Reviewer #1 (Public review):

Summary:

In this study, Tittelmeier et al. explored the role of sphingolipid metabolism in maintaining endolysosomal membrane integrity and its downstream effects on tau aggregation and toxicity, using both worms and human cell models. The authors showed that knockdown of sphingolipid metabolism genes reduced endolysosomal membrane fluidity, as revealed by FRAP and C-Laurdan imaging, leading to increased vesicle rupture. Furthermore, tau aggregates accumulated in endolysosomes and exacerbated membrane rigidity and damage, promoting seeded tau aggregation, likely by enabling tau seed escape into the cytosol. Importantly, unsaturated fatty acid supplementation restored membrane fluidity, suppressed tau propagation, and alleviated neurotoxicity in C. elegans. These findings provide insight into how lipid dysregulation contributes to tau pathology and highlight membrane fluidity restoration as a potential therapeutic avenue for Alzheimer's disease.

Strengths:

The study addresses the connection between sphingolipid metabolism, endolysosomal membrane integrity, and tau pathology, which is a relevant topic in the context of Alzheimer's disease and related tauopathies.

The use of both C. elegans and human cell models provides cross-species perspectives that help frame the findings in a broader biological context.

The combination of FRAP and C-Laurdan dye imaging offers a biophysical approach to investigate changes in membrane properties, which is a technically interesting aspect of the study.

The observation that unsaturated fatty acid supplementation can modulate membrane fluidity and influence tau-related phenotypes adds an element of potential therapeutic interest.

The study presents multiple experimental approaches to address the proposed mechanism, and efforts were made to examine both membrane behavior and tau aggregation dynamics.

Weaknesses:

In Figure 3, the authors used C-Laurdan imaging to assess membrane fluidity and showed that knockdown of SPHK2, the human ortholog of sphk-1, led to increased membrane rigidity. However, the authors did not co-stain with a lysosomal marker, making it unclear whether the observed effect is specific to lysosomal membranes or reflects general membrane changes. Co-staining with LysoTracker or applying segmentation masks to isolate lysosomal signals would significantly improve interpretation.

Line 173 states that Lipofectamine 2000 increases membrane fluidity based on GP index changes, but this is incorrect. A higher GP index indicates increased membrane order (i.e., reduced fluidity), so the statement should be revised. Additionally, Lipofectamine 2000 can itself alter membrane rigidity, posing a risk of false-positive interpretations. To confirm the role of SPHK2 in this phenotype, the authors should use a CRISPR/Cas9 knockout model instead of relying solely on siRNA transfection, which may be confounded by the delivery reagent. Without lysosomal co-staining and SPHK2 KO validation, the authors cannot conclusively claim that SPHK2 loss affects endolysosomal membrane integrity.

The section titled "Fibrillar tau increases membrane rigidity and exacerbates endolysosomal damage" (lines 177-215) requires substantial revision. The narrative jumps abruptly between worms and cell models, making it hard to follow the logic. The use of the F3ΔK281::mCherry strain is introduced without explanation or context. It is unclear whether this strain is relevant to lysosomal membrane rupture, as no reference or justification is provided. The authors should clarify whether this reporter is intended to detect lysosomal membrane permeabilization (LMP). If so, it would be more appropriate to use established LMP reporters, such as lysosome-targeted fluorescent sensors, galectin-based reporters, or dextran leakage assays. Based on the current data in Figure 3G, it is difficult to draw firm conclusions regarding membrane rupture levels.

To support the conclusion that sphingolipid metabolism gene knockdown alters membrane properties, the study would benefit from direct lipidomic analysis. Measuring changes in sphingolipid profiles in both C. elegans and cell models would provide biochemical evidence for the proposed disruption of lipid homeostasis. Given the availability of lipidomics platforms, this type of analysis should be feasible in both worms and human cells and would significantly strengthen the mechanistic claims regarding membrane fluidity and integrity.

The conclusions of the study rely heavily on imaging-based assays, including FRAP, C-Laurdan, and fluorescence microscopy. While these approaches provide valuable spatial and qualitative insights, they are inherently indirect and subject to interpretive limitations. To strengthen the mechanistic claims, the authors should incorporate additional biochemical or quantitative approaches. For example, lipidomics would allow direct measurement of membrane lipid composition changes, and western blotting or quantitative proteomics could assess levels of membrane-associated proteins involved in endolysosomal function or stress responses. Including such data would significantly improve the robustness and reproducibility of the study's conclusions.

The human cell experiments were performed exclusively in HEK293T cells, which are not physiologically relevant for modeling Alzheimer's disease or lysosomal function in neurons. Given that the study aims to draw conclusions related to tau aggregation and lysosomal membrane integrity, the use of a more disease-relevant cellular model is essential. There are several established AD-relevant cell models, including iPSC-derived neurons, neuroblastoma lines expressing tau, or microglial models, which would better reflect the cellular context of tauopathies. Validation of key findings in at least one of these systems would substantially enhance the biological relevance and translational impact of the study.

The authors reported that PUFA supplementation rescues neurotoxic phenotypes by increasing membrane fluidity. However, the data supporting this claim rely entirely on confocal imaging, shown in both the main and supplemental figures. To substantiate the mechanistic link between PUFA treatment and improved lysosomal membrane properties, the authors should include functional assays demonstrating that PUFAs are indeed incorporated into lysosomal membranes. Additionally, lipidomics analysis would be valuable to identify which lipid species are altered upon supplementation and correlate these changes with the observed phenotypic rescue. Furthermore, the conclusion that PUFAs rescue "neurotoxic phenotypes" is not appropriate based on data derived solely from HEK293T cells, which are not neuronal. To make claims about tau-related neurotoxicity, the authors should validate their findings in a more relevant neuronal model, such as SH-SY5Y neuroblastoma cells expressing tau or iPSC-derived neurons. This would better reflect the cellular environment of Alzheimer's disease and provide stronger support for the proposed therapeutic potential of PUFA supplementation.

While the authors demonstrate that ALA supplementation mitigates neurotoxicity in C. elegans expressing aggregated tau (F3ΔK281::mCherry), the current data are not sufficient to conclude that ALA directly rescues tau aggregation toxicity via a lysosome-specific mechanism. It remains unclear how lipid composition is altered upon ALA treatment and whether these changes correlate with functional improvement of lysosomal pathways. The manuscript does not provide mechanistic insight into how ALA enhances lysosomal health or attenuates endolysosomal damage. Moreover, supplementation with PUFAs like ALA can activate a wide range of cellular processes beyond lysosomal function, including alterations in membrane fluidity, signaling cascades, and oxidative stress responses. The authors should clarify how they distinguish the lysosome-related effects from these alternative pathways. For example, did they observe specific lysosomal markers or structural improvements in lysosomes upon ALA treatment? Additional data or controls would be necessary to support a lysosome-specific protective mechanism and to exclude the involvement of other PUFA-responsive pathways in the observed phenotypes.

Reviewer #2 (Public review):

Tittelmeier et al. investigated the role of sphingolipid (SL) metabolism in the maintenance of endolysosomal vesicle integrity. They find that both impaired SL biosynthesis and degradation in C. elegans, decrease the fluidity of endolysosomal membranes and promote their rupture, while it has little effect on plasma membrane fluidity. Endolysosomal membrane fluidity is also negatively affected in human cells upon knockdown (KD) of a gene (SPHK2) involved in the SL degradation pathway. Aggregated forms of tau in both models (C. elegans and human cells) can also cause rigidification of the endolysosomal membrane, with SL homeostasis disruption having an additive effect, exacerbating endolysosomal rupture. Notably, KD of SPHK2 also increased the formation of tau foci, suggesting that compromised endolysosomal integrity may promote tau aggregation. These data provide a clearer understanding of how genetic manipulation of SL metabolism affects endolysosomal membranes and their rigidification in the context of tau aggregation. Supplementation of polyunsaturated fatty acids (PUFAs), which has a beneficial effect on Alzheimer's patients, improved membrane fluidity and reduced tau propagation in human cells and tau-associated neurotoxicity in C. elegans, suggesting a possible mechanism of action.

Overall, the conclusions of this paper are supported by the data, with a few aspects requiring further clarification and elaboration.

(1) A reference to Figure S2E-G, which shows that KD of SL biosynthesis genes do not affect the plasma membrane, is missing from the main text.

(2) In Figure 3C, lipofectamine alone shows that it increases membrane rigidity (increased GP values), not membrane fluidity.

(3) In Figure 3F, the EV cntl condition expressing F3:mCh tau should have increased LGALS3 foci compared to the mCh EV cntl according to Ref (20) and its Figure 2G (at least for Day 5 animals), which would be indicative of the tau spreading in hypodermal tissue. What C. elegans age was examined in Figure 3F? Can the authors provide evidence of the transmission of the F3:mCh tau from the touch receptor neurons to the hypodermis in the EV [similar to Figure 2C & D from Ref (20)] and compare it to the KDs? Otherwise, it seems that KD of SL genes impacts not only endolysosomal rupture but significantly affects tau accumulation/spreading as well (e.g., shown later in HEK cells, where SPHK2 KD increases the formation of tau-Venus foci).

(4) Sphingolipids are essential membrane components and signaling molecules. Does KD of SL genes in C. elegans and the subsequent endolysosomal rupture cause any major, intermediate, or minor defects/phenotypes (in non-aggregation prone models, w/t..)?

Reviewer #3 (Public review):

Summary:

The authors set off with an analysis of the lysosomal integrity upon knockdown of genes of the sphingolipid metabolic pathway that they identified in a previous (yet unpublished) work of an RNA screen using a new C. elegans Tau model. They then used cell culture and C. elegans experiments to study the link between lysosomal rupture and Tau propagation.

Strengths:

The authors use two complementary model systems and use probes to assess membrane rigidity that allow a quick assessment of the membrane dynamics and offer the opportunity to treat the cells with lipids, RNAi. Tau seeds, etc.

Weaknesses:

The main weakness is that this work builds on not-yet-peer-reviewed manuscript that established a new C. elegans Tau model and RNAi screen that aimed to identify genes involved in the propagation of Tau.

This reviewer misses essential information of the C. elegans Tau strain (not included in the method section): e.g., promoter used for the expression, information on the used Tau variant, expression pattern, and aggregation, etc.

Throughout the study, I missed data on:

(1) Effect of the knockdown on Tau expression, localisation (with lysosomal membrane?), aggregation, and proteotoxicity. The effect of the RNAi-mediated knockdown could also simply lead to a reduced expression of Tau that, in turn, leads to suppressed propagation.

(2) A quantification of RNAi knockdown is needed to judge the efficiency of the RNAi, in particular for the combinatorial RNAi experiments involving 2 and even 4 genes in parallel. Ideally, these analyses should be validated with mutants for these genes.

Further:

(3) Figure 4 H, I: Would Tau also aggregate in the absence of externally added Tau?

(4) How specific is the effect for Tau? It would help if the authors could assess other amyloid proteins.

(5) The connection between sphingolipids and AD is not new. See He et al, 2010, Neurobiol. Aging + numerous publications and also not between Tau seeding and lysosomal rupture: Rose et al., PNAS 2024 (that has been cited by the authors).

eLife Assessment

Using advanced CryoEM and mass spectrometry, the authors provide compelling evidence of how tubule formation occurs in an oxygen-dependent manner. These fundamental findings offer a novel mechanism by which rubrerythrin tubules encapsulate encapsulin to prevent oxidative stress in Pyrococcus furiosus. However, there are a few reasonable concerns about biochemical validations and the lack of adequate description of results and methodology.

