and gene regulation (see e.g., How Molecular Competition Influences Fluxes in Gene Expression Networks, De Vos et al, Plos One 2011). Even though this literature does not frame questions in terms of "polygenicity of traits", these analyses address the same basic questions: to what extent do perturbations in gene expression at one gene affect other genes, or to what extent is there crosstalk between different genes or pathways?

We have expanded our introduction to refer to De Vos et al, as well as a few other papers we have recently become aware of. (e.g., Jie Lin & Ariel Amir Nature Communications volume 9, Article number: 4496 (2018))

• Lines 88-89: "supports the network component of the model" is a vague phrase that does not convey much. It would be useful to clarify and make this more precise.

We have clarified this phrasing in the text.

• Lines 113-114: In the context of "selective constraint", it may also be worth discussing previous work by one of the authors: "A population genetic interpretation of GWAS findings for human quantitative traits". What implications would stabilizing selection on multiple traits (as opposed to simple purifying selection) have for the distribution of variances across trait loci and the extent to which trait architectures appear to be polygenic?

While most definitely of great interest to some of the authors, the distribution of variance across loci does not affect our results.

References: Barton and Etheridge 2018 in line 54 is not the correct reference; it should be Barton et al 2017 (paper with Amandine Veber). Fisher 1919 in line 52 is actually Fisher 1918. The formatting of references in the next paragraph (and in various other places in the paper) is also a bit unusual, with some authors referred to by their full names and others only by their last. I believe that it may be useful to crosscheck references throughout the paper.

We have crosschecked the references in the paper.

Line 164: Some word appears to be missing here. Maybe bound -> bound to ?

Fixed

Reviewer #2 (Public Review):

The question the authors pose is very simple and yet very important. Does the fact that many genes compete for Pol II to be transcribed explain why so many trans-eQTL contribute to the heritability of complex traits? That is, if a gene uses up a proportion of Pol II, does that in turn affect the transcriptional output of other genes relevant or even irrelevant for the trait in a way that their effect will be captured in a genome-wide association study? If yes, then the large number of genetic effects associated with variation in complex traits can be explained but such trans-propagating has effects on the transcriptional output of many genes.

This is a very timely question given that we still don't understand how, mechanistically, so many genes can be involved in complex traits variation. Their approach to this question is very simple and it is framed in classic enzyme-substrate equations. The authors show that the trans-propagating effect is too small to explain the ~70% of heritability of complex traits that are associated with trans-effects. Their conclusion relies on the comparison of the order of magnitude of a) the quantifiable transcriptional effects due to Pol II competition, and b) the observed percentage of variance explained by trans effects (data coming from Liu et al 2019, from the same lab).

The results shown in this manuscript rule out that competition for limited resources in the cell (not restricted to Pol II, but applicable to any other cellular resource like ribosomes, etc) could explain the heritability of complex traits.

We thanked the Reviewer for his resounding support of our paper!

Reviewer #2 (Recommendations For The Authors):

The authors rely on simulated data, and although the conclusions hold in a biologically-realistic scenario given the big difference in effect sizes, I wonder if the authors could provide data from the literature (if available) that give the reader a point of reference for the steady state of cells in terms of free/occupied Pol II molecules and/or free/occupied transcription binding sites. This information won't change the conclusion of the manuscript, but it will put it in the context of real biological data.

We have scoured the literature, but have not found readily available data with which to validate our results (beyond that which is already referenced).

Reviewer #3 (Public Review):

Human complex traits including common diseases are highly polygenic (influenced by thousands of loci). This observation is in need of an explanation. The authors of this manuscript propose a model that competition for a single global resource (such as RNA polymerase II) may lead to a highly polygenic architecture of traits. Following an analytical examination, the authors reject their hypothesis. This work is of clear interest to the field. It remains to be seen if the model covers the variety of possible competition models.

We thank the Reviewer for his assessment, support and comments.

Reviewer #3 (Recommendations For The Authors):

This manuscript provides a straightforward and elegant quantitative argument that the competition for the RNA polymerase is not a significant source of trans-eQTLs and, more generally, of genetic variance of complex polygenic phenotypes. This is an unusual manuscript because the authors propose a hypothesis that they confidently reject based on a calculation. This negative result is intuitive. Still, the manuscript is of interest. Progress in understanding the highly polygenic architecture of complex traits is welcome, and the resource competition hypothesis is quite natural. I have three specific comments/concerns listed below.

(1) The manuscripts states that V(x_i)=V(g_i/g_tot). Unless I am missing something, this seems to result from a very strong implicit assumption that all genetic variance is due to variation in the binding of RNA polymerase, while x_i_max is a constant. I would expect that x_i_max may also be genetically variable due to many effects unrelated to the Pol II binding (e.g. transcription rate, bursting, presence of R-loops etc.). I guess that the assumption made by the authors is conservative.

Indeed. We made conservative assumptions throughout, aiming to consider the most extreme scenario in which resource competition may affect trait variation. Our logic being that if even under the most extreme scenario resource competition is a small effect then it is a small effect in all scenarios. We put more emphasis on this point in our revised text.

(2) The manuscript focuses on the competition for RNA polymerase but suggests that the lesson learned is highly generalizable. However, it is an example of a single global limiting resource resulting in first-order kinetics. What happens in a realistic scenario of competition for multiple resources associated with transcription and with downstream processes (free ribonucleotides, spliceosome, polyadenylation machinery, ribosome, post-translational modifications)? It is possible that in most cases a single resource is a limiting factor, but an investigation (or even a brief discussion) of this question would support the claim that the results are generalizable.

We expect competition for multiple resource to result in similarly weak effects. Since there is not a great number of such resources, we do not expect it to change our qualitative result. We added language to that effect in the main text.

(3) Alternatively, what happens in a scenario of competition for multiple local resources shared by a few genes (co-factors, substrates, chaperones, micro-RNAs, post-translational modification factors such as kinases, degradation factors, scaffolding proteins)? In this case, each gene would compete for resources with a few other genes increasing polygenicity without a global competition with all other genes. Intuitively, a large set of such local competitions may lead to a highly polygenic architecture.

This is indeed a scenario in which competition may be a large effect which we mention in our discussion. “the conclusions may differ in contexts where a very small number of genes compete for a highly limited resource, such as access to a particular molecular transporter”

Reviewer #1 (Public Review):

This study explores whether the extreme polygenicity of common traits (the fact that variation in such traits is explained by a very large number of genetic variants) could be explained in part by competition among genes for limiting molecular resources involved in gene regulation, which would cause the expression of most genes to be correlated. While the hypothesis is interesting, I still have some concerns about the analysis and interpretation.

As the authors say in their rebuttal, assuming extreme resource limitation, i.e., going from equation 2 to 5 essentially assumes assuming that 1/(gtot [G] ) <<1 and that terms that are order [ 1/(gtot [G] ) ] can neglected. However, then the authors derive so-called resource competition terms that are order (1/m) where m is the number of genes, so that gtot is proportional to m. My main criticism (which I am not sure was addressed) is thus: can we reliably derive small order (1/m) effects while neglecting order [ 1/(gtot [G] ) ] terms, when both are presumably similar in order of magnitude? Is this mathematically sound?

I do not think the supplement that the authors have added actually gets to this. For example, section 7.1 just gives the textbook derivation of Michelis-Menten kinetics, and does not address my earlier criticism that the terms neglected in going from eq. 16 to eq. 17 (or from eq. 2 to 3) may be similar in magnitude to the terms being derived and interpreted in eqs. 6 and 7.<br /> Similarly, it is unclear from section 7.2 how the authors are doing the simulations. Are these true Michelis-Menten simulations involving equation 2? If yes, then what is the value of [G] and [P_0] in the simulations? If these are not true Michelis-Menten simulations, but instead something that already uses equation 5, then this still does not address my earlier criticism.

Reviewer #2 (Public Review):

The question the authors pose is very simple, and yet very important. Does the fact that many genes compete for Pol II to be transcribed explain why so many trans-eQTL contribute to the heritability of complex traits? That is, if a gene uses up a proportion of Pol II, does that in turn affect the transcriptional output of other genes relevant or even irrelevant for the trait in a way that their effect will be captured in a genome-wide association study? If yes, then the large number of genetic effects associated with variation in complex traits can be explained but such trans-propagating effects on transcriptional output of many genes.

This is a very timely question given that we still don't understand how, mechanistically, so many genes can be involved in complex traits variation. Their approach to this question is very simple and it is framed in classic enzyme-substrate equations. The authors show that the trans-propagating effect is too small to explain the ~70% of heritability of complex traits that is associated with trans-effects. Their conclusion relies on the comparison of the order of magnitude of a) the quantifiable transcriptional effects due to Pol II competition, and b) the observed percentage of variance explained by trans effects (data coming from Liu et al 2019, from the same lab).

The results shown in this manuscript rule out that competition for limiting resources in the cell (not restricted to Pol II, but applicable to any other cellular resource like ribosomes, etc) could explain heritability of complex traits.

Reviewer #3 (Public Review):

Human complex traits including common diseases are highly polygenic (influenced by thousands of loci). This observation is in need of an explanation. The authors of this manuscript propose a model that a competition for a single global resource (such as RNA polymerase II) may lead to a highly polygenic architecture of traits. Following an analytical examination the authors reject their hypothesis. This work is of clear interest to the field. It remains to be seen if the model covers the variety of possible competition models.

eLife assessment

This important study examines the ancestral function of Hippo pathway kinases in contractility and cell density in the ameboid organism Capsaspora owczarzaki, a unicellular animal that is a close relative of multicellular animals. There is convincing evidence for Hippo kinases regulating contractility and cell density but not proliferation in C. owczarzaki. The work complements previous work on the Hippo effector Yorkie homolog in this species, although the unavailability of extensive genetic tools in this species precludes informative epistasis experiments. The work would be of interest to evolutionary and developmental biologists.

Author Response

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

Reviewer #1 (Public Review):

(1) Fig. 3C needs the "still" for the movie of control C. owczarzaki (in Movie S1).

We have now added a WT control in this figure panel.

(2) The elongated cell shape is seen infrequently in control cells, and I wonder whether these events are transient inactivation of coHpo or coWts in these cells. Perhaps the authors could comment on this in the discussion.

This is an interesting possibility and we have now included it in our discussion (Lines 401403).

(3) Does C. owczarzaki normally aggregate or this is a lab-specific phenotype? For example, the slime mold Dictyostelium discoideum forms aggregates during its life cycle. Could some additional information about C. owczarzaki be added to the introduction?

Unfortunately little is known about Capsaspora “in the wild”, as it was isolated as an endosymbiont from a laboratory strain of snails. However, some related filasterians isolated from natural environments also show aggregatve ability, indicating that aggregation is in fact a physiological process in this group of organisms. We have updated our introduction to include this fact (Line 78-80).

Reviewer #2 (Recommendations For The Authors):

The studies on Hippo signalling in Capsaspora are currently limited to genetic experiments and analysis of Yki/YAP localisation. Biochemical evidence that Co Wts phosphorylates Co Yki/YAP on a conserved serine residue(s) would give important further evidence that this essential signalling step in the animal Hippo pathway is conserved in Capsaspora. However, such experiments require antibodies that detect specific phosphorylation events, which might not be available at present. Is mass spectrometry of the phospho-proteome a potential approach that could be employed to investigate this? The benefit of this approach is it would give information on other Hippo pathway proteins and could be used to probe signalling events under different culture conditions (e.g., aggregate, non-aggregate).

In response to this recommendation, we attempted to detect Phospho-coWts and PhosphocoHpo using commercial antibodies against mammalian their homologs, in the hope of cross-species reactivity. However, we could not detect a signal by Western blot. Thus better reagents or refinement of techniques beyond the scope of this article may be required to examine the phosphorylation of these Capsaspora proteins. There was a published report of Capsaspora phosphoproteome analysis (Sebe Pedros et al., 2016 Dev Cell), although phosphorylation of the conserved sites on coYki, coWts, and coHpo was not reported in this analysis, suggesting more targeted approaches may be needed to examine phosphorylation of these core Hippo pathway components.

The following statement that Wts LOF is stronger than Hpo LOF Capsaspora is consistent with overgrowth phenotypes in flies and mammals:

"Interestingly, we found that coWts-/- cells were significantly more likely to show nuclear mScarlet-coYki localization than coHpo-/- cells (Figure 1D), which is consistent with Hpo/MST independent activity of Wts/LATS previously reported in Drosophila and mammals (Zheng et al., 2015)."

However, the following statement describes a stronger phenotype in Hpo LOF Capsaspora than Wts LOF:

"As contractile cells in the coHpo mutant background tended to show a more extreme elongated morphology than the coWts mutant, we focused on the coHpo mutant for further analysis."

Does this mean that Hpo can regulate actomyosin contractility in both Wts/Yki-dependent and independent manners? A genetic experiment, similar to those that have been performed in Drosophila and mammals could help to address this, e.g., what is the phenotype of Hpo, Yki Capsaspora and Wts, Yki double mutant Capsaspora? Do they phenocopy Yki LOF Capsaspora and are the actomyosin phenotypes associated with Hpo and Wts mutant Capsaspora completely or partially suppressed? The authors indicate that generation of double mutant Capsaspora is not technically possible at present, however.

Indeed given available techniques the generation of such double mutants is not currently possible. With this phenotype (aberrant cytoskeletal dynamics), it is hard to say what a “stronger” phenotype is, and which mutant has the “stronger” phenotype. We have edited this statement to try and reflect this point (Line 208-209).

Another outstanding question is whether the Hpo/Wts/Yki-related actomyosin phenotypes are linked to regulation of transcription by Yki, or are regulated non-transcriptionally. Indeed, a non-transcriptional role for Drosophila Yki in promoting actomyosin contractility has been reported (Fehon lab, Dev Cell, 2018). Generation of Scalloped/TEAD mutant Capsaspora would allow this question to be investigated. Alternatively, this could be explored using variant Co Yki transgenes, e.g., one a Co Yki transgene does not form a physical complex with Co Sd/TEAD and a Co Yki transgene that is targeted to the cell cortex.

To address this point, we tested whether a conserved amino acid residue in coYki (F123) that is required for transcriptional activity of human YAP (in this case, F95) is required for the phenotypic effects of the coYki 4SA mutant. We found that, in contrast to expression of coYki 4SA, expression of a coYki 4SA F123A mutant showed no effect on cell or aggregate morphology. These new results, which support a requirement for transcriptional activity for coYki function, have now been added to Figure 7.

Reviewer #3 (Recommendations For The Authors):

Repetition from previous publication:

(1) ej: last sentences of the abstract in both works: From Phillips et al. eLife 2022;0:e77598: "Taken together, these findings implicate an ancestral role for the Hippo pathway in cytoskeletal dynamics and multicellular morphogenesis predating the origin of animal multicellularity, which was co-opted during evolution to regulate cell proliferation".

From this manuscript: "Together, these results implicate cytoskeletal regulation but not proliferation as an ancestral function of the Hippo pathway and uncover a novel role for Hippo signaling in regulating cell density in a proliferation-independent manner "

Our two papers deal with different components of the Hippo pathway: Yorkie/YAP/coYki in Phillips et al. eLife 2022;0:e77598 and upstream kinases in the current paper. The fact that perturbing different components of the pathway leads to similar conclusions actually strengthens the overall conclusion. Nevertheless, to be more clear about the novelty of the current manuscript, we have now changed the current text from “Hippo pathway” to “Hippo kinase cascade”, to emphasize that the current analysis deals with kinases upstream of Yorkie/YAP/coYki (Lines 35, 368-371).

(2) The authors claim that the change in localization of coYki in Hpo -/- and Wts -/- , being now able to enter the nucleus, is the demonstration that the nuclear regulation of Yki by the Hippo pathway is ancestral to animals. Nevertheless, the authors had already made this claim in their publication of eLife 2022, when they made a mutant version of Yki with the four conserved phosphorylation sites (Sebé-Padrós 2012) mutated. Figure 5 A to F in Phillips et al. eLife 2022;0:e77598. In their words "This regulation of coYki nuclear localization, along with the previous finding that coYki can induce the expression of Hippo pathway genes when expressed in Drosophila (Sebé-Pedrós et al., 2012), suggests that the function of coYki has a transcriptional regulator and Hippo pathway effector is conserved between Capsaspora and animals. ".

I understand that the localization of Yki in the coHpo-/- and coWts-/- is needed as part of final proof that Hpo and Wts are the kinases that control Yki phosphorylation in C. owczarzaki, but does not constitute a completely new message and should be written like that. Figure 1C of the actual manuscript drives to the same conclusion as Figure 5 A to F in Phillips et al. eLife 2022;0:e77598

We think that demonstrating that Hippo and Warts orthologs specifically are responsible for regulation of coYki localization is a very important finding: Many unicellular organisms encode Hippo, Warts, and/or Yorkie’s transcriptional factor partner Sd, but not Yorkie. Our understanding is that in these earlier-branching unicellular organisms, the Hippo/Warts kinase module and Sd-like proteins functioned in distinct signaling modules. Thus Yorkie has the interesting property of “fusing” these two distinct signaling modules when it emerged. In this framework, it is interesting to show that this “fusion” occurred in Capsaspora, the most distant known relative of animals with a Yorkie ortholog, indicating that this “fusion” event is very ancient. Although fleshing out of this idea is beyond the scope of this manuscript and we plan to write about it elsewhere, we have modified our discussion to point out the importance that Hippo and Warts specifically are upstream regulators of coYki.

In Drosophila among the genes transcriptionally regulated by Yki, are the positive regulators of the Hippo pathway in order to down regulate the Yki production.

(1) The authors don't explain if these upstream regulators of the Hippo pathway are conserved in C. owczarzaki.

We have now indicated the conservation of some upstream Hippo pathway components (Line 69-71).

(2) Also it would be important to know how much coYki is being active in the C. owczarzaki in the mutant lines of coHpo-/- and coWts-/- in respect to wt and also in respect to coYki 4SA, and how this is impacting the transcription and protein production of down stream genes of coYki. I think some transcriptional and proteomic data would be informative. At least for those genes related with cytoskeleton.

We have now performed RNA-seq on the coHpo and coWts mutants to address the concerns above (See Figure 8 and the final section of Results).

Related with the above. Among the downstream targets of coYki, the authors mentioned in their previous work (Phillips et al. eLife 2022;0:e77598) that B-integrins were up regulated in coYki -/- suggesting that B-integrins could be behind the stronger cell-substrate attachment observed in the coYki-/- mutant. It would be important to investigate if the integrin adhesome is now down regulated and how previous and new results are related to the stronger cellsubstrate attachment in the coHpo-/- and coWts-/- lines. It would be important that previous results on coYki-/-, a mutant line of the same pathway, are discussed in these two new mutant contexts.

Two Capsaspora integrin beta genes were previously found to be upregulated in the coYki mutant (CAOG_05058 and CAOG_01283, from Phillips et al., 2022 eLife). In our coWts and coHpo mutant RNAseq data, we see that CAOG_05058 is upregulated in both coHpo and coWts mutants, whereas CAOG_01283 does not show significantly different expression in either the coHpo or coWts mutant. Because the CAOG_05058 expression data seems to go in the “opposite” direction than you might expect (i.e. not “down regulated” as the reviewer predicts), and because we see no change in expression in CAOG_01283, these results are difficult to interpret. Therefore the role of integrins in Capsaspora Hippo pathway mutant phenotypes is thus still an open question.

Some cells from the coHpo-/- and coWts-/- mutant lines, show higher attachment to the substrate, which results in an elongated shape while the cell detaches from the substrate. The authors claim this phenotype as a contractile behavior in these cells. This behavior would be caused by changes in cytoskeleton regulation or increased number of microvilli or a change in the distribution of microvilli.

(1) In my opinion, this phenotype can not be considered a behavior per se (the cells become round once they are free from the substrate, so the elongation is temporal and the contractile behavior is a consequence from this attachment to the substrate), so I would not say that the Hippo pathway controls a contractile behavior as the authors state as one of the main conclusions of the manuscript.

Many cell behaviors are known to depend on external conditions, such as substrates, growth factors, nutrients, chemokines, etc., and are therefore “temporal” by the reviewer’s criteria. We therefore feel that the phenotype we describe here can be considered a cell behavior.

(2) On the other hand I think that further efforts on microscopy or immunocytochemistry could be performed in order to discern among the different causes; more microvilli? change in microvilli distribution? change in the acto-myosin cytoskeleton? Moreover these options are not mutually exclusive and very likely the explanation is multifactorial.

(3) coWts-/- has a different phenotype at the periphery of the aggregates than coHpo-/-. The authors use stable transfected lines with NMM-Venus to visualize microvilli. It would be interesting that further experiments using this tool would be performed in order to visualize putative differences of the cell membrane at the periphery in the two mutant genotypes.

We have now performed experiments examining filopodia in round vs elongated cells using the NMM-venus marker, as well as differences in filopodial morphology within aggregates in the different genotypes. Our data and conclusions are included in our updated manuscript (Figure 3- figure supplement 1).

The authors nicely inspect the consequences of the mutant lines coHpo-/- and coWts-/- in the formation of the aggregates. They find that the aggregates in these cases are more densely packed likely due to the higher attachment from microvilli, which they are able to revert by using myosin inhibitors.

(1) As mentioned above, it would be interesting that further experiments are performed by using NMM-Venus transfection into the coHpo-/- and coWts-/-genotypes in order to visualize putative differences of the strength and distribution of the microvilli in the aggregates of these two mutant genotypes. These experiments would inform if more or less microvilli contacts are created in these lines and support a mechanical explanation of the denser aggregates in the mutant lines, as they now suggest in the discussion.

We have now performed these experiments, and our data and conclusions are described in the updated manuscript (Figure 5- figure supplement 1).

(2) On the other hand, myosin inhibition through blebbistatin increases the number of elongated cells in the mutant lines, demonstrating that myosin is necessary for the cells to resolve their substrate attachment and become round. In my view is confusing that myosin is needed for cells to become round again (wt phenotype) and at the same time myosin inhibition is needed for aggregates to become less dense (wt phenotype). Do they lose density because more elongated cells are now in the aggregate? These results look confusing to me and I think they should be better discussed. Again the above transfections of NMM-Venus into the coHpo-/- and coWts-/-genotypes could be informative.

We have attempted to detect cells with an “elongated” morphology within WT and mutant aggregates but so far have been unable to visualize such cells. More advanced microscopy techniques at extended time scales may allow us to observe such things, but we believe such studies are beyond the scope of this manuscript.

The authors do not connect and discuss their results with a very relevant study done in Drosophila, Xu J et al. Dev Cell. 2018; 46(3): 271-284.e5, where a transcriptionally independent role of Yki is characterized. In Drosophila, Yki has an important role in a positive feedback loop with myosin at the cortical part of the cell, which is especially relevant for cytoskeleton regulation.

The results encountered by the authors in their previous study with coYki-/-, indicated that coYki was important for proper actin dynamics and cell shape in C. owczarzaki. At that moment they did not interrogate if this phenotype could be due to the lack of a possible role of coYki in the cortex and they argue that the phenotype was caused by the lack of transcription regulation of downstream genes of coYki, which actually many were cytoskeleton related.

Because the cortex function of Yki is independent of regulation of Hpo and Wts, the authors could use these genotypes by comparing them with WT (where the cortical role of Yki should be the same) and coYki-/- to investigate if the cortex role of Yki, is conserved in C. owczarzaki. In Drosophila the cortex role of Yki has been suggested to control tension at the cell surface. Drosophila Yki at the cortex activates myosin II through the N-terminal part of the protein and establishes a positive feedback loop by down regulating the Hippo pathway and obtaining therefore more active DmYki into the nucleus. This mechanism has been proposed by Xu et al. to work as the link between sensing cell tensions at the surface with control of tissue proliferation.

In my opinion these are relevant results in the field that should be addressed in this study or at least well discussed. Actually, I think they could be a great opportunity for investigating if a putative cortex role of Yki is ancestral to its role linked to the Hippo pathway.

We have now addressed this study in our manuscript- please see our response to reviewer #2’s last comment above.

It would be informative to understand how stable expression through hygromicin selection is achieved in the transfection experiments. Is the recombinant plasmid integrated in the genome? Or is it stable as an episome?

We believe that the plasmids stably integrate, as we never lose fluorescent signal once established in a clonal line, even after extended culturing (>6 months). It may be worthwhile to definitely determine integration vs. episome in future studies.

The authors do not speculate or discuss how cell tension and cell proliferation is different for a unicellular organism or a tissue (multicellular) and I think should be addressed since the contexts are different.

This is an interesting and important point, which we plan to discuss in detail in an upcoming review article, as a proper discussion of this idea, we think, is beyond the scope of this manuscript.

Minor point. The study should cite other unicellular holozoans that have been also developed into treatable organisms such as Monosiga brevicollis (Woznica A, Kumar A, et al 2021eLife 10:e70436) and Abeoforma whisleri (Faktorová, D., Nisbet, R.E.R., Fernández Robledo, J.A. et al. Nat Methods17, 481-494 (2020) in line 83 of the manuscript. I am sure the authors appreciate how much effort there is behind every non-model organism put forward as experimentally treatable and should be properly acknowledged.

We agree, and we have now included these examples of non-model organism development in our manuscript.

Reviewer #1 (Public Review):

Summary:

This Research Advance is an extension of this group's prior eLife paper published in 2022 on the conserved roles of the Hippo pathway effector Yorkie in C. owczarzaki (PMID: 35659869). This species is an amoeba that holds an important phylogenetic position as a close relative of multicellular animals. The prior study used genome editing to delete the C. owczarzaki Yki (termed coYki) and found that Yki is not required for proliferation but instead regulates cell contractility and cell aggregation. In the current study, the authors wanted to address whether Hippo pathway kinases - coHippo (coHpo) and coWarts (coWts) - regulate coYki and whether they are dispensable for proliferation but instead regulate cell contractility and cell aggregation. They used genome editing to delete coHpo and coWts singly in C. owczarzaki. Both mutant strains had increased nuclear location of transfected coYki (tagged with Scarlet), suggesting that Hippo kinase pathway regulation of Yki is conserved in this organism. Neither kinase is required for proliferation. Either kinase mutant strain had a significantly increased percentage of cells that were elongated, which was relatively rare in a control population. The incident of elongation could be enhanced in both kinase-mutant and in control cells when myosin inhibitors were added to the media. coHpo and coWts-mutant aggregates were more tightly packed than control cell aggregates, which they hypothesize is due to the increased contractility seen in kinase-mutant cells. They could reduce the density of packing in kinase-mutant aggregates when they treated the cells with myosin or F-actin inhibitors. To test whether the effects observed in kinase-mutant strains were due to increased Yki activation, they generated a coYki with four S to A substitutions (termed coYki4SA), which should produce a dominant-active Yki impervious to phosphorylation and hence inactivation by Hippo kinases. Control C. owczarzaki cells transfected with coYki4SA had increased cell density in aggregates and elongation in adherent cells. These results support their conclusions that coHpo and coWts regulate cell contractility and cell packing through coYki.

Strengths:

The major strengths of the paper include high quality data, robust analyses of the data, and a well-written manuscript. The combination of genome editing in C. owczarzaki, transfection of C. owczarzaki, and time-lapse movies of adherent cells generally support the conclusions (1) that control of cell density is an ancient function of the Hippo pathway; (2) that Hippo pathway control of cytoskeletal properties and contractile behavior underlie its regulation of cell density, and (3) that Hippo kinase control of Yki localization is likely an ancient function of the pathway.

Weaknesses:

There are no weaknesses.

Reviewer #3 (Public Review):

Summary:

The authors aimed to study the activation of gliogenesis and the role of newborn astrocytes in a post-ischemic scenario. Combining immunofluorescence, BrdU-tracing, and genetic cellular labelling, they tracked the migration of newborn astrocytes (expressing Thbs4) and found that Thbs4-positive astrocytes modulate the extracellular matrix at the lesion border by synthesis but also degradation of hyaluronan. Their results point to a relevant function of SVZ newborn astrocytes in the modulation of the glial scar after brain ischemia. This work's major strength is the fact that it is tackling the function of SVZ newborn astrocytes, whose role is undisclosed so far.

Strengths:

The article is innovative, of good quality, and clearly written, with properly described Materials and Methods, data analysis, and presentation. In general, the methods are designed properly to answer the main question of the authors, being a major strength. Interpretation of the data is also in general well done, with results supporting the main conclusions of this article.

Weaknesses:

However, there are some points of this article that still need clarification to further improve this work.

- As a first general comment, is it possible that the increase in Thbs4-positive astrocytes can also happen locally close to the glia scar, through the proliferation of local astrocytes or even from local astrocytes at the SVZ? As it was shown in published articles most of the newborn astrocytes in the adult brain actually derive from proliferating astrocytes, and a smaller percentage is derived from NSCs. How can the authors rule out a contribution of local astrocytes to the increase of Thbs4-positive astrocytes? The authors also observed that only about one-third of the astrocytes in the glial scar derived from the SVZ.

- It is known that the local, GFAP-reactive astrocytes at the scar can form the required ECM. The authors propose a role of Thbs4-positive astrocytes in the modulation, and perhaps maintenance, of the ECM at the scar, thus participating in scar formation likewise. So, this means that the function of newborn astrocytes is only to help the local astrocytes in the scar formation and thus contribute to tissue regeneration. Why do we need specifically the Thbs4-positive astrocytes migrating from the SVZ to help the local astrocytes? Can you discuss this further?

- The authors observed that the number of BrdU- and DCX-positive cells decreased 15 dpi in all OB layers (Fig. S5). They further suggest that ischemia-induced a change in the neuroblasts ectopic migratory pathway, depriving the OB layers of the SVZ newborn neurons. Are the authors suggesting that these BrdU/DCX-positive cells now migrate also to the ischemic scar, or do they die? In fact, they see an increase in caspase-3 positive cells in the SVZ after ischemia, but they do not analyse which type of cells are dying. Alternatively, is there a change in the fate of the cells, and astrogliogenesis is increased at the expense of neurogenesis? The authors should understand which cells are Cleaved-caspase-3 positive at the SVZ and clarify if there is a change in cell fate. Also please clarify what happens to the BrdU/DCX-positive cells that are born at the SVZ but do not migrate properly to the OB layers.

- The authors showed decreased Nestin protein levels at 15 dpi by western blot and immunostaining shows a decrease already at 7div (Figure 2). These results mean that there is at least a transient depletion of NSCs due to the promotion of astrogliogenesis. However, the authors show that at 30dpi there is an increase of slow proliferating NSCs (Figure 3). Does this mean, that there is a reestablishment of the SVZ cytogenic process? How does it happen, more specifically, how NSCs number is promoted at 30dpi? Please explain how are the NSCs modulated throughout time after ischemia induction and its impact on the cytogenic process.

- The authors performed a classification of Thbs4-positive cells in the SVZ according to their morphology. This should be confirmed with markers expressed by each of the cell subtypes.

- In Figure S6, the authors quantified HABP spots inside Thbs4-positive astrocytes. Please show a higher magnification picture to show how this quantification was done.

eLife assessment

The authors analyze the role of newborn astrocytes in the modulation of glial scar formation in a middle carotid artery occlusion (MCAO) hypoxia lesion model. The findings are valuable because they have implications for understanding the molecular and cellular processes contributing to brain repair in response to ischemia. The results are currently incomplete, in the absence of data showing: (i) cell-target specificity of molecular markers for newborn astrocytes; (ii) consistent phenomena across different rodent species.

Reviewer #1 (Public Review):

Summary:

The authors show that SVZ-derived astrocytes respond to a middle carotid artery occlusion (MCAO) hypoxia lesion by secreting and modulating hyaluronan at the edge of the lesion (penumbra) and that hyaluronan is a chemoattractant to SVZ astrocytes. They use lineage tracing of SVZ cells to determine their origin. They also find that SVZ-derived astrocytes express Thbs-4 but astrocytes at the MCAO-induced scar do not. Also, they demonstrate that decreased HA in the SVZ is correlated with gliogenesis. While much of the paper is descriptive/correlative they do overexpress Hyaluronan synthase 2 via viral vectors and show this is sufficient to recruit astrocytes to the injury. Interestingly, astrocytes preferred to migrate to the MCAO than to the region of overexpressed HAS2.

Strengths:

The field has largely ignored the gliogenic response of the SVZ, especially with regard to astrocytic function. These cells and especially newborn cells may provide support for regeneration. Emigrated cells from the SVZ have been shown to be neuroprotective via creating pro-survival environments, but their expression and deposition of beneficial extracellular matrix molecules are poorly understood. Therefore, this study is timely and important. The paper is very well written and the flow of results is logical.

Weaknesses:

The main problem is that they do not show that Hyaluronan is necessary for SVZ astrogenesis and or migration to MCAO lesions. Such loss of function studies have been carried out by studies they cite (e.g. Girard et al., 2014 and Benner et al., 2013). Similar approaches seem to be necessary in this work.

Major points:

(1) How good of a marker for newborn astrocytes is Thbs4? Did you co-label with B cell markers like EGFr? Is the Thbs4 gene expressed in B cells? Do scRNAseq papers show it is expressed in B cells? Are they B1 or B2 cells?

(2) It is curious that there was no increase in Type C cells after MCAO - do the authors propose a direct NSC-astrocyte differentiation?

(3) The paper would be strengthened with orthogonal views of z projections to show co-localization.

(4) It is not clear why the dorsal SVZ is analysed and focused on in Figure 4. This region emanates from the developmental pallium (cerebral cortex anlagen). It generates some excitatory neurons early postnatally and is thought to have differential signalling such as Wnt (Raineteau group).

(5) Several of the images show the lesion and penumbra as being quite close to the SVZ. Did any of the lesions contact the SVZ? If so, I would strongly recommend excluding them from the analysis as such contact is known to hyperactivate the SVZ.

(6) The authors switch to a rat in vitro analysis towards the end of the study. This needs to be better justified. How similar are the molecules involved between mouse and rat?

(7) Similar comment for overexpression of naked mole rat HA.

Reviewer #2 (Public Review):

Summary:

In their manuscript, Ardaya et al have addressed the impact of ischemia-induced gliogenesis from the adult SVZ and their effect on the remodeling of the extracellular matrix (ECM) in the glial scar. They use Thbs4, a marker previously identified to be expressed in astrocytes of the SVZ, to understand its role in ischemia-induced gliogenesis. First, the authors show that Thbs4 is expressed in the SVZ and that its expression levels increase upon ischemia. Next, they claim that ischemia induces the generation of newborn astrocyte from SVZ neural stem cells (NSCs), which migrate toward the ischemic regions to accumulate at the glial scar. Thbs4-expressing astrocytes are recruited to the lesion by Hyaluronan where they modulate ECM homeostasis.

Strengths:

The findings of these studies are in principle interesting and the experiments are in principle good.

Weaknesses:

The manuscript suffers from an evident lack of clarity and precision in regard to their findings and their interpretation.

Reviewer #3 (Public Review):

This paper considers a challenging motor control task - the critical stability task (CST) - that can be performed equally well by humans and macaque monkeys. This task is of considerable interest since it is rich enough to potentially yield important novel insights into the neural basis of behavior in more complex tasks that point-to-point reaching. Yet it is also simple enough to allow parallel investigation in humans and monkeys, and is also easily amenable to computational modeling. The paper makes a compelling argument for the importance of this type of parallel investigation and the suitability of the CST for doing so.

Behavior in monkeys and in human subjects suggests that behavior seems to include two qualitatively different kinds of behavior - in some cases, the cursor oscillates about the center of the screen, and in other cases, it drifts more slowly in one direction. The authors argue that these two behavioral regimes can be reliably induced by instructing human participants to either maintain the cursor in the center of the screen (position control objective), or keep the cursor still anywhere in the screen (velocity control objective) - as opposed to the usual 'instruction' to just not let the cursor leave the screen. A computational model based on optimal feedback control can reproduce the different behaviors under these two instructions.

Overall, this is a creative study that leverages experiments in humans and computational modeling to gain insight into the nature of individual differences in behavior across monkeys (and people). The authors convincingly demonstrate that they can infer the control objectives from participants who were instructed how to perform the task to emphasize either position or velocity control, based on the RMS cursor position and RMS cursor velocity. The authors show that, while other behavioral metrics do contain similar information about the control objective, RMS position and velocity are sufficient, and their approach classifies control objectives for simulated data with high accuracy (~95%).

The authors also convincingly show that the range of behaviors observed in the CST task cannot be explained as emerging from variations in effort cost, motor execution noise, or sensorimotor delays.

One significant issue, however relates to framing the range of possible control objectives as a simple dichotomy between 'position' and 'velocity' objectives. The authors do clearly state that this is a deliberate choice made in order to simplify their first attempts at solving this challenging problem. However, I do think that the paper at times gives a false impression that this dichotomous view of the control objectives was something that emerged from the data, rather than resulting from a choice to simplify the modeling/inference problem. For instance, line 115: "An optimal control model was used to simulate different control objectives, through which we identified two different control objectives in the experimental data of humans and monkeys."

In the no-instruction condition - which is the starting point and which the ultimate goal of the paper is to understand - there is a lot of variability in behavior across trials (even within an individual) and generally no clear correspondence to either the position or velocity objective. This variability is largely interpreted as the monkeys (and people) switching between control objectives on a trial-to-trial basis. If the behavior were truly a bimodal mixture of these two different behaviors, this might be a convincing interpretation. However, there are a lot of trials that fall in-between the patterns of behavior expected under the position and velocity control objectives. The authors do mention this issue in the discussion. However, it's not clearly examined whether these are simply fringe trials that are ambiguous (like some trials generated by the model are), or whether they reflect a substantial proportion of trials that require some other explanation (whether that is blended position/velocity control, or something else). The existence of these 'in-between' trials (which possibly amount to more than a third of all trials) makes the switching hypothesis a lot less plausible.

Overall, while I think the paper introduces a promising approach and overall helps to improve our understanding of the behavior in this task, I'm not fully convinced that the core issue of explaining the variability in behavior in the no-instruction condition (in monkeys especially) has been resolved. The main explanation put forward is that the monkeys are switching between control objectives on a trial-by-trial basis, but there is no real evidence in the data for this, and I don't think there is yet a good explanation of what is occurring in the 'in-between' trials that aren't explained well by velocity or position objectives.

Reviewer #1 (Public Review):

The present study examines whether one can identify kinematic signatures of different motor strategies in both humans and non-human primates (NHP). The Critical Stability Task (CST) requires a participant to control a cursor with complex dynamics based on hand motion. The manuscript includes datasets on performance of NHPs collected from a previous study, as well as new data on humans performing the same task. Further human experiments and optimal control models highlight how different strategies lead to different patterns of hand motion. Finally, classifiers were developed to predict which strategy individuals were using on a given trial.

There are several strengths to this manuscript. I think the CST task provides a very useful behavioural task to explore the neural basis of voluntary control. While reaching is an important basic motor skill and commonly studied, there is much to learn by looking at other motor actions to address many fundamental issues on the neural basis of voluntary control.

I also think the comparison between human and NHP performance is important as there is a common concern that NHPs can be overtrained in performing motor tasks leading to differences in their performance as compared to humans. The present study highlights that there are clear similarities in motor strategies of humans and NHPs. While the results are promising, I would suggest that the actual use of these paradigms and techniques likely need some improvement/refinement. Notably, the threshold or technique to identify which strategy an individual is using on a given trial needs to be more stringent given the substantial overlap in hand kinematics between different strategies.

The most important goal of this study is to set up future studies to examine how changes in motor strategies impact neural processing. The revised manuscript has improved the technique for identifying which strategy appears to be performed by the individual. A pivotal assumption is that one can identify control strategies from differences in behaviour. As I'm sure the authors know, this inversion of the control problem is not trivial and so success requires that there are only a few 'reasonable' strategies to solve the control problem, and that these strategies lead to distinct patterns of behavior. Many of the concerns raised by myself and the other reviewers relate to this challenge. The revised manuscript now uses a more strict criteria which is good improvement.

One of the values of this paper is to start to develop the tools and approaches to address neural basis of control. The strength of the present manuscript is that it includes modelling, explicit strategy instructions in humans, and then analysis of free-form performance in humans and non-human primates. Given the novelty of this question and approach, there likely are many ways that the techniques and approaches could be improved, but I think they've done a great start. Their approach is quite clever and provides an important blueprint for future studies.

One weakness at this point is that there is still substantial overlap in behavoural performance predicted between strategies, as some human participants given an explicit strategy were almost equally categorized as reflecting the other strategy. I'm glad to see the addition of the model performance on perturbation trials as this additional figure clearly highlights much greater separation in performance than when observing natural behavior. While it is not reasonable to expand beyond this for the present manuscript, I think it is essential for this group to develop the perturbation paradigm (and potentially other approaches) that can better isolate behavioral signatures of different control strategies. I think future work will be strengthened by having multiple experimental angles to interpret the neural activity.

eLife assessment:

This valuable work substantially advances our understanding of the organization of neural dynamics relative to behavioral outputs within recurrent neural networks, with implications for biological neural networks. Evidence supporting the conclusions is convincing, with rigorous analyses of neural variance and robustness to noise. The work will be of broad interest to neuroscientists studying computation through dynamics.

Reviewer #1 (Public Review):

Summary:

In this work, the authors utilize recurrent neural networks (RNNs) to explore the question of when and how neural dynamics and the network's output are related from a geometrical point of view. The authors found that RNNs operate between two extremes: an 'aligned' regime in which the weights and the largest PCs are strongly correlated and an 'oblique' regime where the output weights and the largest PCs are poorly correlated. Large output weights led to oblique dynamics, and small output weights to aligned dynamics. This feature impacts whether networks are robust to perturbation along output directions. Results were linked to experimental data by showing that these different regimes can be identified in neural recordings from several experiments.

Strengths:

A diverse set of relevant tasks.

A well-chosen similarity measure.

Exploration of various hyperparameter settings.

Weaknesses:

One of the major connections found BCI data with neural variance aligned to the outputs. Maybe I was confused about something, but doesn't this have to be the case based on the design of the experiment? The outputs of the BCI are chosen to align with the largest principal components of the data.

Proposed experiments may have already been done (new neural activity patterns emerge with long-term learning, Oby et al. 2019). My understanding of these results is that activity moved to be aligned as the manifold changed, but more analyses could be done to more fully understand the relationship between those experiments and this work.

Analysis of networks was thorough, but connections to neural data were weak. I am thoroughly convinced of the reported effect of large or small output weights in networks. I also think this framing could aid in future studies of interactions between brain regions.

This is an interesting framing to consider the relationship between upstream activity and downstream outputs. As more labs record from several brain regions simultaneously, this work will provide an important theoretical framework for thinking about the relative geometries of neural representations between brain regions.

It will be interesting to compare the relationship between geometries of representations and neural dynamics across connected different brain areas that are closer to the periphery vs. more central.

It is exciting to think about the versatility of the oblique regime for shared representations and network dynamics across different computations.

The versatility of the oblique regime could lead to differences between subjects in neural data.

Reviewer #2 (Public Review):

Summary:

This paper tackles the problem of understanding when the dynamics of neural population activity do and do not align with some target output, such as an arm movement. The authors develop a theoretical framework based on RNNs showing that an alignment of neural dynamics to output can be simply controlled by the magnitude of the read-out weight vector while the RNN is being trained. Small magnitude vectors result in aligned dynamics, where low-dimensional neural activity recapitulates the target; large magnitude vectors result in "oblique" dynamics, where encoding is spread across many dimensions. The paper further explores how the aligned and oblique regimes differ, in particular, that the oblique regime allows degenerate solutions for the same target output.

Strengths:

- A really interesting new idea that different dynamics of neural circuits can arise simply from the initial magnitude of the output weight vector: once written out (Eq 3) it becomes obvious, which I take as the mark of a genuinely insightful idea.

- The offered framework potentially unifies a collection of separate experimental results and ideas, largely from studies of the motor cortex in primates: the idea that much of the ongoing dynamics do not encode movement parameters; the existence of the "null space" of preparatory activity; and that ongoing dynamics of the motor cortex can rotate in the same direction even when the arm movement is rotating in opposite directions.

- The main text is well written, with a wide-ranging set of key results synthesised and illustrated well and concisely.

- The study shows that the occurrence of the aligned and oblique regimes generalises across a range of simulated behavioural tasks.

- A deep analytical investigation of when the regimes occur and how they evolve over training.

- The study shows where the oblique regime may be advantageous: allows multiple solutions to the same problem; and differs in sensitivity to perturbation and noise.

- An insightful corollary result that noise in training is needed to obtain the oblique regime.

- Tests whether the aligned and oblique regimes can be seen in neural recordings from primate cortex in a range of motor control tasks.

Weaknesses:

- The magnitude of the output weights is initially discussed as being fixed, and as far as I can tell all analytical results (sections 4.6-4.9) also assume this. But in all trained models that make up the bulk of the results (Figures 3-6) all three weight vectors/matrices (input, recurrent, and output) are trained by gradient descent. It would be good to see an explanation or results offered in the main text as to why the training always ends up in the same mapping (small->aligned; large->oblique) when it could, for example, optimise the output weights instead, which is the usual target (e.g. Sussillo & Abbott 2009 Neuron).

- It is unclear what it means for neural activity to be "aligned" for target outputs that are not continuous time-series, such as the 1D or 2D oscillations used to illustrate most points here. Two of the modelled tasks have binary outputs; one has a 3-element binary vector.

- It is unclear what criteria are used to assign the analysed neural data to the oblique or aligned regimes of dynamics.

eLife assessment

This work presents a valuable data-driven method to extract the "true" synaptic plasticity rule (or learning rule) operating in a neural circuit from empirical measurements of neural activity. The approach aims to train a generative adversarial network (GAN) to match neural activity measurements in terms of statistics, learning them from the data, rather than being pre-determined by the experimenter. The main conclusion is that the extracted learning rules are not unique, but rather degenerate, meaning that multiple plasticity rules can produce the same neural activity. Although the paper presents a thorough investigation using one learning rule as a case study (the Oja rule), the evidence that the results can be inferred beyond the specific numerical experiments presented in the paper is incomplete.

Reviewer #1 (Public Review):

Summary:

The pioneering work of Eve Marder on central pattern generators in the stomatogastric ganglion (STG) has made a strong case for redundancy as a biological mechanism for ensuring functional robustness, where multiple configurations of biophysical parameters are equivalent in terms of their ability to generate desired patterns of periodic circuit activity. In parallel, normative theories of synaptic plasticity have argued for functional equivalences between learning objectives and corresponding plasticity rules in implementing simple unsupervised learning (see Brito & Gerstner 2016, although similar arguments have been made long before e.g. in Aapo Hyvarinen's ICA book). This manuscript argues that similar notions of redundancy need to be taken into account in the study of synaptic plasticity rules in the brain, more specifically in the context of data-driven approaches to extract the "true" synaptic plasticity rule operating in a neural circuit from neural activity recordings. Concretely, the modeling approach takes a set of empirical measurements of the evolution of neural activity and trains a flexibly parametrized model to match that in statistical terms. Instead of being predefined by the experimenter, the features that determine this match are themselves extracted from data using a generative adversarial network framework (GAN). They show that the flexible models manage to reproduce the neural activity to a reasonable degree (though not perfectly), but lead to very different synaptic trajectories.

Strengths:

The idea of learning rule redundancy is a good one, and the use of GANs for the learning rule estimation is a good complement to other data-driven approaches to extract synaptic plasticity ruled from neural data.

Weaknesses:

(1) Numerics provide only partial support to the statements describing the results.

(2) Even if believing the results, I don't necessarily agree with the interpretation. First: unlike the Marder example where there is complementary evidence to argue that the parameter variations actually reflect across animal biophysical variations, here the statements are really about uncertainty that the experimenter has about what is going on in a circuit observed through a certain measurement lens. Second, while taking into account this uncertainty when using the outcomes of this analysis for subsequent scientific goals is certainly sensible, the biggest punchline for me is that simply observing neural activity in a simple and very restricted context does not provide enough information about the underlying learning mechanism, especially when the hypothesis space is very large (as is the case for the MLP). So it seems more useful to use this framework to think about how to enrich the experimental design/ learning paradigms/ or the measurements themselves to make the set of hypotheses more discriminable (in the spirit of the work by Jacob Portes et al, 2022 for instance). Conversely, one should perhaps think about other ways in which to use other forms of experimental data to reasonably constrain the hypothesis space in the first place.

Reviewer #2 (Public Review):

Summary:

This paper poses the interesting and important question of whether plasticity rules are mathematically degenerate, which would mean that multiple plasticity rules can give rise to the same changes in neural activity. They claim that the answer is "yes," which would have major implications for many researchers studying the biological mechanisms of learning and memory. Unfortunately, I found the evidence for the claim to be weak and confusing, and I don't think that readers can currently infer much beyond the results of the specific numerical experiments reported in the paper.

Strengths:

I love the premise of the paper. I agree with the authors that neuroscientists often under-emphasize the range of possible models that are consistent with empirical findings and/or theoretical demands. I like their proposal that the field is shifting its thinking towards characterizing the space of plasticity rules. I do not doubt the accuracy of most reported numerical results, just their meaning and interpretation. I therefore think that readers can safely use most of the the numerical results to revise their thinking about plasticity mechanisms and draw their own conclusions.

Weaknesses:

Unfortunately, I found many aspects of the paper to be problematic. As a result, I did not find the overarching conclusions drawn by the authors to be convincing.

First, the authors aren't consistent in how they mathematically define and conceptually interpret the "degeneracy" of plasticity mechanisms. In practice, they say that two plasticity mechanisms are "degenerate" if they can't build a neural network to distinguish between a set of neural trajectories generated by them. Their interpretation extrapolates far beyond this, and they seem to conclude that such plasticity rules are in principle indistinguishable. I think that this conclusion is wrong. Plasticity rules are simply mathematical functions that specify how the magnitude of a synaptic weight changes due to other factors, here presynaptic activity (x), postsynaptic activity (y), and the current value of the weight (w). Centuries-old mathematics proves that very broad classes of functions can be parameterized in a variety of non-degenerate ways (e.g., by their Taylor series or Fourier series). It seems unlikely to me that biology has developed plasticity rules that fall outside this broad class. Moreover, the paper's numerical results are all for Oja's plasticity rule, which is a third-order polynomial function of x, y, and w. That polynomial functions cannot be represented by any other Taylor series is a textbook result from calculus. One might wonder if this unique parameterization is somehow lost when many synapses combine to produce neural activity, but the neuron model used in this work is linear, so the function that specifies how the postsynaptic activity changes is simply a fourth-order polynomial in 3N+1 variables (i.e., the presynaptic activities of N neurons prior to the plasticity event, the weights of N synapses prior to the plasticity event, the postsynaptic activity prior to the plasticity event, the presynaptic activities of N neurons after the plasticity event). The same fundamental results from calculus apply to the weight trajectories and the activity trajectories, and a non-degenerate plasticity rule could in principle be inferred from either. What the authors instead show is that their simulated datasets, chosen parameterizations for the plasticity rule, and fitting procedures fail to reveal a non-degenerate representation of the plasticity rule. To what extent this failure is due to the nature of the simulated datasets (e.g., their limited size), the chosen parameterization (e.g., an overparameterized multi-layer perceptron), and their fitting procedure (e.g., their generative adversarial network framework) is unclear. I suspect that all three aspects contribute.

Second, I am concerned by the authors' decision to use a generative adversarial network (GAN) to fit the plasticity rule. Practically speaking, the quality of the fits shown in the figures seems unimpressive to me, and I am left wondering if the authors could have gotten better fits with other fitting routines. For example, other authors fit plasticity rules through gradient descent learning, and these authors claimed to accurately recover Oja's rule and other plasticity rules (Mehta et al., "Model-based inference of synaptic plasticity rules," bioRxiv, 2023). Whether this difference is one of author interpretation or method accuracy is not currently clear. The authors do include some panels in Figure 3A and Figure 8 that explore more standard gradient descent learning, but their networks don't seem to be well-trained. Theoretically speaking, Eqn. (7) in Section 4.4 indicates that the authors only try to match p(\vec y) between the data and generator network, rather than p(\vec x, \vec y). If this equation is an accurate representation of the authors' method, then the claimed "degeneracy" of the learning rule may simply mean that many different joint distributions for \vec x and \vec y can produce the same marginal distribution for \vec y. This is true, but then the "degeneracy" reported in the paper is due to hidden presynaptic variables. I don't think that most readers would expect that learning rules could be inferred by measuring postsynaptic activity alone.

Third, it's important for readers to note that the 2-dimensional dynamical systems representations shown in figures like Figures 2E are incomplete. Learning rules are N-dimensional nonlinear dynamical systems. The learning rule of any individual synapse depends only on the current presynaptic activity, the current postsynaptic activity, and the current weight magnitude, and slices through this function are shown in figures like Figure 2D. However, the postsynaptic activity is itself a dynamical variable that depends on all N synaptic weights. It's therefore unclear how one is supposed to interpret figures like Figure 2E, because the change in y is not a function of y and any single w. My best guess is that figures like Figure 2E are generated for the case of a single presynaptic neuron, but the degeneracies observed in this reduced system need not match those found when fitting the larger network.

Reviewer #3 (Public Review):

Summary:

The authors show that a GAN can learn to reproduce the distribution of outputs of a neuron endowed with Oja's plasticity rule throughout its learning process by learning a plasticity rule. The GAN does not, however, learn Oja's rule. Indeed, the plasticity dynamics it infers can differ dramatically from the true dynamics. The authors propose this approach as a way to uncover families of putative plasticity rules consistent with observed activity patterns in biological systems.

Oja's rule was a great choice for the comparison because it makes explicit, I think, the limitations of this approach. As is well known, Oja's rule allows a (linear) neuron to learn the first principal component of its inputs; the synaptic weights converge to the first eigenvector of the input covariance. After this learning process, the response of a neuron to a particular input sample measures the weighted angle between that input and that principal component.

The other, meta-learned plasticity rules that the authors' GAN uncovers notably do not learn the same computation as Oja's rule (Figure 2). This is, to me, the central finding of the paper and fleshed out nicely. It seems to me that this may be because the objective of the GAN is only to reproduce the marginal output statistics of the neuron. It is, if I understand correctly, blind to the input samples, the inputs' marginal statistics, and to correlations between the input and output. I wonder if a GAN that also had some knowledge of the correlation between input and outputs might be more successful at learning the underlying true dynamics.

The focus on reproducing output statistics has some similarity to some types of experiments (e.g., in vivo recordings) but also seems willfully blind to other aspects of these experiments. In my experience, experimentalists are well aware that the circuits they record receive external inputs. Those inputs are often recorded (perhaps in separate experiments or studies). The point being that I'm not sure that this is an entirely fair comparison to the field.

Finally, the plasticity models studied by theoreticians are not only constructed by intuition and hand-tuning. They also draw, often heavily, on biological data and principles. Oja's rule, for example, is simply the combination of Hebbian learning with a homeostatic constraint on the total synaptic weight amplitude (under the choice of a Euclidean norm).

To me, this study very nicely exposes the caveats and risks associated with a blind machine-learning approach to model specification in biology and highlights the need for understanding underlying biological mechanisms and principles. I agree with the authors that heterogeneity and degeneracy in plasticity rules deserve much more attention in the field.

eLife assessment

This paper presents useful results that extend our understanding of how the visual cortex encodes temporal structure, providing new information about sequence representations in superficial layers of the visual cortex. The evidence for prediction errors is solid, however, support for other claims regarding sparsification and simplification of activity following training is incomplete.

Reviewer #1 (Public Review):

Summary:

Knudstrup et al. use two-photon calcium imaging to measure neural responses in the mouse primary visual cortex (V1) in response to image sequences. The authors presented mice with many repetitions of the same four-image sequence (ABCD) for four days. Then on the fifth day, they presented unexpected stimulus orderings where one stimulus was either omitted (ABBD) or substituted (ACBD). After analyzing trial-averaged responses of neurons pooled across multiple mice, they observed that stimulus omission (ABBD) caused a small, but significant, strengthening of neural responses but observed no significant change in the response to stimulus substitution (ACBD). Next, they performed population analyses of this dataset. They showed that there were changes in the correlation structure of activity and that many features of sequence ordering could be reliably decoded. This second set of analyses is interesting and exhibited larger effect sizes than the first results about predictive coding. However, concerns about the design of the experiment temper my enthusiasm.

Strengths:

(1) The topic of predictive coding in the visual cortex is exciting, and this task builds on previous important work by the senior author (Gavornik and Bear 2014) where unexpectedly shuffling sequence order caused changes in LFPs recorded from the visual cortex.

(2) Deconvolved calcium responses were used appropriately here to look at the timing of the neural responses.

(3) Neural decoding results showing that the context of the stimuli could be reliably decoded from trial-averaged responses were interesting. However I have concerns about how the data was formatted for performing these analyses.

Weaknesses:

(1) All analyses were performed on trial-averaged neural responses that were pooled across mice. Owing to differences between subjects in behavior, experimental preparation quality, and biological variability, it seems important to perform at least some analyses on individual analyses to assess how behavioral training might differently affect each animal.

(2) The correlation analyses presented in Figure 3 (labeled the second Figure 2 in the text) should be conducted on a single-animal basis. Studying population codes constructed by pooling across mice, particularly when there is no behavioral readout to assess whether learning has had similar effects on all animals, appears inappropriate to me. If the results in Figure 3 hold up on single animals, I think that is definitely an interesting result.

(3) On Day 0 and Day 5, the reordered stimuli are presented in trial blocks where each image sequence is shown 100 times. Why wasn't the trial ordering randomized as was done in previous studies (e.g. Gavornik and Bear 2014)? Given this lack of reordering, did neurons show reduced predictive responses because the unexpected sequence was shown so many times in quick succession? This might change the results seen in Figure 2, as well as the decoder results where there is a neural encoding of sequence order (Figure 4). It would be interesting if the Figure 4 decoder stopped working when the higher-order block structure of the task was disrupted.

(4) A primary advantage of using two-photon calcium imaging over other techniques like extracellular electrophysiology is that the same neurons can be tracked over many days. This is a standard approach that can be accomplished by using many software packages-including Suite2P (Pachitariu et al. 2017), which is what the authors already used for the rest of their data preprocessing. The authors of this paper did not appear to do this. Instead, it appears that different neurons were imaged on Day 0 (baseline) and Day 5 (test). This is a significant weakness of the current dataset.

Reviewer #2 (Public Review):

Knudstrup et al set out to probe prediction errors in the mouse visual cortex. They use a variant of an oddball paradigm and test how repeated passive exposure to a specific sequence of visual stimuli affects oddball responses in layer 2/3 neurons. Unfortunately, there are problems with the experimental design which make it difficult to interpret the results in light of the question the authors want to address. The conceptual framing, choice of block design structure, and not tracking the same cells over days, are just some of the reasons that make this work difficult to interpret. Specific comments are as follows:

(1) There appears to be some confusion regarding the conceptual framing of predictive coding. Assuming the mouse learns to expect the sequence ABCD, then ABBD does not probe just for negative prediction errors, and ACBD is not just for positive prediction errors. With ABBD, there is a combination of a negative prediction error for the missing C in the 3rd position, and a positive prediction error for B in the 3rd. Likewise, with ACBD, there is a negative prediction error for the missing B at 2nd and missing C at 3rd, and a positive prediction error for the C in 2nd and B in 3rd. Thus, the authors' experimental design does not have the power to isolate either negative or positive prediction errors. Moreover, looking at the raw data in Figure 2C, this does not look like an "omission" response to C, but more like a stronger response to a longer B. The pitch of the paper as investigating prediction error responses is probably not warranted - we see no way to align the authors' results with this interpretation.

(2) Related to the interpretation of the findings, just because something can be described as a prediction error does not mean it is computed in (or even is relevant to) the visual cortex. To the best of our knowledge, it is still unclear where in the visual stream the responses described here are computed. It is possible that this type of computation happens before the signals reach the visual cortex, similar to mechanisms predicting moving stimuli already in the retina (https://pubmed.ncbi.nlm.nih.gov/10192333/). This would also be consistent with the authors' finding (in previous work) that single-cell recordings in V1 exhibit weaker sequence violation responses than the author's earlier work using LFP recordings.

(3) Recording from the same neurons over the course of this paradigm is well within the technical standards of the field, and there is no reason not to do this. Given that the authors chose to record from different neurons, it is difficult to distinguish representational drift from drift in the population of neurons recorded.

(4) The block paradigm to test for prediction errors appears ill-chosen. Why not interleave oddball stimuli randomly in a sequence of normal stimuli? The concern is related to the question of how many repetitions it takes to learn a sequence. Can the mice not learn ACBD over 100x repetitions? The authors should definitely look at early vs. late responses in the oddball block. Also, the first few presentations after the block transition might be potentially interesting. The authors' analysis in the paper already strongly suggests that the mice learn rather rapidly. The authors conclude: "we expected ABCD would be more-or-less indistinguishable from ABBD and ACBD since A occurs first in each sequence and always preceded by a long (800 ms) gray period. This was not the case. Most often, the decoder correctly identified which sequence stimulus A came from." This would suggest that whatever learning/drift could happen within one block did indeed happen and responses to different sequences are harder to interpret.

(5) Throughout the manuscript, many of the claims are not statistically tested, and where they are the tests do not appear to be hierarchical (https://pubmed.ncbi.nlm.nih.gov/24671065/), even though the data are likely nested.

(6) The manuscript would greatly benefit from thorough proofreading (not just in regard to figure references).

(7) With a sequence of stimuli that are 250ms in length each, the use of GCaMP6s appears like a very poor choice.

(8) The data shown are unnecessarily selective. E.g. it would probably be interesting to see how the average population response evolves with days. The relevant question for most prediction error interpretations would be whether there are subpopulations of neurons that selectively respond to any of the oddballs. E.g. while the authors state they "did" not identify a separate population of omission-responsive neurons, they provide no evidence for this. However, it is unclear whether the block structure of the experiments allows the authors to analyze this.

Reviewer #3 (Public Review):

Summary:

This work provides insights into predictive coding models of visual cortex processing. These models predict that visual cortex neurons will show elevated responses when there are unexpected changes to learned sequential stimulus patterns. This model is currently controversial, with recent publications providing conflicting evidence. In this work, the authors test two types of unexpected pattern variations in layer 2/3 of the mouse visual cortex. They show that pattern omission evokes elevated responses, in favor of a predictive coding model, but find no evidence for prediction errors with substituted patterns, which conflicts with both prior results in L4, and with the expectations of a predictive coding model. They also report that with sequence training, responses sparsify and decorrelate, but surprisingly find no changes in the ability of an ideal observer to decode stimulus identity or timing.

These results are an important contribution to the understanding of how temporal sequences and expectations are encoded in the primary visual cortex. However, there are several methodological concerns with the study, and some of the authors' interpretations and conclusions are unsupported by data.

Major concerns:

(1) Experimental design using a block structure. The use of a block structure on test days (0 and 5) in which sequences were presented in 100 repetition blocks leads to several potential confounds. First, there is the potential for plasticity within blocks, which could alter the responses and induce learned expectations. The ability of the authors to clearly distinguish blocks 1 and 2 on Day 0 with a decoder suggests this change over time may be meaningful.

Repeating the experiments with fully interleaved sequences on test days would alleviate this concern. With the existing data, the authors should compare responses from the first trials in a block to the last trials in a block.

This block design likely also accounts for the ability of a decoder to readily distinguish stimulus A in ABCD from A in ABBD. As all ABCD sequences were run in a contiguous block separate from ABBD, the recent history of experience is different for A stimuli in ABCD versus ABBD. Running fully interleaved sequences would also address this point, and would also potentially mitigate the impact of drift over blocks (discussed below).

(2) The computation of prediction error differs significantly for omission as opposed to substitutions, in meaningful ways the authors do not address. For omission errors, PE compares the responses of B1 and B2 within ABBD blocks. These responses are measured from the same trial, within tens of milliseconds of each other. In contrast, substitution PE is computed by comparing C in ABCD to C in ACBD. As noted above, the block structure means that these C responses were recorded in different blocks, when the state of the brain could be different. This may account for the authors' detection of prediction error for omission but not substitution. To address this, the authors should calculate PE for omission using B responses from ABCD.

(3) The behavior of responses to B and C within the trained sequence ABCD differs considerably, yet is not addressed. Responses to B in ABCD potentiate from d0-> d5, yet responses to C in the same sequence go down. This suggests there may be some difference in either the representation of B vs C or position 2 vs 3 in the sequence that may also be contributing to the appearance of prediction errors in ABBD but not ACBD. The authors do not appear to consider this point, which could potentially impact their results. Presenting different stimuli for A,B,C,D across mice would help (in the current paper B is 75 deg and C is 165 deg in all cases). Additionally, other omissions or substitutions at different sequence positions should be tested (eg ABCC or ABDC).

(4) The authors' interpretation of their PCA results is flawed. The authors write "Experience simplifies activity in principal component space". This is untrue based on their data. The variance explained by the first set of PCs does not change with training, indicating that the data is not residing in a lower dimensional ("simpler") space. Instead, the authors show that the first 5 PCs better align with their a priori expectations of the stimulus structure, but that does not mean these PCs necessarily represent more information about the stimulus (and the fact that the authors fail to see an improvement in decoding performance argues against this case). Addressing such a question would be highly interesting, but is lacking in the current manuscript. Without such analysis, referring to the PCs after training as "highly discretized" and "untangled" are largely meaningless descriptions that lack analytical support.

(5) The authors report that activity sparsifies, yet provide only the fraction of stimulus-selective cells. Given that cell detection was automated in a manner that takes into account neural activity (using Suite2p), it is difficult to interpret these results as presented. If the authors wish to claim sparsification, they need to provide evidence that the total number of ROIs drawn on each day (the denominator for sparseness in their calculation) is unbiased. Including more (or less) ROIs can dramatically change the calculated sparseness.

The authors mention sparsification as contributing to coding efficiency but do not test this. Training a decoder on variously sized subsets of their data on days 0 and 5 would test whether redundant information is being eliminated in the network over training.

(6) The authors claim their results show representational drift, but this isn't supported in the data. Rather they show that there is some information in the structure of activity that allows a decoder to learn block ID. But this does not show whether the actual stimulus representations change, and could instead reflect an unrelated artifact that changes over time (responsivity, alertness, bleaching, etc). To actually assess representational drift, the authors should directly compare representations across blocks (one could train a decoder on block 1 and test on blocks 2-5). In the absence of this or other tests of representational drift over blocks, the authors should remove the statement that "These findings suggest that there is a measurable amount of representational drift".

(7) The authors allude to "temporal echoes" in a subheading. This term is never defined, or substantiated with analysis, and should be removed.

Author Response

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

eLife assessment

This study presents a useful inventory of immune signatures that are correlated with cancer treatment-related pneumonitis. The data were collected and analysed using solid and validated methodology and can be used as a starting point for further functional studies.

We sincerely thank the editor for their encouraging comments regarding our study. As rightly pointed out, this study indeed serves as a pivotal starting point for subsequent functional studies.

Reviewer #2 (Recommendations For The Authors):

I greatly appreciate the authors diligence in addressing all the suggested points. The paper now presents significantly stronger evidence to support the findings.

I do have one final question: Could you clarify how the correlation presented in Supplementary Figure 3 was calculated? Is it a Pearson correlation of CTCAE grade directly to marker expression? Additionally, could you explain how the significance was determined? The authors mention a significant correlation for CCR7, but the heatmap displays similarly high values for CD7 and CD57. Finally, I'm curious about the absence of CD16 in the heatmap.

Thank you for your insightful query. To clarify, the correlation shown in Supplementary Figure 3 was indeed calculated using the Pearson correlation coefficient. This involved correlating the CTCAE grade directly with the mean expression levels of each marker. The computations were conducted using GraphPad Prism version 9. Regarding the statistical significance, we defined a threshold of P < 0.05 as significant. Specifically, the P-values for CCR7, CD7, and CD57 were found to be 0.009, 0.035, and 0.039, respectively. Hence, while CCR7 showed a significant correlation, CD7 and CD57 also exhibited relatively high values, as correctly observed. We have added CD7 and CD57 along with CCR7 in the discussion section, though not to mention much for better focusing on CD16.

CD16 was initially omitted from Supplementary Figure 3 to prevent redundancy and preserve data clarity. Nonetheless, in light of your query, we have included CD16 in the correlation matrix to provide a comprehensive view of its association with other markers.

We hope this adequately addresses your question and further clarifies our findings.

Reviewer #3 (Recommendations For The Authors):

General suggestions for presentation in the future:

It is essential to concretely define the numbers presented in all figures and plots. For example, in Figure 6 (I), what does it mean by "percentage representation of FCGR3A (CD16)"? Percentage of what? How did you calculate that? It is also important to show more statistics in general, for example, in dot plots like Figure 6 (H), where are the means and p-values? Little things like that completely change the impact of the figures. For the narrative of this paper, it is OK, but in the future, fine-tuning the presentation would massively improve the impact of the work which the contents deserve.

Thank you for your insightful feedback. Addressing your concerns, I have revised Figure 6H and Figure 6I to provide a more precise and informative presentation of our data. In Figure 6H, the violin plots illustrate the expression intensity of FCGR3A (CD16) on CD4+ and CD8+ T cells. Each dot represents an individual cell within the BALF from both healthy controls (HC) and COVID-19 patients. This data was derived from the single-cell RNA-seq dataset GSE145926. To enhance clarity and statistical robustness, I have now included p-values directly in Figure 6H. Additionally, for a more comprehensive understanding, the means ± standard deviation (SD) have been incorporated into the main text of the manuscript.

Regarding Figure 6I, it depicts the proportion of FCGR3A (CD16)-positive cells within the CD4+ and CD8+ T cell populations in BALF from HC and COVID-19 patients. The threshold for FCGR3A expression was set at 0.5. Upon further review and in response to your feedback, I realized an error in the calculation of the proportion of FCGR3A-positive cells among CD4+ and CD8+ T cells. Initially, the proportion of FCGR3A-positive CD4+ T cells was calculated in relation to the entire CD4+ T cell population, without differentiation between the groups. This has now been corrected, and the adjusted figures are presented in Figure 6I.

I am grateful for the opportunity to refine these figures, as your suggestions have not only helped to correct the error but have also significantly enhanced the impact and clarity of our work. Your guidance has been instrumental in improving the overall quality and presentation of our research, ensuring that the findings are communicated effectively and accurately.

Reviewer #3 (Public Review):

The authors collected BALF samples from lung cancer patients newly diagnosed with PCP, DI-ILD or ICI-ILD. CyTOF was performed on these samples, using two different panels (T-cell and B-cell/myeloid cell panels). Results were collected, cleaned-up, manually gated and pre-processed prior to visualisation with manifold learning approaches t-SNE (in the form of viSNE) or UMAP, and analysed by CITRUS (hierarchical clustering followed by feature selection and regression) for population identification - all using Cytobank implementation - in an attempt to identify possible biomarkers for these disease states. By comparing cell abundances from CITRUS results and qualitative inspection of a small number of marker expressions, the authors claimed to have identified an expansion of CD16+ T-cell population in PCP cases and an increase in CD57+ CD8+ T-cells, FCRL5+ B-cells and CCR2+ CCR5+ CD14+ monocytes in ICI-ILD cases.

By the authors' own admission, there is an absence of healthy donor samples and, perhaps as a result of retrospective experimental design and practical clinical reasons, also an absence of pre-treatment samples. The entire analysis effectively compares three yet-established disease states with no common baseline - what really constitutes a "biomarker" in such cases? These are very limited comparisons among three, and only these three, states.

By including a new scRNA-Seq analysis using a publicly available dataset, the authors addressed this fundamental problem. Though a more thorough and numerical analysis would be appreciated for a deeper and more impactful analysis, this is adequate for the intended objectives of the study.

eLife assessment

This study presents a valuable inventory of immune signatures that are correlated with cancer treatment-related pneumonitis. The data were collected and analysed using validated methodology and can be used as a starting point for further prospective studies. The authors have provided a scRNA-Seq analysis with an HD baseline using publicly available dataset and the evidence for their claims is convincing.

Reviewer #2 (Public Review):

Yanagihara and colleagues investigated the immune cell composition of bronchoalveolar lavage fluid (BALF) samples in a cohort of patients with malignancy undergoing chemotherapy and with lung adverse reactions including Pneumocystis jirovecii pneumonia (PCP) and immune-checkpoint inhibitors (ICIs) or cytotoxic drug induced interstitial lung diseases (ILDs). Using mass cytometry, their aim was to characterize the cellular and molecular changes in BAL to improve our understanding of their pathogenesis and identify potential biomarkers and therapeutic targets. In this regard, the authors identify a correlation between CD16 expression in T cells and the severity of PCP and an increased infiltration of CD57+ CD8+ T cells expressing immune checkpoints and FCLR5+ B cells in ICI-ILD patients.

The conclusions of this paper are mostly well supported by data, but some aspects of the data analysis need to be clarified and extended.

The authors should elaborate on why different sets of markers were selected for each analysis step. E.g., Different sets of markers were used for UMAP, CITRUS and viSNE in the T cell and myeloid analysis.

Author Response

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

eLife assessment

This study presents valuable findings on diabetogenic risk from colorectal cancer (CRC) treatment. The authors claim that postoperative screening for type 2 diabetes should be prioritized in CRC survivors with overweight/obesity, irrespective of the oncological treatment received. The evidence supporting the claims is solid but requires confirmation in different populations. These results have theoretical or practical implications and will be of interest to endocrinologists, oncologists, general practitioners, gastrointestinal surgeons, and policymakers working on CRC and diabetes.

Author response: We thank you for taking the time to provide constructive feedback on our manuscript and for the useful suggestions. We have provided a point-by-point response to each of the reviewers’ comments with clearly marked changes to the manuscript.

Public reviews

Reviewer #1 (Public Review):

Summary:

In this study, the authors set out to determine whether colorectal cancer surgery site (right, left, rectal) and chemotherapy impact the subsequent risk of developing T2DM in the Danish national health register.

Strengths:

• The research question is conceptually interesting

• The Danish national health register is a comprehensive health database

• The data analysis was thorough and appropriate

• The findings are interesting, and a little surprising that there was no impact of chemotherapy on the development of T2DM

Weaknesses:

This is not a weakness as such, but in the discussion, I would consider adding some brief comment on the international generalizability of the findings - e.g. demographic make up of the Danish population health register and background rates of DM and obesity in this population with CRC compared to countries on other continents.

Author response: We agree that this information would be valuable. It has now been added in the Discussion section.

Changes in manuscript: "In Denmark, the overall T2D prevalence is 6.9%25, lower than the global average in 2021 (10.5%) and also falls below the estimate of high-income countries (11.1%).26 Similarly, the obesity rate of 20% aligns with other Scandinavian countries and is below that of most high-income nations.27” (Page 8, line 256-258)

A little more information would be helpful regarding how T2DM was diagnosed in the registry.

Author response: We have now added a more thorough explanation of how T2D was diagnosed in the Methods section.

Changes in manuscript: “Diabetes is defined as the second occurrence of any event across three types of inclusion events: 1) Diabetes diagnosed during hospitalisation 2) diabetes-specific services received at podiatrist 3) purchases of glucose lowering. Thus, if a patient developed transient T2D during chemotherapy treatment, it will only be an inclusion event if they purchase glucose lowering drugs. Individuals were classified as having T1D if they had received prescriptions for insulin combined with a diagnosis of type 1 from a medical hospital department. Otherwise, diabetes was classified as type 2.22” (Page 5, line 154-160)

If someone did develop transient hyperglycemia requiring DM medications during chemotherapy, would the investigators have been able to identify these people?

Author response: Yes, we have added a sentence in the Methods section.

Changes in manuscript: “Thus, if a patient developed transient T2D during chemotherapy treatment, it will only be an inclusion event if they purchase glucose lowering drugs.” (Page 5, line 156-158)

Would they have been classified as T2DM based on filling a prescription for DM meds for a period of time? Also, did the authors have information regarding time to development of T2DM after surgery?

Author response: Yes, if they have 2 (or more) prescriptions of oral glucose lowering drugs. Yes, we have information regarding time to development of T2DM after surgery and found no difference between the groups.

Changes in manuscript: Information on mean time to develop T2D post-surgery has now been added to Table 2.

In the adjusted Models, the authors did not adjust for cancer stage, even though cancer stage appears to be very different between the chemo and no chemo groups. It would be interesting to know if it affects the results if the model adjusted for cancer stage

Author response: We agree that adjustment for cancer stage would be a valuable information and we have performed the analysis and added a sentence in the Result section.

Changes in manuscript: An adjusted analysis of cancer stage now appears in the Supplementary table 1.

“Moreover, adjusting for cancer stage did not affect the results (Supplementary table 1).” (Page 7, line 219-220)

It would be worthwhile to report if mortality rates were different between the groups during follow up, and if the authors investigated whether perhaps differences in mortality rates led to specific groups living longer, and therefore having more time to develop DM

Author response: This situation is accounted for in the analysis by using Cox-regression analysis. This method accounts for the potential competing effect of mortality.

Changes in manuscript: None.

Overall, the authors achieved their aims, and the conclusions are supported by their results as reported.

The results are unlikely to significantly change patient treatment or T2DM screening in this population. With some additional information, as described above, the results would be of interest to the community.

Reviewer #2 (Public Review):

Summary:

The study showed the impact of cancer treatment on new onset of diabetes among patients with colorectal cancer using the national database. Findings reported that individuals with rectal cancer without chemotherapy were less likely to develop diabetes but among other groups, treatment didn't show any impact on the development of diabetes. BMI still played a significant role in developing diabetes regardless of treatment types.

Strengths:

One of the strengths of this study is innovative findings about the prognosis of colorectal cancer treatment stratified by treatment types. Especially, as it examined the impact of treatment on the risk of new chronic disease after diagnosis, it became significant evidence that suggests practical insights in developing a proper monitoring system for patients with colorectal cancer and their outcomes after treatment and diagnosis. It is imperative for providers to guide patients and caregivers to prevent adverse outcomes like new onset of chronic disease based on BMI and types of treatment. The next strength is the national database. As the study used the national database, the generalizability is validated.

Weaknesses:

Even though the study attempted to examine the impact of each treatment option, the dosage of chemotherapy and the types of chemotherapy were not able to be examined due to the data source.

Author response: No unfortunately not. We agree that this would have been valuable information. This is stated in the original manuscript as a limitation. Please refer to page 10 line 305-306.

Changes in manuscript: None.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

Minor things:

There are minor inconsistencies in the methods and results regarding BMI. In the methods, the authors state that BMI <18.5 and >/=40 were excluded, but these groups are included in Table 2.

Author response: This has been corrected

Changes in manuscript: BMI groups <18.5 and >/=40 are now excluded in Table 2. (Page 18)

Line 204, I believe should be BMI 18.5-24.9, not 20-24.9.

Author response: This has been corrected

Changes in manuscript: “For each group (type of surgery ± chemotherapy), the HR for developing T2D depending on BMI subgroups was calculated by using Cox regression analysis adjusted for age, sex, year of surgery, and ASA score using normal weight (BMI:18.5-24.9) as the reference group.” (Page 6, line 184-186)

Rather than showing the BMI mean in Table 1, it would be interesting to see the BMI breakdown by category.

Author response: Yes, we agree. This analysis has now been added to Table 1

Changes in manuscript: Please refer to Table 1

Re line 215, I would consider rewriting to remove the multiple negatives -e.g. Radiation therapy in rectal resected had did not impact the incidence rate of T2D in the Rectal-No-Chemo group or Rectal-Chemo group

Author response: This has been corrected. Please refer to the Result section.

Changes in manuscript: “Radiation therapy in the rectal resected groups had no impact on the incidence rate of T2D (Table 2); and the unadjusted/adjusted HR of developing T2D was non-significant when comparing Rectal-No-Radiation patients with Rectal-Radiation patients (Table 3).” (Page 7, 223-225)

Consider changing some of the "didn't"s in the discussion to "did not"

Author response: This has been corrected.

Changes in manuscript: Revised and corrected throughout the discussion.

Reviewer #2 (Recommendations For The Authors):

Some points need to be clarified and improved.

In the method, patients with Type 1 Diabetes were excluded in the baseline but some patients were diagnosed with Type 1 diabetes after treatment and they were included in your analysis. It is interesting to identify Type 1 Diabetes after the treatment as an outcome, do you think that this diagnosis is caused by the treatment? And incidence rate or other HRs did not seem to include Type 1 Diabetes as stated in the methods. Did you exclude every Type 1 diabetes? If not, It needs to give further explanation about this outcome since the mechanism of Type 1 Diabetes and Type 2 Diabetes is different.

Author response: This matter has now been clarified in the Methods section.

Changes in manuscript: “Additionally, individuals diagnosed with Type 1 diabetes (T1D) either before or after surgery were excluded, along with those diagnosed with T2D preoperatively or within the first 2 weeks postoperatively, as the last group probably represents patients with preoperatively unknown pre-existing prediabetes or diabetes.22” (Page 4, line: 125-128)

Despite limited existing findings, some studies actually reported the incidence rates of Type 2 Diabetes among patients with CRC (Singh S, Earle CC, Bae SJ, et al. Incidence of Diabetes in Colorectal Cancer Survivors. J Natl Cancer Inst. 2016;108(6):djv402. Published 2016 Feb 2. doi:10.1093/jnci/djv402; Khan NF, Mant D, Carpenter L, Forman D, Rose PW. Long-term health outcomes in a British cohort of breast, colorectal and prostate cancer survivors: a database study. Br J Cancer. 2011;105 Suppl 1(Suppl 1):S29-S37. doi:10.1038/bjc.2011.420; Jo A, Scarton L, O'Neal LJ, et al. New onset of type 2 diabetes as a complication after cancer diagnosis: A systematic review. Cancer Med. 2021;10(2):439-446. doi:10.1002/cam4.3666) whereas your study examined the impact of the different types of treatments.

Author response: Our findings of T2D rate among CRC patients are now commented on in discussion section, and the abovementioned studies are included as references.

Changes in manuscript: “This national cohort study demonstrated an IR of developing T2D after CRC surgery similar to previous studies.5,11” (Page 8, line 237-238)

To strengthen the presentation, some places should be revised.

• Line 216: it says that Table 1 showed no impact of radiation therapy on the incidence rate of T2D. However, either the interpretation or the table number seems wrong. Table 1 does not have this information. Correct this statement.

• Line 239: There are typo and incomplete sentence. Check the sentence and correct the sentence.

• Line 257-261: It may be a systematic issue to separate these two paragraphs. But two paragraphs seem related so make them one paragraph.

Author response: These suggested changes have been made. Regarding line 216 the paragraph has been adjusted to the following:

Changes in manuscript: “Radiation therapy in the rectal resected groups had no impact on the incidence rate of T2D (Table 2); and the unadjusted/adjusted HR of developing T2D was non-significant when comparing Rectal-No-Radiation patients with Rectal-Radiation patients (Table 3).” (Page 7, 223-225)

Reference

(1) Araghi M, Soerjomataram I, Jenkins M, et al. Global trends in colorectal cancer mortality: projections to the year 2035. Int J Cancer. 2019;144(12):2992-3000. doi:10.1002/ijc.32055

(2) Arnold M, Sierra MS, Laversanne M, Soerjomataram I, Jemal A, Bray F. Global patterns and trends in colorectal cancer incidence and mortality. Gut. 2017;66(4):683-691. doi:10.1136/gutjnl-2015-310912

(3) González N, Prieto I, del Puerto-Nevado L, et al. 2017 Update on the Relationship between Diabetes and Colorectal Cancer: Epidemiology, Potential Molecular Mechanisms and Therapeutic Implications. Vol 8.; 2017. www.impactjournals.com/oncotarget

(4) Mills KT, Bellows CF, Hoffman AE, Kelly TN, Gagliardi G. Diabetes mellitus and colorectal cancer prognosis: A meta-analysis. Dis Colon Rectum. 2013;56(11):1304-1319. doi:10.1097/DCR.0b013e3182a479f9

(5) Singh S, Earle CC, Bae SJ, et al. Incidence of Diabetes in Colorectal Cancer Survivors. J Natl Cancer Inst. 2016;108(6). doi:10.1093/jnci/djv402

(6) Xiao Y, Wang H, Tang Y, et al. Increased risk of diabetes in cancer survivors: a pooled analysis of 13 population-based cohort studies. ESMO Open. 2021;6(4). doi:10.1016/j.esmoop.2021.100218

(7) Colorectal D, Nordcan 2019. 5-Year Age-Standardised Relative Survival (%), Males and Females. Accessed September 12, 2022. “https://nordcan.iarc.fr/en/dataviz/survival?cancers=520&set_scale=0&sexes=1_2&populations=208”" has been copied into your clipboard

(8) Nano J, Dhana K, Asllanaj E, et al. Trajectories of BMI Before Diagnosis of Type 2 Diabetes: The Rotterdam Study. Obesity. 2020;28(6):1149-1156. doi:10.1002/oby.22802

(9) Maddatu J, Anderson-Baucum E, Evans-Molina C. Smoking and the risk of type 2 diabetes. Translational Research. 2017;184:101-107. doi:10.1016/j.trsl.2017.02.004

(10) Lega IC, Lipscombe LL. Review: Diabetes, Obesity, and Cancer-Pathophysiology and Clinical Implications. Endocr Rev. 2020;41(1). doi:10.1210/endrev/bnz014 (11) Jo A, Scarton L, O’Neal LTJ, et al. New onset of type 2 diabetes as a complication after cancer diagnosis: A systematic review. Cancer Med. 2021;10(2):439-446. doi:10.1002/cam4.3666

(12) Feng JP, Yuan XL, Li M, et al. Secondary diabetes associated with 5-fluorouracil-based chemotherapy regimens in non-diabetic patients with colorectal cancer: Results from a single-centre cohort study. Colorectal Disease. 2013;15(1):27-33. doi:10.1111/j.1463-1318.2012.03097.x

(13) Lee EK, Koo B, Hwangbo Y, et al. Incidence and disease course of new-onset diabetes mellitus in breast and colorectal cancer patients undergoing chemotherapy: A prospective multicenter cohort study. Diabetes Res Clin Pract. 2021;174. doi:10.1016/j.diabres.2021.108751

eLife assessment

This valuable study presents findings that suggest the need for postoperative type 2 diabetes screening and that this should be prioritized in colorectal cancer survivors with overweight/obesity regardless of the type of colorectal cancer treatment applied. The evidence supporting the claims of the authors is solid and the authors use a population-based cohort study including all Danish colorectal patients who had undergone colorectal cancer surgery between 2001-2018. The work will be of interest to medical biologists, endocrinologists and oncologists working on colorectal cancer.

Reviewer #1 (Public Review):

Summary:<br /> In this study, the authors set out to determine whether colorectal cancer surgery site (right, left, rectal) and chemotherapy impact the subsequent risk of developing T2DM in the Danish national health register.

Strengths:<br /> - The research question is conceptually interesting<br /> - The Danish national health register is a comprehensive health database<br /> - The data analysis was thorough and appropriate<br /> -The findings are interesting, and a little surprising that there was no impact of chemotherapy on the development of T2DM<br /> - The authors have addressed my previous clarifications and questions.

- Regarding the generalizability of this study, as the authors discuss the prevalence of T2DM and obesity are lower in Denmark than in a number of other high income countries. Therefore, similar studies in other populations would be of interest.<br /> - The study includes individuals who filled a prescription for diabetes medication, so likely includes some individuals with transient hyperglycemia/steroid induced diabetes during chemotherapy, rather than those with new onset longterm T2DM.

Overall, the authors achieved their aims, and the conclusions are supported by their results as reported.<br /> The results are unlikely to significantly impact clinical practice or T2DM screening in this population, however are of interest to the community.

Author Response

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

Reviewer 1

Summary:

In the present study, authors found the ternary complex formed by NCAN, TNC, and HA as an important factor facilitating the multipolar to bipolar transition in the intermediate zone (IZ) of the developing cortex. NCAM binds HA via the N-terminal Link modules, meanwhile, TNC cross-links NCAN through the CDL domain at the C-terminal. The expression and right localization of these three factors facilitate the multipolar-bipolar transition necessary for immature neurons to migrate radially. TNC and NCAM are also involved in neuronal morphology. The authors used a wide range of techniques to study the interaction between these three molecules in the developing cortex. In addition, single and double KO mice for NCAN and TNC were analyzed to decipher the role of these molecules in neuronal migration and morphology.

Strengths:

The study of the formation of the cerebral cortex is crucial to understanding the pathophysiology of many neurodevelopmental disorders associated with malformation of the cerebral cortex. In this study, the authors showed, for the first time, that the ternary complex formed by NCAN, TNC, and HA promotes neuronal migration. The results regarding the interaction between the three factors forming the ternary complex are convincing.

We appreciate the reviewers' positive assessment of our research.

Weaknesses:

However, regarding the in vivo experiments, the authors should consider some points for the interpretation of the results:

• The authors did not use the proper controls in their experiments. For embryonic analysis, such as cortical migration, neuronal morphology, and protein distribution (Fig. 6, 7, and 9), mutant mice should be compared with control littermates, since differences in the results could be due to differences in embryonic stages. For example, in Fig. 6 the dKO is more developed than the WT embryo.

It was challenging to compare double knockout mice with control littermates. When crossing Ncan and Tcn double heterozygous mice, the probability of obtaining double knockout mice is 1/16. Given an average litter size of around 8, acquiring a substantial number of double knockout mice would necessitate an impractical number of breeding pairs. Consequently, we were constrained to use non-littermate control mice. To address potential differences in developmental stages, we analyzed 19-20 embryos obtained from five individuals in each group, demonstrating that the observed differences between the two groups are more substantial than the inherent variability within each group.

• The authors claim that NCAM and TNC are involved in neuronal migration from experiments using single KO embryos. This is a strong statement considering the mild results, with no significant difference in the case of TNC KO embryos, and once again, using embryos from different litters.

We agree with the reviewer's comment that a single deletion of TNC has a minimal impact on neuronal migration. We have revised the Results section to reflect the mild nature of the TNC KO phenotype more accurately.

Page 8, line 225: "In NCAN KO mice, a significantly lower percentage of labeled cells resided in the upper layer (Bin2), and more cells remained in the lower layer (Bin5) than in WT mice (Figure 7a). In contrast, the impact of a single deletion of TNC on neuronal cell migration was minimal. Although TNC KO mice exhibited a tendency to have a higher proportion of labeled cells in the lower layer (Bin4) than in WT mice, this did not reach statistical significance (Figure 7a). The delay in neuronal migration observed in the single KO mice was milder when compared to that observed in DKO mice (Figure 6a-c), suggesting that simultaneous deletion of both NCAN and TNC is necessary for a more pronounced impairment in neuronal cell migration."

• The measurement of immunofluorescence intensity is not the right method to compare the relative amount of protein between control and mutant embryos unless there is a right normalization.

We agree that measuring immunofluorescence intensity alone is insufficient for comparing the relative amount of protein. In Figure 8, we have employed Western blotting to compare the protein levels, revealing an approximately 50% reduction in NCAN and TNC following hyaluronidase digestion. In Figures 7b and 7c, we demonstrated alterations in the localization patterns of TNC and NCAN in Ncan KO and Tnc KO mice; however, we did not mention their quantity.

• Page 7, line 206. "No significant abnormalities were observed in the laminar structure in 4-week-old DKO mice". The authors should be more careful with this statement since they did not check the lamination of the adult cortex. I would recommend staining, control and mutant mice, with markers of different cortical populations, such as Cux1, Ctip2, Tbr1, to asses this point.

In response to the suggestion, we have conducted additional experiments to provide a more detailed examination of the laminar structure in the cerebral cortex. The results have been incorporated into the revised manuscript as follows:

Page 7, line 209: "To investigate the laminar organization of the postnatal cerebral cortex, we analyzed the distribution of NeuN-positive postmitotic neurons in DKO mice at 2 weeks of age. No notable abnormalities were observed in the laminar structure of DKO mice (Figure 6-figure supplement 3a, b). Additionally, the laminar distribution of Ctip2-positive deep layer neurons showed no significant differences between WT and DKO mice (Figure 6-figure supplement 3a, c)."

• The authors do not explain how they measured the intensity of TNC around the transfected Turbo-RFP-positive neurons.

We added the following description to the Materials and Methods:

Page 18, line 608: "Images were captured in the IZ region containing Turbo-RFP-positive neurons using a 100X magnification objective lens with 3.0X optical zoom on an AX R confocal microscope (Nikon). A total of 10 optical sections were acquired with a step size of 190 nm. Z-projection views were generated, and the staining intensity of TNC around Turbo-RFP-positive neurons was measured in a 59 × 59 µm area using ImageJ FIJI."

• The loading control of the western blots should be always included.

In Figure 6-figure supplement 1, we have incorporated western blot data using a GAPDH antibody as a loading control. We have added an explanation in the figure legend of Figure 3c, stating that we analyzed the same samples as those used in Figure 1e.

• For Fig. 3e, I think values are represented relative to E18 instead to P2.

Thank you for pointing that out. As suggested, we have corrected the representation in Fig. 3e to be relative to E18 instead of P2.

• I would recommend authors use the standard nomenclature for the embryonic stages. The detection of the vaginal plug is considered as E0.5 and therefore, half a day should be added to embryonic stages (E14.5...).

We have revised our manuscript to designate the detection of the vaginal plug as E0.5, and subsequently, we have adjusted all embryonic stages by adding half a day, such as E14.5.

• Fig 10K: I do not see the differences in the number of neurites in the graph.

We have modified the presentation from a box-and-whisker plot to a bar graph to enhance the visibility of differences in the average number of neurites.

• Line 37: Not all of the cerebral cortex is structured in 6 layers but the neocortex.

We have changed 'cerebral cortex' to 'cerebral neocortex.'

Reviewer 2

Summary:

ECM components are prominent constituents of the pericellular environment of CNS cells and form complex and dynamic interactomes in the pericellular spaces. Based on bioinformatic analysis, more than 300 genes have been attributed to the so-called matrisome, many of which are detectable in the CNS. Yet, not much is known about their functions while increasing evidence suggests important contributions to developmental processes, neural plasticity, and inhibition of regeneration in the CNS. In this respect, the present work offers new insights and adds interesting aspects to the facets of ECM contributions to neural development. This is even more relevant in view of the fact that neurocan has recently been identified as a potential risk gene for neuropsychiatric diseases. Because ECM components occur in the interstitial space and are linked in interactomes their study is very difficult. A strength of the manuscript is that the authors used several approaches to shed light on ECM function, including proteome studies, the generation of knockout mouse lines, and the analysis of in vivo labeled neural progenitors. This multi-perspective approach permitted to reveal hitherto unknown properties of the ECM and highlighted its importance for the overall organization of the CNS.

Strengths:

Systematic analysis of the ternary complex between neurons, TNC, and hyaluronic acid; establishment of KO mouse lines to study the function of the complex, use of in utero electroporation to investigate the impact on neuronal migration;

We appreciate the reviewers' insightful comments.

Weaknesses:

The analysis is focused on neuronal progenitors, however, the potential impact of the molecules of interest, in particular, their removal on differentiation and /or survival of neural stem/progenitor cells is not addressed. The potential receptors involved are not considered. It also seems that rather the passage to the outer areas of the forming cortex is compromised, which is not the same as the migration process. The movement of the cells is not included in the analysis.

In this study, we demonstrated that the ternary complex of NCAN, TNC, and HA is predominantly localized in the subplate/intermediate zone. This region lacks neural stem/progenitor cells but serves as the initiation site for the radial migration of postmitotic neurons. Consequently, our study focused on the role of the ternary complex in neuronal migration and polarity formation. We acknowledge that we did not investigate in-depth the potential effects of ECM perturbation on the differentiation and survival of neural stem/progenitor cells. However, as highlighted by the reviewer, it is important to explore the effects on neural stem/progenitor cells. To address this concern, we analyzed Pax6-positive radial glial cells and Tbr2-positive intermediate progenitor cells in the ventricular zone of wild-type and Ncan/Tnc double knockout (DKO) mice. Immunohistochemical analysis revealed no significant differences between WT and DKO mice (Figure 6-figure supplement 4a). Furthermore, the morphology of nestin-positive radial fibers exhibited no distinguishable variations between WT and DKO mice (Figure 6-figure supplement 4b, c).

(1) In the description of the culture of cortical neurons the authors mentioned the use of 5% horse serum as a medium constituent. HS is a potent stimulus for astrocyte differentiation and astrocytes in vitro release neurocan. Therefore, the detection of neurocan in the supernatant of the cultures as shown in Figure 1h might as well reflect release by cultivated astrocytes.

As pointed out by the reviewer, Figure 1h did not conclusively demonstrate that neurons are the sole source of NCAN production. Indeed, in situ hybridization analysis revealed the widespread distribution of Ncan mRNA throughout the cerebral cortex (Figure 2a). This result suggests that the production of NCAN involves not only neurons but also other cell populations, including radial glial cells and astrocytes. While we acknowledge the potential contribution of other cell types to NCAN production, Ncan expression by neurons during radial migration is a crucial aspect of our findings (Figure 1i, j). We have revised the manuscript as follows:

Page 5, line 111: "This result suggested the secretion of NCAN by developing neurons; however, we cannot rule out the involvement of coexisting glial cells in the culture system. To investigate the expression of Ncan mRNA during radial migration in vivo, we labeled radial glial cells in the VZ with GFP through in utero electroporation at E14.5 (Figure 1i, Figure 1-figure supplement 1)."

(2) It is known that neurocan in vivo is expressed by neurons, but may be upregulated in astrocytes after lesion, or in vitro, where the cells become reactive.

We have incorporated the following description into the discussion:

Page 11, line 359: "Previous studies have reported an upregulation of NCAN and TNC in reactive astrocytes, indicating the potential formation of the ternary complex of NCAN, TNC, and HA in the adult brain in response to injury (Deller et al., 1997; Haas et al., 1999)."

(3) Do NCAN KO neurons show an increase in neurite growth on the TNC substrates? The response on POL was changed (Fig. 10h-k), but the ECM substrates were not tested with the KO neurons.

The impact of ECM substrates on NCAN KO neurons has not been investigated, and this remains an avenue for further exploration in our ongoing research. Future studies aim to elucidate the NCAN-TNC connection by identifying TNC cell surface receptors and unraveling the subsequent intracellular signaling pathways.

(4) Do the authors have an explanation for why the ternary complex is concentrated in the SP/IZ zone?