Reviewer #1 (Public review):

Summary:

It is now increasingly becoming clear that macromolecules and their complexes can form larger structures such as filaments or cages in the cells under certain conditions. These can be beneficial for the cells to promote and coordinate metabolic activity or result in protection against stress. Reactive oxygen species (ROS) can be damaging to macromolecules in cells that grow both aerobically and anaerobically, and they have evolved different mechanisms to cope with ROS. Aerobic organisms have a number of enzymes to combat ROS, while anaerobic organisms have evolved other means, and one such mechanism is described by Song et al in the article.<br /> In Pyrococcus furiosus, a hyperthermophilic anaerobic bacterium, Song et al describe the formation of Oxidative stress-induced tubular structures (OSITs). Using proteomics and electron cryomicroscopy (CryoEM), the authors find that the protein Rubrerythrin is upregulated upon exposure to oxygen, and the tetramer of this protein assembles to form these tubules that are varied in length with a consistent diameter of ~480 Å. They further observe that some of these tubules also have spherical viral-like particles. With enriched fraction of the OSITs from the cells and proteomics, it is shown that the predominant protein is encapsulin, which forms a caged structure and traps ferric iron. The combined structures of OSIT by rubreerythrin and the VLPs of encapsulin protect the cells from oxygen radicals by forming a complex.

Strengths:

The combination of proteomics and electron microscopy with the employment of both tomography of cellular sections and single particle cryoEM of enriched samples.

Weaknesses:

Some description of the methods, in particular the workflow of image processing, is not easy to follow and can be described with more clarity and be easier for non-experts to read/understand.

Reviewer #2 (Public review):

The manuscript entitled "Structure of an oxygen-induced tubular nanocompartment in Pyrococcus furiosus" by Wenfei Song et al. employs whole-cell mass spectrometry and cryo-EM (including tomography, helical reconstruction, and single-particle analysis) to investigate the structure and function of the oxidoreductase Rubrerythrin (Rbr) from Pyrococcus furiosus. The study reports that under oxidative stress, Rbr forms a tubular structure, in contrast to its behaviour under anaerobic conditions. Authors characterized oxidoreductase Rubrerythrin (Rbr) from Pyrococcus furiosus under anaerobic conditions and formed a tubular structure when induced with oxidative stress. This study is well-designed. However, I have several questions related to the experimental design and the results obtained from those experiments, which are listed below.

(1) The authors have mentioned that "Under aerobic conditions, Rbr levels are 3 to 13 times higher compared to anaerobic conditions (Figures 1a-d)." Also, they performed whole-cell mass spec to measure the overexpression of the Rbr enzyme under anaerobic conditions. Thus, from the above statement, I consider the authors' claim that P. furiosus cells were cultured under anaerobic conditions and then exposed to oxidative stress. While cell growth under anaerobic conditions appears perfectly fine, the authors conducted the rest of the experiment under aerobic conditions during mass spectrometry and cryo-EM sample preparation. As a baseline, the author first grew the cells in their preferred anaerobic environment and also imaged the same cells that were exposed to air (aerobic) after anaerobic growth. The cell growth in anaerobic conditions is perfectly fine. But how did authors make sure that during anaerobic conditions, the Rbr enzyme is not expressed or not formed? As a control experiment, authors should demonstrate that during mass spec and cryo-EM sample preparations, cells are not exposed to air or maintained in an anaerobic environment. From anaerobic conditions, whenever cells were selected for spec and cryo-EM, cells were exposed to O2, and definitely controlled cells were not in anaerobic conditions anymore.

The authors collected P. furiosus wild-type or Rbr knockout cells in an anaerobic hood, but after that, they centrifuged the cells and plunged them using a Vitrobot. Are the instrument, centrifuge, and Vitrobot kept in an anaerobic environment? Recently, a few studies (anaerobic plunge-freezing in cryo-electron microscopy, Cook et al. (2024), Hands-Portman and Bakker (2022) DOI: 10.1039/D2FD00060A ) have mentioned the anaerobic plunge freeze setup for protein sample or cell freezing. I guess the authors did not use that setup. In these circumstances, the cell is already exposed to O2 during centrifugation and Vitrobot freezing. How were the control experiments properly performed in anaerobic conditions? A similar argument is true for Lamella grid preparation, where the enzyme was already exposed to O2, and single-particle grid preparation, where the purified enzyme is already exposed to O2. How were the control experiments properly performed in anaerobic conditions?

(2) It is important to provide evidence that the overexpressed protein is actually in an anaerobic condition and is later induced with more O2. Also, authors should confirm biochemically that the overexpressed protein in their desired protein "oxidoreductase Rubrerythrin (Rbr)". No biochemical data were provided in this manuscript. During single-particle analysis, the authors had to purify the protein sample and confirm that these were their desired protein samples. No biochemical or biophysical experiments were performed to confirm that the overexpressed protein is the desired protein.

(3) Figure 3, the atomic model looks different in all four tetramers. However, I have fitted the atomic model into the cryo-EM map, which looks reasonable. However, it will be easier for the reader to evaluate the model if the authors show different orientations of the atomic model, as well as if the authors could show that the atomic model fits the cryo-EM map.

(4) How did the authors select initial particle sets like 24 lakhs when forming helices and not forming isolated particles?

(5) The authors proposed a model for electron transfer upon oxidative stress. However, the data is not convincing that VLP is surrounded by Rbr and forms a tube-like structure. Generally, VLP is a sphere-like structure, and Rbr can form a tube-like structure when it interacts with spherical VLP. Rbr will surround VLP, and it will form a Rbr-decorated sphere-like structure.

(6) It will also be important to comment on the diameter of Oxidative stress-induced tubules (OSITs) and 3D reconstruction and/or helical reconstruction of purified protein samples. The spherical cyan densities within the tube are not very clear. If VLP is surrounded by Rbr (Figure 4), extra Rbr densities will be observed on VLP in the tomogram (in Figure 1). However, in the tomogram, VLP is inside Oxidative stress-induced tubules (OSITs). Figure 1 is a contradicting Figure 4. The authors should explain it properly.

(7) The authors performed helical reconstruction. Where is the Layer line calculation in helical reconstruction, and how do authors identify helical parameters for reconstruction?

(8) The authors used an extremely confusing methodology, which was very difficult to follow. The authors performed tomography, helical reconstruction, and single-particle analysis. Why did the authors need 3 different image processing methods to resolve structures that are not clear to me? The authors should also show the proper fitting between the map and the model. In Supplemental Figure 6c, the overall fitting of the subdomain looks ok. However, many peptide chains and side chains are not fitted properly in the EM density map. It will be helpful to show proper side chain fitting. In Supplementary Fig. 6a, the authors binned the data (Bin 8 or Bin 2) but did not mention when they unbinned the data for data processing. Also, the authors implemented C2 symmetry during local refinement. Why do authors suddenly use C2 symmetry expansion?

Minor Comments:

(1) The authors should properly show a schematic diagram of the enzyme subdomains. It will help to understand interactions or tetrameric assembly.

(2) The introduction is poorly written. It will really be helpful for the reader if the authors provide a proper introduction.

(3) The atomic model did not fit into the cryo-EM, so it was hard to determine the overall fitting.

(4) 17.1A pixel size? It's surprising.

(5) It will be better to calculate local resolution and show the map's angular distribution. It is obvious that resolution at the peripheral region will be poorer than core region. Therefore, it will be better to calculate local resolution. Additionally, authors should show the map to model fitting.

Reviewer #3 (Public review):

Summary:

The manuscript authored by Song et al explores the oxidative stress response of Rubrerythrin in Pyrococcus furiosus and the formation of unique tubules that also encapsulate Encapsulin VLPs. This is an excellent study employing diverse methods to comprehensively study the formation of these assemblies under oxidative stress and lays the foundation of understanding oxidative stress through the formation of tubules among redox-sensing proteins like Rubrerythrin. The authors decipher the molecular structure of the tubules and also present a high-resolution reconstruction of the rubrerythin unit that forms the OSITs.

Strengths:

The study is done thoroughly by employing methods like cryoET, single particle cryoEM, mass spectrometry, and expression analyses of knockout strains to delve into an important mechanism to counter oxidative stress. The authors perform comprehensive analyses, and this study represents a vital contribution to understanding how anaerobic organisms can respond to oxidative stress.

Weaknesses:

Not all encapsulin particles seem to be inside the OSITs. Do the authors have any insights into how the tubules sequester these viral particles? Do the VLPs have a role in nucleating the OSIT assembly, and are there interactions between VLP and OSIT surfaces? These could be points that can be discussed in greater detail by the authors.

Can the authors get a subtomogram averaging done for the encapsulin VLPs? A higher resolution reconstruction may provide potential interaction details with the OSITs, if there are any.

The role of the dense granules observed in the rubrerythrin deletion strain is not very well discussed. Is there a way these granules counter oxidative stress? The EDX scanning seems to show a Phosphate increase similar to Ca and Mg. Are these aggregates therefore likely to be calcium and Mg phosphate aggregates? This section of the paper seems incompletely analysed.

The authors should provide density and coordination distances around the diiron ions and provide a comparison with available crystal structures and highlight differences, if any, in Figure 3. Local resolution for the high-res map may be provided for Supplementary Figure 6.

Overall, this is a well-performed study with clear conclusions. The discussion points need to be improved further.

eLife Assessment

This study presents compelling evidence that the denitrosylase SCoR2 regulates cardioprotective metabolic reprogramming in the heart following ischemia/reperfusion injury. The findings are supported by a novel multi-omics approach and the integration of mouse and human data, which provides valuable insights into S-nitrosylation and cardiac metabolism. However, some conclusions remain limited by unresolved methodological issues that warrant clarification.

Reviewer #1 (Public review):

Summary:

This study shows a novel role for SCoR2 in regulating metabolic pathways in the heart to prevent injury following ischemia/reperfusion. It combines a new multi-omics method to determine SCoR2 mediated metabolic pathways in the heart. This paper would be of interest to cardiovascular researchers working on cardioprotective strategies following ischemic injury in the heart.

Strengths:

(1) Use of SCoR2KO mice subjected to I/R injury.

(2) Identification of multiple metabolic pathways in the heart by a novel multi-omics approach.

Weaknesses:

(1) Use of a global SCoR2KO mice is a limitation since the effects in the heart can be a combination of global loss of SCoR2.

(2) Lack of a cell type specific effect.

Reviewer #2 (Public review):

Summary:

This manuscript addresses the gap in knowledge related to the cardiac function of the S-denitrosylase SNO-CoA Reductase 2 (SCoR2; product of the Akr1a1 gene). Genetic variants in SCoR2 have been linked to cardiovascular disease, yet their exact role in the heart remains unclear. This paper demonstrates that mice deficient in SCoR2 show significant protection in a myocardial infarction (MI) model. SCoR2 also affected ketolytic energy production, antioxidant levels, and polyol balance through the S-nitrosylation of crucial metabolic regulators.

Strengths:

(1) Addresses a well-defined gap in knowledge related to the cardiac role of SNO-CoA Reductase 2. Besides the in-depth case for this specific player, the manuscript sheds more light on the links between S-nitrosylation and metabolic reprogramming in the heart.

(2) Rigorous proof of requirement through the combination of gene knockout and in vivo myocardial ischemia/reperfusion.

(3) Identification of precise Cys residue for SNO-modification of BDH1 as SCoR2 target in cardiac ketolysis

Weaknesses:

(1) The experiments with BDH1 stability were performed in mutant 293 cells. Was there a difference in BDH1 stability in myocardial tissue or primary cardiomyocytes from SCoR2-null vs -WT mice? The same question extends to PKM2.