In the mature brain, hyaluronan acts as a scaffold that facilitates the accumulation of ECM components, including proteoglycans and tenascins around neurons. Therefore, it is conceivable that the ECM components bind to hyaluronan in the embryonic brain, resulting in its accumulation in the subplate/intermediate zone. In support of this hypothesis, enzymatic digestion of hyaluronan in the subplate/intermediate zone led to the disappearance of TNC and NCAN accumulation (Figure 8a-c). This result may account for the disparity observed, where Tnc mRNA is expressed in the ventricular zone while the TNC protein localizes to the subplate/intermediate zone.

(5) Are hyaluronic acid synthesizing complexes (HAS) concentrated in the SP/IZ?

According to the reviewer's comment, we have investigated the localization of Has2 and Has3 mRNA using in situ hybridization. However, due to the relatively low expression levels of these enzymes, we encountered challenges in obtaining clear signals (Author response image 1). Further research is needed to understand the mechanisms behind the localization of hyaluronan in the intermediate zone.

Author response image 1.

In situ hybridization analysis of Has2 and 3 mRNA on the E16.5 cerebral cortex. Upper images show results of in situ hybridization using antisense against Has2 and 3. Lower images are in situ hybridization using sense probes as negative controls.

(6) CSPGs as well as TNC are part of the neural stem/progenitors cell niche environment. Does the removal of either of the ECM compounds affect the proliferation, differentiation, and/or survival of NSPCs, or their progeny?

)7) This question relates to the fact that the migration process itself is not visualized in the present study, rather its outcome - the quantitative distribution of labeled neurons in the different bins of the analysis. This could also derive from modified cell numbers.

As pointed out by the reviewer, previous studies have shown the role of CSPGs and TNC as components of the neural stem/progenitor cell niche (see reviews by (Faissner et al., 2017; Faissner and Reinhard, 2015). However, as mentioned in Response #2, based on our analyses, we did not observe a reduction in neural stem/progenitor cells in NCAN/TNC double-knockout mice. While we cannot precisely explain this discrepancy, it is worth noting that many past studies evaluated the activities of the ECM molecules in in vitro systems such as neurospheres. The observed differences may stem from variations in experimental systems.

(8) What is the role of the ECM in the SP/IZ area? Do the cells need the ECM to advance, the reduction would then leave the neuronal progenitors in the VZ area? This somehow contrasts with interpretations that the ECM acts as an obstacle for neurite growth or cell migration, or as a kind of barrier.

The role of the ECM is multifaceted, with certain ECM molecules known to inhibit neurite outgrowth while others facilitate it. Additionally, the effects of ECM can vary depending on the cell type. It is established that after migrating neurons adhere to radial fibers, they utilize these fibers as a scaffold to migrate toward the cortical surface. However, in the subplate/intermediate zone, migrating neurons have not yet adhered to radial fibers. This study provides evidence that multipolar neurons undergo morphological changes into bipolar cells with the assistance of the NCAN, TNC, and HA complex. Subsequently, this facilitates their movement along radial fibers.

(9) A direct visualization of the movement of neural progenitors in the tissue as has been for example performed by the Kriegstein laboratory might help resolve some of these issues.

As suggested by the reviewer, utilizing live imaging techniques to directly observe the movement of neural progenitors within the tissue is indeed a powerful tool. We recognize the significance of addressing these points in future research.

eLife assessment

The solid study addresses the role of extracellular matrix (ECM) in neuronal migration. The authors showed that the interaction between the ternary complex formed by tenascin-C, the chondroitin sulfate proteoglycan neurocan, and hyaluronic acid is important for the multipolar to bipolar transition in the intermediate zone (IZ) of the developing cortex

Reviewer #1 (Public Review):

Summary:<br /> In the present study, authors found the ternary complex formed by NCAN, TNC, and HA as an important factor facilitating the multipolar to bipolar transition in the intermediate zone (IZ) of the developing cortex. NCAM binds HA via the N-terminal Link modules, meanwhile, TNC cross-links NCAN through the CDL domain at the C-terminal. The expression and right localization of these three factors facilitate the multipolar-bipolar transition necessary for immature neurons to migrate radially. TNC and NCAM are also involved in neuronal morphology. The authors used a wide range of techniques to study the interaction between these three molecules in the developing cortex. In addition, single and double KO mice for NCAN and TNC were analyzed to decipher the role of these molecules in neuronal migration and morphology.

Strengths:<br /> The study of the formation of the cerebral cortex is crucial to understanding the pathophysiology of many neurodevelopmental disorders associated with malformation of the cerebral cortex. In this study, the authors showed, for the first time, that the ternary complex formed by NCAN, TNC, and HA promotes neuronal migration. The results regarding the interaction between the three factors forming the ternary complex are convincing.

Reviewer #2 (Public Review):

Summary:

ECM components are prominent constituents of the pericellular environment of CNS cells and form complex and dynamic interactomes in the pericellular spaces. Based on bioinformatic analysis, more than 300 genes have been attributed to the so-called matrisome, many of which are detectable in the CNS. Yet, not much is known about their functions while increasing evidence suggests important contributions to developmental processes, neural plasticity, and inhibition of regeneration in the CNS. In this respect, the present work offers new insights and adds interesting aspects to the facets of ECM contributions to neural development. This is even more relevant in view of the fact that neurocan has recently been identified as a potential risk gene for neuropsychiatric diseases. Because ECM components occur in the interstitial space and are linked in interactomes their study is very difficult. A strength of the manuscript is that the authors used several approaches to shed light on ECM function, including proteome studies, the generation of knockout mouse lines, and the analysis of in vivo labeled neural progenitors. This multi-perspective approach permitted to reveal hitherto unknown properties of the ECM and highlighted its importance for the overall organization of the CNS.

Strengths:

Systematic analysis of the ternary complex between neurone, TNC, and hyaluronic acid; establishment of KO mouse lines to study the function of the complex, use of in utero electroporation to investigate the impact on neuronal migration.

Author Response

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

We would like to thank the reviewer for the constructive comments. We have revised the papers to address the concerns. In summary, here is what we included in the revised version.

• Statistical analysis using biological replicate datasets for WT and K40R doublet microtubule.

• Addition figures for statistical analysis and MIP decorations in MEC17-KO and K40R.

• Revised texts and figures to reflect the new changes, cite proper references and fix small errors throughout the text.

Reviewer #1 (Public Review):

Summary:

The study "Effect of alpha-tubulin acetylation on the doublet microtubule structure" by S. Yang et al employs a multi-disciplinary approach, including cryo-electron microscopy (cryo-EM), molecular dynamics, and mass spectrometry, to investigate the impact of α-tubulin acetylation at the lysine 40 residue (αK40) on the structure and stability of doublet microtubules in cilia. The work reveals that αK40 acetylation exerts a small-scale, but significant, effect by influencing the lateral rotational angle of the microtubules, thereby affecting their stability. Additionally, the study provided an explanation of the relationship between αK40 acetylation and phosphorylation within cilia, despite that the details still remain elusive. Overall, these findings contribute to our understanding of how post-translational modifications can influence the structure, composition, stability, and functional properties of important cellular components like cilia.

Strengths:

(1) Multi-Disciplinary Approach: The study employs a robust combination of cryo-electron microscopy (cryo-EM), molecular dynamics, and mass spectrometry, providing a comprehensive analysis of the subject matter.

(2) Significant Findings: The paper successfully demonstrates the impact of αK40 acetylation on the lateral rotational angles between protofilaments (inter-PF angles) of doublet microtubules in cilia, thereby affecting their stability. This adds valuable insights into the role of post-translational modifications in cellular components.

(3) Exploration of Acetylation-Phosphorylation Relationship: The study also delves into the relationship between αK40 acetylation and phosphorylation within cilia, contributing to a broader understanding of post-translational modifications.

(4) High-quality data: The authors are cryo-EM experts in the field and the data quality presented in the manuscript is excellent.

(5) Depth of analysis: The authors analyzed the effects of αK40 acetylation in excellent depth which significantly improved our understanding of this system.

Thank you for highlighting the strength of our paper.

Weaknesses:

I have no major concerns about this paper, but would recommend that a few minor issues be addressed.

(1) Lack of Statistical Details: The review points out that the paper could benefit from providing more statistical details, such as the number of particles and maps used for analysis, randomization methods, and dataset splitting for statistical analyses.

To address this, we analyzed the true biological replicate datasets (different cultures, cryo-EM vitrification and data collection) from WT and K40R. Since the MEC17-KO was collected as only one dataset, we decided to not divide the MEC-17 using randomization since the division does not lead to independent sets, which tends to yield identical results in the case of cryo-EM. The biological replicates help us to see how consistent is our structure data for interpretation. The information about the replicate dataset is now included in Table 1. The description of the analysis is highlighted in the manuscript and included in the Materials & Methods and Fig. S4.

In summary, the biological replicate between the WT data indicates that the inter-PF rotation angles are significantly consistent between two biological replicates. On the other hand, there are variations in the inter-PF angles between two replicates of K40R data in the B-tubule (Fig. S4B).

Overall, when pooling the data together ( 6 + 6 measurement points for WT dataset 1 & 2 and 6 + 6 measurement points for K40R dataset 1 & 2 and 6 measurement points for MEC17-KO) (Fig. S4), our analysis yields the same statistical significance as the average of all datasets (6 measurement points of the total averages for WT, K40R and MEC17-KO) (Fig. 3).

In addition, the variation in inter-PF rotation angles between certain PF pairs within the K40R replicates (B7B8 and B9B10) is similar to the variation to MEC17-KO. This suggests that the deacetylation induces variation in inter-PF angles while the inter-PF angles are maintained consistently in WT.

(2) Questionable Conclusion Regarding MIPs: The reviewer suggests caution in the paper's conclusion that "Acetylation of αK40 does not affect tubulin and MIPs." The reviewer recommends that this conclusion be more specific or supported by additional evidence to exclude all other possibilities.

We now revised the text to make sure we do not overclaim that “Acetylation of αK40 does not affect tubulin and MIPs.” We now describe more specifically as “Lack acetylation of αK40 does not significantly affect tubulin and MIP interactions”. Also the text was edited to make the statement more specific.

(3) Need for Additional Visual Data: The reviewer recommends that an enlarged local density map along with fitted PDB models be provided in a supplementary figure, such as Figure 4.

We now include the density maps and fitted PDB models in Fig. 4 and Fig. S5. We also include more snapshots of the MIP in K40R and MEC17-KO in Figure S3.

Overall, the paper is strong in its scientific approach and findings but could benefit from additional statistical rigor and clarification of certain conclusions.

Page 11, Line 226: "cluster consists of only ~ acetylated", lacks the percentage. Please correct this.

We corrected it.

eLife assessment

This fundamental study employs a combination of cryo-electron microscopy, molecular dynamics, and mass spectrometry to elucidate the role of α-tubulin acetylation at the lumenal lysine 40 residue (αK40) within the cilium. Compelling evidence shows αK40 acetylation to impact the structure and stability of doublet microtubules in cilia by affecting the lateral rotational angle. The work will be of relevance to those interested in cytoskeleton and structural biology.

Reviewer #1 (Public Review):

Summary:

The study "Effect of alpha-tubulin acetylation on the doublet microtubule structure" by S. Yang et al employs a multi-disciplinary approach, including cryo-electron microscopy (cryo-EM), molecular dynamics, and mass spectrometry, to investigate the impact of α-tubulin acetylation at the lysine 40 residue (αK40) on the structure and stability of doublet microtubules in cilia. The work reveals that αK40 acetylation exerts a small-scale, but significant, effect by influencing the lateral rotational angle of the microtubules, thereby affecting their stability. Additionally, the study provided an explanation of the relationship between αK40 acetylation and phosphorylation within cilia, despite that the details still remain elusive. Overall, these findings contribute to our understanding of how post-translational modifications can influence the structure, composition, stability, and functional properties of important cellular components like cilia.

Strengths:

(1) Multi-Disciplinary Approach: The study employs a robust combination of cryo-electron microscopy (cryo-EM), molecular dynamics, and mass spectrometry, providing a comprehensive analysis of the subject matter.<br /> (2) Significant Findings: The paper successfully demonstrates the impact of αK40 acetylation on the lateral rotational angles between protofilaments (inter-PF angles) of doublet microtubules in cilia, thereby affecting their stability. This adds valuable insights into the role of post-translational modifications in cellular components.<br /> (3) Exploration of Acetylation-Phosphorylation Relationship: The study also delves into the relationship between αK40 acetylation and phosphorylation within cilia, contributing to a broader understanding of post-translational modifications.<br /> (4) High-quality data: The authors are cryo-EM experts in the field and the data quality presented in the manuscript is excellent.<br /> (5) Depth of analysis: The authors analyzed the effects of αK40 acetylation in excellent depth which significantly improved our understanding of this system.

Weaknesses:

I have no major concerns about this paper.

Author Response

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

Public Reviews:

Reviewer #1 (Public Review):

Summary:

The manuscript describes the crystal structures of Streptococcus pneumoniae NOXs. Crystals were obtained for the wild-type and mutant dehydrogenase domain, as well as for the full-length protein comprising the membrane domain. The manuscript further carefully studies the enzyme's kinetics and substrate-specificity properties. Streptococcus pneumoniae NOX is a non-regulated enzyme, and therefore, its structure should provide a view of the NOX active conformation. The structural and biochemical data are discussed on this ground.

Strengths:

This is very solid work. The protein chemistry and biochemical analysis are well executed and carefully described. Similarly, the crystallography must be appreciated given the difficulty of obtaining good enzyme preparations and the flexibility of the protein. Even if solved at medium resolution, the crystal structure of the full-length protein conveys relevant information. The manuscript nicely shows that the domain rotations are unlikely to be the main mechanistic element of NOX regulation. It rather appears that the NADPH-binding conformation is pivotal to enzyme activation. The paper extensively refers to the previous literature and analyses the structures comprehensively with a comparison to previously reported structures of eukaryotic and prokaryotic NOXs.

We thank the referee for these very nice comments about our work.

Weaknesses:

The manuscript is not always very clear with regard to the analysis of NADPH binding. The last section describes a "crevice" featured by the NADPH-binding sites in NOXs. It remains unclear whether this element corresponds to the different conformations of the protein C-terminal residues or more extensive structural differences. This point must be clarified.

We agree with the referee that our terminology was not very clear. Responding to your comment helped us to improve our explanation: we have changed the text to emphasize the differences we observe in the distances between the FAD binding groove and the entire NADPH binding groove, which includes conserved NADPH-contacting motifs as well as the critical aromatic.

A second less convincing point concerns the nature of the electron acceptor. The manuscript states that this NOX might not physiologically act as a ROS producer. A question then immediately arises: Is this protein an iron reductase?

Can the authors better discuss or provide more data about this point?

The referee has a legitimate point, which was also our first idea. In the initial work on SpNOX, where we discovered bacterial NOX enzymes (see Hajjar et al 2017 in mBio), we evaluated its possible role as an iron reductase. There we showed that SpNOX can reduce CytC directly; however, while some reduction of Fe3+-NTA complex (used classically in ferric reductase activity assay) occurred, this reduction was inhibitable by SOD and occurred indirectly by the superoxide produced, so therefore not a true iron reductase activity. This represents a mixed situation of direct and indirect reduction of an iron-containing acceptor that appears to preclude physiological iron reductase activity since it appears that the protein component of CytC allows it to interact with SpNOX. As these questions had been already addressed in a previous paper, we did not add anything here and we prefer to underline this possibility of another acceptor and to leave this question open for future works.

Reviewer #2 (Public Review):

The authors describe the structure of the S. pneumoniae Nox protein (SpNOX). This is a first. The relevance of it to the structure and function of eukaryotic Noxes is discussed in depth.

Strengths and Weaknesses

One of the strengths of this work is the effort put into preparing a pure and functionally active SpNOX preparation. The protein was expressed in E. coli and the purification and optimization of its thermostability and activity are described in detail, involving salt concentration, glycerol concentration, and pH.

This reviewer was surprised by the fact that the purification protocol in the eLife paper differs from those in the mBio and Biophys. J. papers by the absence of the detergent lauryl maltose neopentyl glycol (LMNG). LMNG is only present in the activity assay at a low concentration (0.003%; molar data should be given; by my calculation, this corresponds to 30 μM).

We regret this misunderstanding: our description was not clear enough. As the referee points out, in previous papers we purified the full length SpNOX with the detergent LMNG. In the current paper, we described only the protocol for SpNOX DH domain variant, a soluble cytoplasmic domain. We have now modified the text to clarify the difference between the purification of fulllength SpNOX variants, which were performed with detergent as cited in Vermot et al 2020, and the purification of DH domains, which are soluble and thus did not require detergent in the purification.

In light of the presence of lipids in cryo-EM-solved structures of DUOX and NOX2, it is surprising that the authors did not use reconstitution of the purified SpNOX in phospholipid (nanodisk?). The issue is made more complicated by the statement on p. 18 of "structures solved in detergent like ours" when no use of detergent in the solubilization and purification of SpNOX is mentioned in the Methods section (p. 21-22).

As stated above, detergent used to purify the full-length version of SpNOX. We did in fact perform some preliminary tests of reconstitution in nanodiscs. Different trials of negative staining studies showed heterogeneous size of SpNOX in nanodiscs and the initial images were not promising. Furthermore, in parallel, we had positive results in crystallography relatively quickly with protein in detergent. We thus focused on refining the crystals, which was a fairly long and mobilizing task; we decided to allocate time and resources to the promising avenue and did not further pursue nanodiscs.

We did not go in theCryo-EM direction because the small size of the protein was initially believed to be a significant barrier to successful Cryo-EM. Perhaps we could have pursued this avenue: while our manuscript here was submitted to eLife, another group deposited a preprint in BioRxiv using CryoEM to solve the structure of SpNOX (see comment below). This structure was solved in detergent so even in this CryEM structure there is no information on the potential roles of lipids as asked by the referee.

In this revised version, we have added a comment, in the last paragraph, in reference to the additional data available today thanks to the other structures generated by this other group (Murphy's group).

Can the authors provide information on whether E. coli BL21 is sufficiently equipped for the heme synthesis required for the expression of the TM domain of SpNOX. Was supplementation with δaminolevulinic acid used

The production of His-SpNox in E.coli C41(DE3) is without any δ-aminolevulinic acid supplementation. Supplementation was tested but no change was observed regarding the heme content (UV/Visible spectra) so we settled on the purification described by Vermot et al 2020. Initially, for the mBio paper (Haajar et al 2017), we performed heme titrations which gave stoichiometry between 1.35 to 1.5 heme/protein, indicating 2 hemes (these data were not shown). In the end in this work we observed two hemes in the crystal structure, thus confirming that E.coli, at least for this protein, did not need supplementation with δ-aminolevulinic acid .

The 3 papers on SpNOX present more than convincing evidence that SpNOX is a legitimate Nox that can serve as a legitimate model for eukaryotic Noxes (cyanide resistance, inhibition by DPI, absolute FAD dependence, and NADPH/NADH as the donor or electrons to FAD). It is also understood that the physiological role of SpNOX in S. pneumoniae is unknown and that the fact that it can reduce molecular oxygen may be an experimental situation that does not occur in vivo.

I am, however, linguistically confused by the statement that "SpNOX requires "supplemental" FAD". Noxes have FAD bound non-covalently and this is the reason that, starting from the key finding of Babior on NOX2 back in 1977 to the present, FAD has to be added to in vitro systems to compensate for the loss of FAD in the course of the purification of the enzyme from natural sources or expression in a bacterial host. I wonder whether this makes FAD more of a cosubstrate than a prosthetic group unless what the authors intend to state is that SpNOX is not a genuine flavoprotein.

We believe there is some confusion between SpNOX – the full length transmembran protein -- and SpNOXDH -- the cytosolic domain only. The sentence pinpointed by the referee was in fact “The strict requirement of FAD addition for SpNOXDH activity suggests that the flavin behaves as a cosubstrate”. This statement was about the isolated cytosolic domain that does not contain the TM part of the protein.

We agree that in WT NOX enzymes (including SpNOX) FAD is held within the enzyme structure and thus can be considered, by definition, as a prosthetic group. This is supported by the nanomolar affinity for FAD of SpNOX. We did not intend to say that NOX and SpNOX are not genuine flavoproteins.

On the other hand, when isolated, the affinity of DH domain for flavins drops to the µM level. This µM level of affinity does not allow stable maintenance of the flavin in the active site as illustrated by the spectra of Figure 3. This is instead the typical affinity of a substrate or a co-substrate (similar to that of substrate NADPH) that can be exchangeable and diffuse in and out of the active site. The DH domain recognizes and reduces flavins but, as a consequence of its lower affinity, will release to its environment free reduced flavins. Thus the isolated DH behaves as a flavin reductase that uses flavin as substrate. Such enzymes have already been well described (and some of them are of the FNR family). Such enzymes, using flavin as substrate, typically have affinity for flavin in the µM range and share with the SpNOX DH binding properties centered on the isoalloxazine ring only.

We understand that, in the text, to switch from the SpNOX to the SpNOX DH and for FAD from a prosthetic group to a diffusible co-substrate can be confusing. So, to make it clearer, we modified the following sentences and added references to “some flavin reductases characterization” that could provide support for the reader.

“The strict requirement of FAD addition for SpNOXDH activity and its µM level of affinity suggests that the flavin behaves as a co-substrate rather than a prosthetic group. As an isolated domain, SpNOXDH may work as a flavin reductase enzyme (Gaudu et al, 1994; Fieschi et al 1995; Nivière et al 1996), ..”

We hope that it will help.

I am also puzzled by the statement that SpNOX "does not require the addition of Cyt c to sustain superoxide production". Researchers with a Cartesian background should differentiate between cause and effect. Cyt c serves merely as an electron acceptor from superoxide made by SpNOX but superoxide production and NADPH oxidation occur independently of the presence of added Cyt c.

Thanks to the referee for pointing out this poor wording. We agree and have amended the text to clarify what we originally meant. It is now:

“SpNOXDH requires supplemental FAD to sustain both superoxide production, which can be observed in the presence of Cyt c (Figure 2A), and NADPH oxidation, which can be observed in the absence of Cyt c (Figure 2B).”

The ability of the DH domain of SpNOX (SpNOXDH) to produce superoxide is surprising to this reviewer.The result is based on the inhibition of Cyt c reduction by added superoxide dismutase (SOD) by 40%. In all eukaryotic Noxes superoxide is produced by the one-electron reduction of molecular oxygen by electrons originating from the distal heme, having passed from reduced FAD via two hemes. The proposal that superoxide is generated by direct transfer of electrons from FAD to oxygen deserves a more in-depth discussion and relies too heavily on the inhibitory effect of SOD. A control experiment with inactivated SOD should have been done (SOD is notoriously heat resistant and inactivation might require autoclaving).

The initial reports of a NOX DH-domain-only construct (that of human Nox4) producing superoxide are cited in the text. Moreover, natural flavin reductases are known to produce superoxide due to the release of free reduced flavin in the medium.

As explain above, FAD in full length SpNox is a relay for the electrons from NADPH to heme and is internal to the protein and thus devoted to this specific task.

In the case of SpNOX DH, its flavin reductase behavior leads to the release in the medium of free reduced flavin as a nonspecific diffusible electron carrier. It has been already demonstrated that such free reduced flavin can efficiently reduce soluble O2 and be a source of superoxide.

This has been particularly well documented in (Gaudu et al, 1994. J.Biol.Chem). We have added this reference to the text (see the modified sentence in a reply, 2 comments above).

Furthermore, we want to point to the referee that the link between flavin and superoxide production here is not only based on the inhibition by SOD. When we added the flavin inhibitor DPI we observed no more superoxide production from the DH domain (Figure 2C). This supports the role of free-reduced flavin in both the production of superoxide and also part of direct cyt C reduction as observed.

An unasked and unanswered question is that, since under aerobic conditions, both direct Cyt c reduction (60%) and superoxide production (40%) occur, what are the electron paths responsible for the two phenomena occurring simultaneously?

We thank the referee for dedication to a clear understanding of the mechanism used by the SpNOXDH construct. It pushes us to develop a clear description of the mechanism at work here for the readers. Please find below a proposal mechanism describing the electron transfer from NAD(P)H to free flavin that can, as diffusible species, then reduce non-specifically either the O2 or the Cyt.C encountered.

Author response image 1.

However, it is important to remember that this is not physiological, and rather the result of using a DH domain isolated from the TM of SpNOX. Nonetheless, it shows that the DH domain is fully functional for NAD(P)H as well as the hydride transfer.

This reviewer had difficulty in following the argument that the fact that the kcat of SpNOX and SpNOXDH are similar supports the thesis that the rate of enzyme activation is dependent on hydride transfer from nicotinamide to FAD.

We have amended the text to clarify this point. If the reaction rate is not affected by the presence or absence of the hemes in the TM domain, this inevitably implies that the rate is NOT limited by the electron transfer to the heme, and ultimately to O2, from the FAD, and thus the hydride transfer step that oxidizes the FAD must be the rate limiting step.

The section dealing with mutating F397 is a key part of the paper. There is a proper reference to the work of the Karplus group on plant FNRs (Deng et al). However, later work, addressing comparison with NOX2, should be cited (Kean et al., FEBS J., 284, 3302-3319, 2017). Also, work from the Dinauer group on the minimal effect of mutating or deleting the C-terminal F570 in NOX2 on superoxide production should be cited (Zhen et al., J. Biol. Chem. 273, 6575-6581, 1998).

We thank the reviewer for pointing out our unintended omission of these important works; we have amended the text and added the citations.

It is not clear why mutating F397 to W (both residues having aromatic side chains) would stabilize FAD binding.

In a few words, trp’s double ring can establish larger and stronger vanderWaals contact with the isoalloxazine ring than the phe sidechain. Our discussion regarding this point is extensive in the structural section where we compare the structures with F and W in this position. At this time we do not think it is necessary to add anything to the text.

Also, what is meant by "locking the two subdomains of the DH domain"? What subdomains are meant?

The two subdomains are the NADPH-binding domain and the FAD-binding domain, which we define on p 11 (“SpNOXDH presents a typical fold of the FNR superfamily of reductase domain containing two sub-domains, the FAD-binding domain (FBD) and an NADPH-binding domain (NBD) “) and which are labeled in Fig. 4. By “locking” we meant to convey immobilizing them into a specific conformation; we have amended the text to clarify this point.

Methodological details on crystallization (p. 11) should be delegated to the Methodology section. How many readers are aware that SAD means "Single Wavelength Anomalous Diffraction" or know what is the role of sodium bromide?

We have amended the text to emphasize the intended point, which is the different origins of the two DH structures: the de novo structure was possible through co crystallization with bromide, and the molecular replacement structure used the de novo structure as a model.

The data on the structure of SpNOX are supportive of a model of Nox activation that is "dissident" relative to the models offered for DUOX and NOX2 activation. These latter models suggested that the movement of the DH domain versus the TM domain was related to conversion from the resting to the activated state. The findings reported in this paper show that, unexpectedly, the domain orientation in SpNOX (constitutively active!) is much closer to that of resting NOX2. One of the criteria associated with the activated state in Noxes was the reduction of the distance between FAD and the proximal heme. The authors report that, paradoxically, this distance is larger in the constitutively active SpNOX (9.2 Å) than that in resting state NOX2 (7.6 Å) and the distance in Ca2+-activated DUOX is even larger (10.2 Å).

A point made by the authors is the questioning of the paradigm that activation of Noxes requires DH domain motion.

Instead, the authors introduce the term "tensing", within the DH domain, from a "relaxed" to a more rigid conformation. I believe that this proposal requires a somewhat clearer elaboration

It is clear that the distance between the FAD and NADPH shown in the Duox and Nox2 structures is too large for the chemical reaction of hydride transfer. Wu et al used the terms ‘tense’ and ‘relaxed’ to describe conformations of the DH domain corresponding to ‘short distance’ and ‘longer distance’, respectively, between the two ligand binding sites. We quoted this terminology and have amended the text to clarify that we envision a motion of the NBD relative to the FBD, as distinct from a larger motion of the whole DH domain relative to the TM domain.

The statement on p. 18, in connection to the phospholipid environment of Noxes, that the structure of SpNOX was "solved in detergent" is puzzling since the method of SpNOX preparation and purification does not mention the use of a detergent. As mentioned before, this absence of detergent in the present report was surprising because LMNG was used in the methods described in the mBio and Biophys. J. papers. The only mention of LMNG in the present paper was as an addition at a concentration of 0.003% in the activity assay buffers.

Please see our response to similar points above. Detergent was present for the solubilization of the full-length SpNOX.

The Conclusions section contains a proposal for the mechanism of conversion of NOX2 from the resting to the activated state. The inclusion of this discussion is welcome but the structural information on the constitutively active SpNOX can, unfortunately, contribute little to solving this important problem. The work of the Lambeth group, back in 1999 (cited as Nisimoto et al.), on the role of p67-phox in regulating hydride transfer from NADPH to FAD in NOX2 may indeed turn out to have been prophetic. However, only solving the structure of the assembled NOX2 complex will provide the much-awaited answer. The heterodimerization of NOX2 with p22-phox, the regulation of NOX2 by four cytosolic components, and the still present uncertainty about whether p67-phox is indeed the final distal component that converts NOX2 to the activated state make this a formidable task.

The work of the Fieschi group on SpNOX is important and relevant but the absence of external regulation, the absence of p22-phox, and the uncertainty about the target molecule make it a rather questionable model for eukaryotic Noxes. The information on the role of the C-terminal Phe is of special value although its extension to the mechanism of eukaryotic Nox activation proved, so far, to be elusive.

We really thank the referee for the positive comments on our work and the deep interest shown by this careful evaluation.

We understand the arguments of the referee regarding the relevance of our work here to eukaryotic NOX, but we do not share the reservations expressed. While human NOXes need interactions with other proteins or have EF-hand or other domains that control them, SpNOX corresponds exactly to the minimal core common to any NOX isoform. In fact, because SpNOX has only this conserved core, it is unique in that it can work as a constitutively active NOX without protein-protein interactions or regulatory domains. Thus the fundamentals of electron transfer mechanisms of NOX enzyme are present in SpNOX.

There might be some differences in the internal organization from isoform to isoform (as regarding the relative DH domain vs TM domain orientation) but considering the similarity between NOX2 and SpNOX topology we are rather confident that the SpNOX structure will turn out to be a reasonable model of the activated NOX2 structure. History will tell.

In any case, this work on SpNOX allowed us to highlight hydride transfer as the limiting step and also to highlight some structural differences that could be at the source of the regulation in eukaryotic NOX. In itself, we think this is a significant contribution to the field.

We warmly thank both referees for their constructive remarks and their help in the improvement of this manuscript.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

• The manuscript states that the flavin "behaves" like a co-substrate and thereby reports on the Km for the flavins. I feel that this terminology might be confusing. The flavin is unchanged after the reaction, and what matters is the enzyme's affinity for the flavin and the flavin concentration needed to saturate the enzyme (to have it in the fully holo form).

See above -- answering many questions from referee2, we have extensively commented on that point (substrate, cofactor, affinity, etc..) and made some adjustments in the text to clarify. We hope it is now satisfactory.

• I could not find the methodological description of the experiments performed to measure the Km for the flavins, and the legend of Figure S4 does not help in this regard. I think that the data (left panels of S4) should be interpreted as binding curves with associated Kd values.

We have changed the text to clarify the method used to measure Km for flavins.

• A related point is that the manuscript refers to Km as an "affinity". This is inappropriate and should be avoided, as the Km is not the Kd.

We agree with the referee that the Km is not the Kd. However, under the appropriate conditions, to which our experiments conform, Km is accepted as a relevant approximation of affinity (Srinisivan, FEBS Journal, v 289 pp 6086-6098 2022). We have added a sentence to clarify this point and cite this reference in the text.

• The environment around the putative oxygen site should be shown. The text indicates that "the residues characteristic of the O2 reducing center in eukaryotic FRD domains of NOX and DUOX enzymes are not conserved in SpNOX." How does the site look? This point relates to the more general comment above on the oxidizing substrate used by this bacterial NOX.

This is a really interesting point that contains many potential biological developments for future studies of this prokaryotic family of NOX enzymes. While we were submitting this work to eLife for evaluation, another group (Murphy's lab) filed a pre-publication in BioRXiv, in which they also solved the structure of SpNOX but this time by CryoEM with an unexpected level of resolution for such a small protein (their paper is not yet published but probably under peer review somewhere). In their work, they made a special effort to identify the O2 reducing center (bacterial NOX sequences alignment, mutation studies, …) They were not able to localize such a site with accuracy. There is also other complementary data between their work and ours. So, we will add a paragraph at the end of the discussion to comment on this parallel work and to emphasize on the complementarity of their studies and what it brings to the final understanding of this enzyme.

• The section "A Close-up View of NOX's NAD(P)H Binding Domains vs the FNR Gold Standard" should be clarified.

I found it difficult to understand. Is the different conformation of Phe397 creating the crevice? Could NADPH be modeled in NOX2 and DUOX in the same conformation observed in FNR and modeled in the bacterial NOX? Or would there be clashes, implying the necessity of larger conformational changes to bring the nicotinamide closer to the FAD?

Please see responses above on this point; we have amended the text to clarify. In a few words, we propose that activation in the eukaryotic enzymes would entail NBD subdomain (containing NADPH site) towards the FBD subdomain (containing FAD) through an internal motion within the DH domain. Doing so, they would approach the DH domain topology of SpNOX, which models an active state.

Reviewer #2 (Recommendations For The Authors):

On p. 6, second line, it should be (Figure 1C and 1D). Space is missing between C and "and".

On p. 9, in Figure 3, the labeling A and B are missing. Also, the legend of part B does not correspond to the actual graph colors. Thus, the tracing of F397W is red and not grey as indicated in the legend.

Corrected. Thank you

Author Response

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

Reviewer #1 (Recommendations For The Authors):

(1) V2 epitopes exhibit properties of CD4i epitopes in that they are largely absent from the native Env surface, probably by glycan-occlusion, but become more exposed upon CD4 binding. Although the V2-scaffolds were produced in GnTi- cells to produce highmannose proteins, it appears that no systematic analysis of glycan content or structure was carried out save for enzymatic deglycosylation of the constructs to sharpen bands on SDS-PAGE gels. It would be helpful if the authors could comment on how the lack of this information might impact their conclusions.

We thank the reviewer for this comment.

The lack of native glycan structures is a common phenomenon in all HIV studies involving in vitro cell culture-expressed envelope proteins.

As the reviewer mentioned, it is clear that our V1V2 scaffolds produced in GnTi-cells contain the expected high-mannose glycans, as evident from a significant shift and sharpening of the protein bands on the SDS-PAGE gel upon deglycosylation with the PNGase enzyme.

In our previously published studies by Chand et al.,2017* (ref. below), the V1V2 scaffolds were shown to bind to glycan-dependent PG9 antibody suggesting that the conformation of the PG9 epitope is retained in the high-mannose V1V2 scaffold. This information has also been added to the “Hypothesis and Experimental Design” section of the Results in the revised manuscript.

Additionally, as shown in Results, the human antibodies elicited in study participants against native glycosylated envelope protein due to natural HIV-1 infection distinguished the H173 and Y 173 epitopes in the high-mannose scaffolds, which was also recapitulated in our mouse studies using the GnTi-expressed high-mannose V1V2 scaffolds as antigens.

Therefore, it does not seem likely that differences in glycans per se majorly affected the binding or the conclusions from our studies.

*Chand S, Messina EL, AlSalmi W, Ananthaswamy N, Gao G, Uritskiy G, Padilla-Sanchez V, Mahalingam M, Peachman KK, Robb ML, Rao M, Rao VB. Glycosylation and oligomeric state of the envelope protein might influence HIV-1 virion capture by α4β7 integrin. Virology. 2017 Aug;508:199-212. doi: 10.1016/j.virol.2017.05.016. Epub 2017 May 31. PMID: 28577856; PMCID: PMC5526109.

(2) Similarly, the MD simulations appear to be performed without taking glycan structure/occupancy.

We were unable to perform glycan-dependent MD simulation studies because of the high computational demands and also the technical limitations that existed at the time of the study several years ago. Therefore, we focused on the protein backbone of the short C-strand in the V2 region that lacks glycan sites and in previous published studies has been demonstrated as conformationally polymorphic.

Since this C-strand epitope is the binding site for many V2-directed antibodies identified previously, we hypothesized that it is relevant to explore this small immunogenic epitope for its propensity to change conformation due to an escape mutation discovered at residue 173 in a natural HIV-1 infection. How might this epitope behave in MD simulations in the presence of different glycans requires further investigation.

eLife assessment

This study provides a detailed evaluation of how HIV evades nascent immune pressure from people living with HIV followed nearly immediately after infection. There is convincing evidence that H173 mutations in the V2 loop was a key determinant of selection pressure and escape. These data are congruent with protection in the RV144 clinical trial, the only trial that showed protection from infection. Overall, this study is an important contribution to the field.

Reviewer #1 (Public Review):

Summary:

This study used a unique acute HIV-1 infection cohort, RV217, to study the evolution of transmitted founder viral Envelope sequences under nascent immune pressure. The striking feature of the RV217 cohort is the ability to detect viremia in the first week of infection, which can be followed at discrete Fiebig stages over long time intervals. This study evaluated Env sequences at 1 week, 4 weeks, and 24 weeks to provide a picture of viral and immunological co-evolution from Fiebig Stage I (1 week), Fiebig Stages IV (4 weeks), and Fiebig Stage VI (>24 weeks). This study design enabled lineage tracing of viral variants from a single transmitted founder (T/F) over the Fiebig Stages I, IV, and VI under nascent immune pressure generated in response to the T/F virus and its subsequent mutants.

Strengths:

As expected, there were temporal differences in the appearance of virus quasispecies among the individuals, which were located predominantly in solvent-exposed residues of Env, which is consistent with prior literature. Interestingly, two waves of antibody reactivity were observed for variants with mutations in the V2 region that harbors V2i and V2p epitopes correlated with protection in the RV144 clinical trial. Two waves of antibody response, detected by SPR, were observed, with the first wave being predominated by antibodies specific for the T/F07 V2 epitope associated with H173 located on the C -strand in the V2 region. The second wave was dominated by antibodies specific for an H to Y mutation at 173 that emerged due to antibody-mediated pressure to the original H173 virus. This is a remarkable finding in three ways.

First, the mutation is in the C β-strand, an unlikely paratope contact residue, as this region of the V2 loop is shielded by glycans in Env trimer structures with full glycan representation (see PDB:5t3x). The structure used for modeling in the current study was an earlier structure, PDB:4TVP, that had many truncated glycans. This does not detract from the finding that the H173Y mutation likely causes a conformational shift from a more rigid helical/coil conformation to a more dynamic conformation with a β-stranded and -sheet core preference as indicated by the literature and by the MD simulations carried out by the authors. This observation suggests that using V2 scaffolds with both the H173 and H173Y variants will increase the breadth of potentially protective antibody responses to HIV-1, as indicated in reference 42, cited by the authors. Interestingly, the H173Y mutation abrogates reactivity to mAb CH58 and attenuates reactivity to mAb CH59 isolated from RV144 volunteers. These mAbs recognize conformationally distinct V2 epitopes, adding further credence to the conclusion that the H173Y mutation results in a conformational switch of the V2 region.

Second, the H173Y mutation affects the conformation of V2 epitopes recognized by mAbs that do not neutralize T/F HIV-1 while mediating potent ADCC. The ADCC data in the manuscript provides strong support for this hypothesis and augers for further exploration of the V2 epitopes as HIV-1 vaccine targets.

Third, it is striking that immunogens based on the H173Y mutation successfully recapitulated the observed human antibody responses in wild-type Balb/c mice. The investigators used their extensive knowledge of V2 structures and scaffold-immunogens to create the library of constructs used for this study. In this case, the ΔDSV mutation increased the breadth and magnitude of the murine antibody responses.

Reviewer #2 (Public Review):

Summary:

In this study, researchers aimed to understand how a transmitted/founder (T/F) HIV virus escapes host immune pressure during early infection. They focused on the V1V2 domain of the HIV-1 envelope protein, a key determinant of virus escape. The study involved four participants from the RV217 Early Capture HIV Cohort (ECHO) project, which allowed tracking HIV infection from just days after infection.

The study identified a significant H173Y escape mutation in the V2 domain of a T/F virus from one participant. This mutation, located in the relatively conserved "C" β-strand, was linked to viral escape against host immune pressure. The study further investigated the epitope specificity of antibodies in the participant's plasma, revealing that the H173Y mutation played a crucial role in epitope switching during virus escape. Monoclonal antibodies from the RV144 vaccine trial, CH58, and CH59, showed reduced binding to the V1V2-Y173 escape variant. Additionally, the study examined antibody-dependent cellular cytotoxicity (ADCC) responses and found resistance to killing in the Y173 mutants. The H173Y mutation was identified as the key variant selected against the host's immune pressure directed at the V2 domain.

The researchers hypothesized that the H173Y mutation caused a structural/conformational change in the C β-strand epitope, leading to viral escape. This was supported by molecular dynamics simulations and structural modeling analyses. They then designed combinatorial V2 immunogen libraries based on natural HIV-1 sequence diversity, aiming to broaden antibody responses. Mouse immunizations with these libraries demonstrated enhanced recognition of diverse Env antigens, suggesting a potential strategy for developing a more effective HIV vaccine.

In summary, the study provides insights into the early evolution of HIV-1 during infection, highlighting the importance of the V1V2 domain and identifying key escape mutations. The findings suggest a novel approach for designing HIV vaccine candidates that consider the diversity of escape mutations to induce broader and more effective immune responses.

Strengths:

The article presents several strengths:

(1) The experimental design is well-structured, involving multiple stages from phylogenetic analyses to mouse model testing, providing a comprehensive approach to studying virus escape mutations.

(2) The study utilizes a unique dataset from the RV217 Early Capture HIV Cohort (ECHO) project, allowing for the tracking of HIV infection from the very early stages in the absence of antiretroviral therapy. This provides valuable insights into the evolution of the virus.

(3) The use of advanced techniques such as phylogenetic analyses, nanoscaffold technology, controlled mutagenesis, and monoclonal antibody evaluations demonstrates the application of cutting-edge methodologies in the study.

(4) The research goes beyond genetic analysis and provides an in-depth characterization of the escape mutation's impact, including structural analyses through Molecular Dynamics simulations, antibody responses, and functional implications for virus survival.

(5) The study provides insights into the immune responses triggered by the escape mutation, including the specificity of antibodies and their ability to recognize diverse HIV-1 Env antigens.

(7) The exploration of combinatorial immunogen libraries is a strength, as it offers a novel approach to broaden antibody responses, providing a potential avenue for future vaccine design.

(8) The research is highly relevant to vaccine development, as it sheds light on the dynamics of HIV escape mutations and their interaction with the host immune system. This information is crucial for designing effective vaccines that can preemptively interfere with viral acquisition.

(9) The study integrates findings from virology, immunology, structural biology, and bioinformatics, showcasing an interdisciplinary approach that enhances the depth and breadth of the research.

(10) The article is well-written, with a clear presentation of methods, results, and implications, making it accessible to both specialists and a broader scientific audience.

Author Response

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

Public Reviews:

Reviewer #1 (Public Review):

Summary

Liao et al leveraged two powerful genomics techniques-CUT&RUN and RNA sequencing-to identify genomic regions bound by and activated or inactivated by SMAD1, SMAD5, and the progesterone receptor during endometrial stromal cell decidualization. Additionally, the authors generated novel knock-in HA-SMAD1 and PA-SMAD5 tagged mice to combat antibody issues facing the field, generating a novel model to advance the study of BMP signaling in the female reproductive tract. During decidualization in a murine model, SMAD1/5 are bound to many genomic sites of genes important in decidualization and pregnancy and coregulated responses with progesterone receptor signaling.

Strengths

The authors utilized powerful next generation sequencing and identified important transcriptional mechanisms of SMAD1/5 and PGR during decidualization in vivo.

Weaknesses<br /> None.

Overall, the manuscript and study are well structured and provide critical mechanistic updates on the roles of SMAD1/5 in decidualization and preparation of the maternal endometrium for pregnancy.

We thank you for the summary and consideration.

Reviewer #2 (Public Review):

Summary:

Liao and colleagues generated tagged SMAD1 and SMAD5 mouse models and identified genome occupancy of these two factors in the uterus of these mice using the CUT&RUN assay. The authors used integrative bioinformatic approaches to identify putative SMAD1/5 direct downstream target genes and to catalog the SMAD1/5 and PGR genome co-localization pattern. The role of SMAD1/5 on stromal decidualization was assayed in vitro on primary human endometrial stromal cells. The new mouse models offer opportunities to further dissect SMAD1 and SMAD5 functions without the limitation from SMAD antibodies, which is significant. The CUT&RUN data further support the usefulness of these mouse models for this purpose.

Strengths:

The strength of this study is the novelty of new mouse models and the valuable cistromic data derived from these mice. Overall the present manuscript is an excellent resource paper for the field of reproductive biology.

Weaknesses:

The weakness of the present version of the manuscript includes the self-limited data analysis approaches such as the proximal promoter based bioinformatic filter and an outdated method on inferring the cell type composition. Evidence was provided for potential associations between SMAD1/5 and other major transcription factors. However, causal effects of SMAD1/5 on the genome occupancy of other major uterine transcription factors were discussed but not experimentally examined in the present manuscript, which is understandable.

For data in Figure 2B, the current manuscript fails to elaborate the common and distinct features between clusters 1 and 3 as well as the biological significance of having two separate clusters for SMAD1. In addition, Figure S1A shows overlapping genome occupancy between SMAD1 and SMAD5, which is not clearly demonstrated in Figure 2B.

Thank you for the comments. We’ve added additional interpretations in Lines 281-283, addressing the clustering results mentioned in Figure 2B as suggested. We do appreciate the overlapping genome occupancy in Cluster 1, although the signal intensities may differ between two groups.

Lines 281-283:

“Peaks in cluster 1 exhibit a shared enrichment for both SMAD1 and SMAD5, whereas clusters 2 and 3 demonstrate preferential enrichment for SMAD5 and SMAD1, respectively.”

For data in Figure 5A, the result description does not provide adequate information to guide readers to full understanding of the data. The biological meaning behind the three PR clusters is not stated nor speculated. Moreover, Figure 5A and Figure S1B are inherently connected but fail to be adequately described in the main text.

Thank you for the comments. We’ve added additional interpretations in Lines 415-421 discussing the clustering results mentioned in Figure 5A, together with Supplement Figure 1C (Former Supplement Figure 1B) as suggested.

Lines 415-421:

“Based on the k-means clustering results of the peaks, we demonstrated clusters with shared occupancy between SMAD1/5 and PR (cluster 1), preferential deposition in the SMAD1 (cluster 2), SMAD5 (cluster 4) and PR (clusters 3,5), respectively. Interestingly, between clusters 3 and 5, although the primary enrichment is for PR, overall, the signal intensities for SMAD5 are higher in cluster 5. Together with previous analysis on genes uniquely or commonly bound by SMAD1/5 (Supplement Figure 1A), we speculate such observation can be attributed to a subset of the genes that are potentially co-regulated by SMAD5 and PR.”

Reviewer #3 (Public Review):

Summary:

As SMAD1/5 activities have previously been indistinguishable, these studies provide a new mouse model to finally understand unique downstream activation of SMAD1/5 target genes, a model useful for many scientific fields. Using CUT&RUN analyses with gene overlap comparisons and signaling pathway analyses, specific targets for SMAD1 versus SMAD5 were compared, identified, and interpreted. These data validate previous findings showing strong evidence that SMADs directly govern critical genes required for endometrial receptivity and decidualization, including cell adhesion and vascular development. Further, SMAD targets were overlapped with progesterone receptor binding sites to identify regions of potential synergistic regulation of implantation. The authors report strong correlations between progesterone receptor and SMAD1/5 direct targets to cooperatively promote embryo implantation. Finally, the authors validated SMAD1/5 gene regulation in primary human endometrial stromal cells. These studies provide a data-rich survey of SMAD family transcription, defining its role as a governor of early pregnancy.

Strengths:

This manuscript provides a valuable survey of SMAD1/5 direct transcriptional events at the time of receptivity. As embryo implantation is controlled by extensive epithelial to stromal molecular crosstalk and hormonal regulation in space and time, the authors state a strong, descriptive narrative defining how SMAD1/5 plays a central role at the site of this molecular orchestration. The implementation of cutting-edge techniques and models and simple comparative analyses provide a straightforward, yet elegant manuscript.

Although the progesterone receptor exists as a major regulator of early pregnancy, the authors have demonstrated clear evidence that progesterone receptor with SMAD1/5 work in concert to molecularly regulate targets such as Sox17, Id2, Tgfbr2, Runx1, Foxo1 and more at embryo implantation. Additionally, the authors pinpoint other critical transcription factor motifs that work with SMADs and the progesterone receptor to promote early pregnancy transcriptional paradigms.

Weaknesses:

Although a wonderful new tool to ascertain SMAD1 versus SMAD5 downstream signaling, the importance of these factors in governing early pregnancy is not novel. Furthermore, functional validation studies are needed to confirm interactions at promoter regions. Additionally, the authors presume that all overlapped genes are shared between progesterone receptor and SMAD1/5, yet some peak representations do not overlap. Although, transcriptional activation can occur at the same time, they may not occur in the same complex. Thus, further confirmation of these transcriptional events is warranted.

Thank you for the comments. We recognized this limitation and discussed future options regarding this in Lines 578-583.

Lines 578-583:

“In this study, we determined the overlapped transcriptional control between SMAD1/5 and PR at the gene level, and functionally validated the regulatory effect at the transcript level in a human stromal cell decidualization model. While we observe a subset of peak representations that do not overlap at the base pair level in the promoter regions, future functional screenings at the promoter level, such as luciferase reporter assays to assess transcriptional co-activation by SMAD1/5 and PR, will advance this study.”

Since whole murine uterus was used for these studies, the specific functions of SMAD1/5 in the stroma versus the epithelium (versus the myometrium) remain unknown. Further work is needed to delineate binding and transcriptional activation of SMAD1/5 and the progesterone receptor in the uterine compartments.

We thank the reviewer for the insightful comment. Given the multifaceted roles of SMAD1/5 play the female reproductive tract, we concur that future studies will benefit from a more compartmentalized approach, as discussed in Lines 526-538.

Lines 526-538:

“Published studies have shown that nuclear SMAD1/5 localize to the stroma and epithelium during the decidualization process at 4.5 dpc, during the window of implantation. Conditional deletion of SMAD1/5 exclusively in the uterine epithelium using lactoferrin-icre (Ltf-icre) results in severe subfertility due to impaired implantation and decidual development. Conditional deletion of SMAD1/5/4 exclusively in the cells from mesenchymal lineage (including uterine stroma) using anti-Mullerian hormone type 2 receptor cre (Amhr2-cre) results in infertility with defective decidualization. Given the essential roles of SMAD1/5 in both stroma and epithelium identified by previous studies, we believe that the transcriptional co-regulatory roles of SMAD1/5 and PR reported here using the whole uterus validates a relationship between SMAD1/5 and PR in both the stromal and epithelial compartments. However, it does not rule out potential coregulatory roles of SMAD1/5 and PR in the myometrium, immune cells, and/or endothelium, given that whole uterus was used. The specific transcriptional evaluations of SMAD1/5 in the stroma versus the epithelium would require future validations using single-cell sequencing and/or spatial transcriptomic analysis.”

There are asynchronous gene responses in the SMAD1/5 ablated mouse model compared to the siRNA-treated human endometrial stromal cells. These differences can be confounding. Further investigation is needed to understand the meaning of these differences and as they relate to the entire SMAD transcriptome.

Thank you for the comments. In the current study, we used human endometrial stromal cells as a model to validate our findings functionally, aiming to mimic the specific time point during decidualization. We acknowledge the similarities and differences between the mouse and human cell models, and this information needs to be considered when evaluating genome-wide effects on the transcriptome. This point is discussed ins Lines 589-597.

Lines 589-597:

“Since mice only undergo decidualization upon embryo implantation whilst human stromal cells undergo cyclic decidualization in each menstrual cycle in response to rising levels of progesterone, asynchronous gene responses may occur in comparison between mouse models and human cells. However, cellular transformation during decidualization is conserved between mice and humans, which makes findings in the mouse models a valuable and transferable resource to be evaluated in human tissues. Accordingly, our functional validation studies were performed using human endometrial stromal cells induced to decidualize in vitro for four days, which models the early phases of decidualization. Additional transcriptomic studies of the SMAD1/5 perturbations in human endometrial stromal cells will be of great resource in understanding the entire SMAD1/5 regulomes in humans.”

Recommendations for the authors:

Reviewer #2 (Recommendations For The Authors):

The inference on the cell type composition could use updated bioinformatic tools, which are purely computational without costly and time-consuming wet-lab resources. Perhaps this part of the description could be streamlined if the authors chose to use the method in the current version.

We thank the reviewer for the suggestion. We added the analysis of the cell type composition using the updated tool CIBERSORTx (PMID:31061481) and included the results and discussion regarding the cell type composition changes in Supplement Figure 1B and Lines 392-407.

Lines 392-407

“To explore the major cell types regulated by SMAD1/5, first, we used CIBERSORTx to analyze and depict changes in the cell populations upon SMAD1/5 depletion in the mouse uterus during early pregnancy. By imputing the bulk uterine gene expression profiles to previously published mouse uterine single-cell datasets using CIBERSORTx, we were able to compare changes across both samples and cell types upon the SMAD1/5 perturbation in the mouse uterus. We highlight the proportional increase in the epithelial cells, as well as the decrease in the decidual stromal cells and smooth muscle cells in mice lacking uterine SMAD1/5 during the periimplantation phase (Supplement Figure 1B). Such cell populational changes are in line with the phenotypical observations of decidualization failure and excessive proliferation in the epithelial compartment. In addition, to explore the expression patterns of SMAD1/5 direct targets in human, we profiled the expression levels of the key “up-targets” and “down-targets” in the different cell types of the human endometrium. Using previously published single-cell RNA seq data of human endometrium, we visualized the expression patterns of suppressive targets and activating targets of SMAD1/5 (Figure 4E). Apart from the major epithelial and stromal compartments, SMAD1/5 target genes are also widely expressed in the immune cell populations. Such observations reinforced the importance of the BMP signaling pathways in establishing an immune-privileged environment at the maternal-fetal interface.”

eLife assessment

This study presents two valuable new mouse models that individually tag proteins from the SMAD family to identify distinct roles during early pregnancy. Convincing evidence is provided that SMAD1 and SMAD5 target many of the same genomic regions as each other and the progesterone receptor. Given the broad effect of these signaling pathways in multiple systems, these new tools will most likely interest readers across biological disciplines.

Reviewer #2 (Public Review):

Summary:

Liao and colleagues generated tagged SMAD1 and SMAD5 mouse models and identified genome occupancy of these two factors in the uterus of these mice using the CUT&RUN assay. The authors used integrative bioinformatic approaches to identify putative SMAD1/5 direct downstream target genes and to catalog the SMAD1/5 and PGR genome co-localization pattern. The role of SMAD1/5 on stromal decidualization was assayed in vitro on primary human endometrial stromal cells. The new mouse models offer opportunities to further dissect SMAD1 and SMAD5 functions without the limitation from SMAD antibodies, which is significant. The CUT&RUN data further support the usefulness of these mouse models for this purpose.

Strengths:

The strength of this study is the novelty of new mouse models and the valuable cistromic data derived from these mice. This revised manuscript provides lots of food for thought inside and outside of the field of reproductive biology.

Weaknesses:

Causal effects of SMAD1/5 on the genome occupancy of other major uterine transcription factors were discussed but not experimentally examined in the present manuscript, which is understandable.

Author Response

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

eLife Assessment

This study investigated the factors related to understudied genes in biomedical research. It showed that understudied genes are largely abandoned at the writing stage, and it identified a number of biological and experimental factors that influence which genes are selected for investigation. The study is a valuable contribution to this branch of meta-research, and while the evidence in support of the findings is solid, the interpretation and presentation of the results (especially the figures) needs to be improved.

We thank the editor and reviewers for their detailed and thoughtful assessment of our work. Below, we present detailed responses to reviewers’ comments and suggestions. We are also submitting a version edited for clarity of presentation and precision of interpretation.

Following the eLife assessment, we also tried to identify further statements where results could be presented in a more precise way.

First, in the section Subsequent reception by other scientists does not penalize studies on understudied genes, we now state “This result again opposes the hypothesis that less-investigated genes will yield articles with lower impact.”

Second, in section Identification of biological and experimental factors associated with selection of highlighted genes, we now state:

“We cautiously hypothesize that this might reflect on many different research groups producing reagents surrounding the genes that they actively study. The most informative continuous factor is the number of research articles about a gene (Figure 1B).”, removing claims of causality.

Finally, for improved readability, we have moved all supplemental tables into separate .xlsx files.

Reviewer #1 (Public Review):

Summary and strengths

The authors tried to address why only a subset of genes are highlighted in many publications. Is it because these highlighted genes are more important than others? Or is it because there are non-genetic reasons? This is a critical question because in the effort to discover new genes for drug targets and clinical benefit, we need to expand a pool of genes for deep analyses. So I appreciate the authors' efforts in this study, as it is timely and important. They also provided a framework called FMUG (short for Find My Understudied Gene) to evaluate genes for a number of features for subsequent analyses.

We thank the reviewer for their insightful comments and are pleased that the reviewer shares our appreciation for the gravity of these questions. As the reviewer emphasizes, it is critical to understand whether the choice of genes reflects their importance or non-genetic reasons. Previously we and others demonstrated that this choice does not reflect biological importance, when the latter is assessed through unbiased genome-wide data (e.g.: Haynes et al., 2018; Stoeger et al. 2018). Now we contribute to this critical question by systematically evaluating individual non-genetic reasons. We address the reviewer’s comments below.

Weaknesses

Many of the figures are hard to comprehend, and the figure legends do not sufficiently explain them.

For example, what was plotted in Fig 1b? The number of articles increased from results -> write-ups -> follow-ups in all four categories with different degrees. But it does not seem to match what the authors meant to deliver.

We apologize for the lack of clarity. We identified two interrelated elements that we have now fixed: i) the prior figure legend provided for each genomics approach n number of articles, such as “GWAS (n=450 articles)”; ii) the prior y-axis was labelled “Number of articles”.

Addressing the first element, we now rephrased the legend for clarity:

“b, We identified articles reporting on genome-wide CRISPR screens (CRISPR, 15 focus articles and 18 citing articles), transcriptomics (T-omics, 148 focus articles and 1,678 citing articles), affinity purification–mass spectrometry (AP-MS, 296 focus articles and 1,320 citing articles), and GWAS (450 focus articles and 3,524 citing articles). Focusing only on protein-coding genes (white box plot), we retrieved data uploaded to repositories describing which genes came up as “hits” in each experiment (first colored box plot). We then retrieved the hits mentioned in the titles and abstracts of those articles (second colored box plot) and hits mentioned in the titles and abstracts of articles citing those articles (third colored box plot). Unique hit genes are only counted once.”

The number of genes in each box plot is now reported in the x-axis labels for each step. For example, the results for CRISPR were obtained from 15 focus studies (original research) and 18 subsequent studies (papers citing focus articles). Those 15 studies identified 9,268 genes where loss-of-function changed phenotypes but, in their titles and abstracts, mentioned only 18 of those 9,268 genes. While the 9,268 hit genes have received similar research attention to the entirety of protein-coding genes, the 18 hit genes mentioned in the title or abstract are significantly more well studied. The articles citing the focus articles also only mentioned in their titles and abstracts 19 highly studied hit genes.

Addressing the second element, we updated the axis label to “Number of articles about gene”, to distinguish it from number of articles mentioned in the legend, convey that this is the number of articles about each gene that were published independently of the genomics assays we inspect. To further underscore this point we now label the “20% highest-studied genes” that we mention in the main text, and reworded the figure caption to better capture where the critical increase occurs: “A shift in focus towards well-studied genes occurs during the summarization and write-up of results and remains in subsequent studies.”.

Fig 4 is also confusing. It appears that the genes were clustered by many features that the authors developed. But does it have any relationship with genes being under- or over-studied?

We again apologize for the lack of clarity. As is described in the main text, while the results of Figs. 1-2 suggest that gene popularity may be predict the highlighting of a differentially expressed gene in the title or abstract, we want to conduct a systematically analysis of the factors that correlate with such a decision. We thus build a set of 45 factors that have been discussed as factors explaining why some genes receive increased research attention.

The data in Fig. 4 shows that those 45 factors are not independent but that some are highly correlated. Because of those correlations, we are able to select a smaller number as representative of the full set. Those are the default factors shown to users of FMUG. While users can choose all factors that are significantly correlated with the highlighting in title or abstract, the default of presenting factors representing different clusters of factors enabled us to limit the number of factors that are initially displayed.

Please note that following the suggestion of Reviewer 3, we have now moved this Figure to the supplemental material, as Figure S11.

Reviewer #2 (Public Review)

Summary and strengths

In this manuscript the authors analyse the trajectory of understudied genes (UGs) from experiment to publication and study the reasons for why UGs remain underrepresented in the scientific literature. They show that UGs are not underrepresented in experimental datasets, but in the titles and abstracts of the manuscripts reporting experimental data as well as subsequent studies referring to those large-scale studies. They also develop an app that allows researchers to find UGs and their annotation state. Overall, this is a timely article that makes an important contribution to the field. It could help to boost the future investigation of understudied genes, a fundamental challenge in the life sciences. It is concise and overall well-written, and I very much enjoyed reading it. However, there are a few points that I think the authors should address.

We thank the reviewer for their kind assessment.

Weaknesses

The authors conclude that many UGs "are lost" from genome-wide assay at the manuscript writing stage. If I understand correctly, this is based on gene names not being reported in the title or abstract of these manuscripts. However, for genome-wide experiments, it would be quite difficult for authors to mention large numbers of understudied genes in the abstract. In contrast, one might highlight the expected behaviour of a well-studied protein simply to highlight that the genome-wide study provides credible results.

We agree that it is not reasonable to expect a title or abstract to highlight hundreds or even thousands of differentially expressed genes. We’ve now extended our Study Limitations section to address this:

“we take a gene being mentioned in the title or abstract of an article as a proxy for a gene receiving attention by the article’s authors. The title and abstract are space-limited and thus cannot accommodate discussion of large numbers of genes.”

We also agree that highlighting the expected behavior of a well-studied protein may provide credibility to a study and increase confidence on other results. The soundness of such a strategy was quantitatively studied in a study by Uzzi et al. (Science 2013), which we now include in the section on study limitations as:

“authors beginning manuscripts with something familiar before introducing something new”.

To convey the practical limitation of abstracts needing to be concise, we added the following sentence to our discussion section, when suggesting controlled trials that add genes to abstracts:

“This intervention would need to be carefully designed since abstracts are limited in their size.”

To avoid over-interpretation we have in the discussion also extended the sentence on “lost in a leaky pipeline” to “lost to titles and abstracts of research articles in a leaky pipeline”.

Our focus on titles and abstracts has been equally motivated by their availability (full text still is often behind paywalls and/or not accessible for bulk-download and text-mining) and by abstracts being the most visible and most read parts of research articles (e.g.: bioRxiv estimates that for the preprint for the present manuscript, the abstract was read ~10 times more frequently than full-text HTML and 4 times more frequently than the pdf).

Could this bias the authors' conclusions and, if so, how could this be addressed? For example, would it be worth to normalise studies based on the total number of genes they cover?

We previously described that – in line with the reviewer’s expectations – unstudied genes are preferentially added to the title or abstract of articles that feature more genes in the title or abstract (Stoeger et al., Plos Biology, 2022; Fig. 2B). Normalizing by the total number of genes should thus preserve the pronounced division between well-studied genes and unstudied genes show in Figure 1B. In line with these predictions, we randomly select one gene per title/abstract and find that the effect remains (see new Figure S7).

Author response image 1.

Figure 1B is confusing in its present form. I think the plot and/or the legend need revising. For example, what "numbers to the right of each box plot" are the authors referring to? Also, I assume that the filled boxes are understudied genes and the empty/white box is "all genes", but that's not explained in the legend. In the main text, the figure is referred to with the sentence "we found that hit genes that are highlighted in the title or abstract are strongly over-represented among the 20% highest-studied genes in all biomedical literature ". I cannot follow how the figure shows this. My interpretation is that the y-axis is not showing the number of articles, but represents the percentage of articles mentioning a gene in the title/abstract, displayed on a log scale. If so, perhaps a better axis labels and legend text could be sufficient. But then one would also need to somehow connect this to the statement in the main text about the 20% highest-studied genes (a dashed line?). Alternatively, the authors could consider other ways of plotting these data, e.g. simply plotting the "% of publication in which a gene appears" from 0-100% or so.

Reviewer 1 raised a similar point on overall figure clarity. We identified two interrelated elements that contribute to overall confusion and have now fixed them (see response to Reviewer 1 beginning on page 2 of this document).

We attempted an alternative plotting of Fig 1B according to the reviewer’s suggestion. In the version below, the y-axis instead shows the percent of gene-related articles that are about each gene. We chose to keep the original y-axis (showing number of articles about each gene) as it additionally conveys the absolute scale of scholarship on individual genes.

Author response image 2.

Reviewer #3 (Public Review):

Summary and strengths

The manuscript investigated the factors related to understudied genes in biomedical research. It showed that understudied are largely abandoned at the writing stage and identified biological and experimental factors associated with selection of highlighted genes.

It is very important for the research community to recognize the systematic bias in research of human genes and take precautions when designing experiments and interpreting results. The authors have tried to profile this issue comprehensively and promoted more awareness and investigation of understudied genes.

We thank the reviewer for their kind assessment of our work.

Weaknesses

Regarding result section 1 "Understudied genes are abandoned at synthesis/writing stage", the figures are not clear and do not convey the messages written in the main text. For example, in Figure 1B, figure S5 and S6,

• There is no "numbers to the right of each box plot".

The “numbers to the right” statement in the caption was an erroneous inclusion from an earlier version of the figure. We apologize for our error and have now removed this statement.

• Do these box plots only show understudied genes? How many genes are there in each box plot? The definition and numbers of understudied genes are not clear.

The x-axis describes genes featured in each stage of the publication process (from all protein-coding genes to genes found as hits in genome-wide screen to genes found in the title/abstract to genes found in the title/abstract of citing articles) and the y-axis describes the number of articles annotated to those genes. We have also now added the number of genes in each box plot to the figure. This information is also in Materials and Methods under each technology’s heading (see also response to Reviewer 1 beginning on page 2 of this document).

Author response image 3.

• "We found that hit genes that are highlighted in the title or abstract are strongly over-represented among the 20% highest-studied genes in all biomedical literature (Figure 1B)". This is not clear from the figure.

We have revised Figure 1B and its caption to better communicate the main point of the figure: that genes which make it to the title/abstract of the reporting article tend to be more popular than genes which are hits in genome-wide experiments from those articles. We have added a horizontal line that shows the cutoff for the top 20% most popular genes.

Regarding result section 2 "Subsequent reception by other scientists does not penalize studies on understudied genes", the authors showed in figure 2 that there is a negative correlation between articles per gene before 2015 and median citations to articles published in 2015. Another explanation could be that for popular genes, there are more low-quality articles that didn't get citations, not necessarily that less popular genes attract more citations.

We believe that both explanations for the observed phenomenon are not mutually exclusive. Previously, we focused on the median of citations to articles about a gene to capture the typical effect. In a new analysis, we also find support for the possibility outlined by the reviewer and believe that adding this to our manuscript complements and balances our analysis of citations. Specifically, in the new Figure S8B we find that most popular genes are slightly more likely to be among least cited papers (and in Figure S8A that the least studied genes have been much more likely to be among the most cited papers). In-text, we state:

“Further, since 1990, articles about the least popular genes have at times been 3 to 4 times more likely to be among the most cited articles than articles on the most popular genes whereas articles on the most popular genes have been slightly less to be highly cited than lowly cited (Figure S8)”.

We thank the reviewer for their suggestion, which strengthens our manuscript. The figure caption reads:

“Figure S8: Likelihoods of being highly cited (top 5% of citations among all articles about genes, panel a) or lowly cited (bottom 5% of citations among all articles about genes, panel b) for articles about the most popular genes (top 5% accumulated articles) versus articles about the least popular genes (bottom 5% accumulated articles) by year of publication. Only articles with a single gene in the title/abstract are considered. Shaded regions show ±1 standard error of the proportion."

Author response image 4.

Regarding result section 3 "Identification of biological and experimental factors associated with selection of highlighted genes", in Figure 3 and table s2, the author stated that "hits with a compound known to affect gene activity are 5.114 times as likely to be mentioned in the title/abstract in an article using transcriptomics", The number 5.144 comes out of nowhere both in the figure and the table. In addition, figure 4 is not informative enough to be included as a main figure.

This is the result of both a typo and imprecise terminology. The number should read 4.262 (the likelihood ratio of being mentioned in the title/abstract between genes with and without a compound), which corresponds to an odds ratio of 4.331. We have clarified this in the table caption, stating:

“e.g. hits with a compound known to affect gene activity are 4.262 times as likely to be mentioned in the title/abstract in an article using transcriptomics, corresponding to an odds ratio of 4.331".

We have removed Figure 4 as a main-text figure and added a version, with revised color scheme along comments of Reviewer 1, as Figure S11. We added to the figure caption “Bold indicates FMUG ‘s default factors, which we selected based on this clustering and based on their strength of association with gene selection (Figure 3, Table S2 and Table S3)."

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

• Fig 2a shows that papers highlighting understudied genes are actually cited more. I wonder why authors only looked at data before 2015. Fig 2b shows an increased correlation since 2015. Please consider redrawing Fig 2a to include data from 2015-2020?

We highlight data from 2015 since, from our used version of iCite (v32, released July 2022, covering citations made through most of 2021), papers published in 2015 have had about 6 years to accumulate citations. With fewer years to accumulate citations, insufficient signal may cause correlation to converge toward zero. Below, we repeat the analysis in Figure 2 but only considering citations made within a year of an article’s publication, which substantially reduces correlation (although remaining significant).

Author response image 5.

We added a note to the figure caption:

“We forgo depicting more recent years than 2015 to allow for citations to accumulate over multiple years, providing a more sensitive and robust readout of long-term impact.”

For Figure 2B, we add:

“For more recent years, where articles have had less time to accumulate citations, insufficient signal may cause correlation to converge toward zero.”

• Can FMUG be posted on the web for easy access by researchers with non-computational backgrounds?"

We presently regretfully do not have the resources to create or maintain a web-based version. We hope that the publication of this manuscript will enable us to attract resources to create and maintain a web-based version.

Reviewer #2 (Recommendations for the authors):

• Related to the first weakness in my public review: The observed disparity between CRISPR and GWAS study in terms of which genes they promote to the abstract is interesting. I wonder if this has to do with the application of these techniques. GWAS studies will often highlight that they retrieve known associations between a gene and a phenotype, to show that a screen is working. I guess often the point is to subsequently identify more genes associated with a particular phenotype, but often it is unclear how to validate/verify newly found associations. In contrast, CRISPR screens might be more focussed on functionally/mechanistically understanding unknown processes, e.g. observing a phenotype that appears/disappears in response to a gene deletion. In such studies, the follow-up of a previously unknown gene could be more straightforward and relevant to the outcome. Does that mean CRIPSR screens are better than GWAS studies for addressing the UG problem? Perhaps the authors could briefly discuss this issue.

The number of studies we included featuring CRISPR screens is relatively small (n = 15 compared to n = 450 for GWAS). Thus, it is not possible to conclude in a statistically sound manner whether authors of CRISPR screens are truly more likely to highlight understudied genes.

However, the reviewer raises compelling reasons for why this might be the case, and we now embed the broader discussion point that some techniques might be more powerful toward understudied genes.

The discussion now includes:

“Further, the observed discrepancy between the popularity of hits highlighted by GWAS versus other technologies suggests that some -omics technologies may be more powerful than others for characterizing understudied genes. This possibility merits further research and researchers participating in unknomics should consider the relative strengths of each technology towards providing tractable results for follow-up.”

• Affinity capture mass spectrometry (Aff-MS): Perhaps I misunderstood this, but typically this is referred to as affinity purification MS (AP-MS)

Thank you for the clarification. We have changed ‘Aff-MS’ to ‘AP-MS’ throughout the manuscript.

• Page 3, line 96. The sentence "The first possibility is that seemingly understudied genes are, in fact, not understudied as they would rarely be identified through experiments.". Would they not still be understudied, just not intentionally?

We have rephrased this sentence to:

“The first possibility is that some genes are less studied because they are rarely identified as hits in experiments.”

• Fig 4 is very interesting, but I also found it a bit confusing. First, the choice of colour scheme, where blue shows the absence and white shows the presence of something, seems counterintuitive, especially on a white background. Second, I find it confusing that only some of the experiments are labelled in the heatmap. Could the authors not simply use Fig S9 as Fig 4? Or alternatively, only include the 8 labelled factors in the simplified figure.

In line with this feedback and that of Review #1 and #3, we have removed Figure 4 as a main-text figure and instead include this figure as Supplementary Figure S11. We have reversed the color scheme so that purple indicates one and white indicates zero. We also now label all factors. Previously we had only listed the default features of FMUG. We also now updated the figure legend to convey how it assisted the choice of default factors in FMUG. It reads:

“Bold indicates FMUG ‘s default factors, which we selected based on this clustering and based on their strength of association with gene selection (Figure 3, Table S2 and Table S3)”.

• The FMUG app is fantastic and sounds exactly like something that is required to boost the visibility of understudied genes and overcome the understudied gene bias. However, I did not understand the choice of reporting this in the Discussion section.

We thank the reviewer for their enthusiasm, and have now moved FMUG into the results section.

• To further increase usability of the FMUG app, is there a way it could be deployed online? I appreciate this could require a major amount of coding work, which would not be reasonable to demand. So please consider this a suggestion, potentially for a future implementation.

We presently regretfully do not have the resources to create or maintain a web-based version. We hope that the publication of this manuscript will enable us to attract resources to create and maintain a web-based version.

Reviewer #3 (Recommendations for the authors):

Table s2 and s3: p values are indicated by star signs. However, with so many hypothesis tests, the p values should be corrected for multiple tests.

We have now applied Benjamini-Hochberg multiple hypothesis correction to these tables, correcting p-values within each of the four technologies. We update our significance calling to read:

“We identified 45 factors that relate to genes and found 33 (12 out of 23 binary factors and 21 out of 22 continuous factors) associated with selection in at least one assay type at Benjamini-Hochberg FDR < 0.001.”

Figure S1 - S4

These figures contain too many noninformative boxes. In all the figures, only the last three boxes are informative (reports assessed for eligibility, reports excluded, and studies included in review). The rest boxes convey little information and should be simplified.

We have simplified these diagrams, removing boxes which contained no information.

Figure S6: what does it mean by "prior to the publication of the first article represented in this sample"? What is "this sample"?

“This sample” refers to the collection of 450 GWAS articles, 296 articles using AP-MS, 148 transcriptomics articles, and 15 genome-wide CRISPR screen articles. We have rephrased this sentence to make this clear. It now reads:

“Variant of Figure 1B only considering articles published in 2002 or before, prior to the publication of any of the articles featuring -omics experiments which we considered for this analysis.”

eLife assessment

This study investigated the factors related to understudied genes in biomedical research. It showed that understudied genes are largely abandoned at the writing stage, and it identified a number of biological and experimental factors that influence which genes are selected for investigation. The study is an important contribution to this branch of meta-research, and the evidence in support of the findings is solid.

Reviewer #1 (Public Review):

The authors have addressed most of the concerns I had about the original version in this revised version.

Reviewer #2 (Public Review):

The authors have successfully addressed all of the concerns I had about the original version.

Reviewer #3 (Public Review):

The message conveyed by figure 1b is now clearer, but could still be improved. The authors explained the meaning of this figure well in their response to the reviewers: "For example, the results for CRISPR were obtained from 15 focus studies (original research) and 18 subsequent studies (papers citing focus articles). Those 15 studies identified 9,268 genes where loss-of-function changed phenotypes but, in their titles and abstracts, mentioned only 18 of those 9,268 genes. While the 9,268 hit genes have received similar research attention to the entirety of protein-coding genes, the 18 hit genes mentioned in the title or abstract are significantly more well studied. The articles citing the focus articles also only mentioned in their titles and abstracts 19 highly studied hit genes".<br /> The new Figure S8 is good.

Author Response

Reviewer #1 (Public Review):

Reviewer 1: The structural part of this work is interesting, as it is the first structure of Pin1 with a ligand that bridges both domains. They might want to underline this - all other structures in the PDB have a single domain complex, but never both domains by a single longer peptide.

Done. We have highlighted the novelty of the structure in the abstract, introduction (page 5); and discussion (section “The Pin1-PKC interface is described by a novel bivalent interaction mode”, page 24).

Reviewer 1: I would however question the static representation of this structure - the 90{degree sign} kink in the peptide when complexed is probably one single snapshot, but I hardly believe the PPIase/WW domain orientation to be static. Unless the authors have additional information to stand by this static structure, this point merits being commented on in the manuscript.

Done. Following the reviewer’s suggestion and to avoid the impression of “static” structure, we have added sentences that highlight the dynamic aspects of the complex evident from the entire ensemble representation of Figure 5-figure supplement 2:

Page 15 (Results):

“Of note, the linker region connecting the two domains retains its flexibility in the complex and confers some variability onto the relative positions of the WW and PPIase domains, as is evident from the ensemble representation of Figure 5-figure supplement 2. The complex exhibits novel structural features that distinguish it from all other structures of Pin1 complexes known to date. These features are highlighted in Fig. 6 using the lowest-energy structure of the ensemble.”

Page 24 (Discussion): “Moreover, the retention of linker flexibility in the Pin1::pV5bII complex suggests that Pin1 can potentially adopt minor “extended” states that would not be readily detectable by ensemble-averaged methods such as solution NMR.”

Also, in describing specific interactions in the section “Structural basis of the Pin1-PKCII C-term bivalent recognition mode”, we now note how many structures of the Pin1-pV5bII ensemble have those interactions.

Reviewer 1: I would like to point out to literature that described for example the non-canonical binding (Yeh ES, Lew BO & Means AR (2006) The loss of PIN1 deregulates cyclin E and sensitizes mouse embryo fibroblasts to genomic instability. J Biol Chem 281, 241-251. Pin1 recognizes cyclin E via a noncanonical pThr384- Gly385 motif [33] rather than the pThr380-Pro381 motif.). They mention briefly the absence of isomerase activity in similar TPP motifs, but this information might already come in the Results section.

Done. We have incorporated this information in the Discussion section, page 25 (last paragraph).

Reviewer 1: The expression levels of Pin1 and PKCa are amazingly linear (Fig 7A), but when they overexpress WT Pin1 in a KO line, with 3-4 times higher overexpression, the PKCa levels are hardly higher than in the original WT cell line.

We thank Reviewer 1 for raising this interesting point. Our simple interpretation of the data is that physiological expression of Pin1 in the cell model we use is a limiting factor in the stimulated PKCa degradation pathway, but that Pin1 is no longer a limiting factor at higher expression levels. We now include this point in the Discussion, page 26.

Reviewer 1: Also, the levels in the W34A/R68A/R69A (abolishing both WW and PPIase binding functions) are surprising, why would PKCa levels rise above the level found in the Pin1 KO cells?

This result remains a puzzle but, as we are including all independent biological replicates in the analysis, the data are the data. Moreover, by assessing the functional complementation data to the KO by two-tailed t-test (see last point below), this effect does not reach statistical significance. Nonetheless, as the result is reproducible, we now comment on this effect in the Results, page 21. One speculation is this triple mutant has dominant negative properties imposed on some limiting factor in PKCa degradation that are revealed in the absence of WT Pin1. Considerably more work needs to be done to settle this issue. However, in light of the fact that this result does not conflict with the structural/biochemical data (rather, it is consistent with it), we hope this positive response satisfies the Reviewer.

Reviewer 1: Finally, if even slight overexpression of the C113S catalytically inactive mutant leads to more efficient PKCa degradation than overexpression of the WT Pin1 (Figure 7C), it is hard to interpret. The conclusion that Pin1-mediated regulation of PKCa requires a bivalent interaction mode of Pin1 with PKCa independent of its catalytic activity do depend on these data, so they merit further analysis.

We certainly had no intention of concluding that the C113S catalytically inactive mutant is more efficient with regard to promoting PKCa degradation than overexpression of the WT Pin1. That overstates the data. We concede that our organization of the Pin1 rescue data in the original Fig 7C confused the issue, and that the original text also invited conclusions that overstate the result. To correct this problem, we reorganized Fig. 7C to simplify the presentation by comparing the complementation data to the KO. All statistical comparisons are now to the KO cell line (not to WT as before) and we employ the two-tailed t-test to compare the data. Statistical significance is attained only for reconstituted WT and C113S Pin1 expression. The text is also appropriately revised to describe the results clearly. We trust the Reviewer agrees that the C113S data are compelling and are consistent with a noncanonical (noncatalytic) mode of PKCa regulation by Pin1. This is a major point of Fig 7C as it links the structural/biochemical data to a cellular context.

eLife assessment

Pin1 as an essential prolyl cis/trans isomerase has attracted considerable attention because this enzyme family is implicated in cancer and neurodegenerative diseases. However, the requirement for its catalytic function remains a matter of dispute. The authors provide solid evidence that Pin1 modulates the activity of an important cell signaling kinase, Protein Kinase C, by a non-catalytic mechanism, acting as a chaperone to regulate the stability of this kinase.

Reviewer #1 (Public Review):

When writing a short review on the function of Pin1 some 15 years ago (Lippens et al., Febs J 2007), we concluded the introduction by the following sentence: "..., it seems that further analysis is required to determine whether binding or catalysis is the primary mechanism through which Pin1 affects cell cycle progression." In the present manuscript, the authors provide experimental evidence for the Pin1/PKC interaction that tips the balance towards interaction and not catalysis.

Their main data concern the interaction between the V5 domains of two PKC isoenzymes (alpha and betaII) and Pin1. This V5 domain can be further separated into a Turn Motif (TM) and a Hydrophobic Motif (HM), that both can be phosphorylated on specific positions. Phosphorylation in the TM occurs on a TPP motif, and in agreement with previous results on the same motif in Tau, Pin1 cannot isomerize efficiently the TP amide bond when the residue following the proline is another proline. Phosphorylation of the HM is not proline directed but occurs on a serine flanked by 2 aromatic residues (FSF or FSY, according to the isoenzyme). They dissect in detail the interaction of both motifs with the WW and PPIase domains and conclude that the fully phosphorylated V5 peptide binds Pin1 in a directional mode, with the TM binding to the WW domain and the HM to the PPIase domain.

In the absence of crystals of the complex, they solve a structure by NMR, and use selectively labeled peptides (and probably a lot of NMR time) to obtain a structural model. Finally, they provide functional data by silencing/overepxressing Pin1 and inactive mutants (both at the level of its WW domain and the PPIase domain) in HEK293T cells and evaluating the PKCalpha homeostasis.

The structural part of this work is interesting, as it is the first structure of Pin1 with a ligand that bridges both domains. They might want to underline this - all other structures in the PDB have a single domain complex, but never both domains by a single longer peptide. I would however question the static representation of this structure - the 90{degree sign} kink in the peptide when complexed is probably one single snapshot, but I hardly believe the PPIase/WW domain orientation to be static. Unless the authors have additional information to stand by this static structure, this point merits being commented on in the manuscript.

I would like to point out to literature that described for example the non-canonical binding (Yeh ES, Lew BO & Means AR (2006) The loss of PIN1 deregulates cyclin E and sensitizes mouse embryo fibroblasts to genomic instability. J Biol Chem 281, 241-251. Pin1 recognizes cyclin E via a noncanonical pThr384- Gly385 motif [33] rather than the pThr380-Pro381 motif.). They mention briefly the absence of isomerase activity in similar TPP motifs, but this information might already come in the Results section.

The weakest part seems the in vivo data. Although this is not the main focus of this lab, there is some issues that could be addressed. The expression levels of Pin1 and PKCa are amazingly linear (Fig 7A), but when they overexpress WT Pin1 in a KO line, with 3-4 times higher overexpression, the PKCa levels are hardly higher than in the original WT cell line. Also, the levels in the W34A/R68A/R69A (abolishing both WW and PPIase binding functions) are surprising, why would PKCa levels rise above the level found in the Pin1 KO cells? Finally, if even slight overexpression of the C113S catalytically inactive mutant leads to more efficient PKCa degradation than overexpression of the WT Pin1 (Figure 7C), it is hard to interpret. The conclusion that Pin1-mediated regulation of PKCa requires a bivalent interaction mode of Pin1 with PKCa independent of its catalytic activity do depend on these data, so they merit further analysis.

Reviewer #2 (Public Review):

Chen, Dixit et al. report on the first structure of a bivalent interaction between a natural interaction partner of Pin1: the C-terminal tail of PKC phosphorylated at two sites. The biggest strength of the paper is the impressive amount of NMR-based structural data that is sound and clearly reported. The authors strive to propose a novel non-catalytic mechanistic role for Pin1 that is supported by cell culture models and somewhat by the interaction assays, however, in my eyes, they fell short in proving their mechanistic hypothesis. Nevertheless, the potential ways Pin1 may modulate PKC's activity is nicely discussed.

Author Response

eLife assessment

This computational study is a valuable empirical investigation into the common trait of neurons in brains and artificial neural networks: responding effectively to both objects and their mirror im- ages and it focuses on uncovering conditions that lead to mirror symmetry in visual networks and the evidence convincingly demonstrates that learning contributes to expanding mirror symmetry tuning, given its presence in the data. Additionally, the paper delves into the transformation of face patches in primate visual hierarchy, shifting from view specificity to mirror symmetry to view invariance. It empirically analyzes factors behind similar effects in two network architec- tures, and key claims highlight the emergence of invariances in architectures with spatial pooling, driven by learning bilateral symmetry discrimination and importantly, these effects extend be- yond faces, suggesting broader relevance. Despite strong experiments, some interpretations lack explicit support, and the paper overlooks pre-training emergence of mirror symmetry.

As detailed above, we have now analyzed several convolutional architectures and made a direct link between the artificial neural networks and neuronal data to further support our claims (refer to Figure 6, S10- 13).

To address the concern about pre-training emergence of mirror symmetry, we conducted a new analysis inspecting unit-level response profile, following Baek and colleagues (2021). This analysis is described in detail below (response to R3). In brief, we found that the first fully connected layer in trained networks exhibits twice the number of mirror symmetric units found before training. In addition to our population-level observations (Fig. S2) and explicit training- dataset manipulations (Fig. 4), this finding supports the interpretation of training to discriminate among mirror- symmetric object categories as a major factor behind the emergence of mirror symmetric viewpoint tuning.

Reviewer 1 (Public Review):

By using deep convolutional neural networks (CNNs) as model for the visual system, this study aims at understanding and explaining the emergence of mirror-symmetric viewpoint tuning in the brain.

Major strengths of the methods and results:

1) The paper presents comprehensive, insightful and detailed analyses investigating how mirror- symmetric viewpoint tuning emergence in artificial neural networks, providing significant and novel insights into this complex process.

2) The authors analyze reflection equivariance and invariance in both trained and untrained CNNs’ convolutional layers. This elucidates how object categorization training gives rise to mirror-symmetric invariance in the fully-connected layers.

3) By training CNNs on small datasets of numbers and a small object set excluding faces, the authors demonstrate mirror-symmetric tuning’s potential to generalize to untrained categories and the necessity of view-invariant category training for its emergence.

4) A further analysis probes the contribution of local versus global features to mirror-symmetric units in the first fully-connected layer of a network. This innovative analysis convincingly shows that local features alone suffice for the emergence of mirror-symmetric tuning in networks.

5) The results make a clear prediction that mirror-symmetric tuning should also emerge for other bilaterally symmetric categories, opening avenues for future neural studies.

We are grateful for your insightful feedback and the positive evaluation of our study on mirror-symmetric viewpoint tuning in neural networks. Your constructive comments considerably improved the manuscript. We eagerly look forward to exploring the future research avenues you have highlighted.

Major weaknesses of the methods and results:

Point 1.1) The authors propose a mirror-symmetric viewpoint tuning index, which, although innovative, complicates comparison with previous work and this choice is not well motivated. This index is based on correlating representational dissimilarity matrices (RDMs) with their flipped versions, a method differing from previous approaches.

We have revised the Methods section to clarify the motivation for the mirror-symmetric viewpoint tuning index we introduced.

Manuscript changes:

Previous work quantified mirror-symmetry in RDMs by comparing neural RDMs to an idealized mirror- symmetric RDM (see Fig. 3c-iii in [14]). Although highly interpretable, such an idealized RDM encompasses implicit assumptions about representational geometry that are unrelated to mirror-symmetry. For example, consider a neural RDM reflecting perfect mirror-symmetric viewpoint tuning and wherein for each view, the distances among all of the exemplars are equal. Such a neural RDM would fit an idealized mirror- symmetric RDM better than a neural RDM reflecting perfect mirror-symmetric viewpoint tuning but with non-equidistant exemplars. In contrast, the measure proposed in Eq. 2 equals 1.0 in both cases.

Point 1.2> Faces exhibit unique behavior in terms of the progression of mirror-symmetric viewpoint tuning and their training task and dataset dependency. Given that mirror-symmetric tuning has been identified in the brain for faces, it would be beneficial to discuss this observation and provide potential explanations.

We revised the caption of Figure S1 to explicitly address this point:

Manuscript changes:

For face stimuli, there is a unique progression in mirror-symmetric viewpoint tuning: the index is negative for the convolutional layers and it abruptly becomes highly positive when transitioning to the first fully connected layer. The negative indices in the convolutional layers can be attributed to the image-space asymmetry of non-frontal faces; compared to other categories, faces demonstrate pronounced front-back asymmetry, which translates to asymmetric images for all but frontal views (Fig. S8). The features that drive the highly positive mirror-symmetric viewpoint tuning for faces in the fully connected layers are training-dependent (Fig. S2), and hence, may reflect asymmetric image features that do not elicit equivariant maps in low-level representations; for example, consider a profile view of a nose. Note that cars and boats elicit high mirror- symmetric viewpoint tuning indices already in early processing layers. This early mirror-symmetric tuning is independent of training (Fig. S2), and hence, may be driven by low-level features. Both of these object categories show pronounced quadrilateral symmetry, which translates to symmetric images for both frontal and side views (Fig. S8).

Point 1.3: 3. Previous work reported critical differences between CNNs and neural represen- tations in area AL indicating that mirror-symmetric viewpoint tuning is less present than view invariance in CNNs compared to area AL. While such findings could potentially limit the use- fulness of CNNs as models for mirror-symmetric viewpoint tuning in the brain, they are not addressed in the study.

This point is now addressed explicitly in the caption of Figure S9:

Manuscript changes:

Yildirim and colleagues [14] reported that CNNs trained on faces, notably VGGFace, exhibited lower mirror- symmetric viewpoint tuning compared to neural representations in area AL. Consistent with their findings, our results demonstrate that VGGFace, trained on face identification, has a low mirror-symmetric viewpoint tuning index. This is especially notable in comparison to ImageNet-trained models such as VGG16. This difference between VGG16 and VGGFace can be attributed to the distinct characteristics of their training datasets and objective functions. The VGGFace training task consists of mapping frontal face images to identities; this task may exclusively emphasize higher-level physiognomic information. In contrast, training on recognizing objects in natural images may result in a more detailed, view-dependent representation. To test this potential explanation, we measured the average correlation-distance between the fc6 representations of different views of the same face exemplar in VGGFace and VGG16 trained on ImageNet. The average correlation-distance between views is 0.70±0.04 in VGGFace and 0.93±0.04 in VGG16 trained on ImageNet. The converse correlation distance between different exemplars depicted from the same view is 0.84±0.14 in VGGFace and 0.58±0.06 in VGG16 trained on ImageNet. Therefore, as suggested by Yildirim and colleagues, training on face identification alone may result in representations that cannot explain intermediate levels of face processing.

Point 1.4) The study’s results, while informative, are qualitative rather than quantitative, and lack direct comparison with neural data. This obscures the implications for neural mechanisms and their relevance to the broader field.

We addressed this point by conducting a quantitative comparison between the architectures of various networks and neural response patterns in monkey face patches (see Figures 6, S10-S13, appearing above).

Point 1.5) The study provides compelling evidence that learning to discriminate bilaterally symmetric objects (beyond faces) induces mirror-symmetric viewpoint tuning in the networks, qualitatively similar to the brain. Moreover, the results suggest that this tuning can, in principle, generalize beyond previously trained object categories. Overall, the study provides important conclusions regarding the emergence of mirror-symmetric viewpoint tuning in networks, and potentially the brain. However, the conducted analyses and results do not entirely address the question why mirror-symmetric viewpoint tuning emerges in networks or the brain. Specifically, the results leave open whether mirror-symmetric viewpoint tuning is indeed necessary to achieve view invariance for bilaterally symmetric objects.

We believe that mirror-symmetric viewpoint tuning is not strictly necessary for achieving view-invariance. However, it is a plausible path from view-dependence to view invariance. We addressed this point in the updated limitations subsection of the discussion.

Manuscript changes:

A second consequence of the simulation-based nature of this study is that our findings only establish that mirror-symmetric viewpoint tuning is a viable computational means for achieving view invariance; they do not prove it to be a necessary condition. In fact, previous modeling studies [10, 19, 61] have demonstrated that a direct transition from view-specific processing to view invariance is possible. However, in practice, we observe that both CNNs and the face-patch network adopt solutions that include intermediate representations with mirror-symmetric viewpoint tuning.

Taken together, this study moves us a step closer to uncovering the origins of mirror-symmetric tuning in networks, and has implications for more comprehensive investigations into this neural phenomenon in the brain. The methods of probing CNNs are innovative and could be applied to other questions in the field. This work will be of broad interest to cognitive neuroscientists, psychologists, and computer scientists.

We appreciate your acknowledgment of our study’s contribution to understanding mirror-symmetric tuning in networks and its wider implications in the field.

Reviewer 2 (Public Review);

Strengths

1) The statements made in the paper are precise, separating observations from inferences, with claims that are well supported by empirical evidence. Releasing the underlying code repository further bolsters the credibility and reproducibility. I especially appreciate the detailed discussion of limitations and future work.

2) The main claims with respect to the two convolutional architectures are well supported by thorough analyses. The analyses are well-chosen and overall include good controls, such as changes in the training diet. Going beyond ”passive” empirical tests, the paper makes use of the fully accessible nature of computational models and includes more ”causal” insertion and deletion tests that support the necessity and sufficiency of local object features.

3) Based on modeling results, the paper makes a testable prediction: that mirror-symmetric viewpoint tuning is not specific to faces and can also be observed in other bilaterally symmetric objects such as cars and chairs. To test this experimentally in primates (and potentially other model architectures), the stimulus set is available online.

We express our gratitude for your constructive feedback. Your acknowledgment of the clarity of our statements and the robustness of our empirical evidence is greatly appreciated. We are also thankful for your recognition of our comprehensive analyses and the testable predictions arising from our work.

Point 2.1: Weaknesses

My main concern with this paper is in its choice of the two model architectures AlexNet and VGG. In an earlier study, Yildirim et al. (2020) found an inverse graphics network ”EIG” to better correspond to neural and behavioral data for face processing than VGG. All claims in the paper thus relate to a weaker model of the biological effects since this work does not analyze the EIG model. Since EIG follows an analysis-by-synthesis approach rather than standard classification training, it is unclear whether the claims in this paper generalize to this other model architecture. It is also unclear if the claims will hold for: 1) transformer architectures, 2) the HMAX architecture by Leibo et al. (2017) which has also been proposed as a computational explanation for mirror-symmetric tuning, and, as the authors note in the Discussion, 3) deeper architectures such as ResNet-50 which tend to better align to neural and behavioral data in general. These architectures include different computational motifs such as skip connections and a much smaller proportion of fully-connected layers which are a major focus of this work.

Overall, I thus view the paper’s claims as limited to AlexNet- and VGG-like architectures, both of which fall behind state-of-the-art in their alignment to primates in general and also specifically for mirror-symmetric viewpoint tuning.

We understand your concern regarding the choice of AlexNet and VGG architectures. The decision to focus on these models was driven by the need for a straightforward macroscopic correspondence between the layer structure of the artificial networks and the ventral visual stream. However, acknowledging this potential limitation of generality, we have expanded our analysis to include the EIG model, a transformer architecture, the HMAX model, and deeper convolutional architectures like ResNet-50 and ConvNeXt. Our revised analysis, detailed in Figures S1, S9, and S10-S13, incorporates these additional models and offers a comprehensive evaluation of their brain alignment and mirror-symmetric viewpoint tuning. We found that while the architectures indeed vary in their computational motifs, the emergence of mirror-symmetric viewpoint tuning is not exclusive to AlexNet and VGG. It occurs for every CNN we tested, exactly at the stage where equivariant feature maps are pooled globally. We believe that the new analyses extend the generality of our findings and remove the concern that our claims apply only to older, shallower networks.

For details, please refer to Point 1 in the ’Essential Revisions’ section.

Point 2.2: Minor weaknesses

1) Figure 1A: since the relevance to primate brains is a major motivator of this work, the results from actual neural recordings should be shown and not just schematics. For instance, the mirror symmetry in AL is not as clean as the illustration (compare with Fig. 3 in Yildirim et al. 2020), and in the paper’s current form, this is not easily accessible to the reader.

Thank you for your feedback regarding the presentation of neural recordings in Figure 1A. We have updated Figure 1A to include actual neural RDMs instead of the previous schematic representations.

Point 2.3: 2. Figure 4 L832-845: The claims for the effect of training on mirror-symmetric viewpoint tuning are with respect to the training data only, but there are other differences between the models such as the number of epochs (250 for CIFAR-10 training, 200 for all other datasets), the learning rate (2.5 ∗ 10−4 for CIFAR-10, 10−4 for all others), the batch size (128 vs 64), etc. I do not expect these choices to make a major difference for your claims, but it would be much cleaner to keep everything but the training dataset consistent. Especially the different test accuracies worry me a bit (from 81% to 92%, and they appear different from the accuracy numbers in figure S4 e.g. for CIFAR-10 and asymSVHN), at the very least those should be comparable.