(2) In the absence of tracing experiments, the cross-sectional changes in ketolysis, glycolysis, or polyol intermediates presented in Figures 4 and 5 are suggestive at best. This needs to be stressed while describing and interpreting these results.

(3) The findings from human samples with ischemic and non-ischemic cardiomyopathy do not seem immediately or linearly in line with each other and with the model proposed from the KO mice. While the correlation holds up in the non-ischemic cardiomyopathy (increased SNO-BDH1, SNO-PKM2 with decreased SCoR2 expression), how do the authors explain the decreased SNO-BDH1 with preserved SCoR2 expression in ischemic cardiomyopathy? This seems counterintuitive as activation of ketolysis is a quite established myocardial response to ischemic stress. It may help the overall message clarity to focus the human data part on only NICM patients.

(4) This issue is partially linked to point #(3). Currently, important evidence that is lacking is the demonstration of sufficiency for SCoR2 in S-nytrosylation of targets and cardiac remodeling. Does SCoR2 overexpression in the heart or isolated cardiomyocytes reduce S-nitrosylation of BDH1 and other targets, thus affecting heart function at baseline or under stress?

Reviewer #3 (Public review):

Summary:

This manuscript demonstrates that mice lacking the denitrosylase enzyme SCoR2/AKR1A1 demonstrate robust cardioprotection resulting from reprogramming of multiple metabolic pathways, revealing widespread, coordinated metabolic regulation by SCoR2.

Strengths:

(1) The extensive experimental evidence.

(2) The use of the knockout model.

Weaknesses:

(1) Lack of direct evidence for underlying mechanism(s).

(2) The mouse model used is not tissue-specific.

eLife Assessment

This important study advances our understanding of population-level immune responses to influenza in both children and adults. The strength of the evidence supporting the conclusions is compelling, with high-throughput profiling assays and mathematical modeling. The work will be of interest to immunologists, virologists, vaccine developers, and those working on mathematical modeling of infectious diseases.

Reviewer #1 (Public review):

The authors present exciting new experimental data on the antigenic recognition of 78 H3N2 strains (from the beginning of the 2023 Northern Hemisphere season) against a set of 150 serum samples. The authors compare protection profiles of individual sera and find that the antigenic effect of amino acid substitutions at specific sites depends on the immune class of the sera, differentiating between children and adults. Person-to-person heterogeneity in the measured titers is strong, specifically in the group of children's sera. The authors find that the fraction of sera with low titers correlates with the inferred growth rate using maximum likelihood regression (MLR), a correlation that does not hold for pooled sera. The authors then measure the protection profile of the sera against historical vaccine strains and find that it can be explained by birth cohort for children. Finally, the authors present data comparing pre- and post- vaccination protection profiles for 39 (USA) and 8 (Australia) adults. The data shows a cohort-specific vaccination effect as measured by the average titer increase, and also a virus-specific vaccination effect for the historical vaccine strains. The generated data is shared by the authors and they also note that these methods can be applied to inform the bi-annual vaccine composition meetings, which could be highly valuable.

The following points could be addressed in a revision:

(1) The authors conclude that much of the person-to-person and strain-to-strain variation seems idiosyncratic to individual sera rather than age groups. This point is not yet fully convincing. While the mean titer of an individual may be idiosyncratic to the individual sera, the strain-to-strain variation still reveals some patterns that are consistent across individuals (the authors note the effects of substitutions at sites 145 and 275/276). A more detailed analysis, removing the individual-specific mean titer, could still show shared patterns in groups of individuals that are not necessarily defined by the birth cohort.

(2) The authors show that the fraction of sera with a titer below 138 correlates strongly with the inferred growth rate using MLR. However, the authors also note that there exists a strong correlation between the MLR growth rate and the number of HA1 mutations. This analysis does not yet show that the titers provide substantially more information about the evolutionary success. The actual relation between the measured titers and fitness is certainly more subtle than suggested by the correlation plot in Figure 5. For example, the clades A/Massachusetts and A/Sydney both have a positive fitness at the beginning of 2023, but A/Massachusetts has substantially higher relative fitness than A/Sydney. The growth inference in Figure 5b does not appear to map that difference, and the antigenic data would give the opposite ranking. Similarly, the clades A/Massachusetts and A/Ontario have both positive relative fitness, as correctly identified by the antigenic ranking, but at quite different times (i.e., in different contexts of competing clades). Other clades, like A/St. Petersburg are assigned high growth and high escape but remain at low frequency throughout. Some mention of these effects not mapped by the analysis may be appropriate.

(3) For the protection profile against the vaccine strains, the authors find for the adult cohort that the highest titer is always against the oldest vaccine strain tested, which is A/Texas/50/2012. However, the adult sera do not show an increase in titer towards older strains, but only a peak at A/Texas. Therefore, it could be that this is a virus-specific effect, rather than a property of the protection profile. Could the authors test with one older vaccine virus (A/Perth/16/2009?) whether this really can be a general property?

Reviewer #2 (Public review):

This is an excellent paper. The ability to measure the immune response to multiple viruses in parallel is a major advancement for the field, which will be relevant across pathogens (assuming the assay can be appropriately adapted). I only have a few comments, focused on maximising the information provided by the sera.

Firstly, one of the major findings is that there is wide heterogeneity in responses across individuals. However, we could expect that individuals' responses should be at least correlated across the viruses considered, especially when individuals are of a similar age. It would be interesting to quantify the correlation in responses as a function of the difference in ages between pairs of individuals. I am also left wondering what the potential drivers of the differences in responses are, with age being presumably key. It would be interesting to explore individual factors associated with responses to specific viruses (beyond simply comparing adults versus children).

Relatedly, is the phylogenetic distance between pairs of viruses associated with similarity in responses?

Figure 5C is also a really interesting result. To be able to predict growth rates based on titers in the sera is fascinating. As touched upon in the discussion, I suspect it is really dependent on the representativeness of the sera of the population (so, e.g., if only elderly individuals provided sera, it would be a different result than if only children provided samples). It may be interesting to compare different hypotheses - so e.g., see if a population-weighted titer is even better correlated with fitness - so the contribution from each individual's titer is linked to a number of individuals of that age in the population. Alternatively, maybe only the titers in younger individuals are most relevant to fitness, etc.

In Figure 6, the authors lump together individuals within 10-year age categories - however, this is potentially throwing away the nuances of what is happening at individual ages, especially for the children, where the measured viruses cross different groups. I realise the numbers are small and the viruses only come from a small numbers of years, however, it may be preferable to order all the individuals by age (y-axis) and the viral responses in ascending order (x-axis) and plot the response as a heatmap. As currently plotted, it is difficult to compare across panels

Reviewer #3 (Public review):

The authors use high-throughput neutralisation data to explore how different summary statistics for population immune responses relate to strain success, as measured by growth rate during the 2023 season. The question of how serological measurements relate to epidemic growth is an important one, and I thought the authors present a thoughtful analysis tackling this question, with some clear figures. In particular, they found that stratifying the population based on the magnitude of their antibody titres correlates more with strain growth than using measurements derived from pooled serum data. However, there are some areas where I thought the work could be more strongly motivated and linked together. In particular, how the vaccine responses in US and Australia in Figures 6-7 relate to the earlier analysis around growth rates, and what we would expect the relationship between growth rate and population immunity to be based on epidemic theory.

eLife Assessment

This manuscript focuses on developing a structural model of how the multidomain ECM protein SVEP1 enables Angiopoietin (ANG) binding to the orphan receptor TIE1, resulting in downstream receptor phosphorylation and signaling. This is a potentially important study, however, it currently lacks key controls and is therefore incomplete. The data will be of interest to scientists working in vascular biology and RTK signaling.

Reviewer #1 (Public review):

Summary:

In this manuscript, Uphoff et al. propose a structural and mechanistic model in which the multidomain ECM protein SVEP1 enables Angiopoietin (ANG) binding to the orphan receptor TIE1, thereby promoting downstream receptor phosphorylation and signaling. Using AlphaFold-based modeling, the authors predict that the CCP20 domain of SVEP1 binds to TIE1, creating a composite surface that facilitates Angiopoietin association and TIE1 activation. The resulting ternary model (SVEP1-TIE1-ANG) offers a structural rationale for how SVEP1 converts TIE1 into a functional, ligand-responsive receptor. Additional models and biological assays suggest roles for other domains of SVEP1, such as CCP5-EGF-L7, although these interactions are predicted with low confidence. The authors interpret these findings as the first structural framework for how SVEP1 enables ANG-TIE1 signaling.

Strengths:

(1) The central hypothesis - that SVEP1 enables ANG binding to the orphan receptor TIE1 - is biologically compelling and addresses an important question in vascular biology.

(2) The AlphaFold-predicted ternary complex (SVEP1-TIE1-ANG) is plausible, high-confidence, and structurally consistent with prior functional data (e.g., poly-Ala scanning from Sato-Nishiuchi et al.).

(3) The authors' model offers a potential explanation for the previously observed role of SVEP1 in enhancing ANG signaling through TIE1, and may represent the first structural insight into TIE1's transition from orphan to ligand-activated receptor.

(4) The potential clinical implication - that a combinatorial ligand (ANG+SVEP1) can activate TIE1- could have translational relevance for vascular leak and inflammatory disease.

Weaknesses:

(1) Lack of structural validation and mechanistic follow-up:
Despite the promising AlphaFold model, there are no figures of the predicted interface, no residue-level interactions shown, no ipTM values reported, and no experimental follow-up to test the interface. PAE plots are incorrectly used as confidence justifications, which is not appropriate for complex predictions.

(2) Biophysical validation is missing:
No surface plasmon resonance (SPR), ITC, or biochemical assays are included to confirm ternary complex formation or quantify binding kinetics. Given the manuscript's structural focus, this is a major gap. For instance, an SPR experiment where ANG is immobilized, and TIE1 binding is measured {plus minus} SVEP1, would directly test the model. And allow direct comparison to ANG-TIE2.

(3) Missed opportunity for mutagenesis-driven validation:
 The manuscript does not include any interface-targeted mutations, despite clear opportunities. For example, mutating T2595 in SVEP1 (to R) or mutating the TIE1-specific residues (residues PL 202-203 to LF) could strongly test the model and potentially reveal dominant-negative behaviors. E.g. A T2595 mutant should block ANG binding but not TIE1 binding.

(4) Overinterpretation of weak models:
The additional AlphaFold model involving the CCP5-EGFL7 domains binding TIE1 has extremely low confidence (ipTM < 0.15) when reexamined by this reader and should not be emphasized. There is no biophysical evidence or binding data (SPR) to support this interaction, and its inclusion detracts from the much stronger CCP20 model.

(5) Language around modeling is overstated and potentially misleading:
Terms like "unequivocal," "high-affinity," or "affirms strong binding" in reference to AlphaFold predictions are inappropriate. These are hypotheses -not confirmations - and must be tested at the biochemical level. This should be clarified throughout the manuscript to ensure non-experts do not misinterpret modeling confidence as binding affinity.

(6) Negative stain EM data is not informative due to low resolution and lack of defined interfaces; unless replaced by higher-resolution Cryo-EM, this should be omitted. Better would be co-gel filtration, AUC, or SEC-MALLs with ANG-SVEP1-TIE1.

(7) Disjointed narrative:
The manuscript presents a compelling mechanism involving CCP20-driven ANG binding to TIE1, but then becomes fragmented by introducing the low-confidence CCP5-EGFL7 model and speculative higher-order polymerization models that are not experimentally supported.