We addressed this point by retraining the models while holding most of the hyperparameters constant. Specifically, we standardized the number of epochs, batch size, and weight decay. The remaining differences are necessitated by the characteristics of the specific training image sets used (natural images versus digits). Please note that we do not directly contrast models trained on CIFAR-10 and SVHN; the controlled comparisons are conducted while holding the SVHN training images constant, and are not confounded by hyperparameter choice.

Manuscript changes:

The networks’ weights and biases were initialized randomly using the uniform He initialization [70]. We trained the models using 250 epochs and a batch size of 256 images. The CIFAR-10 network was trained using stochastic gradient descent (SGD) optimizer starting with a learning rate of 10−3 and momentum of 0.9. The learning rate was halved every 20 epochs. The SVHN/symSVHN/asymSVHN networks were trained using the Adam optimizer. The initial learning rate was set to 10−5 and reduced by half every 50 epochs. The hyper-parameters were determined using the validation data. The models reached around 83% test accuracy (CIFAR-10: 81%, SVHN: 89%, symSVHN: 83%, asymSVHN: 80%). Fig. S4 shows the models’ learning curves.

Point 2.4: 3. L681-685: The general statement made in the paper that ”deeper models lose their advantage as models of cortical representations” is not supported by the cited limited comparison on a single dataset. There are many potential confounds here with respect to prior work, e.g. the recording modality (fMRI vs electrodes), the stimulus set (62 images vs thousands), the models that were tested (9 vs hundreds), etc.

We agree that the recording modality and stimulus set may play a critical role in determining model ranking. Since we generalized the analyses to deeper models, we removed this statement from the paper. While we still believe that shallower networks may prove to be better models of the visual cortex, this empirical question is out of the scope of the current manuscript.

Reviewer 3

This study aimed to explore the computational mechanisms of view invariance, driven by the observation that in some regions of monkey visual cortex, neurons show comparable responses to (1) a given face and (2) to the same face but horizontally flipped. Here they study this known phenomenon using AlexNet and other shallow neural networks, using an index for mirror symmetric viewpoint tuning based on representational similarity analyses. They find that this tuning is enhanced at fully connected- or global pooling layers (layers which combine spatial information), and that the invariance is prominent for horizontal- but not vertical- or rotational transformations. The study shows that mirror tuning can be learned when a given set of images are flipped horizontally and given the same label, but not if they are flipped and given different labels. They also show that networks learn this tuning by focusing on local features, not global configurations.

We are grateful for your thorough reading, reflected by the comprehensive summary of our study and its main findings.

Point 3.1) I found the study to be a mixed read. Some analyses were fascinating: for example, it was satisfying to see the use of well-controlled datasets to increase or decrease the rate of mirror-symmetry tuning. The insertion- and deletion¬ experiments were elegant tests to probe the mechanisms of mirror symmetry, asking if symmetry could arise from (1) global feature configurations (in a holistic sense) vs. (2) local features, with stronger evidence for the latter. These two sets of results were successful and interpretable. They stand in contrast with the first analysis, which relies on observations that do not seem justified. Specifically, Figure 2D shows mirror-symmetry tuning across 11 stages of image processing, from pixels space to fully connected layers. It shows that images from different object categories evoke considerably different tuning index values. The explanation for this result is that some categories, such as ”tools,” have ”bilaterally symmetric structure,” but this is not explicitly measured anywhere. ”Boats” are described as having ”front-back symmetry,” more so than flowers. One imagines flowers being extremely symmetric, but perhaps that depends on the metric. What is the metric? At first I thought it was the mirror-symmetric viewpoint tuning index in the image (pixel) space, but this cannot be, as the index for faces and flowers is negative, cars have no symmetry, and boats are positive. To support these descriptions, one must have an independent variable (for object class symmetry) that can be related to the dependent variable (the mirror-symmetric viewpoint tuning index). If it exists, it is not a part of the Results section. This omission undermines other parts of the Results section: ”some car models have an approximate front-back symmetry...however, a flower typically does not...” ”Some,” ”typically:” how many in the dataset exactly, and how often?

We thank you for your insightful observation. You are correct that we did not refer to pixel-space symmetry; our descriptions relate to the 3D structure of the objects used in the study.

Following this comment, we objectively quantified the symmetry planes of the 3D objects. Unfortunately, we do not have direct access to the proprietary 3D meshes of these objects, only to their renders. Therefore, we devised measures that assess the symmetry of the 3D objects through the symmetry they elicit in the different 2D renders.

This analysis is described in the new supplemental figure S8. We believe that these measurements support the qualitative claims we made in the previous draft.

Point 3.2) The description of CIFAR-10 as having bilaterally symmetric categories - are all these categories equally symmetric? If not, would such variability matter in terms of these results?

When considering their 3D structure, all ten CIFAR10 categories exhibit pronounced left-right symmetry. These categories encompass vertebrate animals (birds, cats, deer, dogs, frogs, horses); They also include man-made vehicles (airplanes, cars, ships, and trucks), which, at least externally, are nearly perfectly symmetric by design. It is important to note that this symmetry pertains to the photographed 3D objects, rather than the images themselves, which could be highly asymmetric. Other axes of symmetry (e.g., back-front) in CIFAR10 cannot be measured without 3D representations of the objects.

Point 3.3) These assessments of object category symmetry values are made before experiments are presented, so they are not interpretations of the results, and it would be circular to write it otherwise.

We have changed the order so that the explanations follow the experimental results. This includes the relevant main text paragraph, as well as the relevant figure—both the order of panels and the phrasing of the figure caption.

Point 3.4) Overall, my bigger concern is that the framing is misleading or at best incomplete. The manuscript successfully showed that if one introduces left-right symmetry to a dataset, the network will develop population-level representations that are also bilaterally symmetric. But the study does not explain that the model’s architecture and random weight distribution are sufficient for symmetry tuning to emerge, without training, just to a much more limited degree. Baek et al. showed in 2021 that viewpoint-invariant face-selective units and mirror-symmetric units emerge in untrained networks (”Face detection in untrained deep neural networks”; this current manuscript cites this paper but does not mention that mirror symmetry is a feature of the 2021 study). This current study also used untrained networks as controls (Fig. 3), and while they were useful in showing that learning boosts symmetry tuning, the results also clearly show that horizontal-reflection invariance is far from zero. So, the simple learning-driven explanation for the mirror-symmetric viewpoint tuning for faces is wrong: while (1) network training and (2) pooling are mechanisms that charge the development of mirror-symmetric tuning, the lottery ticket hypothesis is enough for its emergence. Faces and numbers are simple patterns, so the overparameterization of networks is enough to randomly create units that are tuned to these shapes and to wire many of them together. How learning shapes this process is an interesting direction, especially now that this current study has outlined its importance.

We agree with the reviewer that random initialization may result in units that show mirror-symmetric viewpoint tuning for faces in the absence of training. In the revised manuscript, we quantify the occurrence of such units, first reported by Baek et al, in detail, and discuss the relation between Baek et al., 2021 and our work. In brief, our analysis affirms that units with mirror-symmetric viewpoint tuning for faces appear even in untrained CNNs, although we believe their rate is lower than previously reported. Regardless of the question of the exact proportion of such units, we believe it is unequivocal that at the population level, mirror-symmetric viewpoint tuning to faces (and other objects with a single plane of symmetry) is strongly training-dependent.

First, we refer the reviewer to Figure S2, which directly demonstrates the effect of training on the population-level mirror symmetric viewpoint tuning:

Note the non-mirror-symmetric reflection invariant tuning profile for faces in the untrained network.

Second, the above-zero horizontal reflection-invariance referred by the reviewer (Figure 3) is distinct from mirror- symmetric viewpoint tuning; the latter requires both reflection-invariance and viewpoint tuning. More importantly, it was measured with respect to all of the object categories grouped together; this includes objects with quadrilateral symmetry, which elicit mirror-symmetric viewpoint tuning even in shallow layers and without training. To clarify the confusion that this grouping might have caused, we repeated the measurement of invariance in fc6, separately for each 3D object category:

Disentangling the contributions of different categories to the reflection-invariance measurements, this analysis under-scores the necessity of training for the emergence of mirror-symmetric viewpoint symmetry.

Last, we refer the reviewer to Figure S5, which shows that the symmetry of untrained convolutional filters has a narrow, zero-centered distribution. Indeed, the upper limit of this distribution includes filters with a certain degree of symmetry. This level of symmetry, however, becomes the lower limit of the filters’ symmetry distribution following training.

Therefore, we believe that training induces a shift in the tuning of the unit population that is qualitatively distinct from, and not explained by, random-lottery-related mirror-symmetric viewpoint tuned units. In the revised manuscript, we clarify the distinction between mirror-symmetric viewpoint tuning at the population level and the existence of individual units showing pre-training mirror symmetric viewpoint tuning, as shown by Baek et al.

Manuscript changes: (Discussion section)

Our claim that mirror-symmetric viewpoint tuning is learning-dependent may seem to be in conflict with findings by Baek and colleagues [17]. Their work demonstrated that units with mirror-symmetric viewpoint tuning profile can emerge in randomly initialized networks. Reproducing Baek and colleagues’ analysis, we confirmed that such units occur in untrained networks (Fig. S15). However, we also identified that the original criterion for mirror-symmetric viewpoint tuning employed in [17] was satisfied by many units with asymmetric tuning profiles (Figs. S14 and S15). Once we applied a stricter criterion, we observed a more than twofold increase in mirror-symmetric units in the first fully connected layer of a trained network compared to untrained networks of the same architecture (Fig. S16). This finding highlights the critical role of training in the emergence of mirror-symmetric viewpoint tuning in neural networks also at the level of individual units.

Point 3.5) Finally, it would help to cite other previous demonstrations of equivariance and mirror symmetry in neural networks. Chris Olah, Nick Cammarata, Chelsea Voss, Ludwig Schubert, and Gabriel Goh of OpenAI wrote of this phenomenon in 2020 (Distill journal).

We added a reference to the study by Olah and colleagues (2020).

Manuscript changes: (Discussion section)

(see Olah and colleagues (2020) [60] for an exploration of emergent equivariance using activation maximiza- tion).

Point 3.6) Some other observations that might help:

I am enthusiastic about the experiments using different datasets to increase or decrease the rate of mirror-symmetry tuning (sets including CIFAR10, SVHN, symSVHN, asymSVHN); it is worth noting, however, that the lack of a ground truth metric for category symmetry is a problem here too. In the asymSVHN dataset, images are flipped and given different labels. If some categories are naturally symmetric after horizontal flips, such as images containing ”0” or ”8”, then changing the label is likely to disturb training. This would explain why the training loss is larger for this condition (Figure S4D).

We now acknowledge that the inclusion of digits 0 and 8 reduces the accuracy of asymSVHN:

Manuscript changes: (Figure S4 caption)

Note that the accuracy of asymSVHN might be negatively affected by the inclusion of relatively symmetric categories such as 0 and 8.

Our rationale for retaining these digits in the dataset was to manipulate the symmetry of the learned categories (compared to symSVHN) while keeping the images themselves constant.

Regarding ground-truth symmetry of these dataset: For CIFAR-10, the relevant measure of symmetry pertains to the 3D structure of the photographed objects, which we believe is unequivocally symmetric (see Point 3.2). Note that 2D, pixel-space image symmetry is not directly indicative of symmetry in 3D.

For SVHN, which consists of two-dimensional characters, the pixel-space symmetry of the images indeed reflects the objects’ symmetry. However, since we are worried that some readers might confuse our claims that relate to the symmetry of objects with claims (we did not make) about symmetry of 2D images, we prefer to avoid reporting measurements of image-space symmetry. We believe that our interpretation of the experiments with SVHN/symSVHN/asymSVHN holds even in the absence of such measurements.

For your reference, we include here a quantification of image-space horizontal symmetry for each category of CIFAR-10 and SVHN:

Point 3.7) It is puzzling why greyscale 3D rendered images are used. By using greyscale 3D render (at least as shown in the figures) the study proceeds as if the units are invariant under color transformations. Unfortunately, this is not true and using greyscale images impact the activations of different layers of Alexnet in a way that is not fully defined. Moreover, many units in shallow networks focus on color and exactly these units could be invariant to other transformation like the mirror symmetry, but grey scaling the images makes them inactive.

We use grayscale 3D rendered images to align with the setting in other studies investigating mirror- symmetric viewpoint tuning, including Freiwald et al. (2010), Leibo et al. (2017), and Yildirim et al. (2020). The choice of using grayscale images in these studies is motivated by the need to dissociate face-processing from lower-level, hue-specific responses.

eLife assessment

This computational study is a valuable empirical investigation into the common trait of neurons in brains and artificial neural networks: responding effectively to both objects and their mirror images and it focuses on uncovering conditions that lead to mirror symmetry in visual networks and the evidence convincingly demonstrates that learning contributes to expanding mirror symmetry tuning, given its presence in the data. Additionally, the paper delves into the transformation of face patches in primate visual hierarchy, shifting from view specificity to mirror symmetry to view invariance. It empirically analyzes factors behind similar effects in two network architectures, and key claims highlight the emergence of invariances in architectures with spatial pooling, driven by learning bilateral symmetry discrimination and importantly, these effects extend beyond faces, suggesting broader relevance. Despite strong experiments, some interpretations lack explicit support, and the paper overlooks pre-training emergence of mirror symmetry.

Reviewer #1 (Public Review):

By using deep convolutional neural networks (CNNs) as model for the visual system, this study aims at understanding and explaining the emergence of mirror-symmetric viewpoint tuning in the brain.

Major strengths of the methods and results:

(1) The paper presents comprehensive, insightful and detailed analyses investigating how mirror-symmetric viewpoint tuning emergence in artificial neural networks, providing significant and novel insights into this complex process.<br /> (2) The authors analyze reflection equivariance and invariance in both trained and untrained CNNs' convolutional layers. This elucidates how object categorization training gives rise to mirror-symmetric invariance in the fully-connected layers.<br /> (3) By training CNNs on small datasets of numbers and a small object set excluding faces, the authors demonstrate mirror-symmetric tuning's potential to generalize to untrained categories and the necessity of view-invariant category training for its emergence.<br /> (4) A further analysis probes the contribution of local versus global features to mirror-symmetric units in the first fully-connected layer of a network. This innovative analysis convincingly shows that local features alone suffice for the emergence of mirror-symmetric tuning in networks.<br /> (5) The results make a clear prediction that mirror-symmetric tuning should also emerge for other bilaterally symmetric categories, opening avenues for future neural studies.

Major weaknesses of the methods and results:

(1) The authors propose a mirror-symmetric viewpoint tuning index, which, although innovative, complicates comparison with previous work and this choice is not well motivated. This index is based on correlating representational dissimilarity matrices (RDMs) with their flipped versions, a method differing from previous approaches.<br /> (2) Faces exhibit unique behavior in terms of the progression of mirror-symmetric viewpoint tuning and their training task and dataset dependency. Given that mirror-symmetric tuning has been identified in the brain for faces, it would be beneficial to discuss this observation and provide potential explanations.<br /> (3) Previous work reported critical differences between CNNs and neural representations in area AL indicating that mirror-symmetric viewpoint tuning is less present than view invariance in CNNs compared to area AL. While such findings could potentially limit the usefulness of CNNs as models for mirror-symmetric viewpoint tuning in the brain, they are not addressed in the study.<br /> (4) The study's results, while informative, are qualitative rather than quantitative, and lack direct comparison with neural data. This obscures the implications for neural mechanisms and their relevance to the broader field.

The study provides compelling evidence that learning to discriminate bilaterally symmetric objects (beyond faces) induces mirror-symmetric viewpoint tuning in the networks, qualitatively similar to the brain. Moreover, the results suggest that this tuning can, in principle, generalize beyond previously trained object categories. Overall, the study provides important conclusions regarding the emergence of mirror-symmetric viewpoint tuning in networks, and potentially the brain. However, the conducted analyses and results do not entirely address the question why mirror-symmetric viewpoint tuning emerges in networks or the brain. Specifically, the results leave open whether mirror-symmetric viewpoint tuning is indeed necessary to achieve view invariance for bilaterally symmetric objects.

Taken together, this study moves us a step closer to uncovering the origins of mirror-symmetric tuning in networks, and has implications for more comprehensive investigations into this neural phenomenon in the brain. The methods of probing CNNs are innovative and could be applied to other questions in the field. This work will be of broad interest to cognitive neuroscientists, psychologists, and computer scientists.

Reviewer #2 (Public Review):

Strengths

(1) The statements made in the paper are precise, separating observations from inferences, with claims that are well supported by empirical evidence. Releasing the underlying code repository further bolsters the credibility and reproducibility. I especially appreciate the detailed discussion of limitations and future work.

(2) The main claims with respect to the two convolutional architectures are well supported by thorough analyses. The analyses are well-chosen and overall include good controls, such as changes in the training diet. Going beyond "passive" empirical tests, the paper makes use of the fully accessible nature of computational models and includes more "causal" insertion and deletion tests that support the necessity and sufficiency of local object features.

(3) Based on modeling results, the paper makes a testable prediction: that mirror-symmetric viewpoint tuning is not specific to faces and can also be observed in other bilaterally symmetric objects such as cars and chairs. To test this experimentally in primates (and potentially other model architectures), the stimulus set is available online.

Weaknesses

My main concern with this paper is in its choice of the two model architectures AlexNet and VGG. In an earlier study, Yildirim et al. (2020) found an inverse graphics network "EIG" to better correspond to neural and behavioral data for face processing than VGG. All claims in the paper thus relate to a weaker model of the biological effects since this work does not analyze the EIG model. Since EIG follows an analysis-by-synthesis approach rather than standard classification training, it is unclear whether the claims in this paper generalize to this other model architecture. It is also unclear if the claims will hold for: 1) transformer architectures, 2) the HMAX architecture by Leibo et al. (2017) which has also been proposed as a computational explanation for mirror-symmetric tuning, and, as the authors note in the Discussion, 3) deeper architectures such as ResNet-50 which tend to better align to neural and behavioral data in general. These architectures include different computational motifs such as skip connections and a much smaller proportion of fully-connected layers which are a major focus of this work.

Overall, I thus view the paper's claims as limited to AlexNet- and VGG-like architectures, both of which fall behind state-of-the-art in their alignment to primates in general and also specifically for mirror-symmetric viewpoint tuning.

Minor weaknesses

(1) Figure 1A: since the relevance to primate brains is a major motivator of this work, the results from actual neural recordings should be shown and not just schematics. For instance, the mirror symmetry in AL is not as clean as the illustration (compare with Fig. 3 in Yildirim et al. 2020), and in the paper's current form, this is not easily accessible to the reader.

(2) Figure 4 / L832-845: The claims for the effect of training on mirror-symmetric viewpoint tuning are with respect to the training data only, but there are other differences between the models such as the number of epochs (250 for CIFAR-10 training, 200 for all other datasets), the learning rate (2.5 * 10^-4 for CIFAR-10, 10^-4 for all others), the batch size (128 vs 64), etc. I do not expect these choices to make a major difference for your claims, but it would be much cleaner to keep everything but the training dataset consistent. Especially the different test accuracies worry me a bit (from 81% to 92%, and they appear different from the accuracy numbers in figure S4 e.g. for CIFAR-10 and asymSVHN), at the very least those should be comparable.

(3) L681-685: The general statement made in the paper that "deeper models lose their advantage as models of cortical representations" is not supported by the cited limited comparison on a single dataset. There are many potential confounds here with respect to prior work, e.g. the recording modality (fMRI vs electrodes), the stimulus set (62 images vs thousands), the models that were tested (9 vs hundreds), etc.

Reviewer #3 (Public Review):

This study aimed to explore the computational mechanisms of view invariance, driven by the observation that in some regions of monkey visual cortex, neurons show comparable responses to (1) a given face and (2) to the same face but horizontally flipped. Here they study this known phenomenon using AlexNet and other shallow neural networks, using an index for mirror symmetric viewpoint tuning based on representational similarity analyses. They find that this tuning is enhanced at fully connected- or global pooling layers (layers which combine spatial information), and that the invariance is prominent for horizontal- but not vertical- or rotational transformations. The study shows that mirror tuning can be learned when a given set of images are flipped horizontally and given the same label, but *not* if they are flipped and given different labels. They also show that networks learn this tuning by focusing on local features, not global configurations.

I found the study to be a mixed read. Some analyses were fascinating: for example, it was satisfying to see the use of well-controlled datasets to increase or decrease the rate of mirror-symmetry tuning. The insertion- and deletion¬ experiments were elegant tests to probe the mechanisms of mirror symmetry, asking if symmetry could arise from (1) global feature configurations (in a holistic sense) vs. (2) local features, with stronger evidence for the latter. These two sets of results were successful and interpretable. They stand in contrast with the first analysis, which relies on observations that do not seem justified. Specifically, Figure 2D shows mirror-symmetry tuning across 11 stages of image processing, from pixels space to fully connected layers. It shows that images from different object categories evoke considerably different tuning index values. The explanation for this result is that some categories, such as "tools," have "bilaterally symmetric structure," but this is not explicitly measured anywhere. "Boats" are described as having "front-back symmetry," more so than flowers. One imagines flowers being extremely symmetric, but perhaps that depends on the metric. What is the metric? At first I thought it was the mirror-symmetric viewpoint tuning index in the image (pixel) space, but this cannot be, as the index for faces and flowers is negative, cars have no symmetry, and boats are positive. To support these descriptions, one must have an independent variable (for object class symmetry) that can be related to the dependent variable (the mirror-symmetric viewpoint tuning index). If it exists, it is not a part of the Results section. This omission undermines other parts of the Results section: "some car models have an approximate front-back symmetry...however, a flower typically does not..." "Some," "typically:" how many in the dataset exactly, and how often? The description of CIFAR-10 as having bilaterally symmetric categories - are all these categories equally symmetric? If not, would such variability matter in terms of these results? These assessments of object category symmetry values are made before experiments are presented, so they are not interpretations of the results, and it would be circular to write it otherwise.

Overall, my bigger concern is that the framing is misleading or at best incomplete. The manuscript successfully showed that if one introduces left-right symmetry to a dataset, the network will develop population-level representations that are also bilaterally symmetric. But the study does not explain that the model's architecture and random weight distribution are sufficient for symmetry tuning to emerge, without training, just to a much more limited degree. Baek et al. showed in 2021 that viewpoint-invariant face-selective units and mirror-symmetric units emerge in untrained networks ("Face detection in untrained deep neural networks"; this current manuscript cites this paper but does not mention that mirror symmetry is a feature of the 2021 study). This current study also used untrained networks as controls (Fig. 3), and while they were useful in showing that learning boosts symmetry tuning, the results also clearly show that horizontal-reflection invariance is far from zero. So, the simple learning-driven explanation for the mirror-symmetric viewpoint tuning for faces is wrong: while (1) network training and (2) pooling are mechanisms that charge the development of mirror-symmetric tuning, the lottery ticket hypothesis is enough for its emergence. Faces and numbers are simple patterns, so the overparameterization of networks is enough to randomly create units that are tuned to these shapes and to wire many of them together. How learning shapes this process is an interesting direction, especially now that this current study has outlined its importance.

Finally, it would help to cite other previous demonstrations of equivariance and mirror symmetry in neural networks. Chris Olah, Nick Cammarata, Chelsea Voss, Ludwig Schubert, and Gabriel Goh of OpenAI wrote of this phenomenon in 2020 (Distill journal).

Some other observations that might help:

- I am enthusiastic about the experiments using different datasets to increase or decrease the rate of mirror-symmetry tuning (sets including CIFAR10, SVHN, symSVHN, asymSVHN); it is worth noting, however, that the lack of a ground truth metric for category symmetry is a problem here too. In the asymSVHN dataset, images are flipped and given different labels. If some categories are naturally symmetric after horizontal flips, such as images containing "0" or "8", then changing the label is likely to disturb training. This would explain why the training loss is larger for this condition (Figure S4D).

- It is puzzling why greyscale 3D rendered images are used. By using greyscale 3D render (at least as shown in the figures) the study proceeds as if the units are invariant under color transformations. Unfortunately, this is not true and using greyscale images impact the activations of different layers of Alexnet in a way that is not fully defined. Moreover, many units in shallow networks focus on color and exactly these units could be invariant to other transformation like the mirror symmetry, but grey scaling the images makes them inactive.

Author Response

Reviewer #1 (Public Review):

In this manuscript, the authors perform a very thorough, extensive characterization of the impact of an iron-rich diet on multiple phenotypes in a wide range of inbred mouse strains. While a work of this type does not offer mechanistic insights, the value of the study lies not only in its immediate results but also in what it can offer to future researchers as they explore the genetic basis of iron levels and other related phenotypes in rodent studies. The creation of a web resource and the offer from the authors to share all available samples is particularly laudable, and helps to increase the accessibility of the work to other scientists. There is one shortcoming to the work however. To induce iron overload in mice in the main study in this work, mice were placed on an iron-rich diet that differed in its composition from the baseline diet in more than just iron. This could influence some of the phenotypes observed in this study.

We thank the reviewer for their comments. We hope that this work can provide insight and/or support for a wide variety of future studies. Regarding the diets, yes, in our initial pilot study with 6 strains, the baseline diet was inadvertently not isocaloric with the high iron diet, and it also used a different source of cellulose and contained individual amino acids in ratios found in casein, instead of casein, which was used as the protein source for the high iron diet. The baseline metal composition however was the same. We included data from the pilot study in this manuscript because it provided some important early insight, but made sure to note this caveat since it could potentially affect some results. We added some additional text to the Methods section to help clarify this further. The other subsequently performed studies in this paper were not affected, for example the Control study performed in C57BL/6J has a baseline diet that matches the high iron diet except for iron. For our HMDP genetic study with 114 strains, we did not have a baseline group, so all mice were on the same high iron diet.

Reviewer #2 (Public Review):

Here, the authors tried to identify the genes and biological pathways underlying iron overload and its associated pathologies in mice. Several wet lab experiments and measurements alongside many bioinformatic analyses like GWAS, RNA-seq data analysis (DEG), eQTL analysis, TWAS, and gene-set enrichment analysis have been performed. The study design is good enough and the author tried to validate the results. The data have been submitted (Accession #: GSE230674) but are not public yet.

Thank you very much for your detailed and thoughtful review and for helping us to improve our manuscript.

1) The main issue of this manuscript is its length. It's too long, especially the result section. It's hard for readers to follow the paper. Moreover, you added results about other minerals, mostly copper, which seems too much (considering the fact that this study is about iron). The text doesn't have the required Integrity and focus. You should decide where you want to put the focus of this manuscript and I strongly recommend shortening the manuscript, try to be short and sweet as much as you can.

Thank you for this helpful suggestion. We have moved or removed excess discussion from the Results section. We moved the specific GWAS results for copper and related red cell traits to the Supplementary text file “Supplementary File 24” so that only iron and triglyceride GWAS results are described in the main text. We kept in the discussion about the copper findings in the Discussion section, since we believe the deficiency is an important phenotype induced by the high iron diet that may impact other studies of dietary iron overload. We also believe that the copper and anemia GWAS loci may be of interest to some readers. We considered putting the copper and anemia findings in a separate manuscript, but ultimately decided to include it here, although we do agree it makes the manuscript longer.

2) Also, the "Methods" section is long, some parts are over-detailed (mostly wet lab procedures) and some parts are not detailed enough. It seems the "Statistical analyses" part doesn't have extra information. I recommend removing the first paragraph and moving some of the information from the second paragraph to the right place in the Method section.

We reorganized the first part of the statistical analyses section for clarity, and as mentioned further below, added in more detail regarding the GWAS significance thresholds:

“Analyses were performed using GraphPad Prism (GraphPad Software, La Jolla, CA) and in R. P < 0.05 was considered significant for these tests and for bicor analyses. All reported P values are based on a two-sided hypothesis. The initial number of mice per group in the pilot (N = 6 per group) and Control studies (N = 8 per group) were determined based on previous studies where similar phenotypes were measured. For the HMDP study, permutation and simulation studies were previously used to test the statistical power of the HMDP using parameters including the variance explained by SNPs, genetic background, random errors, and the number of repeated measurements per strain (Bennett, Farber et al. 2010). Appropriate sample sizes to achieve adequate statistical power were determined based on previous analyses. Differences in sample sizes among the HMDP strains were due to differences in strain availability as determined by breeding success and losses. For GWAS, thresholds for significant (P < 4.1e-6; -log10P > 5.387) loci were defined using permutation as previously described (Bennett, Farber et al. 2010). The suggestive locus threshold (P < 4.1e-5; -log10P > 4.387) was based on reducing the significance threshold by one log unit. The cis eQTL GWAS threshold (P < 1e-4) was based on a calculated 1% FDR threshold of 1.73e-3, adjusted to 1e-4 to be slightly more conservative. The trans-eQTL threshold (P < 1e-6) was based on the 4.1e-6 threshold, adjusted to 1e-6 to be more conservative as well.”

We tried moving the missing values notes in the second paragraph to the various method sections in the paper they apply to, but this led to much repetition and was in some cases not clear, so we decided to keep this information together in the statistical analyses section.

3) Some part of your discussion section, is retelling the results. Please discuss your results and compare them with previous findings.

We have revised the discussion to remove several parts that mostly just summarized the results and agree this improves the text. As mentioned above, we moved some discussion that was in the Results section to the Discussion section as well.

4) Add detail about your GWAS model. As you had repeated samples from each strain, it's good to mention how you considered this. Also, show how you determined the significance threshold.

Thank you for this suggestion. The GWAS software we used (FaST-LMM) derives a kinship matrix from the genotypes of the individuals considered in the analysis; this kinship matrix is used to correct for population structure including multiple individuals per strain.

The trait GWAS significance threshold was determined using permutation analysis (Bennett, Farber et al. 2010). The suggestive GWAS threshold was based on reducing the significance threshold by one log unit. The cis eQTL GWAS threshold was based on a calculated 1% FDR threshold of 1.73e-3, adjusted to 1e-4 to be slightly more conservative. The trans-eQTL threshold was based on the 4.1e-6 threshold, adjusted to 1e-6 to be more conservative as well.

To improve the text, we added to the Methods section under the “Genome-wide association analysis and heritability estimation” header the following:

“Traits were quantile transformed to normalize the distribution and then GWAS was performed using the FaST-LMM program (Lippert, Listgarten et al. 2011), which corrects for population structure (including multiple samples per strain) by using a kinship matrix derived from the genotypes to be analyzed.”

We also revised the GWAS threshold text to include more detail:

“Analyses were performed using GraphPad Prism (GraphPad Software, La Jolla, CA) and in R. P < 0.05 was considered significant for these tests and for bicor analyses. All reported P values are based on a two-sided hypothesis. For GWAS, thresholds for significant (P < 4.1e-6; -log10P > 5.387) loci were defined using permutation as previously described (Bennett, Farber et al. 2010). The suggestive locus threshold (P < 4.1e-5; -log10P > 4.387) was based on reducing the significance threshold by one log unit. The cis eQTL GWAS threshold (P < 1e-4) was based on a calculated 1% FDR threshold of 1.73e-3, adjusted to 1e-4 to be slightly more conservative. The trans-eQTL threshold (P < 1e-6) was based on the 4.1e-6 threshold, adjusted to 1e-6 to be more conservative as well. “

5) The abstract could be better. It also doesn't have a conclusion.

We revised the abstract and added in a conclusion:

“Tissue iron overload is a frequent pathologic finding in multiple disease states including non-alcoholic fatty liver disease (NAFLD), neurodegenerative disorders, cardiomyopathy, diabetes, and some forms of cancer. The role of iron, as a cause or consequence of disease progression and observed phenotypic manifestations, remains controversial. In addition, the impact of genetic variation on iron overload related phenotypes is unclear, and the identification of genetic modifiers is incomplete. Here, we used the Hybrid Mouse Diversity Panel (HMDP), consisting of over 100 genetically distinct mouse strains optimized for genome-wide association studies (GWAS) and systems genetics, to characterize the genetic architecture of dietary iron overload and pathology. Dietary iron overload was induced by feeding male mice (114 strains, 6-7 mice per strain on average) a high iron diet for six weeks, and then tissues were collected at 10-11 weeks of age. Liver metal levels and gene expression were measured by ICP-MS/ICP-AES and RNASeq, and lipids were measured by colorimetric assays. FaST-LMM was used for genetic mapping, and Metascape, WGCNA, and Mergeomics were used for pathway, module, and key driver bioinformatics analyses. Across the HMDP, we identified many traits that exhibited high inter-strain variability on the high iron diet, and we found a substantial contribution of genetics to many traits. Mice on the high iron diet accumulated iron in the liver, with a 6.5 fold difference across strain means. The iron loaded diet also led to a spectrum of copper deficiency and anemia, with liver copper levels highly positively correlated with red blood cell count, hemoglobin, and hematocrit. Hepatic steatosis of various severity was also observed histologically, with 52.5 fold variation in triglyceride levels across the strains. Most clinical traits examined had at least one significant GWAS locus, and notably, liver triglyceride and iron mapped most significantly to an overlapping locus on chromosome 7 that has not been previously associated with either trait. By genetically mapping liver mRNA expression, we identified cis- and trans-eQTL for thousands of genes, and we integrated this with trait correlation data to identify candidate causal genes at many trait loci. Using network modeling, significant key drivers for both iron and triglyceride accumulation were found to be involved in cholesterol biosynthesis and oxidative stress management. To make the full data set accessible and useable by others, we have made our data and analyses available on a resource website. Overall, our study confirms and expands upon the contribution of mouse genetic background to dietary iron overload and associated pathology. The numerous GWAS loci, candidate genes, and biological pathways identified here provide a rich public resource to drive further investigation.”

6) Page 8, lines 4-7: Please remove these lines or move them to the Method section. The last paragraph of the introduction should clearly explain the goal of the study.

We removed these lines and revised this paragraph for clarity:

In order to gain further insight into genetic contributors to iron overload and associated pathology, we measured clinical traits and hepatic mRNA expression in 114 mouse strains fed a high iron diet. The mice are from a genetically diverse cohort known as the Hybrid Mouse Diversity Panel (HMDP), a panel optimized for systems genetics studies that has previously been used to examine numerous complex traits, including obesity, diabetes, atherosclerosis, heart failure, carbon tetrachloride induced liver fibrosis, and fatty liver disease (Lusis, Seldin et al. 2016; Seldin, Yang et al. 2019; Tuominen, Fuqua et al. 2021; Cao, Wang et al. 2022).

7) Page 68, line 13: Explain the abbreviation (RINe) before use. Also, most probably it is RIN (RNA Integrity Number).

Thank you for pointing this out. We updated the methods text as follows: “All samples had RNA integrity number equivalents (RINe) values greater than 8 as measured on an Agilent 2200 TapeStation (Agilent, Santa Clara, CA).” We also added RINe to the abbreviations section.

8) The heritability estimates seem high and the 1% difference between broad- and narrow-sense heritability means there is almost no dominant and epistatic genetic variance between alleles affecting the studied trait (which is hard to accept). I recommend considering a within-group (strain) variance (common environmental effect) component in the model to absorb this source of variation in this component, so the genetic variance and consequently the heritability estimates would be more accurate. You also can consider this source of variance in your GWAS model.

Thank you for bringing up these points. While we try to minimize environmental effects by keeping these mice and samples in as similar environmental and experimental conditions as feasible, some will remain. Thus, in our analyses, we try to factor in remaining environmental variation by using data from multiple mice per strain. The programs we used for GWAS and heritability calculations take into account within-group (strain) variance. We added the following sentence to the Methods section just after mention of the programs used to calculate heritability:

“Both of the software packages used for heritability estimation account for environmental variance within strains.”

We agree that the broad-sense and narrow-sense estimates are close to each other for many traits and that this suggests low levels of dominance and epistasis. A low level of non-additive genetic variance is not uncommon and theoretically predicted for complex traits, as has been reported previously and discussed in the references below:

Hill WG, Goddard ME, Visscher PM. Data and theory point to mainly additive genetic variance for complex traits. PLoS Genet. 2008 Feb 29;4(2):e1000008. doi: 10.1371/journal.pgen.1000008. PMID: 18454194

Hivert V, Sidorenko J, Rohart F, Goddard ME, Yang J, Wray NR, Yengo L, Visscher PM. Estimation of non-additive genetic variance in human complex traits from a large sample of unrelated individuals. Am J Hum Genet. 2021 May 6;108(5):786-798. doi: 10.1016/j.ajhg.2021.02.014. Epub 2021 Apr 2. Erratum in: Am J Hum Genet. 2021 May 6;108(5):962. PMID: 33811805

It has also been argued that many human GWAS studies, as well as studies using populations of mice designed for complex trait analyses, including the HMDP population, inherently lack the statistical power to detect epistasis:

Buchner DA, Nadeau JH. Contrasting genetic architectures in different mouse reference populations used for studying complex traits. Genome Res. 2015 Jun;25(6):775-91. doi: 10.1101/gr.187450.114. Epub 2015 May 7. PMID: 25953951

Taking all this together we would argue that it is not surprising to see the little difference between the narrow and broad heritability estimates for many traits in our study. To provide more context to the reader regarding how to interpret our heritability findings, we added the following text to the discussion section, under limitations:

“Finally, in our study with the HMDP population, estimated broad and narrow sense heritabilities were similar for many traits, suggesting modest non-additive contributions (e.g dominance and epistasis) to the variance in these traits. While such results are common and theoretically predicted for complex traits (Hill, Goddard et al. 2008; Hivert, Sidorenko et al. 2021), our study population may also not be optimal for detection of these effects (Buchner and Nadeau 2015).”

eLife assessment

This manuscript presents a detailed phenotyping of the role of dietary iron in a large number of genetically distinct mouse strains. There are exciting and convincing data that could be valuable in their impact on the fields of nutrition, iron metabolism and anemia.

Reviewer #1 (Public Review):

In this manuscript, the authors perform a very thorough, extensive characterization of the impact of an iron-rich diet on multiple phenotypes in a wide range of inbred mouse strains. While a work of this type does not offer mechanistic insights, the value of the study lies not only in its immediate results but also in what it can offer to future researchers as they explore the genetic basis of iron levels and other related phenotypes in rodent studies. The creation of a web resource and the offer from the authors to share all available samples is particularly laudable, and helps to increase the accessibility of the work to other scientists. There is one shortcoming to the work however. To induce iron overload in mice in the main study in this work, mice were placed on an iron-rich diet that differed in its composition from the baseline diet in more than just iron. This could influence some of the phenotypes observed in this study.

Reviewer #2 (Public Review):

Here, the authors tried to identify the genes and biological pathways underlying iron overload and its associated pathologies in mice. Several wet lab experiments and measurements alongside many bioinformatic analyses like GWAS, RNA-seq data analysis (DEG), eQTL analysis, TWAS, and gene-set enrichment analysis have been performed. The study design is good enough and the author tried to validate the results. The data have been submitted (Accession #: GSE230674) but are not public yet.

(1) The main issue of this manuscript is its length. It's too long, especially the result section. It's hard for readers to follow the paper. Moreover, you added results about other minerals, mostly copper, which seems too much (considering the fact that this study is about iron). The text doesn't have the required Integrity and focus. You should decide where you want to put the focus of this manuscript and I strongly recommend shortening the manuscript, try to be short and sweet as much as you can.<br /> (2) Also, the "Methods" section is long, some parts are over-detailed (mostly wet lab procedures) and some parts are not detailed enough. It seems the "Statistical analyses" part doesn't have extra information. I recommend removing the first paragraph and moving some of the information from the second paragraph to the right place in the Method section.<br /> (3) Some part of your discussion section, is retelling the results. Please discuss your results and compare them with previous findings.<br /> (4) Add detail about your GWAS model. As you had repeated samples from each strain, it's good to mention how you considered this. Also, show how you determined the significance threshold.<br /> (5) The abstract could be better. It also doesn't have a conclusion.<br /> (6) Page 8, lines 4-7: Please remove these lines or move them to the Method section. The last paragraph of the introduction should clearly explain the goal of the study.<br /> (7) Page 68, line 13: Explain the abbreviation (RINe) before use. Also, most probably it is RIN (RNA Integrity Number).<br /> (8) The heritability estimates seem high and the 1% difference between broad- and narrow-sense heritability means there is almost no dominant and epistatic genetic variance between alleles affecting the studied trait (which is hard to accept). I recommend considering a within-group (strain) variance (common environmental effect) component in the model to absorb this source of variation in this component, so the genetic variance and consequently the heritability estimates would be more accurate. You also can consider this source of variance in your GWAS model.

Author Response

Reviewer #1 (Public Review):

Combining functional MRI with a decoder, the authors probe the neural substrate of the double drift illusion in visual cortex. Their elegant behavioural paradigm keeps the actual retinal position of the stimulus stable while inducing the illusion with a combination of smooth pursuit and visual motion. The results show that the illusory drift path can be decoded from a signal in extrastriate visual area hMT+ but not other visual areas. Importantly, this can be done in the absence of spatial attention to the stimulus location.

The particular strengths of this study lie in the elegant paradigm and the clear attentional control. The methodology of the decoder is powerful and at the same time straightforward, well explained, and well accepted in the literature. A potential weakness of the study is the lack of simultaneous eye movement recordings in the scanner. Such data could have provided further clarification of the potential underlying neural mechanism and whether differences in eye movements could contribute to the decoding of the visual illusion path. There are some controls that mitigate this.

We have addressed the Reviewer's comment by repeating the fMRI experiment in a new group of subjects in which we were able to also obtain concurrent, high-quality eye tracking. When we initially conducted the experiment, it was not possible to perform eye tracking in the 7T scanner at NIH. Because of this limitation, we were forced to depend on careful eye tracking in a pre-scan behavioral experiment. But in the ensuing period of time, we have developed a protocol for obtaining high quality eye tracking with an Eyelink 1000 mounted in the bore of the scanner. Now that we have the ability to collect concurrent eye tracking, we repeated the fMRI experiment and found that our main fMRI result replicated (i.e, it was possible to decode the direction of the illusion from fMRI responses in hMT+). Additional, the concurrent fMRI eye tracking enabled us to make four important observations (see new Fig 4):

First, subjects maintained stable fixation when the target was stationary during fixation and accurately pursued the vertically moving target during illusion (Fig 4). This analysis confirms that the drifting Gabor remained at a relatively fixed position on the retina during the illusory period.

Second, there were no differences in microsaccades between any of the conditions. We quantified the direction, amplitude, and frequency of all saccades for each condition. While we did observe small rightward microsaccades, none of the microsaccade characteristics differed between conditions. The rightward microsaccades may have been due to the sustained eccentric leftward fixation. Or, it may have been due to attention to the right visual field stimulus (despite the foveal attention task). Or it may have reflected the known horizontal microsaccade bias. Regardless, we do not believe our fMRI results are related to microsaccades because these small saccades did not differ across condition.

Finally, we wondered if small not-easily-quantified ocular deviations could have differed between conditions, and somehow result in differences in fMRI activity picked up by the decoding analysis. To test for this possibility, we trained a classier to discriminate condition based on the raw eye traces (just as we did in the main fMRI data analysis). But unlike the fMRI analysis, we found that it was not possible to decode the direction of the illusion from the eye traces themselves.

We conclude that the ability to decode the illusion from fMRI responses were not due to differences in eye movements caused by the illusion.

The authors provide important evidence for a potential neural substrate in the extrastriate visual cortex for encoding the perceived spatial location of a moving stimulus. This significantly extends previous studies that showed relevant spatiotopic signals outside visual cortex. Understanding the neural substrate and the underlying neural mechanisms for encoding perceived spatiotopic location are of broad importance for our understanding of the neural basis of sensory perception.

We thank the Editor for this positive assessment of our work.

Reviewer #3 (Public Review):

The authors studied the neural basis of the double drift illusion, an illusion in which a Gabor drifting both horizontally within an aperture and moving vertically along a path appears to follow a diagonal trajectory, perceptually displaced off its true vertical path in the direction of the horizontal drift. The illusion is strong and its neural basis is intriguing. The authors suggest it can be used to address the locus of spatiotopic processing in the brain. They find that fMRI BOLD activity in hMT+ can be used to decode the illusory drift direction of the stimulus, even under conditions of withdrawn attention. They internally replicate this result and ensure it is not due to local motion. They interpret the finding to indicate that hMT+ contains spatiotopic information. This was a carefully designed and conducted study, and the manuscript writing and figures are clear.

Despite the care that went into the study design and control experiments, I see some potential interpretational issues, and I am uncertain about the scientific advance. My main questions are about the interpretation of the findings, the possible confound of smooth pursuit eye movements, and the relation to previous studies, including previous fMRI studies of the same illusion. I also would like to see more thorough reporting of behavior.

Major comments

1) The authors motivate the study by saying that there have been conflicting results about which brain areas are involved in spatiotopic coding, but they did not give an indication about why there might be conflicting results or why the current study is suitable to address the previous discrepancies. Is this study simply adding another observation to the existing body of literature, or does it go beyond previous studies in a critical theoretical way?

There have indeed been conflicting results in the literature. One idea that has received some prior support in the literature is that spatiotopic location information can depend on the task. Our experiment tests this idea by measuring cortical responses during an illusion that involves spatiotopic coding. Previous human fMRI studies reporting spatiotopic coding have not really linked cortical activity with the perception of spatiotopic coordinates. Hence, we feel that our results make a unique contribution to the field.

2) The authors interpret the finding of illusory drift direction encoding in hMT+ to mean that hMT+ is coding the illusory spatial position of the stimulus. But could an alternative explanation be that hMT+ is coding the illusory global motion direction, and not the spatial position per se? If this is a possible account, then the result would still indicate that an illusory motion percept is reflected in hMT+ but it would seem not to answer the question about spatiotopic coding which motivated the paper.

Here, the Reviewer suggests an interesting alternative explanation—that responses in MT pertain to the direction of global motion rather than stimulus position. However, this alternative possibility would still involve spatiotopic coding. In order for the brain to compute the direction of global motion of a stimulus that is at a fixed retinal position, some spatiotopic computation must occur. So, we do not agree with the Reviewers suggestion that this alternative explanation undermines the motivation of this study.

3) It is good that the authors sought to rule out the possibility that smooth pursuit eye movements were driving the decoding results in hMT+, but I'm not sure they have yet convincingly done so. Decoding based on the pursuit selective voxels alone was very nearly significant (p = 0.052), which was not acknowledged in the text of the paper. Furthermore, because voxels that were both pursuit and stimulus selective were excluded from the pursuit selective ROI, decoding performance in that ROI may have been underestimated.

To clarify, voxels that were identified by both localizers were NOT excluded from either ROI. When we repeated decoding (from Expt 2, Fig 3B) using disjoint voxel selection (i.e., analyzing voxels that only responded in the stim localizer, or only responded in the pursuit localizer, and excluding voxels that responded to both), we obtained qualitatively similar results, although the magnitude of the effects were smaller, which is not surprising given the much smaller number of voxels remaining in the ROI, and hence the disjoint ROIs only proved marginally significant in MT for the stim localizer (p=0.049).

4) A previous fMRI study of the double drift illusion (Liu et al. 2019 Current Biology) also found above chance decoding of illusory drift direction in hMT+. The authors mention this study but do not discuss it, so it was unclear to me what the advance is of the current study over that study. The main differences I see are that in the current study, 1) the observer is also moving their eyes so that the double drift stimulus is theoretically stabilized on the retina, and 2) attention is withdrawn from the stimulus. But in both studies, hMT+ contains information about the illusory drift direction even though retinotopic information is the same, so it's not clear to me that the differences between these studies lead to fundamentally different interpretations.

The results of Liu et al. are not relevant to the reference frame used to encode the stimulus. Because subjects were fixating in Liu et al., the encoding of the illusion could have been in either retinal or spatiotopic coordinates. In our study, the stimulus must have been encoded in spatiotopic coordinates. One interesting feature of Liu et al. is the issue of cross decoding the illusion and actual percept (training the decoder on veridical motion of different angles, and then testing the decoder on data collected during the illusion). One potentially interesting extension of the cross decoding approach would be to train the decoder on a version of the illusion involving fixation (as in Liu et al), but then testing the decoder on the illusion during pursuit. One would expect cross decoding if spatiotopic coordinates are used in both cases. We now discuss this possibility (Discussion: Relationship to a previous study of the double-drift illusion).

Author Response

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

We thank the reviewers for their reading of the manuscript, and their suggestions. We have extensively addressed all these concerns in the text, and also included several new data and figures in the revised version of the manuscript. We hope that our response and the new experimental data fully address the concerns raised by the reviewers. We include a detailed, pointby-point response to each of the reviewer concerns, pointing to new data and specific changes made in the main manuscript.

Note: Do note that these new data have resulted in a new figure-figure 6, a new supplementary figure -figure 2-figure supplement 2, and an increase in the number of panels in each figure, as well as supplementary figures.