Reviewer #2 (Public review):

Uphoff and colleagues present the results of a study focused on characterizing the binding of SVEP1 to TIE1 along with Angiopoietin-2. Starting with computational prediction of SVEP1 binding to TIE1, the authors identify the region of SVEP1 that serves as a high-affinity ligand for TIE1. Advanced studies identify a weak secondary binding site within SVEP1 that appears to be sufficient but not necessary for its interaction with TIE1 based on in vivo rescue experiments. The most novel contribution of the manuscript seems to be the identification of angiopoietin-1 and -2 as co-factors that seem to enhance the binding of SVEP1 with TIE1 and impact downstream AKT signaling. They propose a complex in which SVEP1 binds to TIE1 and ANG2.

Although the first set of results is essentially confirmatory, the identification of ANG-2 as a "co-factor" enhancing the binding of SVEP1 to TIE1 and associated downstream signaling (i.e., Figures 3 and 4) is novel and is of interest. However, the manuscript and its conclusions would greatly benefit from some clarifying details and additional experiments to ensure rigor and support specific claims.

eLife Assessment

Engineering of adeno-associated virus (AAV) replication proteins may provide new insights into Parvoviral replication. The authors created a useful collection of Rep protein variants with changes that alter the amino acid sequence, but these did not lead to clear improvements in how the virus worked. Instead, their screen showed that changes that do not alter the protein ("synonymous" mutations) and changes to the promoter were more common. As it stands the results are incomplete due to potential issues with the screening design. We encourage a more complete characterization, which may enhance the translational potential of the approach.

Reviewer #1 (Public review):

Engineering of AAV replication proteins may provide new insights into Parvoviral replication and potentially enable improved recombinant AAV vector yield when incorporated into the manufacturing process. Silberg and colleagues report an AAV Rep library, that is an interesting and powerful approach, however, the screening design and subsequent experiments lack rigor and ultimately the results are premature. Overall, the manuscript does not accurately describe state-of-the-art in the field and has significant shortcomings with experimental design/data analysis. Key concerns are noted below:

The high enrichment of P19 variants in the library was likely an artifact of the fact they only transfected 20 ng of RepCap into their 5-plate preps. When such little Rep is provided, any boost in Rep expression levels will have a major on yield. When more RepCap is provided, 10 ug in their later evaluation, small changes in Rep expression are unlikely to have major impacts on yield. A more effective strategy would have been to transfect a normal amount of DNA and then utilize serial passaging through infectious cycling to account for cross packaging.

Introduction:<br /> - There are 7 FDA approved AAV gene therapies.<br /> - The description of "shuffling" when citing Mietsczh et al is misapplied. The cited paper discusses rationally designed hybrids.<br /> - The graphic represents a hybrid capsid, but the focus is rep. As such, this should be depicted differently.

Figures 1 and 2 are validation of previously published findings and general optimization of the experimental conditions. These do not provide the reader any new insight or information.

In Figure 3: The experimental approach is limited. It is unclear how the subpooling of different conditions was performed. As mentioned above, their library transfection strategy will significantly bias the results. The enriched variants have not been evaluated - specifically, the enriched non-synonymous mutations have not been shown to yield higher titers when tested individually.<br /> In Figure 4: The claim is made that "several synonymous mutations within the p19 promoter region increase Rep DNA packaging activity." However, Figure 4c does not show statistically significant differences in support of this claim. Additional supporting data is needed. Further, Authors state that the synonymous mutations are near the P19 promoter. However, looking at the sequence shown in figure 4, their annotation of the P19 promoter is not correct and the mutations are actually within the P19 promoter. Relatedly, the authors note that mutations enriched in the p19 region include additional tetranucleotide repeats. No synthetic variants with additional GCTCs have been generated to test this hypothesis. Further, these results would benefit from a Western blot and transcript analysis to confirm Rep52/40, expression levels of constructs.

Reviewer #2 (Public review):

In the present study the authors have investigated the effects of mutations on Rep protein ability to package DNA within the gene therapy vector, AAV. A detailed investigation of Rep mutants selected from a library has been probed for their ability to produce active virions. While the concept is interesting the outcome effects are very limited.

The major issue is the lack of immediate applicability and relevance in the vector production pipeline for AAV. The authors have found that with the synthetic GFP transgene cargo, mutations of the p19 promoter did not lead to enhanced AAV vector packing. Thus the data is quite preliminary and a complete characterization may be necessary to further enhance the translational potential of the approach.

Reviewer #3 (Public review):

While the AAV capsid has long been the target of protein engineering, its Rep proteins have been comparatively less studied. Since Rep plays a variety of roles for genome replication and virion packaging, gaining a deeper mechanistic understanding of their function and/or engineering versions that enable higher packaging productivity would be of interest to the field. This study generates a library of non-synonymous mutations in AAV2 rep (intended to cover all 19 aa changes at all positions, and coming close), packaged an AAV with AAV9 capsid, and sequenced the results to assess which amino acid changes resulted in an enrichment/depletion of genomes containing that variant rep. They found that proline substitutions are disruptive, well known from protein mutagenesis studies. The most significant enrichment sfound, however, were a set of synonymous mutations (unintended members of the library, as the library was designed to contain non-synonymous mutations) that lie within the p19 promoter. However, attempts to package recombinant vector using these individual rep variants in the AAV helper construct did not increase viral titer.

A previous study conducted analogous mutagenesis on Rep: Jain et al., "Comprehensive mutagenesis maps the effect of all single-codon mutations in the AAV2 rep gene on AAV production" eLife 2024 (cited here as reference 19). It is not clear that this current study is a significant advance relative to the prior, quite comprehensive study. Both generated a library of non-synonymous mutations and observed fitness effects on Rep. Because this study sequenced the full rep, rather than barcodes associated with each rep variant, it found the enrichment in the synonymous mutations. However, these should ideally advance a basic understanding of Rep biology and/or result in better AAV production, but they did neither. It is speculated in the Discussion that the mutations generated additional GCTC motifs in p19, elements that mediate protein-DNA interactions. However, the role of GCTC motifs is speculative, and no transcriptional analysis is conducted. Furthermore, as discussed above, the mutations do not result in higher viral titers. Perhaps there's a transcriptional effect at the much lower copy numbers of vector genome present during library selection vs. rAAV packaging. They also found stop codons in Rep domains thought to be required for viral packaging, but functional studies confirming the screening findings are not conducted. As a result, the biological or technical relevance of the findings are extremely unclear, and thus the impact is relatively low.

The description of herring DNA co-transfection and cross-packaging (which is a well-known pitfall) are somewhat technical and arguably don't merit too much main manuscript attention.

eLife assessment

This important study investigates the adaptability of prey capture by archerfish, which hunt insects by spitting at them and then rapidly turning to reach their landing point on the water surface. The results of elaborate behavioral experiments and measurements show that, even though the visuomotor behavior unfolds very rapidly (in less than 100 ms), it is not hardwired and can adapt to different simulated physics and different prey shapes. The data are convincing and should be of relevance to those interested in rapid decision making in general, beyond the archerfish model.

Reviewer #1 (Public review):

Summary:

The authors test whether the archerfish can modulate the fast response to a falling target. By manipulating the trajectory of the target, they claim that the fish can modulate the fast response. While it is clear from the result that the fish can modulate the fast response, the experimental support for argument that the fish can do it for a reflex like behavior is inadequate.

Strengths:

Overall, the question that the authors raised in the manuscript is interesting.

Weaknesses:

Major comments:

(1) The argument that the fish can modulate reflex-like behavior relies on the claim that the archerfish makes the decision in 40 ms. There is little support for the 40 ms reaction time. The reaction time for the same behavior in Schlegel 2008, is 60-70 ms and in Tsvilling 2012 about 75 ms, if we take the half height of the maximum as estimated reaction time in both cases. If we take the peak (or average) of the distribution as an estimation of reaction time, the reaction time is even longer. This number is critical for the analysis the authors perform since if the reaction time is longer, maybe this is not a reflex as claimed. In addition, mentioning the 40 ms in the abstract is overselling the result. The title is also not supported by the results.

(2) A critical technical issue of the stimulus delivery is not clear. The frame rate is 120 FPS and the target horizontal speed can be up to 1.775 m/s. This produces target jumping on the screen 15 mm each frame. This is not a continuous motion. Thus, the similarity between the natural system where the target experience ballistic trajectory and the experiment here is not clear. Ideally, another type of stimulus delivery system is needed for a project of this kind that requires fast moving targets (e.g. Reiser, J. Neurosci.Meth. 2008). In addition, the screen is rectangular and not circular, so in some directions the target vanishes earlier than others. It must produce a bias in the fish response but there is no analysis of this type.

(3) The results here rely on the ability to measure the error of response in the case of virtual experiment. It is not clear how this is done since the virtual target does not fall. How do authors validate that the fish indeed perceives the virtual target as falling target? Since the deflection is at a later stage of the virtual trajectory, it is not clear what is the actual physics that governs the world of the experiment. Overall, the experimental setup is not well designed.

Comments on revisions:

The authors handled the comments, and the manuscript has improved accordingly. While some issues could still benefit from further clarification and depth, the current version meets the necessary standards.

Reviewer #2 (Public review):

Summary:

This manuscript studies the prey capture by archer fish, which observe the initial values of motion of aerial prey they made fall by spitting on them, and then rapidly turn to reach the ballistic landing point on the water surface. The question raised by the article is whether this incredibly fast decision-making process is hardwired and thus unmodifiable or can be adjusted by experience to follow a new rule, namely that the landing point is deflected from a certain amount from the expected ballistic landing point. The results show that the fish learn the new rule and use it afterwards in a variety of novel situations that include height, side and speed of the prey, and which preserve the speed of the fish's decision. Moreover, a remarkable finding presented in this work is the fact that fish that have learned to use the new rule can relearn to use the ballistic landing point for an object based on its shape (a triangle) while keeping simultaneously the 'deflected rule' for an object differing in shape (a disc); in other words, fish can master simultaneously two decision-making rules based on the different shape of objects.

Strengths:

The manuscript relies on a sophisticated and clever experimental design that allows changing the apparent landing point of a virtual prey using a virtual reality system. Several robust controls are provided to demonstrate the reliability and usefulness of the experimental setup.

Overall, I like very much the idea conveyed by the authors that even stimuli triggering apparently hardwired responses can be relearned in order to be associated to a different response, thus showing the impressive flexibility of circuits that are sometimes considered as mediating pure reflexive responses. This is the case - as an additional example - of the main component of the Nasanov pheromone of bees (geraniol), which triggers immediate reflexive attraction and appetitive responses, and which can, nevertheless, be learned by bees in association with an electric shock so that bees end up exhibiting avoidance and the aversive response of sting extension to this odorant(1), which is a fully unnatural situation, and which shows that associative aversive learning is strong enough to override preprogrammed responding, thus reflecting an impressive behavioral flexibility.

Weaknesses:

As a general remark, there is some information that I missed and that are mandatory in the analysis of behavioral changes: one is the variability in the performances displayed. The authors mentioned that the results reported come from 6 fish (which is a low sample size). How were the individual performances in terms of consistency? Were all fish equally good in adjusting/learning the new rule? How did errors vary according to individual identity? It seems to me that this kind of information should be available as the authors reported that individual fish could be recognized and tracked (see lines 620-635) and is essential for appreciating the flexibility of the system under study.

The other information that I could not find explained in a proper way is referred to the speed of the learning process. Admittedly, fish learn in an impressive way the new rule and even two rules simultaneously; yet, how long did they need to achieve this? In the article, Fig 2 mention that at least 6 training stages (each defined as a block of 60 evaluated turn decisions, which actually shows that the standard term 'Training Block' would be more appropriate) were required for the fish to learn the 'deflected rule'. While this means 360 trials (turning starts), I was left with the question of how long did this process last? How many hours, days, weeks were needed for the fish to learn? And as mentioned above, were al fish equally fast in learning? I would appreciate explaining this very important point because learning dynamics is relevant to understanding the flexibility of the system.

Comments After Revision:

There was consensus among reviewers that the authors addressed the initial critiques adequately and that the manuscript improved accordingly. The revision clarified several methodological aspects, and the addition of the new Fig. 2 was particularly helpful. It elucidates the experimental approach used in the study and offers essential context for understanding points that may have been unclear in the previous version.

Author response:

The following is the authors’ response to the original reviews

eLife Assessment

This valuable study investigates prey capture by archer fish, showing that even though the visuomotor behavior unfolds very rapidly (within 40-70 ms), it is not hardwired; it can adapt to different simulated physics and different prey shapes. Although there was agreement that the model system, experimental design, and main hypothesis are certainly interesting, opinions were divided on whether the evidence supporting the central claims is incomplete. A more rigorous definition and assessment of "reflex speed", more detailed evidence of stimulus control, and a more detailed analysis of individual subjects could potentially increase confidence in the main conclusions.

Thank you very much. There are several points that we had to absolutely make sure that they are very well understood. (1) Explaining in the best possible way the experiment with a fly sliding on top of a glass plate. Here, the virtual ballistic landing point can be calculated using simple high school physics. It turns out that this is where the fish turn to – even though the fly is not falling at all. Once this is understood it becomes clear that we can precisely measure latency and accuracy of the C-start turns. In the new version we explain this essential aspect in more detail and add an extra Figure (new Figure 2). This may, perhaps, help readers to notice this important background (previously covered in Fig. 1C). (2) The full experimental evidence that the VR method works is presented in more detail and all measurements necessary will be clear after the new Figure 2. They will however not be clear if this Figure is ignored. (3) We have rewritten the manuscript to make it easier to understand what we wanted to show, why we needed VR to proceed and why the archerfish highspeed decision lent itself so readily to tackle the problem. (4) We emphasize the importance of speed-accuracy tradeoffs in standard decision-making and also include data on the absence of such a relation in the archerfish highspeed decisions.

So, in summary, we have emphasized what we wanted to show and what we did not want to show, we have rewritten the text to make it easier for future readers and we have tried to add more guidance to the figures. We do hope very much that the beauty of the quite unexpected findings is more easily visible to those who take the trouble of actually reading the paper.  

Public Reviews:

Reviewer #1 (Public review):

Summary:

The authors test whether the archerfish can modulate the fast response to a falling target. By manipulating the trajectory of the target, they claim that the fish can modulate the fast response. While it is clear from the result that the fish can modulate the fast response, the experimental support for the argument that the fish can do it for a reflex-like behavior is inadequate.

Please note that we have not simply tested whether archerfish can 'modulate the fast response'. We quantitatively test specific hypotheses on the rules used by the fish. For this the accuracy of the decisions is analyzed with respect to specific points that can be calculated precisely in each of the experiments. These points are shown on the figures and in the movies that were meant to illustrate this important aspect. We had to make sure that the way we calculate the predicted point(s) is made as clear as possible in the text. We added more text and separated the fundamentally important aspects in a separate Figure 2 to make it more difficult to overlook the fundamental aspects that lay the foundation for everything that follows.

Strengths:

Overall, the question that the authors raised in the manuscript is interesting.

Thank you and we do hope very much that, with our revision, you will see the beauty of the findings.

Weaknesses:

(1) The argument that the fish can modulate reflex-like behavior relies on the claim that the archerfish makes the decision in 40 ms. There is little support for the 40 ms reaction time. The reaction time for the same behavior in Schlegel 2008, is 6070 ms, and in Tsvilling 2012 about 75 ms, if we take the half height of the maximum as the estimated reaction time in both cases. If we take the peak (or average) of the distribution as an estimation of reaction time, the reaction time is even longer. This number is critical for the analysis the authors perform since if the reaction time is longer, maybe this is not a reflex as claimed. In addition, mentioning the 40 ms in the abstract is overselling the result. The title is also not supported by the results.

Although the minimum latency is indeed 40 ms (it can be slightly less: e.g., see the evidence in the paper, for instance the plots in the new Fig. 4) the paper's statements are not dependent on a specific number. Even if minimum latency was 100 ms (which it is not) the speed of the response and the absence of a speedaccuracy relation (now shown directly in Fig. 4) is what is of importance. To show this we have completely rewritten large parts of the manuscript.

(2) A critical technical issue of the stimulus delivery is not clear. The frame rate is 120 FPS and the target horizontal speed can be up to 1.775 m/s. This produces a target jumping on the screen 15 mm in each frame. This is not a continuous motion. Thus, the similarity between the natural system where the target experiences ballistic trajectory and the experiment here is not clear. Ideally, another type of stimulus delivery system is needed for a project of this kind that requires fast-moving targets (e.g. Reiser, J. Neurosci.Meth. 2008). In addition, the screen is rectangular and not circular, so in some directions, the target vanishes earlier than others. It must produce a bias in the fish response but there is no analysis of this type.

Please note that the new Fig. 3 (former Fig. 2) reports all the evidence that is needed to just show this and in a way that could in no way have been better. We have rewritten the text to explain what needs to be shown experimentally in order to be able to proceed, what critical tests were done and what results were obtained. We also add a short comment on another unsuccessful attempt that we have tried before.

(3) The results here rely on the ability to measure the error of response in the case of a virtual experiment. It is not clear how this is done since the virtual target does not fall. How do the authors validate that the fish indeed perceives the virtual target as the falling target? Since the deflection is at a later stage of the virtual trajectory, it is not clear what is the actual physics that governs the world of the experiment. Overall, the experimental setup is not well designed.

Understanding this aspect is essential. If the glass plate experiment is not thoroughly understood (new Fig. 2 with new text to emphasize that this is absolutely essential) nothing that follows makes any sense, including what is meant by the statement that the decision could be hardwired to ballistic motion.

Reviewer #2 (Public review):

Summary:

This manuscript studies prey capture by archer fish, which observe the initial values of motion of aerial prey they made fall by spitting on them, and then rapidly turn to reach the ballistic landing point on the water surface. The question raised by the article is whether this incredibly fast decision-making process is hardwired and thus unmodifiable or can be adjusted by experience to follow a new rule, namely that the landing point is deflected from a certain amount of the expected ballistic landing point. The results show that the fish learn the new rule and use it afterward in a variety of novel situations that include height, side, and speed of the prey, and which preserve the speed of the fish's decision. Moreover, a remarkable finding presented in this work is the fact that fish that have learned to use the new rule can relearn to use the ballistic landing point for an object based on its shape (a triangle) while keeping simultaneously the 'deflected rule' for an object differing in shape (a disc); in other words, fish can master simultaneously two decisionmaking rules based on the different shape of objects.

Strengths:

The manuscript relies on a sophisticated and clever experimental design that allows changing the apparent landing point of a virtual prey using a virtual reality system. Several robust controls are provided to demonstrate the reliability and usefulness of the experimental setup.

Overall, I very much like the idea conveyed by the authors that even stimuli triggering apparently hardwired responses can be relearned in order to be associated with a different response, thus showing the impressive flexibility of circuits that are sometimes considered mediating pure reflexive responses.

Thank you so much for this precise assessment of what we have shown!

This is the case - as an additional example - of the main component of the Nasanov pheromone of bees (geraniol), which triggers immediate reflexive attraction and appetitive responses, and which can, nevertheless, be learned by bees in association with an electric shock so that bees end up exhibiting avoidance and the aversive response of sting extension to this odorant (1), which is a fully unnatural situation, and which shows that associative aversive learning is strong enough to override preprogrammed responding, thus reflecting an impressive behavioral flexibility.

That's very interesting, thanks and we are very happy to mention this important study in the revised version.

Weaknesses:

As a general remark, there is some information that I missed and that is mandatory in the analysis of behavioral changes.

Firstly, the variability in the performances displayed. The authors mentioned that the results reported come from 6 fish (which is a low sample size). How were the individual performances in terms of consistency? Were all fish equally good in adjusting/learning the new rule? How did errors vary according to individual identity? It seems to me that this kind of information should be available as the authors reported that individual fish could be recognized and tracked (see lines 620-635) and is essential for appreciating the flexibility of the system under study.

Secondly, the speed of the learning process is not properly explained. Admittedly, fish learn in an impressive way the new rule and even two rules simultaneously; yet, how long did they need to achieve this? In the article, Figure 2 mentions that at least 6 training stages (each defined as a block of 60 evaluated turn decisions, which actually shows that the standard term 'Training Block' would be more appropriate) were required for the fish to learn the 'deflected rule'. While this means 360 trials (turning starts), I was left with the question of how long this process lasted. How many hours, days, and weeks were needed for the fish to learn? And as mentioned above, were all fish equally fast in learning? I would appreciate explaining this very important point because learning dynamics is relevant to understanding the flexibility of the system.

First, it is very important to keep the question in mind that we wanted to clarify: Does the system have the potential to re-tune the decisions to other non-ballistic relations between the input variables and the output? This would have been established if one fish was found capable of doing that. We have rewritten the introduction and discussion to specifically say what our aim was. We feel that the paper is already extremely long and difficult to understand (even after we tried very hard in this revision to explain everything in detail and as good as we could), requires the establishment of a method whose success was really unexpected and finding a degree of plasticity that we did not expect at all. We also have added a section in the discussion stating what we can, and we cannot say given the number of fish examined. For instance, we do not know if there are differences in the speed at which the different individuals mastered the new rules and if social learning could play a role to speed up the acquisition. That is a brilliant idea and we are very interested in checking this - but we wanted to stick with the (quite ambitious) goal of the present study.

Reference:

(1) Roussel, E., Padie, S. & Giurfa, M. Aversive learning overcomes appetitive innate responding in honeybees. Anim Cogn 15, 135-141, doi:10.1007/s10071011-0426-1 (2012).

Thanks for this reference!

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

Minor comments:

(1) What is the difference between Reinel, J. Exp. Bio. 2016 and the current study?

Clearly in that study all objects were strictly falling ballistically, and latency and accuracy of the turn decisions were determined when the initial motion was not only horizontal but had an additional vertical component of speed. The question of that study was if the need to account to an additional variable (vertical speed) in the decision would affect its latency or accuracy. The study showed that also then archerfish rapidly turn to the later impact point. It also showed that accuracy and latency were not changed by the added degree of freedom.

(2) How do Figures 2 F and G demonstrate that an accurate start is possible?

See above.

(3) Figure 4 is hard to follow, it is not clear what is presented and how it supports the claim that the new rule is represented in a way that allows immediate generalization.

Yes, this is not at all an easy experiment. Briefly, fish were re-trained at only one height level and then are tested at other levels. The strategy is as in the experiments Schuster et al. 2004 Current Biology, Vol. 14, 1565–1568, Figure 5. We have changed text and Figure (new Figure 5) to show how the predictions were reached.

Reviewer #2 (Recommendations for the authors):

Minor remarks

Lines 88-90: I was surprised to see that in this section, the authors did not mention the speed-accuracy trade-off off which has inspired numerous experiments in animal behavior (1). This could be used to back their point, namely, that speed comes with an apparent cost of a loss in accuracy.