General response comments, highlighting a few aspects missed by the reviewers

This manuscript has an enormous amount of data in it. This is understandable, since in part we are proposing an entirely new hypothesis, and way to think about mitochondrial repression, built around substantial circumstantial evidences from diverse literature sources. But to keep the narrative readable and the main idea understandable, a lot of information had to be only very briefly mentioned in the text, and is therefore included as supplemental information. Due to that, it may not always be apparent that this study has set several technical benchmarks. These experiments are extremely challenging to perform, took many iterations to standardize, and in themselves are a first in the field. Yeast cells have the highest known rate of glycolytic flux for any organism. Measuring this glycolytic rate using the formation of intermediates is hard, and all current estimates have been in vitro, and using a stop-flow type set up. In this study, we optimized and directly measured the glycolytic flux using isotope labelled glucose (13C-glucose), which has never been reported before in highly glycolytic cells such as yeast. This is due to the very rapid label saturation (within seconds) after 13C glucose pulse (as is now shown in the figure 2-figure supplement 1). For brevity, this is summarized in this study with sufficient information to reproduce the method, but we will put out a more detailed, associated methodology paper describing several challenges, infrastructure requirements, and resources to be able to carry out these types of experiments using yeast. An added highlight of these experiments with WT and Ubp3 deletion strains is the most direct till date experimental demonstration that glycolytic flux in yeast in high glucose follows zero-order kinetics, and depends entirely on the amounts of the glycolytic enzymes (presumably operating at maximal activity). This nicely complements the recent study by Grigatis 2022 (cited in the discussion), that suggests this possibility.

Separately, this study required the estimation of total inorganic phosphates, as well as mitochondrial pools of phosphates. Till date, there are no studies that have estimated mitochondrial pools of phosphate (for a variety of reasons). In this study, we also experimentally determined the changes in mitochondrial phosphate pools. For this, we had to establish and standardize a rapid mitochondrial isolation method in yeast. Thus, this study provides the first quantitative estimates of mitochondrial Pi amounts (in the context of measured mitochondrial outputs), as shown now in Figure 4. This component on mitochondrial isolation in yeast to assess metabolites may also be explored in future as a methods paper.

Specific responses to the Reviews:

Reviewer #1 (Public Review):

The study by Vengayil et al. presented a role for Ubp3 for mediating inorganic phosphate (Pi) compartmentalization in cytosol and mitochondria, which regulates metabolic flux between cytosolic glycolysis and mitochondrial processes. Although the exact function of increased Pi in mitochondria is not investigated, findings have valuable implications for understanding the metabolic interplay between glycolysis and respiration under glucose-rich conditions. They showed that UBP3 KO cells regulated decreased glycolytic flux by reducing the key Pidependent-glycolytic enzyme abundances, consequently increasing Pi compartmentalization to mitochondria. Increased mitochondria Pi increases oxygen consumption and mitochondrial membrane potential, indicative of increased oxidative phosphorylation. In conclusion, the authors reported that the Pi utilization by cytosolic glycolytic enzymes is a key process for mitochondrial repression under glucose conditions.

(1) However, the main claims are only partially supported by the low number of repeats and utilizing only one strain background, which decreased the overall rigor of the study. The fullpower yeast model could be utilized with testing findings in different backgrounds with increased biological repeats in many assays described in this study. In the yeast model, it has been well established that many phenotypes are genotype/strain dependent (Liti 2019, Gallone 2016, Boekout 2021, etc...). with some strains utilizing mitochondrial respiration even under high glucose conditions (Kaya 2021). It would be conclusive to test whether wild strains with increased respiration under high glucose conditions would also be characterized by increased mitochondrial Pi.

“However, the main claims are only partially supported by the low number of repeats and utilizing only one strain background, which decreased the overall rigor of the study. The full-power yeast model could be utilized with testing findings in different backgrounds with increased biological repeats in many assays described in this study.”

Thank you for the suggestion. We agree that a larger, universal statement cannot be made with data from a single strain, since yeasts do have substantial diversity. In this study, we had originally used a robust, prototrophic industrial strain (CEN.PK background). We have now utilized multiple, diverse strains of S. cerevisiae to test our findings. This includes strains from the common laboratory backgrounds – W303 and BY4742 – which have different auxotrophies, as well as another robust, highly flocculent strain from a prototrophic Σ1278 background. Using all these strains, we now comprehensively find that the role of altered Pi budgeting as a constraint for mitochondrial respiration, and the role of Ubp3 as a regulator of mitochondrial repression is very well conserved. In all tested strains of S. cerevisiae the loss of Ubp3 increases mitochondrial activity (as shown by increased mitochondrial membrane potential and increased Cox2 levels in Figure 6A, B). These data now expand the generality of our findings, and strengthen the manuscript. These results are included in the revised manuscript as a new figure- Figure 6 and the associated text.

Some of the included data in the revised manuscript are shown below:

Author response image 1.

Mitochondrial activity and Cox2 levels in ubp3Δ in different genetic backgrounds

We also used the W303 strain to assess Pi levels, and its role in increasing mitochondrial respiration. We find that the loss of Ubp3 in this genetic background also increases Pi levels and that the increased Pi is necessary for increasing mitochondrial respiration (Figure 6C, D).

Author response image 2.

Basal OCR in WT vs ubp3Δ (W303 strain background) in normal vs low Pi

These experiments collectively have strengthened our findings on the critical role of intracellular Pi budgeting as a general constraint for mitochondrial respiration in high glucose.

“It would be conclusive to test whether wild strains with increased respiration under high glucose conditions would also be characterized by increased mitochondrial Pi.”

Addressed partially above. Right now the relative basal respiration in glucose across different strains is not well known. We measured mitotracker activity in high glucose in multiple WT strains of S. cerevisiae (W303, Σ1278, S288C and BY4742, compared to the CEN.PK strain). These strains all largely had similar mitotracker potential, except for a slight increase in mitochondrial membrane potential in Σ1278 strain, but not in other strains. We further characterized this using Cox2 protein levels as well as basal OCR, and found that these do not increase. These data is shown below, and is not included in the main text since it does not add any new component to the study.

Author response image 3.

Mitochondrial respiration in different WT strains

We did find this suggestion very interesting though, and are exploring directions for future research based on this suggestion. Since we have now identified a role for intracellular Pi allocation in regulating the Crabtree effect, an interesting direction can be to understand the glucose dependent mitochondrial Pi transport in Crabtree negative yeast strains. We will have to bring in a range of new tools and strains for this, so these experiments are beyond the focus of this current study.

We hope that these new experiments in different genetic backgrounds increases the breadth and generality of our findings, and stimulates new lines of thinking to address how important the role of Pi budgeting as a constraint for mitochondrial repression in high glucose might be.

(2) It is not described whether the drop in glycolytic flux also affects TCA cycle flux. Are there any changes in the pyruvate level? If the TCA cycle is also impaired, what drives increased mitochondrial respiration?

Thank you for pointing this out, and we agree this should be included. We have addressed these concerns in the revised version of the manuscript

Since glucose derived pyruvate must enter the mitochondrial TCA cycle, one possibility is that a decrease in glycolytic rate could decrease the TCA flux. An alternate possibility is that the cells coincidently increase the pyruvate transport to mitochondria, to thereby maintain the TCA cycle flux comparable to that of WT cells. To test both these possibilities, we first measured the steady state levels of pyruvate and TCA cycle intermediates in WT vs ubp3Δ cells. We do not observe any significant change in the levels of pyruvate, or TCA cycle intermediates (except malate, which showed a significant decrease in ubp3Δ cells). This data is now included in the revised manuscript as Figure 2 – figure supplement 1D and figure supplement 2 A, along with associated text.

Author response image 4.

Pyruvate levels in WT vs ubp3Δ

Author response image 5.

Steady state TCA cycle intermediate levels

Next, in order to address if the TCA cycle flux is impaired in ubp3Δ cells, we also measured the TCA cycle flux in WT vs ubp3Δ cells by pulsing the cells with 13C glucose and tracking 13C label incorporation from glucose into TCA cycle intermediates. This experiment first required substantial standardization, for the time of cell collection and quenching post 13C glucose addition, by measuring the kinetics of 13C incorporation into TCA cycle intermediates at different time points after 13C glucose addition. The standardization of this method is now included in the revised manuscript as Figure 2 – figure supplement 2 C, along with associated text, and is shown below for reference.

Author response image 6.

Kinetics of 13C labelling in TCA cycle intermediates

Actual TCA cycle flux results: For measuring the TCA cycle flux, cells were treated with 1% 13C glucose, quenched and samples were collected at 7 mins post glucose addition which is in the linear range of 13C label incorporation (Figure 2- Figure 2 – figure supplement 2 C).

Result: We did not observe any significant changes in the relative 13C label incorporation in TCA cycle intermediates. This data is included in the revised manuscript as Figure 2 – figure supplement 2 D, along with associated text, and is below for your reference.

Author response image 7.

TCA cycle flux

What these data show is that the TCA cycle flux itself is not altered in ubp3Δ. A likely interpretation of this data is that this is due to the increase in the pyruvate transport to mitochondria in ubp3Δ cells, as indicated by the ~10-fold increase in Mpc3 (mitochondrial pyruvate transporter) protein levels (shown in Figure 5-figure supplement 5H), allowing the net same amount of pyruvate into the mitochondria. This increased mitochondrial pyruvate transport could support maintaining the TCA flux in ubp3Δ cells, and supporting the increased respiration. Putting a hierarchy together, the increased respiration in ubp3Δ cells could therefore be primarily due to increased Pi transport, followed by a consequent increase in ETC proteins. We leave it to the readers of this study to make this conclusion.

We hope that we have addressed all concerns that the reviewer has with respect to TCA cycle flux in ubp3Δ cells.

(3) In addition, some of the important literature was also missed in citation and discussion. For example, in a recent study (Ouyang et al., 2022), it was reported that phosphate starvation increases mitochondrial membrane potential independent of respiration in yeast and mammalian cells, and some of the conflicting results were presented in this study.

We are very aware of the recent study by Ouyang et al, which reports that Pi starvation increases mitochondrial membrane potential independent of respiration. However, this study is distinct from the context of our case due to the reasons listed below.

(a) The reviewer may have misinterpreted our low Pi condition as Pi starvation. There is no Pi ‘starvation’ in this study. Here, we cultured ubp3Δ and tdh2Δtdh3Δ cells in a low Pi medium with 1 mM Pi concentration in order to bring down the intracellular free Pi to that of WT levels. These cells are therefore not Pi-starved, but have been manipulated to have the same intracellular Pi levels as that of WT cells, as shown in Figure 4-figure supplement 1D. The Pi concentration in the medium is still in the millimolar range, and the cells are grown in this medium for a short time (~4 hrs) till they reach OD600 ~ 0.8. This is entirely different from the conditions used in Ouyang et al., 2022, where the cells were grown in a Pi-starvation condition with 1-100 micromolar Pi in the medium for a time duration of 6-8 hrs. Since cells respond differentially to changes in Pi concentrations over time (Vardi et al., 2014), the response to low Pi vs Pi starvation will be completely different.

(b) In our study, mitochondrial membrane potential is used as only one of the readouts for mitochondrial activity. Our estimations of mitochondrial respiration are established by including other measurements such as Cox2 protein levels (as an indicator of the ETC) and basal OCR measurements (measuring respiration), all of which provide distinct information. The mitochondrial membrane potential can be regulated independent of mitochondrial respiration state (Liu et al., 2021), using membrane potential alone as a readout to estimate mitochondrial respiration can therefore be limiting in the information it provides. As indicated earlier, mitochondrial membrane potential can change, independent of mitochondrial respiration (Ouyang et al., 2022) and ATP synthesis (Liu et al., 2021). Since the focus of our study is mitochondrial respiration, and not just the change in membrane potential, making conclusions based on potential alone are ambiguous. Most studies in the field have in fact not used the comprehensive array of distinct estimates that we use in this study, and we believe the standards set in this study should become a norm for the field.

(c) The only mutant that is similar to the Ouyang et al study is the Mir1 deletion mutant, which results in acute Pi starvation in mitochondria. In this strain, we find an increase in mitochondrial membrane potential. The data is not included in the manuscript but is shown below.

Author response image 8.

Mitochondrial potential in WT vs mir1Δ

As clear from this data, mitochondrial membrane potential is significantly high in mir1Δ cells. However, the basal OCR and Cox2 protein levels clearly show decreased mitochondrial respiration which is expected in this mutant (Figure 5 A,B). This in fact highlights the limitations of solely relying on mitochondrial membrane potential measurements to draw conclusions, as doing so will lead to a misinterpretation of the actual mitochondrial activity in these cells. We do not wish to highlight limitations in other studies, but hope we make our point clear.

(4) An additional experiment with strains lacking mitochondrial DNA under phosphate-rich and restricted conditions would further strengthen the result.

Strains lacking mitochondrial DNA (Rho0 cells) cannot express the mitochondrially encoded ETC subunit proteins. These strains are therefore incapable of performing mitochondrial respiration. Since Rho0 cells are known to utilize alternate mechanisms to maintain their mitochondrial membrane potential (Liu et al., 2021), using mitotracker fluorescence as a readout of mitochondrial respiration in these strains under different Pi conditions is inconclusive and misleading due to the reasons mentioned in point number 3(b and c). However, since this was a concern raised by the reviewer, we now measured basal OCR in WT and Rho0 strains with Ubp3 deletion under normal vs low Pi medium. As expected, Rho0 cells show extremely low basal OCR values, an entire order of magnitude lower than WT cells. At these very low (barely detectable) levels the deletion of Ubp3 or change in Pi concentration in the medium does not change basal OCR, since these strains are not capable of respiration. We have included this data as Figure 4-figure supplement 1G.

Author response image 9.

Basal OCR in Rho0 cells

(5) Western blot control panels should include entire membrane exposure, and non-cut western blots should be submitted as supplementary.

The non-cut western blot images and the loading controls are now included in the revised manuscript as a supplementary file 2.

(6) In Figure 4, it is shown that Pi addition decreases basal OCR to the WT level. However, the Cox2 level remains significantly higher. This data is confusing as to whether mitochondrial Pi directly regulates respiration or not.

As described in the previous point, the Cox2 levels and the OCR provide distinct pieces of information. In figure 4, we show that culturing ubp3Δ in low Pi significantly decreases both Cox2 protein levels and basal OCR. Since Cox2 protein levels and basal OCR are different readouts for mitochondrial activity, there could be differences in the extent by which Pi availability controls each of these factors. Basal OCR is a direct readout for mitochondrial respiration, and is regulated by multiple factors including ETC protein levels, rate of ATP synthesis, rate of Pi transport etc. In figure 4, we find that culturing ubp3Δ in low Pi decreases basal OCR to WT level. This strongly suggests that high Pi levels are necessary to increase basal OCR in ubp3Δ.

(7) Representative images of Ubx3 KO and wild-type strains stained with CMXRos are missing.

Thank you for noticing this. This data is now included in the revised manuscript as Figure 1figure supplement 1C.

Author response image 10.

(8) Overall, mitochondrial copy number and mtDNA copy number should be analyzed in WT and Ubo3 KO cells as well as Pi-treated and non-treated cells, and basal OCR data should be normalized accordingly. The reported normalization against OD is not appropriate.

This is a valid concern raised by the reviewer, and something we had extensively considered during the study. To normalize the total mitochondrial amounts in each strain, we always measure the protein levels of the mitochondrial outer membrane protein Tom70. While we had described this in the methods, it may not have been obvious in the text. But this information is included in Figure 1-figure supplement 1G. We did not observe any significant change in Tom70 levels, suggesting that the total mitochondrial amount does not change in ubp3Δ, and we have noted this in the manuscript (results section relevant to Figure 1). As an additional control, to directly measure the mitochondrial amount in these conditions, we have now measured the mitochondrial volume in ubp3Δ cells and WT cells treated with Pi. For this, we used a strain which encodes mitochondria targeted with mNeon green protein (described in Dua et al., JCB, 2023), and which can therefore independently assess total mitochondrial amount. We do not observe any changes in mitochondrial volume or amounts in ubp3Δ cells or WT+Pi, compared to that of WT cells. Therefore, the change in mitochondrial respiration in Ubp3 deletion and Pi addition are not due to changes in total amounts of mitochondria in these conditions. Given all these, the normalization of basal OCR using total cell number is therefore the most appropriate way for normalization. This is also conventionally used for basal OCR normalization in multiple studies.

We have now included these additional data on mitochondrial volumes and amounts in the revised manuscript as Figure1-figure supplement 1F and Figure5-figure supplement 1D, and associated text, and is shown below.

Author response image 11.

Mitochondrial volume in WT vs ubp3Δ cells

Author response image 12.

Mitochondrial volume in WT and WT+Pi

These data collectively address the reviewer’s concerns regarding changes in mitochondrial amounts in all the conditions and strains used in this study.

Reviewer #2 (Public Review):

Summary:

Cells cultured in high glucose tend to repress mitochondrial biogenesis and activity, a prevailing phenotype type called Crabree effect that is observed in different cell types and cancer. Many signaling pathways have been put forward to explain this effect. Vengayil et al proposed a new mechanism involved in Ubp3/Ubp10 and phosphate that controls the glucose repression of mitochondria. The central hypothesis is that ∆ubp3 shifts the glycolysis to trehalose synthesis, therefore leading to the increase of Pi availability in the cytosol, then mitochondria receive more Pi, and therefore the glucose repression is reduced.

Strengths:

The strength is that the authors used an array of different assays to test their hypothesis. Most assays were well-designed and controlled.

Weaknesses:

I think the main conclusions are not strongly supported by the current dataset.

(1) Although the authors discovered ∆ubp3 cells have higher Pi and mitochondrial activity than WT in high glucose, it is not known if WT cultured in different glucose concentration also change Pi that correlate with the mitochondrial activity. The focus of the research on ∆ubp3 is somewhat artificial because ∆ubp3 not only affects glycolysis and mitochondria, but many other cellular pathways are also changed. There is no idea whether culturing cells in low glucose, which derepress the mitochondrial activity, involves Ubp3 or not. Similarly, the shift of glycolysis to trehalose synthesis is also not relevant to the WT cells cultured in a low-glucose situation. “The focus of the research on ∆ubp3 is somewhat artificial because ∆ubp3 not only affects glycolysis and mitochondria, but many other cellular pathways are also changed. There is no idea whether culturing cells in low glucose, which de-repress the mitochondrial activity, involves Ubp3 or not.”

We would like to clarify that the focus of this research is not on Ubp3, or to address mechanistic aspects of how Ubp3 regulates mitochondrial activity, or to identify the targets of Ubp3. That would be an entirely distinct study, with a very different approach.

In this study, while carrying out a screen, we serendipitously found that ubp3Δ cells showed an increase in mitochondrial activity in high glucose. Subsequently, we used this observation, bolstered by diverse orthogonal approaches, to identify a general, systems-level principle that governs mitochondrial repression in high glucose. Through this, we identify a role of phosphate budgeting as a controller of mitochondrial repression in high glucose. In this study, our entire focus has been to use orthogonal approaches, as well as parsimonious interpretations, to establish this new hypothesis as a possibility. We hope this idea, supported by these data, will now enable researchers to pursue other experiments to establish the generality of this phenomenon.

We have not focused our effort in identifying the role of Ubp3, or its regulation upon changes in glucose concentration in this context. That is a very specific, and separate effort, and misses the general point we address here. It is entirely possible that Ubp3 might also regulate mitochondrial activity by additional mechanisms other than mitochondrial Pi availability (such as via the reduction of key glycolytic enzymes at nodes of glycolysis, resulting in reduced glycolytic flux and rerouted glucose metabolism). Had the goal been to identify Ubp3 substrates, it is very likely that we would not have found the role of Pi homeostasis in controlling mitochondrial respiration. This is particularly because the loss of Ubp3 does not result in an acute disruption of glycolysis, unlike say a glycolytic enzyme mutant, which would have resulted in severe effects on growth and overall metabolic state. This would have made it difficult to dissect out finer details of metabolic principles that regulate mitochondrial respiration.

In order to further corroborate our findings, we used the glycolysis defective mutant tdh2Δtdh3Δ cells, where we find a similar change in Pi balance. This complements the key observations made using ubp3Δ cells. Distinctly, we utilized the glycolytic inhibitor 2DG to independently assess the role of mitochondrial Pi transport in regulating respiration. Together, in this study we do not just relying on genetic mutants, but combine the Ubp3 deletion strain with a reduced GAPDH activity strain, and pharmacologic inhibition of glycolysis. Distinctly, we find that mitochondrial Pi transporter levels are repressed under high glucose (Figure 5C, Figure 5-figure supplement 1B). Further, we find that mitochondrial Pi transport is important in increasing mitochondrial respiration upon shift to low glucose and glycolytic inhibition by 2-DG. Therefore, we collectively unravel a more systems level principle that regulates glucose mediated mitochondrial repression, as opposed to a mechanistic study of Ubp3 targets.

Of course, given the conservation of Ubp3, we are very excited to pursue a mechanistic study of Ubp3 targets in future. This is a general challenge for deubiquitinase enzymes, and till date there are very few bona fide substrates known for any deubiquitinase enzyme, from any cellular system (due to challenges in the field that we discuss separately, and have included in the discussion section of this text).

“Similarly, the shift of glycolysis to trehalose synthesis is also not relevant to the WT cells cultured in a low-glucose situation”

The reviewer is correct in pointing out that in low-glucose, the shift to trehalose synthesis might not be as relevant. We observe that the glycolysis defective mutant tdh2Δtdh3Δ cells does not show an increase in trehalose synthesis (Figure 3-figure supplement 1E). However, in this context, the decrease in the rate of GAPDH catalysed reaction alone appears to be sufficient to increase the Pi levels (Figure 3F) even without an increase in trehalose. Therefore, there might be differences in the relative contributions of these two arms towards Pi balance, based on whether it is low glucose in the environment, or a mutant such as ubp3 that modulates glycolytic flux. In ubp3Δ cells, the combination of low rate of GAPDH catalyzed reaction and high trehalose will happen (based on how glycolytic flux is modulated), vs only the low rate of GAPDH catalyzed reaction in tdh2Δtdh3Δ cells. As an end point the increase in Pi happens in both cases, but with slightly differing outcomes. It is also to be noted that in terms of free Pi sources a low-glucose condition (with low glycolytic rate) is very different from a no-glucose, respiratory condition (where cells perform very high gluconeogenesis). In high respiration conditions such as ethanol, cells switch to high gluconeogenesis, where there is a huge increase trehalose synthesis as a default (eg see Varahan et al 2019). In this condition, trehalose synthesis could be a major source for Pi (eg see Gupta 2021), and could support the increased mitochondrial respiration. In an ethanol medium, the directionality of GAPDH reaction is reversed. Therefore, this reaction will also now become an added source of Pi, instead of a consumer of Pi (see illustration in Figure 3G). Therefore, a reasonable interpretation is that a combination of increased trehalose and increased 1,3 BPG to G3P conversion can be a major Pi source to increasing mitochondrial respiration in a non-glucose, respiratory medium.

“it is not known if WT cultured in different glucose concentration also change Pi that correlate with the mitochondrial activity”

This is valid point raised by the reviewer. We have already found that the protein levels of mitochondrial Pi transporter is increased in a non-glucose respiratory (ethanol) medium and a low (0.1%) glucose medium (see Figure 5C, Figure5-figure supplement 1B). In addition, we have tried measuring mitochondrial Pi levels in cells grown in a high glucose medium vs a respiratory, ethanol medium. The results are shown below for the reviewer’s reference. Reviewer response image 3 – Mitochondrial Pi levels in ethanol vs glucose

Author response image 13.

We observe a clear trend where mitochondrial Pi levels are high in cells grown in ethanol medium compared to that of cells grown in high glucose. However, the estimation of Pi, and normalising the Pi levels in isolated mitochondria is extremely difficult in this condition (note that this has never been done before). This is likely due to a rapid rate of conversion of ADP and Pi to ATP (in ethanol) which increases the variation in the estimation of steady state Pi levels, and the high amounts of mitochondria in ethanol grown cells. Since the date shows high variation, we have not included this data in the manuscript, but we are happy to include it here in the response.

Indeed, this study opens up the exciting question of addressing how intracellular Pi allocation is regulated in different conditions of glucose. This can be further extended to Crabtree negative strains such as K. lactis which do not show mitochondrial repression in high glucose. All of these are rich future research programs.

(2) The central hypothesis that Pi is the key constraint behind the glucose repression of mitochondrial biogenesis/activity is supported by the data that limiting Pi will suppress mitochondrial activity increase in these conditions (e.g., ∆ubp3). However, increasing the Pi supply failed to increase mitochondrial activity. The explanation put forward by the authors is that increased Pi supply will increase glycolysis activity, and somehow even reduce the mitochondrial Pi. I cannot understand why only the increased Pi supply in ∆ubp3, but not the increased Pi by medium supplement, can increase mitochondrial activity. The authors said "...that ubp3Δ do not increase mitochondrial Pi by merely increasing the Pi transporters, but rather by increasing available Pi pools". They showed that ∆ubp3 mitochondria had higher Pi but WT cells with medium Pi supplement showed lower Pi, it is hard to understand why the same Pi increase in the cytosol had a different outcome in mitochondrial Pi. Later on, they showed that the isolated mito exposed to higher Pi showed increased activity, so why can't increased Pi in intact cells increase mito activity? Moreover, they first showed that ∆ubp3 had a Mir1 increase in Fig3A, then showed no changes in FigS4G. It is very confusing.

“I cannot understand why only the increased Pi supply in ∆ubp3, but not the increased Pi by medium supplement, can increase mitochondrial activity.”

This is an interesting point, that requires a nuanced explanation, which we try to provide below.

For mitochondrial respiration to increase in the presence of high Pi, the cytosolic Pi has to be transported to the mitochondria sufficiently. In ubp3Δ the increased free Pi (as a consequence of rewired glycolysis) is transported to the mitochondria (Figure 4). This increased mitochondrial Pi can therefore increase mitochondrial respiration in ubp3Δ.

In case of WT+Pi, the externally supplemented Pi cannot further enter mitochondria (as shown in Figure 5-Figure supplement 1C) and is most likely restricted to the cytosol. Because of this inability of the Pi to access mitochondria, the mitochondrial respiration does not increase in WT+Pi (Figure 5-Figure supplement 1E).

The likely reason for this difference in mitochondrial Pi transport in ubp3Δ vs WT+Pi is the relative difference in their glycolytic rate. The glycolytic rate is inherently decreased in ubp3Δ, but not in WT+Pi. To dissect this possibility of glycolytic rate itself contributing to the Pi availability in the mitochondria, we inhibited glycolysis in WT cells (using 2DG), and then supplemented Pi. Compared to cells in the same glucose condition (with 2DG, but without supplementing excess Pi), now the WT+Pi (+2DG) has higher mitochondrial respiration (Figure 5-Figure supplement 1F). This suggests that a combination of low glycolysis and high Pi is required for increasing mitochondrial respiration (as elaborated in the discussion section of the manuscript).

An obvious question that arises out of this observation is how does the change in glycolytic rate regulate mitochondrial Pi transport. One consequence of altering the glycolytic rate is a change in cytosolic pH. This itself will bear on the extent of Pi transport into mitochondria, as discussed in detail below.

In mitochondria, Pi is co-transported along with protons. Therefore, changes in cytosolic pH (which changes the proton gradient) can control the mitochondrial Pi transport (Hamel et al., 2004). Glycolytic rate is a major factor that controls cytosolic pH. The cytosolic pH in highly glycolytic cells is ~7, and decreasing glycolysis results in cytosolic acidification (Orij et al., 2011). Therefore, under conditions of decreased glycolysis (such as loss of Ubp3), cytosolic pH becomes acidic. Since mitochondrial Pi transport is dependent on the proton gradient, a low cytosolic pH would favour mitochondrial Pi transport. Therefore, under conditions of decreased glycolysis (2DG treatment, or loss of Ubp3), where cytosolic pH would be acidic, increasing cytosolic Pi might indirectly increase mitochondria Pi transport, thereby leading to increased respiration.

To explain this and integrate all these points, we have extended a discussion section in this manuscript. We include this section below:

“Supplementing Pi under conditions of low glycolysis (where mitochondrial Pi transport is enhanced), as well as directly supplementing Pi to isolated mitochondria, increases respiration (Figure 5, Figure 5-figure supplement 1). Therefore, in order to derepress mitochondria, a combination of increased Pi along with decreased glycolysis is required. An additional systems-level phenomenon that might regulate Pi transport to the mitochondria is the decrease in cytosolic pH upon decreased glycolysis (60, 61). The cytosolic pH in highly glycolytic cells is ~7, and decreasing glycolysis results in cytosolic acidification (60, 61). Therefore, under conditions of decreased glycolysis (2DG treatment, deletion of Ubp3, and decreased GAPDH activity), cytosolic pH becomes acidic. Since mitochondrial Pi transport itself is dependent on the proton gradient, a low cytosolic pH would favour mitochondrial Pi transport (62). Therefore, under conditions of decreased glycolysis (2DG treatment, or loss of Ubp3, or decreased GAPDH activity), where cytosolic pH would be acidic, increasing cytosolic Pi might indirectly increase mitochondria Pi transport, thereby leading to increased respiration. Alternately, increasing mitochondrial Pi transporter amounts can achieve the same result, as seen by overexpressing Mir1 (Figure 5).”

This possibility of changes in cytosolic pH regulating mitochondrial Pi transport and thereby respiration is a really interesting future research question, and an idea that has not yet been explored till date. This can stimulate new lines of thinking towards finding conserved biochemical principles that control mitochondrial repression in high glucose.

“Moreover, they first showed that ∆ubp3 had a Mir1 increase in Fig3A, then showed no changes in FigS4G. It is very confusing”

increase in Mir1 in ubp3Δ shown in figure 3A comes from the analysis of the proteomics dataset from a previous study (Isasa et al., 2015). Subsequently, we more systematically experimentally assessed Mir1 levels directly, and did not observe an increase in Mir1 (Figure 4figure supplement 1H in revised manuscript). It is entirely possible that in a large-scale study (as in Isasa 2015), some specific proteomic targets might not fully reproduce when tested very specifically (as is described in Handler et al., 2018 and Mehta et al., 2022). We do clearly indicate this in the text, but given the density of information in this study, it is understandable that this point was missed by the reviewer.

(3) Given that there is no degradation difference for these glycolytic enzymes in ∆ubp3, and the authors found transcriptional level changes, suggests an alternative possibility where ∆ubp3 may signal through unknown mechanisms to parallelly regulate both mitochondrial biogenesis and glycolytic enzyme expression. The increase of trehalose synthesis usually happens in cells under proteostasis stress, so it is important to rule out whether ∆ubp3 signals these metabolic changes via proteostasis dysregulation. This echoes my first point that it is unknown whether wild-type cells use a similar mechanism as ∆ubp3 cells to regulate the glucose repression of mitochondria.

We appreciate this point raised by the reviewer, but this again requires some clarification (as made earlier). The goal of this study was to identify systems-level principles that explain mitochondrial repression in high glucose. Although we started by performing a screen to identify proteostatic regulators of mitochondrial activity in high glucose, and identified Ubp3 as a mediator of mitochondrial activity, our approach was to use ubp3Δ cells as a model to understand the metabolic principles that regulate mitochondrial repression. This has been reiterated repeatedly in the manuscript – for example lines 123-124 “We therefore decided to use ubp3Δ cells to start delineating requirements for glucose-mediated mitochondrial repression.” and again in the discussion section – lines 442-460, where we discuss some unique advantages of using ubp3Δ cells to understand a general basis of mitochondrial regulation. To test this hypothesis, we also used orthogonal approaches, as well as other mutants and conditions with defective glycolysis, such as tdh2Δtdh3Δ cells and 2DG treatments. Only with these multiple converging evidences do we infer that there might be a role of the change in Pi balance (due to changes in glycolytic rate) in regulating mitochondrial activity.

We certainly agree that there is great value in identifying the mechanistic details of how Ubp3 regulates mitochondria. But this requires very distinct approaches not pursued in this study. This is not the question that we are addressing in this story. Separately, identifying targets of DUBs is one of the exceptional challenges in biology, since there are currently no straightforward chemicalbiology approaches to do so for this class of proteins. Unlike kinase/phosphatase systems, or even ubiquitin ligases, substrate trapping mutants etc have proven to be abject failures in identifying direct targets of DUBs. A quantitative proteomics study might suggest some proteins/cellular processes regulated by Ubp3. This has been attempted for several DUBs, but rarely have any direct substrates of DUBs every been identified, in any system. A high quality quantitative, descriptive proteome dataset of ubp3Δ cells is already available from a previous study (Isasa et al., 2015), which we cite extensively in this manuscript, and indeed was invaluable for this study. We cannot improve the outstanding quality dataset already available. Interestingly, the findings of this study actually help substantiate our idea of an increased mitochondrial activity and change in Pi homeostasis in ubp3Δ cells. The Isasa et al dataset finds proteins involved in mitochondrial respiration that are high in ubp3Δ cells, and the glycolytic enzymes and PHO regulon proteins are reduced. In our study, using these data references, we were able to conceptually piece together how changes in glycolytic flux can alter Pi balance.

Apart from identifying changes in protein levels, a separate challenge in making sense of this quantitative proteomics data is the difficulty in pinpointing any target of Ubp3 that specifically regulates these processes. A single DUB can have multiple substrates, and this could regulate the cellular metabolic state in a combinatorial manner. This is the essence of all signaling regulators in how they function, and it is therefore important to understand what their systems-level regulation of cell states are (separate from their specific individual substrates). Therefore, identifying the specific target of Ubp3 responsible for this metabolic rewiring can be very challenging. These experiments are well beyond the scope or interest of the current manuscript.

If we had pursued that road in this study, we would not have made any general findings related to Pi balance, nor would this more general hypothesis have emerged.

(4) Other major concerns:

(a) The authors selectively showed a few proteins in their manuscript to support their conclusion. For example, only Cox2 and Tom70 were used to illustrate mitochondrial biogenesis difference in line 97. Later on, they re-analyzed the previous MS dataset from Isasa et al 2015 and showed a few proteins in Fig3A to support their conclusion that ∆ubp3 increases mitochondrial OXPHOS proteins. However, I checked that MS dataset myself and saw that many key OXPHOS proteins do not change, for example, both ATP1 and ATP2 do not change, which encode the alpha and beta subunits of F1 ATPase. They selectively reported the proteins' change in the direction along with their hypothesis.

To clarify, we observe an increase in Cox2 protein levels but not in Tom70 levels which suggests that there is no increase in mitochondrial biogenesis. The increase is specific to some respiration related mitochondrial proteins such as Cox2 (Figure 1E, Figure 3A). We have clearly pointed out this in the manuscript. We used Cox2 protein levels as an additional readout for ETC activity, to validate our observations coming from the potentiometric mitotracker readouts, and basal oxygen consumption rate (OCR) measurements. This was for 3 reasons: Cox2 is a mitochondrial genome encoded subunit of the complex IV (cytochrome c oxidase) in the ETC, and has a redox centre critical for the cytochrome c oxidase activity. The biogenesis and assembly of complex IV subunits have been studied with respect to multiple conditions such as glucose availability and hypoxia and the expression and stability of the mitochondrial encoded complex IV subunits are exceptionally well correlated to changes in mitochondrial respiration (Fontanesi et al., 2006). Cox2 is very well characterised in S. cerevisiae, and the commercially available Cox2 antibodies are outstanding, which makes estimating Cox2 levels by western blotting unambiguous and reproducible.

We re-analyzed the proteomic dataset from Isasa et al to find out additional information regarding the key nodes that are differentially regulated in ubp3Δ. We have not claimed at any point in the manuscript that all OXPHOS related proteins are upregulated in ubp3Δ, nor is there any need for that to be so. We identified Ubp3 from our screen, observed an increase in mitochondrial potential, basal OCR, and Cox2 levels. We later found out that the proteomic data set for ubp3Δ also supports our observations that mitochondrial respiration is upregulated in ubp3Δ. The reviewer points out that we “showed a few proteins in Fig3A to support their conclusion that ∆ubp3 increases mitochondrial OXPHOS proteins”. Our conclusion is that the deletion of Ubp3 increases mitochondrial respiration. The combined readouts which we used to reach this conclusion (OCR, mitochondrial potential, mitochondrial ATP production, Cox2 levels) are far more direct, comprehensive and conclusive than showing an increase in a few proteins related to OXPHOS, as also explained earlier toward a distinct reviewer query. Since different mitochondrial proteins are regulated by different mechanisms, we need not see an increase in all the OXPHOS proteins in a mutant like ubp3Δ where mitochondrial respiration is high. An increase in some key proteins would be sufficient to increase the respiration as seen in our case.

To summarise, the proteomic dataset supports our observation, but our conclusions are not dependent on the increase in OXPHOS proteins observed in the dataset.

(b) The authors said they deleted ETC component Cox2 in line 111. I checked their method and table S1, I cannot figure out how they selectively deleted COX2 from mtDNA. This must be a mistake.

Yes, we understand that for mitochondrially encoded proteins, a simple knock-out strategy has limitations. However, we first tried to generate the Cox2 deletion mutant by a standard PCR mediated gene deletion strategy (Longtine 1998), with the optimistic assumption that even if all Cox2 is not lost, a substantial fraction of the Cox2 genes would be lost via recombination. We selected the transformants after strong antibiotic selection, and then we measured the Cox2 protein levels. Gratifyingly, we found that the mutant strain had substantially decreased Cox2 protein levels (but not a complete loss), and this was retained across generations. The data is shown below.

Author response image 14.

Cox2 levels in WT vs Cox2 mutants

Since the mutants have decreased Cox2 levels, we went ahead and performed growth assays using this strain, in a WT or Ubp3 deletion background. Deletion of Ubp3 in the Cox2 mutant resulted in a more severe growth defect.

However, we fully agree that this strain is not a complete Cox2 knockout, and it is possible that the decrease in Cox2 is due to modifications in some other unelated gene. In the text, we should also not have named this cox2Δ. Since we are not sure of the exact genetic modification in this mutant, we have removed this data from the revised manuscript.

Instead, we have now repeated all experiments, utilizing a fully characterised Cox2 mutant -cox262, described in (5) which has defective respiration. In this revised version, we find that deletion of Ubp3 in this strain retains the originally observed severe growth defect in glucose. This is consistent with our conclusion that a functional mitochondria is required for proper growth in ubp3Δ mutant. To separately validate this conclusion, we also utilized a Rho0 strain which does not have mitochondrial DNA and thereby cannot perform mitochondrial respiration. We show that deletion of Ubp3 results in a more severe growth defect in a Rho0 strain. These results are included in the revised manuscript as figure 1-figure supplement 1 I.

Author response image 15.

Also, we further confirmed that the Rho0 strain and Rho0 ubp3 strain is incapable of respiration, using seahorse assay. This data is included in the revised manuscript as Figure 4-figure supplement 1G.

Author response image 16.

Basal OCR in Rho0 cells

We hope that these new data address the reviewer’s concerns about the Cox2 mutant.

(c) They used sodium azide in a lot of assays to inhibit complex IV. However, this chemical is nonspecific and broadly affects many ATPases as well. Not sure why they do not use more specific inhibitors that are commonly used to assay OCR in seahorse.

We have now performed growth assays for WT and ubp3Δ cells in the presence of specific mitochondrial OXPHOS inhibitors - oligomycin and FCCP. We observe a more severe growth defect in ubp3Δ cells compared to WT cells in the presence of oligomycin and FCCP, similar to the results observed with sodium azide. All these data are now included in the revised manuscript as Figure 1I, Figure1-figure supplement 1H, along with associated text.

Author response image 17.

Growth rate in the presence of FCCP

Author response image 18.

Figure1-figure supplement 1H- Growth rate in the presence of oligomycin

We hope that these new data addresses the reviewer’s concerns.

(d) The authors measured cellular Pi level by grinding the entire cells to release Pi. However, this will lead to a mix of cytosolic and vacuolar Pi. Related to this caveat, the cytosol has ~50mM Pi, while only 1-2mM of these glycolysis metabolites, I am not sure why the reduction of several glycolysis enzymes will cause significant changes in cytosolic Pi levels and make Pi the limiting factor for mitochondrial respiration. One possibility is that the observed cytosolic Pi level changes were caused by the measurement issue mentioned above.

The Pi estimation shown in figure 3 C, E, F and G is a measure of total Pi in the cells. The vacuole is a major storehouse of phosphate in cells. However, unlike plant cells where free phosphate is stored in vacuoles, yeast vacuoles store phosphate only in the form of polyphosphates (Yang et al., 2017, Hürlimann et al., 2007). The free Pi formed from the hydrolysis of polyphosphate is subsequently transported to cytosol via the exporter Pho91 (Hürlimann et al., 2007). This therefore makes cytosol and mitochondria the major storage of usable free Pi in yeast. Since the malachite green assay that we use for phosphate estimation is specific to free Pi, and not polyphosphate, the Pi estimates that we show in figure 3 come from a combination of cytosolic and mitochondrial Pi. As explained earlier, in order to specifically measure mitochondrial Pi, we have established methods to rapidly isolate mitochondria, and then followed this by estimating Pi in these isolated mitochondria (Figure 4B). Here we clearly see a large increase in mitochondrial Pi in the Ubp3 deletion cells. This allows us to estimate the changes in Pi levels that specific to mitochondria, without relying only on total Pi changes.

“the cytosol has ~50mM Pi, while only 1-2mM of these glycolysis metabolites, I am not sure why the reduction of several glycolysis enzymes will cause significant changes in cytosolic Pi levels and make Pi the limiting factor for mitochondrial respiration”

The reviewer has completely missed the fact that the glycolytic rate in yeast is the highest known for any cell. While the steady state levels of glycolytic metabolites might be ~2 mM, the process of glycolysis is not static but is rapid and continuous. Glucose is continuously broken down and converted to pyruvate, along with the consumption of Pi and generation of ATP. This is the reason for the rapid 13C label saturation (within seconds of 13C glucose addition) in yeast cells (Figure 2-figure supplement 1F). This instantaneous label saturation makes accurate flux measurements arduous because of which we had to optimize a method for measuring glycolytic flux in yeast cells (Figure 2-D, Figure 2-figure supplement 1F). Indeed, for that reason, our measurements of glycolytic flux in yeast are the first time this is being reported in the field. This in itself is an enormously challenging experiment, and establishes a new benchmark.

In highly glycolytic cells, most of the ATP is synthesized via glycolysis and the rate of glycolysis and ATP synthesis is very high. In the reaction catalysed by GAPDH, Pi and ADP is converted to ATP. This ATP formed acts as a Pi donor to most of the Pi consuming reactions in the cells. Some of these processes such a protein translation utilizes ATP, but releases Pi and ADP and this Pi enters the cellular Pi pool. Several other reactions such as nucleotide biosynthesis, polyphosphate biosynthesis and protein phosphorylation use ATP as a Pi donor and the Pi is fixed in biomolecules. Increasing the rates of these ‘Pi sinks’ therefore can result in a decrease in Pi pools. This is a concept we have earlier tried to clarify more elaborately in (Gupta and Laxman, 2021). In fact, increasing nucleotide biosynthesis and polyphosphate synthesis has earlier been suggested to decrease available free Pi (Austin and Mayer 2020, Desfougères et al., 2016). When glycolytic flux is high, this is coupled/tuned to the consumption of Pi which will be correspondingly high due to increased ATP, nucleotide and polyphosphate synthesis. Pi levels rapidly decrease upon glucose addition, due to the continuous Pi consumption during glycolysis (Hohmann et al., 1996, Van Heerden et al., 2014 , Koobs et al., 1972). Therefore, changes in glycolytic rate due to change in glycolytic enzyme levels can result in significant changes in Pi levels due to changes in Pi consumption rate.

Our results also show that the apart from Pi levels, the glycolytic state can regulate mitochondrial Pi transport as well. This is the reason for mitochondrial Pi levels and basal OCR not increasing merely by adding Pi to cells. We show that basal OCR can be increased by adding Pi in the presence of 2DG. This regulation of mitochondrial Pi transport is a major limiting factor for mitochondrial respiration and could be mediated partly by the regulating of Mir1 levels and also by the changes in the cytosolic pH which regulates the rate of mitochondrial Pi transport. We have discussed these points in the discussion section in our manuscript.

We hope that this clarifies the reviewer’s concerns regarding how changes in glycolytic rate can regulate changes in cytosolic Pi levels.

(e) The authors used ∆mir1 and MIR1 OE to show that Pi viability in the mitochondrial matrix is important for mitochondrial activity and biogenesis. This is not surprising as Pi is a key substrate required for OXPHOS activity. I doubt the approach of adding a control to determine whether Pi has a specific regulatory function, while other OXPHOS substrates, like ADP, O2 etc do not have the same effect.

To clarify, we only used the mir1Δ cells to understand the requirement for Pi transport from cytosol to mitochondria in controlling respiration. The reviewer is correct in stating that deletion of Mir1 would reduce Pi import to mitochondria and thereby inhibit respiration. This is exactly the conclusion we suggest from this experiment as stated in the manuscript – “These data suggest that mitochondrial Pi transport (via Mir1) is critical for maintaining basal mitochondrial activity even in high glucose”. We have only used these experiments to support the idea that even though glycolysis and mitochondria are in different compartments, a change in Pi balance in one compartment (cytosol) can affect Pi levels in the other (mitochondria) since there is Pi transport between these two compartments. Since mitochondria has its own polyphosphate reserves, in the absence of these experiments with mir1Δ cells it can be imagined that mitochondria PolyP can be an additional source of Pi to support respiration, and therefore changes in cytosolic Pi may have only a minor effect on mitochondrial respiration. Our experiments with mir1Δ and Mir1-OEcells indubitably suggest that Pi transport to mitochondria from cytosol is important for respiration, and therefore changes in cytosolic Pi levels (or maintaining cytosolic Pi at a lower level due to the rate of glycolysis) will have rippling effects in mitochondrial Pi availability. Further, these data suggest that for example under glycolytic inhibition (low glucose, or 2DG), while all factors (signalling, substrate availability etc) favour respiration (and mitochondrial derepression), cells cannot unable to achieve this in the absence of ample Pi transport from cytosol. This therefore places Pi at the centre stage in controlling mitochondrial respiration.

We conclude that Pi is a major, but not the only constraint for mitochondrial respiration. There certainly could be a role for ADP, oxygen availability etc in regulating respiration. However, these are beyond the scope of our study. We have discussed about the potential role of ADP in regulating mitochondrial repression in the discussion section. “An additional consideration is the possible contribution of changes in ADP in regulating mitochondrial activity, where the use of ADP in glycolysis might limit mitochondrial ADP. Therefore, when Pi changes as a consequence of glycolysis, it could be imagined that a change in ADP balance can coincidentally occur. However, prior studies show that even though cytosolic ADP decreases in the presence of glucose, this does not limit mitochondrial ADP uptake, or decrease respiration, due to the very high affinity of the mitochondrial ADP transporter.”

We hope that this clarifies the reviewer’s concerns regarding the use of Mir1 OE and mir1Δ strains.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

Some of the experiments should be repeated in other strain backgrounds for reproducibility and rigor.

As discussed in the response to point number 1, we have now utilized multiple strains of S. cerevisiae to test our findings. We now find that our discoveries regarding the role of altered Pi budgeting as a constraint for mitochondrial respiration, and the role of Ubp3 as a regulator of mitochondrial repression are conserved across multiple genetic backgrounds of S. cerevisiae. These results are included in the revised manuscript as a new figure- Figure 6 and associated text. We used the W303, Σ1278 and BY4742 strains of S. cerevisiae to show that deletion of Ubp3 increases mitochondrial activity (as shown by increased mitochondrial membrane potential and increased Cox2 levels). Using the W303 strain we show that the deletion of Ubp3 increases Pi levels and that the increased Pi is necessary for increasing mitochondrial respiration (Figure 6C, D). These added experiments have substantially broadened the generality of our findings.

The number of biological repeats needs to be increased in all experiments.

We have increased the number of biological repeats in key experiments that shows that the increased Pi levels are necessary for the increased mitochondrial respiration in ubp3Δ and tdh2Δtdh3Δ cells (revised Figure 4F). Apart from a few basal OCR measurements and mitotracker data in supplementary figure, all our experiments are performed for 3 biological repeats. In case of basal OCR measurements, yeast cells have to be aliquoted to poly-L-lysine coated seahorse plates and centrifuged to ensure that the cells are properly settled. This is due to the non-adherent nature of yeast cells. During the centrifugation step, the wells in the two end rows cannot be utilized due to uneven settling of cells which affects the basal OCR readings in these wells. In case of several experiments that involve multiple samples, we were therefore limited to restrict the number of biological replicates to 2 (repeated independently), so that all samples could be accommodated in the plate.

Full western blot images should be supplemented along with the other data.

The complete western blot images are now included in the revised manuscript as supplementary file 2.

TCA cycle flux should be analyzed and presented in the study to conclude some of the findings.

As discussed in detail in the response to point number 2, we have performed steady state and flux measurements for TCA cycle intermediates. This data is now included as a new supplement figure- Figure 2-figure supplement 2.

Reviewer #2 (Recommendations For The Authors):

(1) In Fig. 2A, they should also include the gluconeogenesis enzymes (fructose 1,6 bisphosphatase, PEP carboxykinase, and pyruvate carboxylase) to exclude the possibility that glycolytic intermediates are not rerouted to gluconeogenesis.

We measured the protein levels of Fbp1 (fructose 1,6 bisphosphatase) and Pck1 (PEP carboxykinase). We observed an increase in the protein levels in both enzymes in ubp3Δ. The data is shown below.

Author response image 19.

Fbp1 and Pck1 protein levels

While we agree that this is an interesting observation which might help us in understanding the metabolic rewiring in ubp3Δ, we have not included this data in the current revised version of the manuscript due to two main reasons.