Yes, that is a crucial aspect that was completely missing even though it demonstrates a key aspect of 'standard' versus some 'highspeed' decisions! We definitely had to include it and also to show, directly under the conditions of our present experiments (in the new Fig. 4) the absence of a significant speedaccuracy relation for the archerfish highspeed decisions! Thank you very much for emphasizing this crucial aspect!

Lines 182-184: Specify that this situation corresponds to the hatched bar in Figure (this can be specified in the caption of the figure, where the bar is not mentioned).

Thanks!

Lines 187-188: here and elsewhere (e.g. lines 224-225, etc), the error made by the fish is presented in cm (see Figure 2 where the inset shows how the error was computed). I wonder if it would not be more appropriate to present it in terms of the angular difference between the trajectory made by the fish and the food delivery location.

Angles could also be used, but because of the large variation in initial distances (that we wanted to make sure that the fish had to capture a rule, allowing them to respond from various distances) another measure was used that we found somehow more natural: it is simply how close a fish would get to the landing point if it continued in the direction assumed after the turn. Although we describe how we defined accuracy we did not discuss why this measure was used in this (and many previous studies). We are very happy to add this. Please also note that running all tests based on angular errors (which we also have done throughout to ensure that the conclusions are independent on an arbitrary measure of the error) leads to no different conclusion. We have added a brief explanation in the text and in the new Fig. 2.

Lines 299-323: Is it my impression or did fish have more trouble in generalizing their learned rule to the condition untrained larger height (see for instance red curves in Figures 4 D, E, G)? Could the authors elaborate on this point?

We changed the code to make this more clear. The red curves (before marked A to highlight impact point option A) correspond to the errors to the ballistic impact point without deflection, so what would have to be compared are the black curves (marked P to highlight the virtual impact point that should be chosen had the fish immediately generated to the untrained conditions). We have rewritten the text and the labels in the Figure (now Figure 5) to illustrate the predictions and to name them in more helpful ways and so that they can't be confused with panel labels. At any rate, what needs to be compared, to check the idea, are the black curves, and these are not statistically different between both heights (p=0.525, Mann-Whitney). Interestingly, none of the black curves from all panels (D-G) differ (p>0.3).

Line 559: if we are speaking here about luminance contrast, it should read 'Michelson Contrast' rather than 'Michelsen Contrast'.

Absolutely, thanks!

References

(1) Chittka, L., Skorupski, P. & Raine, N. E. Speed-accuracy tradeoffs in animal decision making. Trends Ecol Evol 24, 400-407, doi:10.1016/j.tree.2009.02.010 (2009).

An excellent paper that helps to stress our main question

eLife Assessment

This manuscript reports a useful computational study of information encoding across the monkey prefrontal and pre-motor cortices during decision making. While many of the conclusions are supported with solid analyses, the evidence for the main interpretation of the results, the role of an information bottleneck across areas, is not complete. The results will be of interest to a systems and computational neuroscience audience.

Reviewer #1 (Public review):

In this study the authors aim to understand why decision formation during behavioural tasks is distributed across multiple brain areas. They hypothesize that multiple areas are used in order to implement an information bottleneck (IB). Using neural activity recorded from monkey DLPFC and PMd performing a 2-AFC task, they show that DLPFC represents various task variables (decision, color, target configuration), while downstream PMd primarily represents decision information. Since decision information is the only information needed to make a decision, the authors suggest that PMd has a minimal sufficient representation (as expected from an IB). They then train 3-area RNNs on the same task, and show that activity in the first and third areas resemble the neural representations of DLPFC and PMd, respectively. In order to propose a mechanism, they analyse the RNN and find that area 3 ends up with primarily decision information because feedforward connections between areas primarily propagate decision information.

Overall, the paper reads well and the data analysis and RNN modeling are well done and mostly correct. I agree with the authors that PMd has less information than DLPFC, meaning that some of the target and color information is attenuated. I also agree that this also happens in their multi-area RNN.

However, I find the use of the IB principle here muddles the water rather than clarifying anything. The key problem is that the authors evoke the information bottleneck in a mostly intuitive sense, but they do not actually use it (say, in their modelling). Rather, the IB is simply used to motivate why information will be or should be lost. Since the IB is a generic compressor, however, it does not make any statements about how a particular compression should be distributed or computed across brain areas.

If I ignore the reference to the information bottleneck, I still see a more mechanistic study that proposes a neural mechanism of how decisions are formed, in the tradition of RNN-modelling of neural activity as in Mante et al 2013. Seen through this more limited sense, the present study succeeds at pointing out a good model-data match.

Major points

(1) The IB is a formal, information-theoretic method to identify relevant information. However, in the paper, reference to the information bottleneck method (IB) is only used to motivate why (task-irrelevant) information should be lost in higher areas. The IB principle itself is actually never used. The RNNs are fitted using standard techniques, without reference to the IB. Without a formal link, I think the authors should describe their findings using words (e.g., task-irrelevant information is lost), rather than stating this as evidence for an information-theoretic principle.

(2) The advantage of employing a formal theory is that all assumptions have to be clarified. Since the authors only evoke the IB, but never employ it, they refrain from clarifying some of their assumptions. That is what creates unnecessary confusion.

For instance, the authors cite the following predictions of the IB principle: "(1) There exists a downstream area of cortex that has a minimal and sufficient representation to perform a task ... (2) there exists an upstream area of cortex that has more task information than the minimal sufficient area" - However, since the information bottleneck method is a generic compressor, it does not make any predictions about areas (or neurons). For a given sensory input p(x), a given task output p(y|x), and a given information loss, the IB generates exactly one optimal representation. In other words, the predictions made by the authors relie on other assumptions (e.g. feedforward processing, hierarchy, etc.) and these are not clearly stated.

(3) A corrollary to this problem is that the authors do not formally define task-irrelevant information. It seems the authors simply use the choice or decision as the thing that needs to be computed, and identify all other information as task-irrelevant. That's at least what I glean from the RNN model. However, I find that highly confusing because it suggests the conclusion that color information or target information are task-irrelevant. Surely, that cannot be true, since the decision is based on these quantities!

(4) If we define the output as the only task-relevant information, then any representation that is a pure motor representation would qualify as a minimal sufficient representation to carry out the correct actions. However, it is well-known that sensory information is lost in motor areas. It is not clear to me what exactly we gain by calling motor representations "minimal sufficient representations."

In summary, I think the authors should refrain from evoking the IB - which is a formal, mathematical principle - unless they actually use it formally as well.

Reviewer #2 (Public review):

This study advances our understanding of information encoding in the DLPFC and PMD brain regions. The conclusions are supported with convincing and robust analyses conducted on monkey datasets and trained RNN models. However, there are some concerns regarding the interpretation of findings related to the information bottleneck theory and the mapping of brain areas in the RNN simulations.

The authors' justification regarding mapping between model areas and anatomical areas remains insufficient, in my opinion. However, I recognize that my initial critique may not have been fully clear. The issue I see is this: whichever area is mapped to the first RNN module will trivially exhibit stimulus information, and downstream regions will naturally show a gradual loss of that information if one simply reads out their responses.

Thus, the observed stimulus loss in later modules could be an inevitable consequence of the model's structure, rather than a meaningful analog to the PFC-PMd transition. This point requires more careful justification or a reevaluation of the proposed mapping.

eLife Assessment

This important work substantially advances our understanding of reactive oxygen species (ROS) as a regenerative signal during postnatal cerebellum repair by activating adaptive progenitor reprogramming. The evidence supporting the conclusions is compelling, with rigorous genomic assays and in vivo analyses. This work will be of broad interest to biologists working on stem cells, neurodevelopment and regenerative medicine.

Reviewer #1 (Public review):

Summary:

In this manuscript, Pakula et al. explore the impact of reactive oxygen species (ROS) on neonatal cerebellar regeneration, providing evidence that ROS activates regeneration through Nestin-expressing progenitors (NEPs). Using scRNA-seq analysis of FACS-isolated NEPs, the authors characterize injury-induced changes, including an enrichment in ROS metabolic processes within the cerebellar microenvironment. Biochemical analyses confirm a rapid increase in ROS levels following irradiation and forced catalase expression, which reduces ROS levels, and impairs external granule layer (EGL) replenishment post-injury.

Strengths:

Overall, the study robustly supports its main conclusion and provides valuable insights into ROS as a regenerative signal in the neonatal cerebellum.

Comments on revisions:

The authors have addressed most of the previous comments. However, they should clarify the following response:

*"For reasons we have not explored, the phenotype is most prominent in these lobules, that is why they were originally chosen. We edited the following sentence (lines 578-579):

First, we analyzed the replenishment of the EGL by BgL-NEPs in vermis lobules 3-5, since our previous work showed that these lobules have a prominent defect."*

It has been reported that the anterior part of the cerebellum may have a lower regenerative capacity compared to the posterior lobe. To avoid potential ambiguity, the authors should clarify that "the phenotype" and "prominent defect" refer to more severe EGL depletion at an earlier stage after IR rather than a poorer regenerative outcome. Additionally, they should provide a reference to support their statement or indicate if it is based on unpublished observations.

Reviewer #2 (Public review):

Summary:

The authors have previously shown that the mouse neonatal cerebellum can regenerate damage to granule cell progenitors in the external granular layer, through reprogramming of gliogenic nestin-expressing progenitors (NEPs). The mechanisms of this reprogramming remain largely unknown. Here the authors used scRNAseq and ATACseq of purified neonatal NEPs from P1-P5 and showed that ROS signatures were transiently upregulated in gliogenic NEPs ve neurogenic NEPs 24 hours post injury (P2). To assess the role of ROS, mice transgenic for global catalase activity were assessed to reduce ROS. Inhibition of ROS significantly decreased gliogenic NEP reprogramming and diminished cerebellar growth post-injury. Further, inhibition of microglia across this same time period prevented one of the first steps of repair - the migration of NEPs into the external granule layer. This work is the first demonstration that the tissue microenvironment of the damaged neonatal cerebellum is a major regulator of neonatal cerebellar regeneration. Increased ROS is seen in other CNS damage models, including adults, thus there may be some shared mechanisms across age and regions, although interestingly neonatal cerebellar astrocytes do not upregulate GFAP as seen in adult CNS damage models. Another intriguing finding is that global inhibition of ROS did not alter normal cerebellar development.

Strengths:

This paper presents a beautiful example of using single cell data to generate biologically relevant, testable hypotheses of mechanisms driving important biological processes. The scRNAseq and ATACseq analyses are rigorously conducted and conclusive. Data is very clearly presented and easily interpreted supporting the hypothesis next tested by reduce ROS in irradiated brains.

Analysis of whole tissue and FAC sorted NEPS in transgenic mice where human catalase was globally expressed in mitochondria were rigorously controlled and conclusively show that ROS upregulation was indeed decreased post injury and very clearly the regenerative response was inhibited. The authors are to be commended on the very careful analyses which are very well presented and again, easy to follow with all appropriate data shown to support their conclusions.

Weaknesses:

The authors also present data to show that microglia are required for an early step of mobilizing gliogenic NEPs into the damaged EGL. While the data that PLX5622 administration from P0-P5 or even P0-P8 clearly shows that there is an immediate reduction of NEPs mobilized to the damaged EGL, there is no subsequent reduction of cerebellar growth such that by P30, the treated and untreated irradiated cerebella are equivalent in size. There is speculation in the discussion about why this might be the case. Additional experiments and tools are required to assess mechanisms. Regardless, the data still implicate microglia in the neonatal regenerative response, and this finding remains an important advance.