(1) Since ubp3Δ cells have a defective glycolysis and therefore a defective glucose repression, the mRNA and protein levels of gluconeogenic enzymes which are usually glucose-repressed might increase. This might be a response at the level of transcription and translation of these enzymes and might or might not change the rate of gluconeogenesis in these cells. This is because of multiple other factors that regulate gluconeogenic flux such as allostery, mass action etc. Therefore, to avoid confounding our main points and since we cannot make a conclusive assumption on the gluconeogenic metabolism in these mutants, we don’t include this data. The primary focus of our story is the mitochondrial repression component. Understanding the feedback controls that alter gluconeogenesis in these mutants is beyond the scope of this study and could be addressed in a separate future study.

(2) As we highlight extensively in the response letter and in the manuscript, our aim is not to understand the specific mechanistic role of Ubp3. In this manuscript, we identify the conserved constraints that control mitochondrial repression without focusing just on the role of Ubp3 in regulating this. Whether Ubp3 regulates gluconeogenesis is a question that could be addressed in a future study that focuses on identifying the altered signalling mechanisms in ubp3Δ and the targets of Ubp3.

(2) In line 292, page 10, there is a typo "dermine".

We apologize for this mistake. Corrected.

(3) In Figure 5A, is there a reason why they chose 0.1% glucose condition as a low glucose condition? Also, is there a dose-dependent change in OCR or other mitochondrial functions according to the concentration of glucose?

The glucose concentration of 0.1% was selected to decrease (but not completely remove) the available glucose. 0.1% glucose is considered as a standard low glucose condition in S. cerevisiae (Yin et al., 2003) and the effect of this glucose concentration on cellular processes has been extensively studied (Yin et al., 2003, Takeda et al., 2015 etc). <0.2% glucose is the critical threshold for activating respiratory metabolism (Takeda et al., 2015) and shifting cells to 0.1% glucose in our experiments will activate respiration, as we show in our data. However, this is very different from completely removing glucose or using an alternate carbon source such as ethanol, because this would result in full activation of gluconeogenesis. We further find that when cells are grown in ethanol, the gluconeogenic activation will also change the Pi homeostasis. This will in part be a result of the fully reversed direction of the GAPDH catalysed reaction (Figure 3G). If such a condition is used, it could lead to misinterpretations, and confound the conclusions that we make from these set of experiments where Pi homeostasis play a major role. In 0.1% glucose it has been shown that gluconeogenesis is still partly repressed (Yin et al., 2003). The pathways utilizing alternate carbon sources still remain repressed (even though to a lower extend compared to 2% glucose) in 0.1% glucose (Yin et al., 2003). We hope that this clarifies the concerns regarding the rationale behind using 0.1% glucose in our experiments.

The extent of glucose repression is dependent on the concentration of glucose. Glucose concentration >1% has been shown to activate degradation of mRNAs involved in alternate carbon utilization. Different signaling pathways involved in growth under glucose and glucose repression is regulated by glucose concentration. This is discussed in detail in Yin et al., 2002. We (Figure 5figure supplement 1A) also observe a dose dependent increase in mitochondrial membrane potential in the presence of 2DG. This also suggests that the rate of glycolysis (which could be also mediated by changes in glucose concentration) can regulate the extent of mitochondrial derepression.

eLife assessment

This valuable study informs on the regulation of metabolic flux between glycolysis and respiration in yeast, mediated by inorganic phosphate. The authors propose a novel mechanism involving Ubp3/Ubp10 that potentially mitigates the Crabtree effect, based on an array of assays that offer substantial insight into mitochondrial biogenesis under high glucose conditions. However, the evidence supporting the conclusions remains incomplete because of a limited number of replicates, particularly in protein blot analyses with insufficient normalization methods, which undermines the robustness of the findings. This work could influence the subfield significantly if the methodological weaknesses are addressed to provide more support for the proposed model.

Reviewer #1 (Public Review):

The study by Vengayil et al. presented a role for Ubp3 for mediating inorganic phosphate (Pi) compartmentalization in cytosol and mitochondria, which regulates metabolic flux between cytosolic glycolysis and mitochondrial processes. Although the exact function of increased Pi in mitochondria is not investigated, findings have valuable implications for understanding the metabolic interplay between glycolysis and respiration under glucose-rich conditions. They showed that UBP3 KO cells regulated decreased glycolytic flux by reducing the key Pi-dependent-glycolytic enzyme abundances, consequently increasing Pi compartmentalization to mitochondria. Increased mitochondria Pi increases oxygen consumption and mitochondrial membrane potential, indicative of increased oxidative phosphorylation. In conclusion, the authors reported that the Pi utilization by cytosolic glycolytic enzymes is a key process for mitochondrial repression under glucose conditions.

Comments on revised version:

This reviewer appreciates the author's responses addressing some of the concerns.

However, the concern of reproducibility and experimental methods applied to the study is still valid, particularly considering that many conclusions were drawn from western blot analysis. The authors used separate gel loading controls for western blot analysis, which is not a valid method. Considering loading and other errors/discrepancies during the transfer phase of the assay, the direct control should be analyzing the membrane after transfer or using an internal control antibody on the same membrane. None of the western blots are indicated with marker sizes, and it isn't very clear how many repeats there are and whether those repeats are biological or technical repeats.

Concern regarding citing the Ouyang et al. paper is still valid. This paper is an essential implication in phosphate metabolism and is directly related to some of the findings associated with mitochondrial function, along with conflicting results, which should be discussed in the discussion section. As a reviewer, I do not request citing any paper from the authors in general; however, considering some of the conflicting results here, citing and discussing paper from Ouyang et al. will improve the interoperation/value of their findings.

Considering these factors, the presented results do not fully support the findings.

Reviewer #2 (Public Review):

Summary:

Cells cultured in high glucose tend to repress mitochondrial biogenesis and activity, a prevailing phenotype type called Crabree effect that observed in different cell types and cancer. Many signaling pathways have been put forward to explain this effect. Vengayil et al proposed a new mechanism involved in Ubp3/Ubp10 and phosphate that controls the glucose repression of mitochondria. The central hypothesis is that ∆ubp3 shift the glycolysis to trehalose synthesis, therefore lead to the increase of Pi availability in the cytosol, then mitochondrial received more Pi and therefore the glucose repression is reduced.

Strengths:

The strength is that the authors used an array of different assays to test their hypothesis. Most assays were well designed and controlled.

Weaknesses:

I think the main conclusions are not strongly supported by the current dataset. Here are my comments on authors' response and model.

(1) The authors addressed some of my concerns related to ∆ubp3. But based on the results they observed and discussed, the ∆ubp3 redirect some glycolytic flux to gluconeogenesis while the 0.1% glucose in WT does not. Similarly, the shift of glycolysis to trehalose synthesis is also not relevant to the WT cells cultured in low glucose situation. This should be discussed in the manuscript to make sure readers are not misled to think ∆ubp3 mimic low glucose. It is likely that ∆ubp3 induce proteostasis stress, which is known to activate respiration and trehalose synthesis.

(2) Pi flux: it is known that vacuole can compensate the reduction of Pi in the cytosol. The paper they cited in the response, especially the Van Heerden et al., 2014 showed that the pulse addition of glucose caused transient Pi reduction and then it came back to normal level after 10min or so. If the authors mean the transient change of glycolysis and respiration, they should point that out clearly in the abstract and introduction. If the authors are trying to put out a general model, then the model must be reconsidered.

The cytosol has ~50mM Pi (van Eunen et al., 2010 FEBSJ), while only 1-2mM of glycolysis metabolites, not sure why partial reduction of several glycolysis enzymes will cause significant changes in cytosolic Pi level and make Pi the limiting factor for mitochondrial respiration. In response to this comment, the authors explained the metabolic flux that the rapid, continuous glycolysis will drain the Pi pool even each glycolytic metabolite is only 1-2mM. However, the metabolic flux both consume and release Pi, that's why there is such measurement of overall free Pi concentration amid the active metabolism. One possibility is that the observed cytosolic Pi level changes was caused by the measurement fluctuation, as they showed in "Reviewer response image 3".

Importantly, the authors measured Pi inside mito for ethanol and glucose, but not the cytosolic Pi, which is the key hypothesis in their model. The model here is that the glycolysis competes with mito for free cytosolic Pi, so it needs to inhibit glycolysis to free up cytosolic Pi for mitochondrial import to increase respiration. I don't see measurement of cytosolic Pi upon different conditions, only the total Pi or mito Pi. The fact is that in Fig.3C they saw WT+Pi in the medium increase total free Pi more than the ∆ubc3, while WT decrease mito Pi compared to WT control and ∆ubc3 and therefore decrease basal OCR upon Pi supplement. A simple math of Pitotal = Pi cyto + Pi mito tells us that if WT has more Pitotal (Fig.3C) but less Pi mito (fig.5 supp 1C), then it has higher Pi cyto. This is contradictory to what the authors tried to rationalize. Furthermore, as I pointed out previously, the isolated mitochondria can import more Pi when supplemented, so if there is indeed higher Picyto, then the mito in WT should import more Pi. So, to address these contradictory points, the authors must measure Pi in the cytosol, which is a critical experiment not done for their model. For example, they hypothesized that adding 2-DG, or ∆ubp3, suppress glycolysis and thus increase the supply of cytosolic Pi for mito to import, but no cytosolic Pi was measured (need absolute value, not the relative fold changes). It is also important to specific how the experiments are done, was the measurement done shortly after adding 2-DG. Given that the cells response to glucose changes/pulses differently in transient vs stable state, the authors are encouraged to specify that.

The most likely model to me is that, which is also the consensus in the field, is that no matter 2-DG or ∆ubp3, the cells re-wiring metabolism in both cytosol and mitochondria, and it is the total network shift that cause the mitochondrial respiration increase, which requires the increase of mito import of Pi, ADP, O2, and substrates, but not caused/controlled by the Pi that singled out by the authors in their model.

(3) The explanation that cytosolic pH reduction upon glucose depletion/2DG is a mistake. There are a lot of data in the literature showing the opposite. If the authors do think this is true, then need to show the data. Again, it is important to distinguish transient vs stable state for pH changes.

Reviewer #1 (Public Review):

Summary:<br /> In this manuscript, Fister et. al. investigate how amputational and burn wounds affect sensory axonal damage and regeneration in a zebrafish model system. The authors discovered that burn injury results in increased peripheral axon damage and impaired regeneration. Convincing experiments show altered axonal morphology and increased Ca2+ fluxes as a result of burn damage. Further experimental proof supports that early removal of the burnt tissue by amputation rescues axonal damage. Burn damage was also shown to markedly increase keratinocyte migration and increase localized ROS production as measured by the dye Pfbsf. These responses could be inhibited by Arp 2/3 inhibition and isotonic treatment.

Strengths:<br /> The authors use state-of-the-art methods to study and compare transection and burn-induced tissue damage. Multiple experimental approaches (morphology, Ca2+ fluxing, cell membrane labeling) confirm axonal damage and impaired regeneration time. Furthermore, the results are also accompanied by functional response tests of touch sensitivity. This is the first study to extend the role of tissue-damage-related osmotic exposure beyond wound closure and leukocyte migration to a novel layer of pathology: axonal damage and regeneration.

Weaknesses:<br /> The conclusions of the paper claiming a link between burn-induced epithelial cell migration, spatial redox signaling, and sensory axon regeneration are mainly based on correlative observations. Arp 2/3 inhibition impairs cell migration but has no significant effect on axon regeneration and restoration of touch sensitivity.

Pharmacological or genetic approaches should be used to prove the role of ROS production by directly targeting the known H2O2 source in the system: DUOX.

While the authors provide clear and compelling proof that osmotic responses lie at the heart of the burn-induced axonal damage responses, they did not consider the option of further exploring any biology related to osmotic cell swelling. Could osmotic ATP release maybe play a role through excitotoxicity? Could cPLA2 activation-dependent eicosanoid production relate to the process? Pharmacological tests using purinergic receptor inhibition or blockage of eicosanoid production could answer these questions.

The authors provide elegant experiments showing that early removal of the burnt tissue can rescue damage-induced axonal damage, which could also be interpreted in an osmotic manner: tail fin transections could close faster than burn wounds, allowing for lower hypotonic exposure time. Axonal damage and slow regeneration in tail fin burn wounds could be a direct consequence of extended exposure time to hypotonic water.

eLife assessment

This important study identifies a novel link between the early keratinocyte response to wounds and the subsequent regenerative capacity of local sensory neurons. The evidence supporting the claims of the authors is solid, although the inclusion of additional pharmacological and genetic manipulations might have strengthened the mechanistic aspects. The work will be of interest to cell and developmental biologists interested in tissue regeneration and cell interactions in a broader context.

Reviewer #2 (Public Review):

This is an interesting study in which the authors show that a thermal injury leads to extensive sensory axon damage and impaired regrowth compared to a mechanical transection injury. This correlates with increased keratinocyte migration. That migration is inhibited by CK666 drug treatment and isotonic medium. Both restrict ROS signalling to the wound edge. In addition, the isotonic medium also rescues the regrowth of sensory axons and recovery of sensory function. The findings may have implications for understanding non-optimal re-innervation of burn wounds in mammals.

The interpretation of results is generally cautious and controls are robust.

Here are some suggestions for additional discussion:<br /> The study compares burn injury which produces a diffuse injury to a mechanical cut injury which produces focal damage. It would help the reader to give a definition of wound edge in the burn situation. Is the thermally injured tissue completely dead and is resorbed or do axons have to grow into damaged tissue? The two-cut model suggests the latter. Also giving timescales would help, e.g. when do axons grow in relation to keratinocyte movement? An introductory cartoon might help.

Could treatment with CK666 or isotonic solution influence sensory axons directly, or through other non-keratinocyte cell types, such as immune cells?

Reviewer #3 (Public Review):

Fister and colleagues use regeneration of the larval zebrafish caudal fin to compare the effects of two modes of tissue damage-transection and burn-on cutaneous sensory axon regeneration. The authors found that restoration of sensory axon density and function is delayed following burn injury compared to transection.

The authors hypothesized that thermal injury triggers signals within the wound microenvironment that impair sensory neuron regeneration. The authors identify differences in the responses of epithelial keratinocytes to the two modes of injury: keratinocytes migrate in response to burn but not transection. Inhibiting keratinocyte migration with the small-molecule inhibitor of Arp2/3 (CK666) resulted in decreased production of reactive oxygen species (ROS) at early, but not late, time points. Preventing keratinocyte migration by wounding in isotonic media resulted in increased sensory function 24 hours after burn.

Strengths of the study include the beautiful imaging and rigorous statistical approaches used by the authors. The ability to assess both axon density and axon function during regeneration is quite powerful. The touch assay adds a unique component to the paper and strengthens the argument that burns are more damaging to sensory structures and that different treatments help to ameliorate this.

A weakness of the study is the lack of genetic and cell-autonomous manipulations. Additional comparisons between transection and burns, in particular with manipulations that specifically modulate ROS generation or cell migration without potentially confounding effects on other cell types or processes would help to strengthen the manuscript. In terms of framing their results, the authors refer to "sensory neurons" and "sensory axons" throughout the text - it should be made clear what type of neuron(s)/axon(s) are being visualized/assayed. Along these lines, a broader discussion of how burn injuries affect sensory function in other systems - and how the authors' results might inform our understanding of these injury responses - would be beneficial to the reader.

In summary, the authors have established a tractable vertebrate system to investigate different sensory axon wound healing outcomes in vivo that may ultimately allow for the identification of improved treatment strategies for human burn patients. Although the study implicates differences in keratinocyte migration and associated ROS production in sensory axon wound healing outcomes, the links between these processes could be more rigorously established.

Reviewer #2 (Public Review):

Summary:

Peptidoglycan remodeling, particularly that carried out by enzymes known as amidases, is essential for the later stages of cell division including cell separation. In E. coli, amidases are generally activated by the periplasmic proteins EnvC (AmiA and AmiB) and NlpD (AmiC). The ABC family member, FtsEX, in turn, has been implicated as a modulator of amidase activity through interactions with EnvC. Specifically how FtsEX regulates EnvC activity in the context of cell division remains unclear.

Strengths:

Li et al. make two primary contributions to the study of FtsEX. The first, the finding that ATP binding stabilizes FtsEX in vitro, enables the second, structural resolution of full-length FtsEX both alone (Figure 2) and in combination with EnvC (Figure 3). Leveraging these findings, the authors demonstrate that EnvC binding stimulates FtsEX-mediated ATP hydrolysis approximately two-fold. The authors present structural data suggesting EnvC binding leads to a conformational change in the complex. Biochemical reconstitution experiments (Figure 5) provide compelling support for this idea.

Weaknesses:<br /> The potential impact of the study is curtailed by the lack of experiments testing the biochemical or physiological relevance of the model which is derived almost entirely from structural data.

Altogether the data support a model in which interaction with EnvC, results in a conformational change stimulating ATP hydrolysis by FtsEX and EnvC-mediated activation of the amidases, AmiA and AmiB. However, the study is limited in both approach and scope. The importance of interactions revealed in the structures to the function of FtsEX and its role in EnvC activation are not tested. Adding biochemical and/or in vivo experiments to fill in this gap would allow the authors to test the veracity of the model and increase the appeal of the study beyond the small number of researchers specifically interested in FtsEX.

eLife assessment

This is a useful study that provides solid, yet confirmatory findings about the complex (FtsEX) that controls peptidoglycan remodeling during bacterial cell division. The authors capitalize on the finding that ATP binding stabilizes the FtsEX complex allowing structural characterization for this system. A model is then developed using biochemical approaches to explain ATP regulation. The resulting model would be strengthened were the authors to incorporate their structural findings as well as previously published work.

Reviewer #1 (Public Review):

Summary:

In this paper, Li and colleagues overcome solubility problems to determine the structure of FtsEX bound to EnvC from E. coli.

Strengths:

The structural work is well done and the work is consistent with previous work on the structure of this complex from P. aerugionsa.

Weaknesses:

The model does not take into account all information that the authors obtained as well as known in vivo data.

The work lacks a clear comparison to the Pseudomonas structure highlighting new information that was obtained so that it is readily available to the reader.

The authors set out to obtain the structure of FtsEX-EnvC complex from E. coli. Previously, they were unable to do so but were able to determine the structure of the complex from P. aeruginosa. Here they persisted in attacking the E. coli complex since more is known about its involvement in cell division and there is a wealth of mutants in E. coli. The structural work is well done and recapitulates the results this lab obtained with this complex from P. aeruginosa. It would be helpful to compare more directly the results obtained here with the E. coli complex with the previously reported P. aeruginosa complex - are they largely the same or has some insight been obtained from the work that was not present in the previous complex from P. aeruginosa. This is particularly the case in discussing the symmetrical FtsX dimer binding to the asymmetrical EnvC, since this is emphasized in the paper. However, Figures 3C & D of this paper appear similar to Figures 2D & E of the P. aeruginosa structure. Presumably, the additional information obtained and presented in Figure 4 is due to the higher resolution, but this needs to be highlighted and discussed to make it clear to a general audience.

The main issue is the model (Figure 6). In the model ATP is shown to bind to FtsEX before EnvC, however, in Figure 1c it is shown that ADP is sufficient to promote binding of FtsEX to EnvC.

The work here is all done in vitro, however, information from in vivo needs to be considered. In vivo results reveal that the ATP-binding mutant FtsE(D162N)X promotes the recruitment of EnvC (Proc Natl Acad Sci U S A 2011 108:E1052-60). Thus, even FtsEX in vivo can bind EnvC without ATP (not sure if this mutant can bind ADP).

Perhaps the FtsE protein from E. coli has to have bound nucleotides to maintain its 3D structure.

Reviewer #2 (Public Review):

Summary:

Li et al. investigated the mechanism of action of an important herbicide, caprylic acid (CAP). The authors used untargeted metabolomics to find out differently expressed metabolites (DEM). It led to the identification of metabolites involved in amino acid metabolism, carbon fixation, carbon, glyoxylate, and dicarboxylate metabolism. Using previously published proteomics data and the newly conducted metabolomics data, the authors identified a serine hydroxymethyl transferase in Conyza canadensis (CcSHMT1) to be a likely candidate for CAP inhibition.

The authors conducted a series of in vitro and in vivo tests to elucidate the effect of CAP on SHMT1 inhibition. Plants overexpressing SHMT1 were used to analyze the effect of SHMT1 expression, activity, and inhibition, among others. Purified SHMT1 was used to elucidate enzyme kinetics in the presence or absence of inhibitors. CRISPR-based editing was a powerful method of investigating the effect of SHMT1 mutants on CAP application and complements the overexpression and in vitro studies. Finally, computational docking of CAP on SHMT1 was conducted to identify key interacting residues. The results are overall consistent with one another and present a unified framework for CAP activity as an herbicide. Unexpected variations in SHMT1 expression and activity levels upon CAP treatment suggest complex biological compensatory mechanisms in response to SHMT1 deficiency. Further studies are needed to understand the effect of these perturbations that will be required to successfully develop and deploy CAP-resistant crops for widespread use in agriculture. In conclusion, the authors did a commendable job of elucidating SHMT1 as a biologically relevant target for CAP.

Strengths:

- Combines computational docking, enzyme kinetics using purified proteins, and several different model plant species and two different methods of testing (overexpression and base editing) to establish plant response and survival.

- Sound experimental designs and the presence of controls validate the results and provide additional confidence in the authors' conclusions.

Weaknesses:

- Relied too heavily on the study of plants overexpressing SHMT1, which do not have native gene regulation, and this might limit the generalizability of their conclusions.

-The authors did not leverage computational docking analysis to validate or seek corroboration of the performance of plant alleles obtained from the base editing experiments.

eLife assessment

This study presents a valuable contribution towards understanding the protein target and mechanism of action of an herbicide, which could be applied to the development of herbicide-based technologies to improve crop yields. Evidence is gathered using a variety of technical approaches that enrich and support the findings, but the methodology and the presentation of the results are incomplete.

Reviewer #1 (Public Review):

Caprylic acid (CAP), i.e., octanoic acid, is a saturated fatty acid. CAP is commonly used as a food contact surface sanitizer. In mammals, caprylic acid is related to hunger sensation (i.e., food consumption). serine hydroxymethyl transferase (SHMT) has been previously known as a potential herbicidal target. The present study involves a huge amount of work. The results are useful and contribute well to the literature. The data does support the conclusion. It does not seem that SHMT is the only target of CAP though (CAP may act on other proteins as well). A major deficiency of this manuscript is that there are many unclear, inaccurate, or unconcise descriptions.

Reviewer #3 (Public Review):

Summary:

Li et al investigated the initial target of the herbicidal caprilic acid (CAP). Using a combination of proteomic and metabolomic approaches, they generated a list of candidate targets for CAP and identified a Serine hydroxymethyl transferase (SHMT) as the best candidate.

CAP application to Conyza canadensis induces an early and brief increase in SHMT1 protein and transcript. Studies with purified recombinant CcSHMT1 indicate that enzymatic activity is inhibited by CAP. The authors suggest a kinetic mechanism of CAP inhibition but more data should be collected to reach a firm conclusion on this point.

Transgenic Arabidopsis and rice plants expressing CcSHMT1 show increased tolerance to CAP, as measured by biomass reduction 7 days after treatment with CAP. Similar results were obtained with Arabidopsis and rice plants overexpressing AtSHMT2 and OsSHMT1, respectively. OsSHMT1 single and double mutant rice plants showed increased tolerance to CAP. These results strongly link CAP tolerance to the level of SHMT, which can be manipulated by transgenesis, and suggest that engineered SHMT can also lead to higher CAP tolerance.

Finally, structural analysis allowed the identification of three residues close to the active site involved in the binding of CAP. Arabidopsis plants containing AtSHMT2 modified in these three residues are more sensitive to CAP.

Strengths:

The work of Li et al. includes a large number of assays using different methodologies. The evidence suggests that SHMT inhibition by CAP is effective in inhibiting plant growth. In addition, new technologies that manipulate SHMT levels or activity may improve crop yield by controlling weeds. Structural analysis can be the starting point for the design of more complex molecules that exceed the herbicidal activity of CAP.

Weaknesses:

The methods are rather incomplete, lacking many details necessary to fully understand the author's reasoning. It is not possible to reproduce the experiments on the basis of the information provided.

Although the conclusions are generally well supported, the results are presented in an incorrect or confusing manner. In the comparison of wild-type and transgenic plants, the control condition is missing in some experiments (Figures 4A and 5A). In some plots, the scales are not logical, making them difficult to interpret and fit into an equation (Figures 4B, 4C, 4E, 5E, 6E, 6F).

A final concern is the finding that some point mutations in the SHMT1 gene lead to more tolerant plants (Figures 6D, 6E, 6F). The authors could then explain whether this means that resistance to CAP could be easily acquired by weeds.

Author Response

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

We greatly appreciate your positive assessment and the suggestions by Reviewer #2 on the previous version of our manuscript, all of which are very helpful and have greatly improved our manuscript. We have added a description of Biomineralized columnar architecture in the Results section, added a discussion of the Family Eoobolidae, provided more details in the Material and Methods section, and revised other parts of the manuscript based on her/his comments. We are grateful that these comments have enhanced the overall quality of our manuscript. In this letter, we take the opportunity to note and discuss the various changes as below.

Reviewer #2:

(1) Two early Cambrian taxa of linguliform brachiopods are assigned to the family Eoobolidae. The taxa exhibit a columnar shell structure and the phylogenetic implications of this shell structure in relation to other early Cambrian families is discussed. It is the interesting idea regarding the evolution of shell structure.

We thank Reviewer 2 very much for her/his very constructive suggestions. All the comments have been thoroughly considered, and introduced into the revised version of the manuscript.

(2) The early record of shell structures of linguliform brachiopods is incomplete and partly contradictory. The authors maintain silence regarding contradictory information throughout the article to an extent that information is cited wrongly.

We agree with Reviewer #2 that the early record of shell structure of linguliform brachiopods is incomplete and potentially in some instances contradictory. This situation is well demonstrated in the Introduction and Systematic Palaeontology sections of this paper. This is also the reason why we think the detailed investigation of early linguliform shell architectures is so important, and we hope this work will be useful for further comparative studies on brachiopod biomineralization. We also understand that more detailed studies of the complexity and diversity of linguliform brachiopod architectures (especially their early fossil representatives) require further investigation.

(3) The article is written under the assumption that all eoobolids have a columnar shell structure. Thus, the previously claimed columnar structure of Eoobolus incipiens which has been re-illustrated in the paper is not convincing and could be interpreted in other ways.

Yes, the type specimen of Eoobolus is poorly known and we do not know its shell structure, but the ornamentation, pseudointerarea etc. are well preserved and promote a character diagnosis. In this paper, we focus on the detailed study of Cambrian eoobolids with exquisitely well-preserved columns from the Cambrian Series 2 based on the collection of more than 30 thousand early Cambrian brachiopod specimens in China and Australia. With the wide preservation of columnar shells in early eoobolid specimens, it is likely that Eoobolus has columnar shell architecture, although there is no documentation of the shell structure from every single Eoobolus specimen.

The secondary columns of Eoobolus incipiens is well demonstrated in Fig. 4a. The size of these columns can be well compared with the columns from other Eoobolus species and acrotretide brachiopods, which are quite different from the criss-cross baculae. As we noted in the manuscript, the columnar structure Eoobolus incipiens is very simple (short columns and less number of columnar units) and can be readily secondarily phosphatised. This is also the reason why it is hard to find the columnar shell architecture in early eoobolids.

(4) The article needs a proper results section. The Discussion is mainly a review of published data. Other potential results are hidden in this "discussion".

I would recommend to reorganize the paper and make it a solid presentation of the new taxa and other new results, i.e., have a solid Results section. The Discussion should discuss relevant points that relate to the new results rather than reviewing shell structure in general but skipping relevant parts such as the tertiary shell layer.

We have reorganised the manuscript based on these comments. A general description of the biomineralized columnar architecture is added in the Results section. As the Supplementary section (main results) includes 7 figures and 3 tables, it will increase the size of the current paper if they are moved to the main text. We would prefer to keep the main results in the Supplementary based on the style and format of eLife.

As the current information on the shell structures of early linguliform brachiopods is unclear, we need to review most of the previous studies on brachiopod shells in the first part of Discussion section. It will help the readers to follow our results and conclusion. So, we think some of the review content is necessary and helps build the Discussion section. The tertiary shell layer, which is not developed in our studied material, is not discussed in the current research.

(5) In addition, a more elaborate Methods section is needed in which it is explained how the data for shell thicknesses and numbers of laminae was obtained.

The potential evolutionary patterns that are discussed towards the end (summarized in Fig 6) are interesting but rather unconvincing as the way the data has been obtained has never been clarified. Shell thicknesses and numbers of laminae that built up the shell of several taxa are compared, but at no point it is stated where these measurements were taken. Shell thicknesses vary within a shell and also the presence of the never mentioned tertiary layer is modifying shell thicknesses. Hence, the presented data appears random and is not comparable. The obtained evolutionary patterns must be considered as dubious.

A proper Methods section would be needed that explains how the data presented in Fig. 6 has been obtained. Plus it needs to be convincingly explained that the obtained data is in fact comparable and represents, e.g., equivalent areas of the shell in all involved taxa.

All the information is added in the Material and Methods section. We are aware of the marginal accretionary secretion of brachiopod shells. It is well known that the shell at the posterior is thicker (usually the thickest) than that at the anterior, we did not note this in the previous manuscript. We have measured all the shell data (shell thickness and number of columnar unit) from the posterior part of the adult shell for all the studied taxa. And the measurements of diameter and height of orthogonal columns are performed on available adult specimens from this study and previously published literature. Consequently, the obtained data are comparable and represent equivalent areas of the shell on all involved taxa.

In term of the tertiary shell layer, we do not find any evidence of this tertiary shell layer from our studied material. The tertiary shell layer is well developed in some recent and Palaeozoic lingulides (Holmer, 1989), but it is not recognised in the early eoobolides and acrotretides.

(6) A critical revision of the family Eoobolidae and Lingulellotretidae including a revision of the type species of Eoobolus and Lingulellotreta is needed.

Concerning the families Eoobolidae and Lingulellotretidae, we are aware of the current problematic situation of these families, and we have added more remarks regarding the Eoobolidae in the Systematic Palaeontology section of the manuscript. However, the revision of the families Eoobolidae and Lingulellotretidae falls outside the scope of this paper. We prefer to exclude it just now, as it will be part of an upcoming publication based on more material from China, Australia, Sweden and Estonia that we are currently working on.

eLife assessment

This valuable study examines the evolution of the pillars in the shell architecture of organo-phosphatic brachiopods. The phylogenetic implications of this shell structure in relation to other early Cambrian brachiopod families are interpreted based on solid evidence. As such, this paper with interesting ideas regarding the evolution of brachiopod shell structure contributes to our understanding of the ecology and evolution of brachiopods as a whole.

Reviewer #2 (Public Review):

Summary:

Two early Cambrian taxa of linguliform brachiopods are assigned to the family Eoobolidae. The taxa exhibit a columnar shell structure and the phylogenetic implications of this shell structure in relation to other early Cambrian families is outlined.

Strengths:

Interesting idea regarding the evolution of shell structure.

Weaknesses:

The early record of shell structures of linguliform brachiopods is incomplete and partly contradictory. The authors maintain silence regarding contradictory information throughout the article to an extend that information is cited wrongly. The article is written under the assumption that all eoobolids have a columnar shell structure. Thus, the previously claimed columnar structure of Eoobolus incipiens which has been re-illustrated in the paper is not convincing and could be interpreted in other ways.

The article still needs a proper results section. The Discussion is mainly a review of published data. Other potential results are hidden in this "discussion". Large sections of the paper appear irrelevant and can be deleted.

A critical revision of the family Eoobolidae and Lingulellotretidae including a revision of the type species of Eoobolus and Lingulellotreta is needed first.

The potential evolutionary patterns that are presented towards the end (summarized in Fig 6) are interesting but rather unconvincing. The stated numbers of shell laminae, whose origin has now been clarified in a still rather short Methods section, represent a mix of data and are not comparable. Achieved numbers of laminae based on literature data include laminae from the secondary and tertiary shell layer, a distinction between the two would be crucial for the proposed claims.<br /> The obtained evolutionary patterns as presented in Fig. 6 must, after the second revision and clarification of the methods used, be regarded as misleading and reflects a limited understanding of shell growths in linguliform brachiopods (despite the extensive review of the literature).

Author Response

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

eLife assessment

This manuscript provides important insights into the degradation of a host tRNA modification enzyme TRMT1 by SARS-CoV-2 protease nsp5. The data convincingly support the main conclusions of the paper. These results will be of interest to virologists interested in studying the alterations in tRNA modifications, host methyltransferases, and viral infections.

Public Reviews:

Response to Public Reviews

We appreciate the reviewers’ assessment that our findings are well supported and provide important insight to the field. We also thank the reviewers for their comments and suggestions that have improved the quality of this manuscript. Through the requested edits and experiments, we provide additional results in this revision that further support and extend our original findings.

We acknowledge the major questions that remain to be addressed, including the biological relevance of TRMT1 cleavage by Nsp5. We note that elucidating the biological role of host protein cleavage by viral proteases has been a long-standing challenge. For example, several endogenous proteins have been identified as cleavage targets of HIV protease, but the functional relevance for many of these cases took decades to resolve or remain unknown to this day. Nonetheless, we have added additional experiments that suggest a possible role for TRMT1 and TRMT1 cleavage in SARS-CoV-2 pathobiology.

Key additions in the revised manuscript include:

• Subcellular localization of full-length TRMT1 and TRMT1 fragments (Supplemental Figure 4).

• Experiments demonstrating that TRMT1 levels are reduced to near background levels in SARS-CoV-2 infected human cells at higher MOI (Figure 6C and D).

• Results showing that expression of the non-cleavable TRMT1 mutant can promote virion particle infectivity (Figure 8).

• The addition of an “Ideas and Speculation” subsection that is now being offered to authors by eLife.

Reviewer #1 (Public Review):

Zhang et al. investigate the hypothesis that tRNA methyl transferase 1 (TRMT1) is cleaved by NSP5 (nonstructural protein 5 or MPro), the SARS-CoV-2 main protease, during SARS-CoV-2 infection. They provide solid evidence that TRMT1 is a substrate of Nsp5, revealing an Nsp5 target consensus sequence and evidence of TRMT1 cleavage in cells. Their conclusions are exceptionally strong given the co-submission by D'Oliveira et al showing cleavage of TRMT1 in vitro by Nsp5. Separately, the authors convincingly demonstrate widespread downregulation of RNA modifications during CoV-2 infection, including a requirement for TRMT1 in efficient viral replication. This finding is congruent with the authors' previous work defining the impact of TRMT1 and m2,2g on global translation, which is most likely necessary to support infection and virion production. What still remains unclear is the functional relevance of TRMT1 cleavage by Nsp5 during infection. Based on the data provided here, TRMT1 cleavage may be an act by CoV2 to self-limit replication, as the expression of a non-cleavable TRMT1 (versus wild-type TRMT1) supports enhanced viral RNA expression at certain MOIs. Theoretically, TRMT1 cleavage should inactivate the modification activity of TRMT1, which the authors thoroughly and elegantly investigate with rigorous biochemical assays. However, only a minority of TRMT1 undergoes cleavage during infection in this study and thus whether TRMT1 cleavage serves an important functional role during CoV-2 replication will be an important topic for future work. The authors fairly assess their work in this regard. This study pushes forward the idea that control of tRNA expression and functionality is an important and understudied area of host-pathogen interaction.

We thank the reviewer for the thoughtful assessment of our study.

We acknowledge that only a minority of TRMT1 undergoes cleavage during infection at the originally tested MOI. However, the ~40% reduction in TRMT1 levels after infection with SARS-CoV-2 is quite substantial considering that the TRMT1 in the nucleus and mitochondria are likely to be inaccessible to Nsp5. Moreover, we detected a reduction in m2,2G modification in the infected human cells, providing evidence for a functional impact on TRMT1 activity (Figure 1C).

To further test the effects of SARS-CoV-2 infection on endogenous TRMT1, we infected 293T cells at a higher MOI and measured TRMT1 levels. At MOI=5, we found that SARS-CoV-2 infection led to near complete depletion of TRMT1 in human cells. This result suggests that SARS-CoV-2 infection could have a profound impact on TRMT1 levels during pathogenesis. We have added this new experiment as Figures 6C and D.

Weaknesses noted:

The detection of the N-terminal TRMT1 fragment by western blot is not robust. The polyclonal antibody used to detect TRMT1 in this work cross-reacts with a non-specific protein product. Unfortunately, this obstructs the visualization of the predicted N-terminal TRMT1 fragment. It is unclear how the authors were able to perform densitometry, given the interference of the nonspecific band. Additionally, the replicates in the source data make it clear that the appearance of the N-terminal fragment "wisp" under the non-specific band is not seen in every replicate. Though the disappearance of this wisp with mutant Nsp5 and uncleavable TRMT1 is reassuring, the detection of the N-terminal fragment with the TRMT1 antibody should be assessed critically. Considering this group has strong research interests in TRMT1, I assume that attempts to make other antibodies have proved unfruitful. Additionally, N-terminal tagging of TRMT1 is predicted to disrupt the mitochondrial targeting signal, eliminating the potential for using alternative antibodies to see the N-terminal fragment.

We agree that the anti-TRMT1 antibody used here is sub-optimal for detection of the N-terminal TRMT1 fragment. However, as noted by the Reviewer, we provided multiple ways of corroborating that the lower-molecular weight band detected in human cells expressing Nsp5 corresponds to the N-terminal TRMT1 fragment. We have shown that the TRMT1 cleavage band is not detectable in human cells expressing GFP or inactive Nsp5. This indicates that the lower molecular weight TRMT1 band only arises when active Nsp5 protease is expressed. Moreover, the TRMT1 cleavage band is not detectable in TRMT1-KO cell lines, demonstrating that the band arises from TRMT1 cleavage rather than a non-specific protein. We have also detected the C-terminal fragment if TRMT1 is over-expressed with Nsp5. In addition, we have shown that the mutation of the predicted Nsp5 cleavage site in TRMT1 abolishes the appearance of the N- and Cterminal cleavage fragments.

Despite the drawbacks of this antibody, we identified gel running conditions that resolves the non-specific band from the N-terminal TRMT1 cleavage fragment. Thus, for quantification, we measured the total signal of both the cleavage band and the nonspecific band in all lanes (Figure 3). After normalization to actin, the total signal from the cleavage band and the non-specific band in the control lane from cells expressing GFP was subtracted from the lanes with cells expressing Nsp5 to calculate the signal arising from the cleavage band. We have updated our Materials and Methods to provide details on how we quantified the TRMT1 cleavage band.

While we did test other antibodies against TRMT1, none of them were sensitive enough to detect TRMT1 cleavage fragments at endogenous levels. For example, we included results with an antibody targeting the C-terminus of TRMT1 that could not detect TRMT1 cleavage products at endogenous levels (Supplemental Figure 3). However, the antibody could detect the C-terminal TRMT1 fragments if TRMT1 was overexpressed with Nsp5 (Supplemental Figure 3).

These technical issues reiterate the fact that the functional significance of TRMT1 cleavage during CoV-2 infection remains unclear. However, this study demonstrates an important finding that the tRNA modification landscape is altered during CoV-2 infection and that TRMT1 is an important host factor supporting CoV-2 replication.

We agree that the functional relevance of TRMT1 cleavage by Nsp5 remains an open question. Thus, we have added an experiment to test the functional impact of TRMT1 on virion particle production and infectivity (Figure 8). We find that TRMT1 expression is required for optimal virus production, consistent with our observation that TRMT1deficient cells exhibit reduced viral RNA replication. In addition, we find that expression of the non-cleavable TRMT1 mutant can promote virion particle infectivity (Figure 8, TRMT1-Q530N). These results are consistent with the Reviewer’s conclusion that “TRMT1 cleavage may be an act by CoV-2 to self-limit replication, as the expression of a non-cleavable TRMT1 (versus wild-type TRMT1) supports enhanced viral RNA expression at certain MOIs”. We discuss the potential implications of this result and their functional relevance in the “Ideas and Speculation” subsection.

Reviewer #2 (Public Review):

Summary:

The manuscript titled 'Proteolytic cleavage and inactivation of the TRMT1 tRNA modification enzyme by SARS-CoV-2 main protease' from K. Zhang et al. demonstrates that several RNA modifications are downregulated during SARS-CoV-2 infection including the widespread m2,2G methylation, which potentially contributes to changes in host translation. To understand the molecular basis behind this global hypomodification of RNA during infection, the authors focused on the human methyltransferase TRMT1 that catalyzes the m2,2G modification. They reveal that TRMT1 not only interacts with the main SARS-CoV-2 protease (Nsp5) in human cells but is also cleaved by Nsp5. To establish if TRMT1 cleavage by Nsp5 contributes to the reduction in m2,2G levels, the authors show compelling evidence that the TRMT1 fragments are incapable of methylating the RNA substrates due to loss of RNA binding by the catalytic domain. They further determine that expression of full-length TRMT1 is required for optimal SARS-CoV-2 replication in 293T cells. Nevertheless, the cleavage of TRMT1 was dispensable for SARS-CoV-2 replication hinting at the possibility that TRMT1 could be an off-target or fortuitous substrate of Nsp5. Overall, this study will be of interest to virologists and biologists studying the role of RNA modification and RNA modifying enzymes in viral infection.

We thank the reviewer for the thoughtful assessment of our study.

We agree with the possibility that TRMT1 could be a fortuitous substrate of Nsp5 due to the coincidental presence of a Nsp5 cleavage site in TRMT1. As considered in our Discussion section, TRMT1 cleavage could be a collateral effect of SARS-CoV-2 infection. While TRMT1 could be an off-target substrate during viral infection, the subsequent effect on tRNA modification levels could have physiological consequences on downstream processes that affect cellular health. This information could still be useful for understanding the pathophysiological consequences of SARS-CoV-2 infection in tissues.

Strengths:

• The authors use a state-of-the-art mass spectrometry approach to quantify RNA modifications in human cells infected with SARS-CoV-2.

• The authors go to great length to demonstrate that SARS-CoV-2 main protease, Nsp5, interacts, and cleaves TRMT1 in cells and perform important controls when needed. They use a series of overexpression with strategically placed tags on both TRMT1 and Nsp5 to strengthen their observations.

• The use of an inactive Nsp5 mutant (C145A) strongly supports the claim of the authors that Nsp5 is solely responsible for TRMT1 cleavage in cells.

• Although the direct cleavage was not experimentally determined, the authors convincingly show that TRMT1 Q530N is not cleaved by Nsp5 suggesting that the predicted cleavage site at this position is most likely the bona fide region processed by Nsp5 in cells.

• To understand the impact of TRMT1 cleavage on its RNA methylation activity, the authors rigorously test four protein constructs for their capacity not only to bind RNA but also to introduce the m2,2G modification. They demonstrate that the fragments resulting from TRMT1 cleavage are inactive and cannot methylate RNA. They further establish that the C-terminal region of TRMT1 (containing a zinc-finger domain) is the main binding site for RNA.

• While 293T cells are unlikely an ideal model system to study SARS-CoV-2 infection, the authors use two cell lines and well-designed rescue experiments to uncover that TRMT1 is required for optimal SARS-CoV-2 replication.

Weaknesses:

• Immunoblo0ng is extensively used to probe for TRMT1 degradation by Nsp5 in this study. Regretfully, the polyclonal antibody used by the authors shows strong non-specific binding to other epitopes. This complicates the data interpretation and quantification since the cleaved TRMT1 band migrates very closely to a main non-specific band detected by the antibody (for instance Fig 3A). While this reviewer is concerned about the cross-contamination during quantification of the N-TRMT1, the loss of this faint cleaved band with the TRMT1 Q530N mutant is reassuring. Nevertheless, the poor behavior of this antibody for TRMT1 detection was already reported and the authors should have taken better precautions or designed a different strategy to circumvent the limitation of this antibody by relying on additional tags.

We acknowledge the sub-optimal performance of the commercial anti-TRMT1 antibody used in our study. Nevertheless, we have provided multiple lines of evidence indicating that the lower molecular weight band detected using this antibody corresponds to the N-terminal TRMT1 fragment. As noted by the reviewer, we have shown that the lower molecular weight band disappears using the TRMT1-Q530N non-cleavable mutant. The lower molecular weight signal is also absent in TRMT1-KO cell lines expressing Nsp5. Moreover, we have shown that the TRMT1 cleavage band is undetectable in human cells expressing GFP or inactive Nsp5. We have also detected the C-terminal fragment when TRMT1 is over-expressed with Nsp5.

As discussed in the response to Reviewer 1, we did consider alternative approaches for detecting the N-terminal fragment. We thought about tagging TRMT1 at the N-terminus so that we could detect the cleavage band using a different antibody. However, as noted by Reviewer 1, the tagging of TRMT1 at the N-terminus is likely to disrupt the mitochondrial targeting signal and alter the localization of TRMT1. In addition, we spent considerable time and effort testing alternative antibodies against TRMT1. However, none of them were effective at detecting the N- or C-terminal TRMT1 fragments. For example, we included results with a different antibody targeting the C-terminus of TRMT1 that could not detect TRMT1 cleavage products at endogenous levels but could detect them when TRMT1 was overexpressed with Nsp5 (Supplemental Figure 3).

• While 293T cells are convenient to use, it is not a well-suited model system to study SARS-CoV2 infection and replication. Therefore, some of the conclusions from this study might not apply to better-suited cell systems such as Vero E6 cells or might not be observed in patient-infected cells.

We acknowledge the potential caveats associated with using 293T human embryonic cells as a system for testing SARS-CoV2 replication. However, we note that 293T cells have been used as a physiological model for discovering and characterizing key aspects of SARS-CoV-2 biology, including viral replication. For example, SARS-CoV-2 has been shown to exhibit significant replication and virion production in 293T cells expressing ACE2 that can be inhibited by known SARS-CoV-2 antiviral compounds:

• https://www.thelancet.com/journals/lanmic/article/PIIS2666-5247(20)300045/fulltext

• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9444585/

• https://www.science.org/doi/10.1126/sciadv.add3867

• https://www.pnas.org/doi/full/10.1073/pnas.2025866118

293T cells have also been demonstrated to exhibit cytopathic effects upon SARS-CoV-2 infection that are dependent upon the ACE2 receptor and mirror that of infected lung cells in culture and in patient tissues:

• https://www.embopress.org/doi/full/10.15252/embj.2020106267

• https://journals.asm.org/doi/full/10.1128/jvi.00002-22

• https://journals.plos.org/plospathogens/article?id=10.1371/journal.ppat.1009715

• https://www.nature.com/articles/s41559-021-01407-1

In addition to 293T cells, we have demonstrated that infection of MRC5 human pulmonary fibroblast cells with SARS-CoV-2 results in a decrease in TRMT1 levels and m2,2G modification (Figure 1). The reduction in TRMT1 levels in MRC5 cells after SARS-CoV-2 infection is similar to that observed in 293T cells.

• The reduction of bulk TRMT1 levels is minor during infection of MRC5 cells with SARS-CoV-2 (Fig 1). This does not seem to agree with the more dramatic reduction in m2,2G modification levels. Cellular Localization experiments of TRMT1 would help clarify this. While TRMT1 is found in the cytoplasm and nucleus, it is possible that TRMT1 is more dramatically degraded in the cytoplasm due to easier access by Nsp5.

We agree that the processing of newly synthesized TRMT1 in the cytoplasm is likely to be the main cause for the reduction of TRMT1 levels in the infected MRC5 cells. Thus, we followed the Reviewer’s suggestion to conduct cellular localization experiments of TRMT1 (Supplemental Figure 4). Through these experiments, we show that full-length TRMT1 exhibits localization to the cytoplasm, mitochondria, and nucleus, consistent with prior findings from our group and others. This result supports the conclusion that cytoplasmic TRMT1 is the likely target of Nsp5 cleavage while TRMT1 in the nucleus and mitochondria are inaccessible to Nsp5. We also note that the decrease in cytoplasmic TRMT1 could account for the reduction in m2,2G modifications if the cytoplasmic pool of TRMT1 is responsible for modifying any exported tRNAs that were not modified in the nucleus.

• In Fig 6, the authors show that TRMT1 is required for optimal SARS-CoV-2 replication. This can be rescued by expressing TRMT1 (Fig 7). Nevertheless, it is unknown if the methylation activity of TRMT1 is required. The authors could have expressed an inactive TRMT1 mutant (by disrupting the SAM binding site) to establish if the RNA modification by TRMT1 is important for SARS-CoV-2 replication or if it is the protein backbone that might contribute to other processes.

We agree that it would be interesting to test if the methylation activity of TRMT1 is important for optimal SARS-CoV-2 replication. However, the present study focuses on the cleavage of TRMT1 by Nsp5 and the biological effects of this cleavage. Thus, we feel that generating another human cell line lies outside the scope of this paper and would be an excellent idea for future studies. We thank the reviewer for the proposed experiment.

• Fig 7, the authors used the Q530N variant to rescue SARS-CoV-2 replication in TRMT1 KO cells. This is an important experiment and unexpectedly reveals that TRMT1 cleavage by Nsp5 is not required for viral replication. To strengthen the claim of the authors that TRMT1 is required to promote viral replication and that its cleavage inhibits RNA methylation, the authors could express the TRMT1 N-terminal construct in the TRMT1 KO cells to assess if viral replication is restored or not to similar levels as WT TRMT1. This will further validate the potential biological importance of TRMT1 cleavage by Nsp5.