Author response:

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

Reviewer #1 (Public review):

Summary:

In this manuscript, Pakula et al. explore the impact of reactive oxygen species (ROS) on neonatal cerebellar regeneration, providing evidence that ROS activates regeneration through Nestin-expressing progenitors (NEPs). Using scRNA-seq analysis of FACS-isolated NEPs, the authors characterize injury-induced changes, including an enrichment in ROS metabolic processes within the cerebellar microenvironment. Biochemical analyses confirm a rapid increase in ROS levels following irradiation and forced catalase expression, which reduces ROS levels, and impairs external granule layer (EGL) replenishment post-injury.

Strengths:

Overall, the study robustly supports its main conclusion and provides valuable insights into ROS as a regenerative signal in the neonatal cerebellum.

Comments on revisions:

The authors have addressed most of the previous comments. However, they should clarify the following response:

*"For reasons we have not explored, the phenotype is most prominent in these lobules, that is why they were originally chosen. We edited the following sentence (lines 578-579):

First, we analyzed the replenishment of the EGL by BgL-NEPs in vermis lobules 3-5, since our previous work showed that these lobules have a prominent defect."*

It has been reported that the anterior part of the cerebellum may have a lower regenerative capacity compared to the posterior lobe. To avoid potential ambiguity, the authors should clarify that "the phenotype" and "prominent defect" refer to more severe EGL depletion at an earlier stage after IR rather than a poorer regenerative outcome. Additionally, they should provide a reference to support their statement or indicate if it is based on unpublished observations.

Our comment does not refer to a more severe EGL depletion at an earlier stage. There is instead poorer regeneration of the anterior region. The irradiation approach used provides consistent cell killing of GCPs across the cerebellum. This can be seen in Fig. 1c, e, g, i in our previous publication: Wojcinski, et al. (2017) Cerebellar granule cell replenishment post-injury by adaptive reprogramming of Nestin+ progenitors. Nature Neuroscience, 20:1361-1370). Also, Fig 2e, g, k, m in the paper shows that by P5 and P8, posterior lobule 8 recovers better than anterior lobules 1-5.

Reviewer #2 (Public review):

Summary:

The authors have previously shown that the mouse neonatal cerebellum can regenerate damage to granule cell progenitors in the external granular layer, through reprogramming of gliogenic nestin-expressing progenitors (NEPs). The mechanisms of this reprogramming remain largely unknown. Here the authors used scRNAseq and ATACseq of purified neonatal NEPs from P1-P5 and showed that ROS signatures were transiently upregulated in gliogenic NEPs ve neurogenic NEPs 24 hours post injury (P2). To assess the role of ROS, mice transgenic for global catalase activity were assessed to reduce ROS. Inhibition of ROS significantly decreased gliogenic NEP reprogramming and diminished cerebellar growth post-injury. Further, inhibition of microglia across this same time period prevented one of the first steps of repair - the migration of NEPs into the external granule layer. This work is the first demonstration that the tissue microenvironment of the damaged neonatal cerebellum is a major regulator of neonatal cerebellar regeneration. Increased ROS is seen in other CNS damage models, including adults, thus there may be some shared mechanisms across age and regions, although interestingly neonatal cerebellar astrocytes do not upregulate GFAP as seen in adult CNS damage models. Another intriguing finding is that global inhibition of ROS did not alter normal cerebellar development.

Strengths:

This paper presents a beautiful example of using single cell data to generate biologically relevant, testable hypotheses of mechanisms driving important biological processes. The scRNAseq and ATACseq analyses are rigorously conducted and conclusive. Data is very clearly presented and easily interpreted supporting the hypothesis next tested by reduce ROS in irradiated brains.

Analysis of whole tissue and FAC sorted NEPS in transgenic mice where human catalase was globally expressed in mitochondria were rigorously controlled and conclusively show that ROS upregulation was indeed decreased post injury and very clearly the regenerative response was inhibited. The authors are to be commended on the very careful analyses which are very well presented and again, easy to follow with all appropriate data shown to support their conclusions.

Weaknesses:

The authors also present data to show that microglia are required for an early step of mobilizing gliogenic NEPs into the damaged EGL. While the data that PLX5622 administration from P0-P5 or even P0-P8 clearly shows that there is an immediate reduction of NEPs mobilized to the damaged EGL, there is no subsequent reduction of cerebellar growth such that by P30, the treated and untreated irradiated cerebella are equivalent in size. There is speculation in the discussion about why this might be the case. Additional experiments and tools are required to assess mechanisms. Regardless, the data still implicate microglia in the neonatal regenerative response, and this finding remains an important advance.

As stated previously, the suggested follow up experiments while relevant are extensive and considered beyond the scope of the current paper.

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

Public Reviews:

Reviewer #1 (Public review):

Summary:

In this manuscript, Pakula et al. explore the impact of reactive oxygen species (ROS) on neonatal cerebellar regeneration, providing evidence that ROS activates regeneration through Nestin-expressing progenitors (NEPs). Using scRNA-seq analysis of FACS-isolated NEPs, the authors characterize injury-induced changes, including an enrichment in ROS metabolic processes within the cerebellar microenvironment. Biochemical analyses confirm a rapid increase in ROS levels following irradiation, and forced catalase expression, which reduces ROS levels, and impairs external granule layer (EGL) replenishment post-injury.

Strengths:

Overall, the study robustly supports its main conclusion and provides valuable insights into ROS as a regenerative signal in the neonatal cerebellum.

Weaknesses:

(1) The diversity of cell types recovered from scRNA-seq libraries of sorted Nes-CFP cells is unexpected, especially the inclusion of minor types such as microglia, meninges, and ependymal cells. The authors should validate whether Nes and CFP mRNAs are enriched in the sorted cells; if not, they should discuss the potential pitfalls in sampling bias or artifacts that may have affected the dataset, impacting interpretation.

In our previous work, we thoroughly assessed the transgene using RNA in situ hybridization for Cfp, immunofluorescent analysis for CFP and scRNA-seq analysis for Cfp transcripts (Bayin et al., Science Adv. 2021, Fig. S1-2)(1), and characterized the diversity within the NEP populations of the cerebellum. Our present scRNA-seq data also confirms that Nes transcripts are expressed in all the NEP subtypes. A feature plot for Nes expression has been added to the revised manuscript (Fig 1E), as well as a sentence explaining the results. Of note, since this data was generated from FACS-isolated CFP+ cells, the perdurance of the protein allows for the detection of immediate progeny of Nes-expressing cells, even in cells where Nes is not expressed once cells are differentiated. Finally, oligodendrocyte progenitors, perivascular cells, some rare microglia and ependymal cells have been demonstrated to express Nes in the central nervous system; therefore, detecting small groups of these cells is expected (2-4). We have added the following sentence (lines 391-394):

“Detection of Nes mRNA confirmed that the transgene reflects endogenous Nes expression in progenitors of many lineages, and also that the perdurance of CFP protein in immediate progeny of Nes-expressing cells allowed the isolation of these cells by FACS (Figure 1E)”.

(2) The authors should de-emphasize that ROS signaling and related gene upregulation exclusively in gliogenic NEPs. Genes such as Cdkn1a, Phlda3, Ass1, and Bax are identified as differentially expressed in neurogenic NEPs and granule cell progenitors (GCPs), with Ass1 absent in GCPs. According to Table S4, gene ontology (GO) terms related to ROS metabolic processes are also enriched in gliogenic NEPs, neurogenic NEPs, and GCPs.

As the reviewer requested, we have de-emphasized that ROS signaling is preferentially upregulated in gliogenic NEPs, since we agree with the reviewer that there is some evidence for similar transcriptional signatures in neurogenic NEPs and GCPs. We added the following (lines 429-531):

“Some of the DNA damage and apoptosis related genes that were upregulated in IR gliogenic-NEPs (Cdkn1a, Phlda3, Bax) were also upregulated in the IR neurogenic-NEPs and GCPs at P2 (Supplementary Figure 2B-E).”

And we edited the last few sentences of the section to state (lines 453-459):

“Interestingly, we did not observe significant enrichment for GO terms associated with cellular stress response in the GCPs that survived the irradiation compared to controls, despite significant enrichment for ROS signaling related GO-terms (Table S4). Collectively, these results indicate that injury induces significant and overlapping transcriptional changes in NEPs and GCPs. The gliogenic- and neurogenic-NEP subtypes transiently upregulate stress response genes upon GCP death, and an overall increase in ROS signaling is observed in the injured cerebella.”

(3) The authors need to justify the selection of only the anterior lobe for EGL replenishment and microglia quantification.

We thank the reviewers for asking for this clarification. Our previous publications on regeneration of the EGL by NEPs have all involved quantification of these lobules, thus we think it is important to stay with the same lobules. For reasons we have not explored, the phenotype is most prominent in these lobules, that is why they were originally chosen. We edited the following sentence (lines 578-579):

“First, we analyzed the replenishment of the EGL by BgL-NEPs in vermis lobules 3-5, since our previous work showed that these lobules have a prominent defect.”

(4) Figure 1K: The figure presents linkages between genes and GO terms as a network but does not depict a gene network. The terminology should be corrected accordingly.

We have corrected the terminology and added the following (lines 487-489):

“Finally, linkages between the genes in differentially open regions identified by ATAC-seq and the associated GO-terms revealed an active transcriptional network involved in regulating cell death and apoptosis (Figure 1K).”

(5) Figure 1H and S2: The x-axis appears to display raw p-values rather than log10(p.value) as indicated. The x-axis should ideally show -log10(p.adjust), beginning at zero. The current format may misleadingly suggest that the ROS GO term has the lowest p-values.

Apologies for the mistake. The data represents raw p-values and the x-axis has been corrected.

(6) Genes such as Ppara, Egln3, Foxo3, Jun, and Nos1ap were identified by bulk ATAC-seq based on proximity to peaks, not by scRNA-seq. Without additional expression data, caution is needed when presenting these genes as direct evidence of ROS involvement in NEPs.

We modified the text to discuss the discrepancies between the analyses. While some of this could be due to the lower detection limits in the scRNA-seq, it also highlights that chromatin accessibility is not a direct readout for expression levels and further analysis is needed. Nevertheless, both scRNA-seq and ATAC-seq have identified similar mechanisms, and our mutant analysis confirmed our hypothesis that an increase in ROS levels underlies repair, further increasing the confidence in our analyses. Further investigation is needed to understand the downstream mechanisms. We added the following sentence (lines 478-481):

“However, not all genes in the accessible areas were differentially expressed in the scRNA-seq data. While some of this could be due to the detection limits of scRNA-seq, further analysis is required to assess the mechanisms of how the differentially accessible chromatin affects transcription.”

(7) The authors should annotate cell identities for the different clusters in Table S2.

All cell types have been annotated in Table S2.

(8) Reiterative clustering analysis reveals distinct subpopulations among gliogenic and neurogenic NEPs. Could the authors clarify the identities of these subclusters? Can we distinguish the gliogenic NEPs in the Bergmann glia layer from those in the white matter?

Thank you for this clarification. As shown in our previous studies, we can not distinguish between the gliogenic NEPs in the Bergmann glia layer and the white matter based on scRNA-seq, but expression of the Bergmann glia marker Gdf10 suggests that a large proportion of the cells in the Hopx+ clusters are in the Bergmann glia layer. The distinction within the major subpopulations that we characterized (Hopx-, Ascl1-expressing NEPs and GCPs) are driven by their proliferative/maturation status as we previously observed. We have included a detailed annotation of all the clusters in Table S2, as requested and a UMAP for mKi57 expression in Fig 1E. We have clarified this in the following sentence (lines 383-385):

“These groups of cells were further subdivided into molecularly distinct clusters based on marker genes and their cell cycle profiles or developmental stages (Figure 1D, Table S2).”