Indeed, we did not expect to find that human cells expressing the TRMT1-Q530N variant exhibit higher levels of viral replication. This suggests that cleavage of TRMT1 is inhibitory for viral replication. To provide further support for this observation, we analyzed the viral titer and infectivity of supernatants derived from human cells expressing wildtype TRMT1 or TRMT1-Q530N. Consistent with our finding that TRMT1-Q530N cells contain more viral RNA, the media supernatants from TRMT1Q530N expressing cells exhibit higher viral titer and infectivity compared to supernatants from TRMT1-KO cells expressing wildtype TRMT1. These results provide additional evidence that TRMT1 is required to promote viral replication. Moreover, these findings suggest that TRMT1 cleavage and reduced protein synthesis could selflimit viral replication. The additional results have been added as Figure 8.

• Fig 7 shows that the TRMT1 Q530N variant rescues SARS-CoV-2 replication to greater levels then WT TRMT1. The authors should discuss this in greater detail and its possible implications with their proposed statement. For instance, are m2,2G levels higher in Q530N compared to WT? Does Q530N co-elute with Nsp5 or is the interaction disrupted in cells?

These are excellent points brought up by the Reviewer. As noted above, we have added an additional experiment that tests the functional relevance of TRMT1 expression and cleavage on virion production and infectivity (Figure 8). Moreover, we have followed the Reviewer’s suggestion and discussed the potential implications of these findings in the “Ideas and Speculation” subsection.

Reviewer #3 (Public Review):

Summary:

In this manuscript, the authors have used biochemical approaches to provide compelling evidence for the cleavage of TRMT1 by SARS-CoV-2 Nsp5 protease. This work is of wide interest to biochemists, cell biologists, and structural biologists in the coronavirus (CoV) field. Furthermore, it substantially advances the understanding of how CoV's interact with host factors during infection and modify cellular metabolism.

We thank the reviewer for the thoughtful assessment of our study.

Strengths:

The authors provide multiple lines of biochemical evidence to report a TRMT1-Nsp5 interaction during SARS-CoV-2 infection. They show that the host enzyme TRMT1 is cleaved at a specific site and that it generates fragments that are incapable of functioning properly. This is an important result because TRMT1 is a critical player in host protein synthesis. This also advances our understanding of virus-host interactions during SARS-CoV-2 infections.

Weaknesses:

The major weakness is the lack of mechanistic insights into TRMT1-Nsp5 interactions. The authors have provided commendable biochemical data on proving the TRMT1-Nsp5 interaction but without clear mechanistic insights into when this interaction takes place in the context of SARS-CoV-2 propagation, what are the functional consequences of this interaction on host biology, and does this somehow benefit the infecting virus? I feel that the authors played it a bit safe despite having access to several reagents and an extremely promising research direction.

We agree that our findings have prompted questions on the mechanistic and functional relevance of TRMT1 cleavage by Nsp5. To begin addressing the latter point, we have included a new experiment testing the impact of TRMT1 expression and cleavage on SARS-CoV-2 virus production and infectivity (Figure 8). We find that TRMT1-deficient cells infected with SARS-CoV-2 exhibit less virion production and the viruses produced are less infectious. Intriguingly, we find that expression of the non-cleavable TRMT1-Q530N variant in TRMT1-KO cells promotes an increase of viral titer as well as infectivity compared to expression of wildtype TRMT1. These results provide evidence for an unexpected role for TRMT1 expression in virus production and the generation of optimally infectious SARS-CoV-2 particles. We discuss the potential implications of this finding in the “Ideas and Speculation” subsection.

We agree that understanding the timing and effects of Nsp5-TRMT1 interaction will be an important area of investigation moving forward. We would like to include additional time points beyond 24- and 48-hours post-infection. However, we have found that the MRC5-ACE2 cells exhibited increased levels of cell death at 72 and 96-hours postinfection that could confound results (Raymonda et al 2022). Moreover, we would like to know how the reduction in m2,2G modifications affects host tRNA biology and translation. However, these experiments involve large-scale methods such as tRNA sequencing and ribosome profiling which are outside the scope of our current studies and will be the subject of future efforts.

We acknowledge the Reviewer’s assessment that we “played it a bit safe” in discussing the functional consequences of Nsp5-TRMT1 interaction. We aimed for a circumspect interpretation of our results and their biological implications, but might have been too cautious in our conclusions. Thus, we have added an “Ideas and Speculation” subsection that discusses possible reasons for how TRMT1 cleavage and interaction with Nsp5 could benefit the virus. We thank the Reviewer for pointing out this issue in our initial manuscript.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

Having reviewed an earlier version of this manuscript, I appreciated the recent progress made by the authors. I felt the entire body of work is quite solid and the interpretations are clear and not overstated. One piece of data I thought deserved a sentence or two of discussion was the complementation assay with Q530N TRMT1. This experiment suggests the possibility that cleavage of TRMT1 by Nsp5 may be an act to self-limit replication, although this result could also be due to the elevated levels of Q530N TRMT1 expression compared to WT. I still think it is worthy of discussion. Another thing I would recommend is to include the length of infection by SARS-CoV-2 in the figure legends.

We thank the reviewer for their positive response and constructive comments.

We have followed the Reviewer’s suggestion to further discuss how cleavage of TRMT1 may act to self-limit replication in the “Ideas and Speculation” subsection. We have also included the length of infection by SARS-CoV-2 in the figure legends.

Reviewer #2 (Recommendations For The Authors):

In addition to the comments mentioned in the public review, this reviewer encourages the authors to address the following points:

• Please clarify the rationale behind choosing 24 and 48 hours post-infection as time points for the analyses (Fig 1). One would expect even lower levels of TRMT1 and RNA modification after 72 and 96 hours post-infection.

We chose the 24 and 48-hour time points since we have shown that MRC5 cells exhibit elevated accumulation of viral RNA at these time points (Raymonda et al 2022). However, at 72 and 96-hours post-infection, we have found that the MRC5-ACE2 cells exhibited cytopathic effects indicative of cell death that could confound results. We have included the rationale for these time points in our revised manuscript.

• In Supplementary Figure 3, please add in the legend the meaning of the asterisk symbol.

The asterisks denote non-specific bands that are still detectable in the TRMT1-KO cell line. We have updated the Figure Legend and thank the Reviewer for catching this omission.

• In Supplementary Figure 3B, there is an intermediate band in lane 3 with C145A when using the antibody 609-659. The authors should clarify what that band is.

The intermediate band in lane 3 (and in lane 6) of Supplemental Figure 3B represents non-specific detection of the Nsp5-C145A variant that exhibits extremely high levels of expression since it cannot self-cleave. We have clarified the identity of the band in the figure legend.

Reviewer #3 (Recommendations For The Authors):

I have only minor comments:

Although the authors have done a commendable job of providing compelling biochemical evidence of TRMT1 cleavage by Nsp5, it is not clear how this enhances viral infection. The discussion presents the experimental findings and prior publications as a series of correlated observations without clearly specifying the mechanistic benefits of TRMT1 hijacking towards CoV propagation, or even proposing a mechanistic hypothesis to this end.

We agree with the Reviewer that providing a mechanistic hypothesis on how TRMT1 cleavage impacts virus biology will help inform future studies. We have followed the Reviewer’s suggestion and discuss potential mechanisms in the “Ideas and Speculation” subsection.

How do these experiments inform us about the cell biology of SARS-CoV- infections? Does Nsp5-mediated degradation start early in infection? Is the loss of TRMT1 sustained over the course of the infection? Do Nsp5 concentrations or relative amounts correlate with TRMT1 loss during this period? For instance, is there only a modest increase in Nsp5 levels from 24h to 48h? I would suggest adding a few more data points than just 24h and 48h in the cell culture experiments. As the manuscript stands right now, it will be a bit difficult for readers to appreciate the relevance of this study in its present form.

These are excellent questions raised by the Reviewer. The temporal effects of SARSCoV-2 infection on TRMT1 levels will be an important area to dissect moving forward.

As mentioned above, we would like to include additional time points beyond 24- and 48-hours post-infection. However, at 72 and 96-hours post-infection, we have found that the MRC5-ACE2 cells exhibited increased levels of cell death that could confound results.

However, we do observe a correlation between the level of infection and the amount of TRMT1 depletion. In our newly added Figure 6C and 6D, we show that increasing the MOI leads to a concomitant increase in N-protein production that correlates with the amount of TRMT1 depletion. Moreover, we have added additional experiments to explore the biological relevance of our findings in terms of virion particle production and infectivity. We thank the reviewer for these insightful questions that have improved our manuscript and provide a foundation for future studies.

Related to this previous comment: how do the authors rationalize their inference that TRMT1 is essential for SARS-CoV-2 infection, yet it is cleaved during the infection? What seems to be the advantage of this seemingly contradictory but possibly quite intriguing inference?

We acknowledge the paradox that TRMT1 seems to be essential for SARS-CoV-2 replication but is cleaved during the infection. We propose several hypotheses to explain these findings:

Hypothesis 1: TRMT1 could be a bystander target. The loss of TRMT1 expression leads to a decrease in modifications that impacts translation. This decrease in translation capacity of the infected cells would lead to decreased production of viral proteins and reduced viral replication. This could explain why TRMT1-deficient cells exhibit less virus production. This could also account for why the TRMT1-Q530N mutant might produce more virus. In this case, the cleavage of TRMT1 and biological effects on viral replication and virion production are coincidental. However, even if TRMT1 cleavage and inactivation does not impact viral replication or production, it would still be important to know the cellular impacts that contribute to disease pathogenesis.

Hypothesis 2: The slight diminishment of viral replication due to host translation inhibition could outweigh the benefits of shutting down host responses dependent upon protein synthesis. The decrease in TRMT1-catalyzed tRNA modification caused by Nsp5 cleavage could severely inhibit host translation while viral translation can still be maintained through a tRNA pool optimized for viral translation, albeit at a slightly lower rate than if TRMT1 is not cleaved.

Hypotheses 3: The Nsp5-TRMT1 interaction could allow the virus to bind tRNAs that are packaged in viral particles as suggested previously (Pena et al., 2022). The finding that expression of the non-cleavable TRMT1-Q530N variant enhances viral replication and infectivity supports the hypothesis that TRMT1 could facilitate tRNA uptake into viral particles. The packaging of specific tRNAs in viral particles could enhance viral translation in the subsequent round of infection, thereby enhancing infectivity and perhaps facilitating the species jump of SARS-CoV-2 towards hosts with incompatible codon bias.

We have included these hypotheses in the new “Ideas and Speculation” subsection.

eLife assessment

This manuscript provides important insights into the degradation of the host tRNA modification enzyme TRMT1 by the SARS-CoV-2 protease Nsp5 (nonstructural protein 5 or MPro). The data convincingly support the main conclusions of the paper. These results will be of interest to virologists studying the alterations in tRNA modifications, host methyltransferases, and viral infections.

Reviewer #1 (Public Review):

Zhang et al. investigate the hypothesis that tRNA methyl transferase 1 (TRMT1) is cleaved by NSP5 (nonstructural protein 5 or MPro), the SARS-CoV-2 main protease, during SARS-CoV-2 infection. They provide solid evidence that TRMT1 is a substrate of Nsp5, revealing an Nsp5 target consensus sequence and evidence of TRMT1 cleavage in cells. Their conclusions are exceptionally strong given the co-submission by D'Oliveira et al showing cleavage of TRMT1 in vitro by Nsp5. The detection of the N-terminal TRMT1 fragment by western blot is not robust; however, the authors provide corroborating assays and detailed densitometry methods, providing confidence to this reviewer that a TRMT1 fragment is produced under some conditions. Separately, the authors convincingly demonstrate widespread downregulation of RNA modifications during CoV-2 infection, including a requirement for TRMT1 in efficient viral replication. This finding is congruent with the authors' previous work defining the impact of TRMT1 and m2,2g on global translation, which is most likely necessary to support infection and virion production. Based on the data provided here, TRMT1 cleavage may be an act by CoV-2 to self-limit replication, as expression of a non-cleavable TRMT1 (versus wild type TRMT1) supports enhanced viral RNA expression at certain MOIs. The authors propose a few fascinating ideas to why this may be so in "Ideas and Speculation." Theoretically, TRMT1 cleavage should inactivate the modification activity of TRMT1, which the authors thoroughly and elegantly investigate with rigorous biochemical assays. However, only a minority of TRMT1 undergoes cleavage during infection at low MOIs and thus whether TRMT1 cleavage serves an important functional role during CoV-2 replication will be an important topic for future work. The authors fairly assess their work in this regard. In summary, this study demonstrates an important finding that the tRNA modification landscape is altered during CoV-2 infection, and that TRMT1 is an important host factor supporting CoV-2 replication. Their data pushes forward the idea that control of tRNA expression and functionality is an important and understudied area of host-pathogen interaction.

Reviewer #2 (Public Review):

Summary:<br /> The manuscript titled 'Proteolytic cleavage and inactivation of the TRMT1 tRNA modification enzyme by SARS-CoV-2 main protease' from K. Zhang et al., demonstrates that several RNA modifications are downregulated during SARS-CoV-2 infection including the widespread m2,2G methylation, which potentially contributes to changes in host translation. To understand the molecular basis behind this global hypomodification of RNA during infection, the authors focused on the human methyltransferase TRMT1 that catalyzes the m2,2G modification. They reveal that TRMT1 not only interacts with the main SARS-CoV-2 protease (Nsp5) in human cells but is also cleaved by Nsp5. To establish if TRMT1 cleavage by Nsp5 contributes to the reduction in m2,2G levels, the authors show compelling evidence that the TRMT1 fragments are incapable of methylating the RNA substrates due to loss of RNA binding by the catalytic domain. They further determine that expression of full-length TRMT1 is required for optimal SARS-CoV-2 replication in 293T cells. Nevertheless, the cleavage of TRMT1 was dispensable for SARS-CoV-2 replication hinting at the possibility that TRMT1 could be an off-target or fortuitous substrate of Nsp5. Overall, this study will be of interest to virologist and biologists studying the role of RNA modification and RNA modifying enzyme in viral infection.

Strengths:<br /> • The authors use state-of-the-art mass spectrometry approach to quantify RNA modifications in human cells infected with SARS-CoV-2.<br /> • The authors go to great lengths to demonstrate that SARS-CoV-2 main protease, Nsp5, interacts and cleaves TRMT1 in cells and perform important controls when needed. They use a series of overexpression with strategically placed tags on both TRMT1 and Nsp5 to strengthen their observations.<br /> • The use of an inactive Nsp5 mutant (C145A) strongly supports the claim of the authors that Nsp5 is solely responsible for TRMT1 cleavage in cells.<br /> • Although the direct cleavage was not experimentally determined, the authors convincingly show that TRMT1 Q530N is not cleaved by Nsp5 suggesting that the predicted cleavage site at this position is most likely the bona fide region processed by Nsp5 in cells.<br /> • To understand the impact of TRMT1 cleavage on its RNA methylation activity, the authors rigorously test four protein constructs for their capacity not only to bind RNA but also to introduce the m2,2G modification. They demonstrate that the fragments resulting from TRMT1 cleavage are inactive and cannot methylate RNA. They further establish that the C-terminal region of TRMT1 (containing a zinc-finger domain) is the main binding site for RNA.<br /> • While 293T cells are unlikely an ideal model system to study SARS-CoV-2 infection, the authors use two cell lines and well-designed rescue experiments to uncover that TRMT1 is required for optimal SARS-CoV-2 replication.

Weaknesses:<br /> • Immunoblotting is extensively used to probe for TRMT1 degradation by Nsp5 in this study. Regretfully, the polyclonal antibody used by the authors shows strong non-specific binding to other epitopes. This complicates the data interpretation and quantification since the cleaved TRMT1 band migrates very closely to a main non-specific band detected by the antibody (for instance Fig 3A). While this reviewer is concerned about the cross-contamination during quantification of the N-TRMT1, the loss of this faint cleaved band with the TRMT1 Q530N mutant is reassuring. Nevertheless, the poor behavior of this antibody for TRMT1 detection was already reported and the authors should have taken better precautions or designed a different strategy to circumvent the limitation of this antibody by relying on additional tags.<br /> • While 293T cells are convenient to use, it is not a well-suited model system to study SARS-CoV-2 infection and replication. Therefore, some of the conclusions from this study might not apply to better suited cell systems such as Vero E6 cells or might not be observed in patient infected cells.<br /> • The reduction of bulk TRMT1 levels is minor during infection of MRC5 cells with SARS-CoV-2 (Fig 1). This does not seem to agree with the more dramatic reduction in m2,2G modification levels. Cellular Localization experiments of TRMT1 would help clarify this. While TRMT1 is found in the cytoplasm and nucleus, it is possible that TRMT1 is more dramatically degraded in the cytoplasm due to easier access by Nsp5.<br /> • In fig 6, the authors show that TRMT1 is required for optimal SARS-CoV-2 replication. This can be rescued by expressing TRMT1 (fig 7). Nevertheless, it is unknown if the methylation activity of TRMT1 is required. The authors could have expressed an inactive TRMT1 mutant (by disrupting the SAM binding site) to establish if the RNA modification by TRMT1 is important for SARS-CoV-2 replication or if it is the protein backbone that might contribute to other processes.<br /> • Fig 7, the authors used the Q530N variant to rescue SARS-CoV-2 replication in TRMT1 KO cells. This is an important experiment and unexpectedly reveals that TRMT1 cleavage by Nsp5 is not required for viral replication. To strengthen the claim of the authors that TRMT1 is required to promote viral replication and that its cleavage inhibits RNA methylation, the authors could express the TRMT1 N-terminal construct in the TRMT1 KO cells to assess if viral replication is restored or not to similar levels as WT TRMT1. This will further validate the potential biological importance of TRMT1 cleavage by Nsp5.<br /> • Fig 7, shows that the TRMT1 Q530N variant rescues SARS-CoV-2 replication to greater levels then WT TRMT1. The authors should discuss this in greater detail and its possible implications with their proposed statement. For instance, are m2,2G levels higher in Q530N compared to WT? Does Q530N co-elute with Nsp5 or is the interaction disrupted in cells?

Reviewer #3 (Public Review):

Summary:<br /> In this manuscript, the authors have used biochemical approaches to provide compelling evidence for the cleavage of TRMT1 by SARS-CoV-2 Nsp5 protease.<br /> This work is of wide interest to biochemists, cell biologists, and structural biologists in the coronavirus (CoV) field. Furthermore, it substantially advances the understanding of how CoV's interact with host factors during infection and modify cellular metabolism.

Strengths:<br /> The authors provide multiple lines of biochemical evidence to report a TRMT1-Nsp5 interaction during SARS-CoV-2 infection. They show that the host enzyme TRMT1 is cleaved at a specific site, and that it generates fragments that are incapable of functioning properly. This is an important result because TRMT1 is a critical player in host protein synthesis. This also advances our understanding of virus-host interactions during SARS-CoV-2 infections. Furthermore, this revised submission attempts to address the mechanistic role of TRMT1-Nsp5 interaction.

Weaknesses:<br /> The discussion on the enhanced viral infectivity upon expression of the non-cleavable TRMT1 is unclear. As presented, this is a bit contradictory to the suggested function of the TRMT1-Nsp5 interaction in diverting the host tRNA pools towards viral propagation. If the authors' model were correct, then one would expect a non-cleavable TRMT1 to inhibit viral infectivity because the virus would be unable to divert the host tRNA pools towards its propagation. I think this section needs to be written more clearly. But other than this, I have no further questions/suggestions for the authors.

Author Response

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

After revision, I only have a few remaining remarks:

l. 180 The authors write: We were able to process all 4 datasets with minimal adjustments to the default parameter values (Methods).

But they still don't indicate how they vary parameters and how important this is for success or how this affects absolute measurements such as average cell length. Could they give a table of parameter values and some sense of sensitivity for any future user?

We thank the reviewer for the suggestion. We see how this info is valuable for the user. We’ve added a table with the parameter values used for processing each dataset in the supplemental information, along with the default parameters for reference (lines 476 - 496). In that section we also discuss which parameters may affect the output measurements of cell size, etc.

l. 192-193 They write 'The software performed well on BACMMAN, molyso and MoMa datasets.' Naming the datasets after the analysis methods used in the original papers could be confusing, as they analyse data with MM3. Not sure how best to resolve this, maybe using first author names instead.

We thank the reviewer for pointing this out. We now refer to them with the first author names.

Related to the request of ref. #1 for a video tutorial, the video currently displayed under the github readme.md section 'Usage guide' is not functional. And the video at the top of the same page is very short with minimal information.

We thank the reviewer for letting us know the tutorial video was not functional. We’ve tested it on Linux, Mac and Windows machines on both Firefox and Chrome. We were not able to reproduce any problems for the video - could they let us know what browser / OS was used and any other specifics? If it’s easier, we can be reached through the Github page as well.

eLife assessment

This article provides a review and test of image-analysis methods for bacteria growing in the 'mother-machine' microfluidic device, introducing also a new graphical user interface for the computational analysis of mother-machine movies based on the 'Napari' environment. The tool allows users to segment cells based on two previously published methods (classical image transformation and thresholding as well as UNet-based analysis), with solid evidence for their robust performance based on comparison with other methods and use of datasets from other labs. While it was difficult to assess the user-friendliness of the new GUI, it appears to be valuable and promising for the field.

Reviewer #1 (Public Review):

The authors aim to develop an easy-to-use image analysis tool for the mother machine that is used for single-cell time-lapse imaging. Compared with related software, they tried to make this software more user-friendly for non-experts with a design of "What You Put Is What You Get". This software is implemented as a plugin of Napari, which is an emerging microscopy image analysis platform. The users can interactively adjust the parameters in the pipeline with good visualization and interaction interface.

Strengths:

- Updated platform with great 2D/3D visualization and annotation support.<br /> - Integrated one-stop pipeline for mather machine image processing.<br /> - Interactive user-friendly interface.<br /> - The users can have a visualization of intermediate results and adjust the parameters.

Weaknesses:

- Based on the presentation of the manuscript, it is not clear that the goals are fully achieved.<br /> - Although there is great potential, there is little evidence that this tool has been adopted by other labs.<br /> - the diversity of datasets used in this study is limited.<br /> - Some paragraphs in the Discussion section are like blogs with general recommendations. Although the suggestions look pretty useful, it is not the focus of this manuscript. It might be more appropriate to put it in the GitHub repo or a documentation page. The discussion should still focus on the software, such as features, software maintenance, software development roadmap, and community adoption.

A discussion of the likely impact of the work on the field, and the utility of the methods and data to the community.<br /> - The impact of this work depends on the adoption of the software MM3. Napari is a promising platform with an expanding community. With good software user experience and long-term support, there is a good chance that this tool could be widely adopted in the mother machine image analysis community.<br /> - The data analysis in this manuscript is used as a demo of MM3 features, rather than scientific research.

Reviewer #2 (Public Review):

The authors present an image-analysis pipeline for mother-machine data, i.e., for time-lapses of single bacterial cells growing for many generations in one-dimensional microfluidic channels. The pipeline is available as a plugin of the python-based image-analysis platform Napari. The tool comes with two different previously published methods to segment cells (classical image transformation and thresholding as well as UNet-based analysis), which compare qualitatively and quantitatively well with the results of widely accessible tools developed by others (BACNET, DelTA, Omnipose). The tool comes with a graphical user interface and example scripts, which should make it valuable for other mother-machine users, even if this has not been demonstrated yet.

The authors also add a practical overview of how to prepare and conduct mother-machine experiments, citing their previous work, referring to detailed instructions on their github page, and giving more advice on how to load cells using centrifugation.

Finally, the authors emphasize that machine-learning methods for image segmentation reproduce average quantities of training datasets, such as the length at birth or division. Therefore, differences in training can propagate to differences in measured average quantities. This result is not surprising but good to remember before interpreting absolute measurements of cell shape.

Author Response

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

Response to reviewers:

We would like to thank all the reviewers and the editors for their thorough and helpful feedback on our work. Before addressing specific questions and points, we would like to make a general comment on a mechanistic aspect of this study. The reviewers correctly pointed out that our study does not reveal the molecular mechanism that leads to centromeric histone depletion specifically from meiotic chromosomes. Identifying this mechanism requires a deep and thorough understanding of how centromeric histones are loaded and centromeres are established each cell cycle, and how they are maintained over time in different cell types. To our knowledge, these mechanisms have not been described in plants. To add a further layer of complexity, it appears that the mechanisms governing CENH3 maintenance may be (partially) different in plant mitotic and meiotic cells, and the mechanistic basis of this difference is unknown. Obviously, these are interesting but also complex questions and their resolution will require considerable resources and effort, which we believe is beyond the scope of this manuscript. Nevertheless, our finding that CENH3 maintenance and centromere function in meiotic cells are sensitive to heat stress is an unexpected discovery with profound implications for plant adaptation, which provides a strong incentive for further exploration of centromere maintenance mechanisms in plants.

Furthermore, we would like to apologize to reviewers for poor quality of pictures in the original submission. It was decreased by conversion to a pdf format during submission.

eLife assessment

This important study reports how heat stress affects centromere integrity by compromising the loading of the centromere protein CENH3 and by prolonging the spindle assembly checkpoint during male meiosis in Arabidopsis thaliana. The evidence supporting the claims by live cell imaging is convincing, although deeper mechanistic insight is lacking, making the study overall somewhat preliminary in nature. This work will be of interest to a broad audience of biologists working on how chromatin states are affected by stress conditions.

Public Reviews:

Reviewer #1 (Public Review):

Summary:

Khaitova and co-workers present here an analysis of centromere composition and function during elevated temperatures in the plant Arabidopsis. The work relates to the ongoing climate change during which spikes in high temperatures will be found. Hence, the paper addresses a timely subject.

The authors start by confirming earlier studies that high temperatures reduce the fertility of Arabidopsis plants. Interestingly, a hypomorphic mutant of the centromeric histone variant CENH3 (CENP-A), which was previously described by the authors, sensitizes plants to heat and results in a drop in viable pollen and silique length. The drop in fertility coincides with the formation of micronuclei in meiosis and an extension of meiotic progression as revealed by live cell imaging. Based on this finding, the authors then show that at high temperatures, the fluorescence intensity of a YFP:CENH3 declines in meiosis but remarkably not in the surrounding cells (tapetum cells). In addition, the amount of BMF1 (a Bub1 homolog and part of the spindle assembly checkpoint) also appears to decline on the kinetochores of meiocytes as judged by BMF1 reporter line. However, whether this is dependent on a decline of CENH3 or represents a separate pathway is not clear.

We provide new data in Figure S6 showing that BMF1 loading on centromeres is substantially reduced in cenh3-4 mutants. Thus, efficient tethering of BMF1 to centromeres depends on CENH3.

Finally, the authors measure the duration of the spindle checkpoint and find that it is extended under high temperatures from which they conclude that the attachment of spindle fibers to kinetochores is compromised under heat.

Strengths:

This is an interesting and important paper as it links centromere organization/function to heat stress in plants. A major conclusion of the authors is that weakened centromeres, presumably by heat, may be less effective in establishing productive interactions with spindle microtubules.

Weaknesses:

The paper does not explain the molecular reason why CENH3 levels in meiocyctes are reduced or why the attachment of spindle fibers to kinetochore is less efficient at high versus low temperatures.

While we cannot explain the molecular mechanism underlying temperature-dependent depletion of CENH3 in meiocytes, the less efficient attachment of microtubules to the kinetochores at higher temperatures is likely caused by reduced levels of CENH3, which result in smaller centromeres that are less effective in establishing productive microtubule-kinetochore attachments. Here (new Figure S6) and in our previous study (Capitao et al. 2021), we have shown that amount of centromere/kinetochore proteins is reduced at centromeres in cenh3-4 mutants, and that these plants exhibit prolonged SAC and slower chromosome biorientation.

Reviewer #2 (Public Review):

Summary:

This work investigates how increased temperature affects pollen production and fertility of Arabidopsis thaliana plants grown at selected temperature conditions ranging from 16C to 30C. They report that pollen production and fertility decline with increasing temperature. To identify the cause of reduced pollen and fertility, they resort to living cell imaging of male meiotic cells to identify that the duration of meiosis increases with an increase in temperature. They also show that pollen sterility is associated with the increased presence of micronuclei likely originating from heat stress-induced impaired meiotic chromosome segregation. They correlate abnormal meiosis to weakened centromere caused by meiosis-specific defective loading of the centromere-specific histone H3 variant (CenH3) to the meiotic centromeres. Similar is the case with kinetochore-associated spindle assembly checkpoint(SAC) protein BMF1. Intriguingly, they observe a reverse trend of strong CENH3 presence in the somatic cells of the tapetum in contrast to reduced loading of CENH3 in male meiocytes with increasing temperature. In contrast to CENH3 and BMF1, the SAC protein BMF3 persists for longer periods than the WT control, based on which authors conclude that the heat stress prolongs the duration of SAC at metaphase I, which in turn extends the time of chromosome biorientation during meiosis I. The study provides preliminary insights into the processes that affect plant reproduction with increasing temperatures which may be relevant to develop climate-resilient cultivars.

Strengths:

The authors have mastered the live cell imaging of male meiocytes which is a technically demanding exercise, which they have successfully employed to examine the time course of meiosis in Arabidopsis thaliana plants exposed to different temperature conditions. In continuation, they also monitor the loading dynamics and resident time of fluorescently tagged centromere/kinetochore proteins and spindle assembly checkpoint proteins to precisely measure the time duration of respective proteins to study their precise dynamics and function in male meiosis.

Weaknesses:

Here the authors use only one representative centromere protein CENH3, one kinetochore-associated SAC protein BMF1, and the SAC protein BMF3 to conclude that heat stress impairs centromere function and prolongs SAC with increased temperatures. Centromere and its associated protein complex the kinetochores and the SAC contain a multitude of proteins, some of which are well characterized in Arabidopsis thaliana. Hence the authors could have used additional such tagged proteins to further strengthen their claim.

Indeed, several other proteins have recently been characterized as centromere/kinetochore components and could have been included in the study to further validate the results presented. To strengthen our argument, we have added new experimental data (Figure S4) showing temperature-induced depletion of CENH3 in wild-type plants by immunocytology. Thus, we convincingly show that temperature stress reduces the amount of CENH3. This is likely to affect the loading of most kinetochore and centromeric proteins. Here (new Figure S6) and in our previous study (Capitao et al., 2021), we have shown that genetic depletion of CENH3 in cenh3-4 mutants results in reduced loading of CENPC, MIS12 and BMF1 at mitotic centromeres and reduced loading of BMF3 and BMF1 at meiotic centromeres. We also attempted to assess the levels of CENPC and MIS12 on meiotic chromosomes by immunocytology, but our antibodies, which work on mitotic spreads, did not stain meiotic chromosomes.

Though the results presented here are interesting and solid, the study lacks a deeper mechanistic understanding of what causes the defective loading of CenH3 to the centromeres, and why the SAC protein BMF3 persists only at meiotic centromeres to prolong the spindle assembly checkpoint. Also, this observation should be interpreted in light of the fact that SAC is not that robust in plants as several null mutants of plant SAC components are known to grow as healthy as wild-type plants at normal growth conditions without any vegetative and reproductive defects.

Thank you for raising this point. We are of the opinion that SAC operates and it is important in plants - we have added a citation to a preprint from the Schnittger lab (Lampou et al., 2023, BioRxiv) that was published while this manuscript was under review. We think this is the most comprehensive analysis of plant SAC to date, clearly showing that SAC delays progression to anaphase in the presence of spindle inhibitors, although adaptation eventually occurs and the cell cycle progresses. This is very similar to the situation in animals, which also undergo spindle adaptation in similar situations. The difference between plants and animals may be due to subsequent events, where plants are better able to tolerate genome instability and resume cell division in the presence of abnormal chromosome numbers. Robustness and redundancy may be another reason why plant mutants deficient in SAC do not show obvious growth retardation.

One of the immediate responses to heat stress is the production of heat shock proteins(Hsps), which act as molecular chaperones to safeguard the proteome. It will be interesting to see if the expression levels of known HsPs can be correlated with their role in stabilizing the structure of SAC proteins like BMF1 to prolong its presence at the meiotic kinetochores.

Indeed, the heat stress response is likely to be involved in this process. We sought to investigate the role of this pathway by analyzing Arabidopsis mutants deficient in HEAT-SHOCK FACTOR BINDING PROTEIN (HSBP), which acts as a negative regulator of the heat shock response. This experiment was prompted by the observation that hsbp mutants have reduced fertility. We expected that an unrestricted heat stress response might affect meiosis and pollen formation. However, our initial experiments did not show altered pollen viability in response to heat stress in hsbp plants and we did not pursue this line of research further.

Reviewer #3 (Public Review):

Summary:

Khaitova et al. report the formation of micronuclei during Arabidopsis meiosis under elevated temperatures. Micronuclei form when chromosomes are not correctly collected to the cellular poles in dividing cells. This happens when whole chromosomes or fragments are not properly attached to the kinetochore microtubules. The incidence of micronuclei formation is shown to increase at elevated temperatures in wild-type and more so in the weak centromere histone mutant cenH3-4. The number of micronuclei formed at high temperatures in the recombination mutant spo11 is like that in wild-type, indicating that the increased sensitivity of cenh3-4 is not related to the putative role of cenh3 in recombination. The abundance of CENH3-GFP at the centromere declines with higher temperature and correlates with a decline in spindle assembly checkpoint factor BMF1-GFP at the centromeres. The reduction in CENH3-GFP under heat is observed in meiocytes whereas CENH3-GFP abundance increases in the tapetum, suggesting there is a differential regulation of centromere loading in these two cell types. These observations are in line with previous reports on haploidization mutants and their hypersensitivity to heat stress.

Strengths:

This paper is an important contribution to our insights into the impact of heat stress on sexual reproduction in plants.

Weaknesses:

While it is highly significant, I struggled to interpret the results because of the poor quality of the figures and the videos.

We apologize for the poor quality of the figures. The figure resolution was drastically reduced during the conversion of the manuscript to pdf on publisher web site.

Reviewer #1 (Recommendations For The Authors):

To complete the presented analysis, it would be great to analyze the signal strength of the here-presented BMF3 reporter at high temps, see below for further reasoning.

Quantification of the BMF3 signal is difficult - it is only transiently associated with kinetochores and its level changes over time. Nevertheless, analysis of our movies taken under the same microscope settings indicates that the amount of BMF3 decreases with increasing temperature. This is illustrated in the new Figure S6C.

Conversely, how is the BMF1 and BMF3 signal strength in cenh3-4 mutants?

We performed an analysis of BMF1 and BMF3 signal in cenh3-4 mutants and observed a reduced level of signal from both proteins (Figure S6). In the case of BMF1, no signal was detectable in either somatic or meiotic cells.

How do the authors explain the reduction in BMF1 signal at 26 and 30{degree sign}C versus the extension of the duration of the SAC as measured by the persistence of a BMF3 signal (line 192: "...reduces the amount of CENH3 and the kinetochore protein BMF1 on meiotic centromeres, potentially affecting their functionality..." versus line 213: "...We observed that while the BMF3:GFP signal persisted, on average, for about 22.7 min at 21 and 26{degree sign}C, its appearance was prolonged to 40.5 min at 30{degree sign}C..."). Is the BMF3 signal also reduced at high temps (see question above)?

This is a very interesting point. While we see reduced levels of both proteins under heat stress or in cenh3-4 plants, the effect on BMF1 is much more pronounced and becomes undetectable under these conditions. This contrasts with BMF3, which appears to be reduced but is still clearly visible. These data suggest that BMF1 is more sensitive to reduced levels of CENH3 and it further corroborates the findings from the Schnittger lab that BMF1 is not the core component of SAC.

Line 18-20: The observation that heat stress reduces fertility has been made by several research teams before this study. I propose to write "confirm"/"support" etc. instead of "reveal" to avoid a (presumably not intended) false priority claim in the abstract.

We apologize, this was unintentional and we cite the relevant literature in the article. We have rewritten the abstract to avoid this impression.

Figure 2: The panel/legend appears to be a bit mixed up. Panel C is described in legend under A. In addition, I cannot find any blue arrows in panel A (which is described as panel B). Correspondingly, the references to the panels in this figure (lines 134/135 and following) need to be updated. I am also not sure how the meiocytes in this figure were stained. The dots look like centromeres but then their intensity rather increases with increasing temperature. If correct, how can this be reconciled with the authors' statement that centromeres decrease in size at higher temps?

We apologize for the mix up. An early version of the Figure was accidentally submitted and we now corrected it. The Panel B shows DAPI stained meiocytes at the tetrad stage and examples of micronuclei are indicated by arrowheads.

Line 520: Should read "genotype" not "phenotype".

Corrected

Reviewer #2 (Recommendations For The Authors):

(1) It is intriguing that heat stress impairs only the centromeres and segregation of meiotic chromosomes but not the mitotic chromosomes. No analysis of mitotic divisions is provided in the manuscript. As they have generated marker lines, it is reasonable to examine the mitotic time course as well by live monitoring of root tissues exposed to similar temperature conditions as done for meiotic analysis. This will help to address the effect of heat stress on mitotic centromeres and its comparison with meiosis will provide a better picture. There are two likely outcomes during mitosis:

(a) It is possible that the heat stress also slows down mitotic progression as well as is the case in meiosis as shown in this paper and hence it is important to examine those as well to compare and contrast the CENH3/BMF1 dynamics in mitosis and meiosis.

(b) The second scenario is that there is no effect of heat stress on the centromere integrity of mitotic chromosomes. In fact, the authors show indirect evidence in support of this wherein the eYFP: CENH3 showed a strong signal in the tapetal cells (somatic origin) surrounding the male meiocytes (generative origin). It is interesting that somatic cells of the tapetum show a strong signal whereas the meiocytes lack this. The authors should elaborate on this contrasting result.

The effect we observed seems to be specific to meiosis. We analyzed the progression of mitosis in root cells and we see a negligible effect of temperature on mitotic progression and no micronuclei formation. Interestingly, in terms of CENH3 loading, root cells show a slight decrease in CENH3 at 30°C, in contrast to the situation in tapetum cells. These and other data suggest a tissue/cell specific behavior of centromere maintenance and deserve further analysis. We plan to publish data on mitosis and tissue-specific aspects of CENH3 loading in a separate manuscript.

(2) Spindle assembly checkpoint (SAC) comprises several core proteins that are recruited to the kinetochores to correct the errors during the defective cell cycle. Here the authors demonstrate the prolonged presence of BMF3 as the only proof to claim that heat stress prolongs the spindle assembly checkpoint during metaphase I. Have the authors observed the dynamics of any other SAC core components such as MAD1, MAD2, MPS1, BUB3, and the like during heat stress?

No, we did not. We provide several independent lines of evidence that centromere structure and functionality are affected, and spindle checkpoint analysis is only one of them. At the time we designed these experiments, the only experimentally validated and well-characterized component of the SAC was BMF3, and we used only on this protein as SAC reporter because a general analysis of the SAC was not the primary goal of our study. While this paper was under review, a preprint from the Schnittger lab focusing on plant SAC was published that comprehensively analyzed these SAC components in Arabidopsis and provided a solid foundation and resources for further research in this direction. This study also uses BMF3 as a reporter for SAC in meiotic cells. It is noteworthy that despite using different microscopic methods and different plant reporter lines, our labs independently arrived at exactly the same duration of BMF3 association with the kinetochore (i.e. 22 min).

(3) Is BMF1 a component of SAC or the kinetochore? I understand that BMF1 is a part of the core SAC ( Komaki and Schnittger, 2017) although it localizes to the kinetochore. There are well-characterized kinetochore proteins in Arabidopsis such as Mis12, NUF2, NNF1, and SPC24(MUN1) which the authors could have used as a kinetochore marker. Regardless, here the authors used it as a kinetochore marker. Being a part of SAC, one would expect the prolonged presence of BMF1 similar to BMF3 in the meiotic kinetochores but it is the other way. How to explain these contrasting results?

As discussed in the public section of the review, BMF1 does not seem to be the core component of SAC. Furthermore, this protein localizes to centromeres/kinetochore throughout the cell cycle and therefore, it cannot be used as SAC reporter.

(4) Micronuclei can form as a result of chromosome missegregation as shown for spo11-1 and also due to segregation error caused by DNA repair defects. Here it is not clear what is the origin of micronuclei. It is very hard to decipher from live cell imaging. A simple meiotic spread of anthers of different treatments would address the origin of micronuclei.

Cytology cannot easily determine the origin of micronuclei in meiotic cells. Acentric fragments produced from aberrant DNA repair will still be cytologically detectable only after metaphase I as they are tethered to the remaining chromatin via cohesion. Therefore, we took advantage of spo11 mutants that do not form any meiotic breaks, and hence cannot generate acentric fragments by aberrant repair, to discriminate the origin of micronuclei. We reason that all micronuclei produced in spo11 plants originate from chromosome mis-segregation and their increase at elevated temperature support the notion that heat stress further impairs chromosome segregation.

(5) Fig.1 B The microspores are not clearly visible in the alexander-stained anthers. It is not clear which is fertile and which is sterile. A better quality picture would be ideal to appreciate the fact.

Again, we apologize for poor quality of pictures due to manuscript conversion.

Reviewer #3 (Recommendations For The Authors):

(1) In Figure 2, it should be pointed out where the micronuclei are. I see here and there a single bright spot. In Arabidopsis, we have noticed bright spots under stress conditions that are autofluorescent signals. It needs to be shown that these spots are not observed in non-GFP lines. Better image quality may help too.

The micronuclei in Figure 2 are visualized by DAPI staining, not with GFP. The nuclei are now indicated by arrowheads.

(2) It was not possible to see the centromeres in Figure 3 hence I could not verify the fluorescence intensities of CENH3 and BMF1. There is also something wrong with the color codes blue and red in fig3B, C, and D.

Again, we apologize for poor quality of pictures due to manuscript conversion.

(3) Also in the videos it would help to point out where the micronuclei are seen. At what stage were these nuclei quantified? Given that meiosis progression in the cenh3-4 mutant is slower, it may be necessary to wait long enough to see established micronuclei. This information is supposed to be presented in Figure 2C. However, the X-axis shows time, not number. So I presume Fig 2C shows the duration of meiosis stages in the mutant. In Fig 2B, it shows the number of micronuclei per lobe. However, to correlate the incidence of micronuclei formation and the frequency of polyad formation (inviable microspores), one needs the quantification of the numbers of meiocytes carrying micronuclei. Then one can correlate the number of pollen per anther (shown in Fig 1c) with the incidence of micronuclei formation. The question of whether the degree of fertility reduction is due to micronuclei formation is a major issue that should be clarified.

Then micronuclei were not quantified from the movies, but from DAPI stained whole anthers at the tetrad stage as indicated in the main text. We also apologize for confusion with the Figure 2 as we mixed up the panels in the original submission. This has been corrected in the new submission.

eLife assessment

This study is an important contribution to our insights into the impact of heat stress on sexual reproduction in plants and provides information about how centromere integrity is affected by heat stress during male meiosis in Arabidopsis thaliana. The evidence supporting the claims, specifically the dynamics of tagged proteins in meiocytes by live cell imaging is solid, even though a deeper mechanistic understanding is still lacking.

Reviewer #2 (Public Review):

Summary:

Here the authors examine how increased temperature affects pollen production and fertility of Arabidopsis thaliana plants grown at selected temperature conditions ranging from 16C to 30C. They show that pollen production and fertility decline with increasing temperature. To identify the cause of reduced pollen and fertility, they resort to living cell imaging of male meiotic cells to identify that duration of meiosis increases with an increase in temperature. They also show that pollen sterility is associated with the increased presence of micronuclei likely originating from heat stress-induced impaired meiotic chromosome segregation. They correlate abnormal meiosis to weakened centromere caused by meiosis-specific defective loading of the centromere-specific histone H3 variant (CenH3) to the meiotic centromeres. Similar is the case with kinetochore-associated spindle assembly checkpoint(SAC) protein BMF1. Intriguingly, they observe a reverse trend of strong CENH3 presence in the somatic cells of the tapetum in contrast to reduced loading of CENH3 in male meiocytes with increasing temperature. In contrast to CENH3 and BMF1, the SAC protein BMF3 persists for longer periods than the WT control, based on which authors conclude that the heat stress prolongs the duration of SAC at metaphase I, which in turn extends the time of chromosome biorientation during meiosis I. This study provides insights onto the processes that affect plant reproduction with increasing temperatures which may be relevant to develop climate-resilient cultivars.

Strengths:

This study shows that the centromere function is affected under heat stress in meiotic cells by modulating the dynamics of the centromere specific histone H3 (CENH3) that in turn compromises the assembly of kinetochore complex proteins. This they have demonstrated by the way of live cell imaging of male meiocytes by tracking the loading dynamics and resident time of fluorescently tagged centromere/kinetochore proteins and spindle assembly checkpoint proteins.

Weaknesses:

Though the results presented here are interesting and solid, the current study lacks a deeper mechanistic understanding of what causes the defective loading of CenH3 to the centromeres, and why the SAC protein BMF3 persists only at meiotic centromeres to prolong the spindle assembly checkpoint, which will be interesting to delve further to completely understand the process.

Here the authors monitor one representative centromere protein CENH3, one kinetochore-associated SAC protein BMF1, and the SAC protein BMF3 to conclude that heat stress impairs centromere/kinetochore function and prolongs SAC with increased temperatures. Centromere and its associated protein complex the kinetochores and the SAC contains a multitude of proteins, some of which are well characterized in Arabidopsis thaliana. Hence the authors could have used additional such tagged proteins to further strengthen their claim.

Reviewer #3 (Public Review):

Summary:

Khaitova et al. report the formation of micronuclei during Arabidopsis meiosis under elevated temperature. Micronuclei form when chromosomes are not correctly collected to the cellular poles in dividing cells. This happens when whole chromosomes or fragments are not properly attached to the kinetochore microtubules. The incidence of micronuclei formation is shown to increase at elevated temperature in wild type and more so in the weak centromere histone mutant cenH3-4. The number micronuclei formation at high temperature in the recombination mutant spo11 is like that in wild type, indicating that the increased sensitivity of cenh3-4 is not related to the putative role of cenh3 in recombination. The abundance of CENH3-GFP at the centromere declines with higher temperature and correlates with a decline in spindle assembly checkpoint factor BMF1-GFP at the centromeres. The reduction in CENH3-GFP under heat is observed in meiocytes whereas CENH3-GFP abundance increases in the tapetum, suggesting there is a differential regulation of centromere loading in these two cell types. These observations are in line with previous reports on haploidization mutants and their hypersensitivity to heat stress.

Strength:

The paper shows that the kinetochore function during meiosis is sensitive to high temperature and this leads to inequivalent chromosome segregation during meiosis and reduced fertility.

Weakness:

The increased sensitivity to high temperature stress of the hypomorphic mutant cenh3-4 mutant not only reduces fertility but also growth, which is not accompanied with the formation of micronuclei as in meiosis. The impact on mitosis therefore seems to be different from that in meiosis.

eLife assessment

This valuable paper by Yao, Dai, and colleagues describes a viral gene drive against herpes simplex virus 1 in cell culture. The authors provided solid evidence that an engineered gene drive sequence, expressing either spCas9 or Un1Cas12f1 nuclease, could spread efficiently in the population of wild-type viruses and induce fewer drive-resistant mutations than spCas9. Limitations include a mechanistically inaccurate title, several methodologic flaws, and a paucity of descriptions of possible therapeutic applications.

Reviewer #1 (Public Review):

Summary:

The authors developed a new viral 'gene drive' based on an alternate CRISPR Cas system: UNCas12f1. They show in HSV-1 that the gene drive virus can transmit as hypothesized and is superior to Cas9 in terms of evolutionary robustness.

Strengths:

No doubt this is an impressive technological achievement and UNCas12f1 does appear superior to Cas9 in terms of taking longer to develop resistance. This is a strong body of work and Fig 3B is the crux of the paper for me showing that resistance does take longer in terms of % of viruses that are wildtype versus UNCas12f1 gene drive. I applaud the authors and I think this is a nice technological contribution.

Weaknesses:

I will focus on major conceptual issues.

(1) Mechanism. It is not really that clear to me why the UNCas12f1 has a higher barrier to the evolution of resistance. Is this simply a temporal delay or is there something intrinsic about UNCas12f1 that does not allow resistance to arise? There is a some discussion about this but it is speculative and I could not understand why resistance would not develop.

(2) Evolution. Fig 3B is the crux of the paper for me showing that resistance does take longer in terms of % of viruses that are wildtype versus UNCas12f1 gene drive. The authors did a nice job, however, I think they need to temper the claims somewhat as longer studies (other studies typically go out to >40 days) might show resistance arising. Also, I think absolute viral titers need to be shown in addition to percentage of viruses.

(3) Therapeutic Utility. Is this proposed as a therapeutic strategy? If so, how would it work? Could it lower overall total viral burden (i.e., wt + gene drive)? Another issue that I think needs to be specifically addressed is the issue of MOI as typically HSV-1 is thought to be (i.e. shown to be) a low MOI infection in vivo and in patients, whereas this strategy appears to rely on high MOI. Overall, to me, this is probably the major weakness: i.e., whether this strategy has therapeutic potential.

(4) Title. I don't think the subordinate clause of the title "virus that 'infect' viruses" is quite correct. This needs to be be reworded. This strategy converts the viral population from wild type to a gene drive virus but "infect" does not seem accurate.