(9) In the Methods section, the authors mention filtering out genes with fewer than 10 counts. They should specify if these genes were used as background for enrichment analysis. Background gene selection is critical, as it influences the functional enrichment of gene sets in the list.

As requested, the approach used has been added to the Methods section of the revised paper. Briefly, the background genes used by the goseq function are the same genes used for the probability weight function (nullp). The mm8 genome annotation was used in the nullp function, and all annotated genes were used as background genes to compute GO term enrichment. The following was added (lines 307-308):

“The background genes used to compute the GO term enrichment includes all genes with gene symbol annotations within mm8.”

(10) Figure S1C: The authors could consider using bar plots to better illustrate cell composition differences across conditions and replicates.

As suggested, we have included bar plots in Fig. S1D-F.

(11) Figures 4-6: It remains unclear how the white matter microglia contribute to the recruitment of BgL-NEPs to the EGL, as the mCAT-mediated microglia loss data are all confined to the white matter.

We have thought about the question and had initially quantified the microglia in the white matter and the rest of the lobules (excluding the EGL) separately. However, there are very few microglia outside the white matter in each section, thus it is not possible to obtain reliable statistical data on such a small population. We therefore did not include the cells in the analysis. We have added this point in the main text (line 548).

“As a possible explanation for how white matter microglia could influence NEP behaviors, given the small size of the lobules and how the cytoarchitecture is disrupted after injury, we think it is possible that secreted factors from the white matter microglia could reach the BgL NEPs. Alternatively, there could be a relay system through an intermediate cell type closer to the microglia.” We have added these ideas to the Discussion of the revised paper (lines 735-738).

Reviewer #2 (Public review):

Summary:

The authors have previously shown that the mouse neonatal cerebellum can regenerate damage to granule cell progenitors in the external granular layer, through reprogramming of gliogenic nestin-expressing progenitors (NEPs). The mechanisms of this reprogramming remain largely unknown. Here the authors used scRNAseq and ATACseq of purified neonatal NEPs from P1-P5 and showed that ROS signatures were transiently upregulated in gliogenic NEPs ve neurogenic NEPs 24 hours post injury (P2). To assess the role of ROS, mice transgenic for global catalase activity were assessed to reduce ROS. Inhibition of ROS significantly decreased gliogenic NEP reprogramming and diminished cerebellar growth post-injury. Further, inhibition of microglia across this same time period prevented one of the first steps of repair - the migration of NEPs into the external granule layer. This work is the first demonstration that the tissue microenvironment of the damaged neonatal cerebellum is a major regulator of neonatal cerebellar regeneration. Increased ROS is seen in other CNS damage models including adults, thus there may be some shared mechanisms across age and regions, although interestingly neonatal cerebellar astrocytes do not upregulate GFAP as seen in adult CNS damage models. Another intriguing finding is that global inhibition of ROS did not alter normal cerebellar development.

Strengths:

This paper presents a beautiful example of using single cell data to generate biologically relevant, testable hypotheses of mechanisms driving important biological processes. The scRNAseq and ATACseq analyses are rigorously conducted and conclusive. Data is very clearly presented and easily interpreted supporting the hypothesis next tested by reduce ROS in irradiated brains.

Analysis of whole tissue and FAC sorted NEPS in transgenic mice where human catalase was globally expressed in mitochondria were rigorously controlled and conclusively show that ROS upregulation was indeed decreased post injury and very clearly the regenerative response was inhibited. The authors are to be commended on the very careful analyses which are very well presented and again, easy to follow with all appropriate data shown to support their conclusions.

Weaknesses:

The authors also present data to show that microglia are required for an early step of mobilizing gliogenic NEPs into the damaged EGL. While the data that PLX5622 administration from P0-P5 or even P0-P8 clearly shows that there is an immediate reduction of NEPs mobilized to the damaged EGL, there is no subsequent reduction of cerebellar growth such that by P30, the treated and untreated irradiated cerebella are equivalent in size. There is speculation in the discussion about why this might be the case, but there is no explanation for why further, longer treatment was not attempted nor was there any additional analyses of other regenerative steps in the treated animals. The data still implicate microglia in the neonatal regenerative response, but how remains uncertain.

Recommendations for the authors:

Reviewer #2 (Recommendations for the authors):

This is an exemplary manuscript.

The methods and data are very well described and presented.

I actually have very little to ask the authors except for an explanation of why PLX treatment was discontinued after P5 or P8 and what other steps of NEP reprogramming were assessed in these animals? Was NEP expansion still decreased at P8 even in the presence of PLX at this stage? Also - was there any analysis attempted combining mCAT and PLX?

We agree with the reviewer that a follow up study that goes into a deeper analysis of the role of microglia in GCP regeneration and any interaction with ROS signaling would interesting. However, it would require a set of tools that we do not currently have. We did not have enough PLX5622 to perform addition experiments or extend the length of treatment. Plexxikon informed us in 2021 that they were no longer manufacturing PLX5622 because they were focusing on new analogs for in vivo use, and thus we had to use what we had left over from a completed preclinical cancer study. We nevertheless think it is important to publish our preliminary results to spark further experiments by other groups.

References

(1) Bayin N. S. Mizrak D., Stephen N. D., Lao Z., Sims P. A., Joyner A. L. Injury induced ASCL1 expression orchestrates a transitory cell state required for repair of the neonatal cerebellum. Sci Adv. 2021;7(50):eabj1598.

(2) Cawsey T, Duflou J, Weickert CS, Gorrie CA. Nestin-Positive Ependymal Cells Are Increased in the Human Spinal Cord after Traumatic Central Nervous System Injury. J Neurotrauma. 2015;32(18):1393-402.

(3) Gallo V, Armstrong RC. Developmental and growth factor-induced regulation of nestin in oligodendrocyte lineage cells. The Journal of neuroscience : the official journal of the Society for Neuroscience. 1995;15(1 Pt 1):394-406.

(4) Huang Y, Xu Z, Xiong S, Sun F, Qin G, Hu G, et al. Repopulated microglia are solely derived from the proliferation of residual microglia after acute depletion. Nat Neurosci. 2018;21(4):530-40.

eLife Assessment

This study presents a valuable observation of how deletion of a major repair protein in bacteria can facilitate the rise of mutations that confer resistance against a range of different antibiotics. The data presented are convincing, and the authors addressed the concerns raised by the reviewers in their resubmission, improving the strength of their findings.

Reviewer #1 (Public review):

Summary:

Jin et al. investigated how the bacterial DNA damage (SOS) response and its regulator protein RecA affects the development of drug resistance under short-term exposure to beta-lactam antibiotics. Canonically, the SOS response is triggered by DNA damage, which results in the induction of error-prone DNA repair mechanisms. These error-prone repair pathways can increase mutagenesis in the cell, leading to the evolution of drug resistance. Thus, inhibiting the SOS regulator RecA has been proposed as means to delay the rise of resistance.

In this paper, the authors deleted the RecA protein from E. coli and exposed this ∆recA strain to selective levels of the beta-lactam antibiotic, ampicillin. After an 8h treatment, they washed the antibiotic away and allowed the surviving cells to recover in regular media. They then measured the minimum inhibitory concentration (MIC) of ampicillin against these treated strains. They note that after just 8 h treatment with ampicillin, the ∆recA had developed higher MICs towards ampicillin, while by contrast, wild-type cells exhibited unchanged MICs. This MIC increase was also observed in subsequent generations of bacteria, suggesting that the phenotype is driven by a genetic change.

The authors then used whole genome sequencing (WGS) to identify mutations that accounted for the resistance phenotype. Within resistant populations, they discovered key mutations in the promoter region of the beta-lactamase gene, ampC; in the penicillin-binding protein PBP3 which is the target of ampicillin; and in the AcrB subunit of the AcrAB-TolC efflux machinery. Importantly, mutations in the efflux machinery can impact the resistance towards other antibiotics, not just beta-lactams. To test this, they repeated the MIC experiments with other classes of antibiotics, including kanamycin, chloramphenicol, and rifampicin. Interestingly, they observed that the ∆recA strains pre-treated with ampicillin showed higher MICs towards all other antibiotics tested. This suggests that the mutations conferring resistance to ampicillin are also increasing resistance to other antibiotics.

The authors then performed an impressive series of genetic, microscopy, and transcriptomic experiments to show that this increase in resistance is not driven by the SOS response, but by independent DNA repair and stress response pathways. Specifically, they show that deletion of the recA reduces the bacterium's ability to process reactive oxygen species (ROS) and repair its DNA. These factors drive the accumulation of mutations that can confer resistance towards different classes of antibiotics. The conclusions are reasonably well-supported by the data, but some aspects of the data and the model need to be clarified and extended.

Strengths:

A major strength of the paper is the detailed bacterial genetics and transcriptomics that the authors performed to elucidate the molecular pathways responsible for this increased resistance. They systemically deleted or inactivated genes involved in the SOS response in E. coli. They then subjected these mutants to the same MIC assays as described previously. Surprisingly, none of the other SOS gene deletions resulted in an increase in drug resistance, suggesting that the SOS response is not involved in this phenotype. This led the authors to focus on the localization of DNA PolI, which also participates in DNA damage repair. Using microscopy, they discovered that in the RecA deletion background, PolI co-localizes with the bacterial chromosome at much lower rates than wild-type. This led the authors to conclude that deletion of RecA hinders PolI and DNA repair. Although the authors do not provide a mechanism, this observation is nonetheless valuable for the field and can stimulate further investigations in the future.

In order to understand how RecA deletion affects cellular physiology, the authors performed RNA-seq on ampicillin-treated strains. Crucially, they discovered that in the RecA deletion strain, genes associated with antioxidative activity (cysJ, cysI, cysH, soda, sufD) and Base Excision Repair repair (mutH, mutY, mutM), which repairs oxidized forms of guanine, were all downregulated. The authors conclude that down-regulation of these genes might result in elevated levels of reactive oxygen species in the cells, which in turn, might drive the rise of resistance. Experimentally, they further demonstrated that treating the ∆recA strain with an antioxidant GSH prevents the rise of MICs. These observations will be useful for more detailed mechanistic follow-ups in the future.

Weaknesses:

Throughout the paper, the authors use language suggesting that ampicillin treatment of the ∆recA strain induces higher levels of mutagenesis inside the cells, leading to the rapid rise of resistance mutations. However, as the authors note, the mutants enriched by ampicillin selection can play a role in efflux and can thus change a bacterium's sensitivity to a wide range of antibiotics, in what is known as cross-resistance. The current data is not clear on whether the elevated "mutagenesis" is driven by ampicillin selection or by a bona fide increase in mutation rate.

Furthermore, on a technical level, the authors employed WGS to identify resistance mutations in the ampicillin-treated wild-type and ∆recA strains. However, the WGS methodology described in the paper is inconsistent. Notably, wild-type WGS samples were picked from non-selective plates, while ΔrecA WGS isolates were picked from selective plates with 50 μg/mL ampicillin. Such an approach biases the frequency and identity of the mutations seen in the WGS and cannot be used to support the idea that ampicillin treatment induces higher levels of mutagenesis.

Finally, it is important to establish what the basal mutation rates of both the WT and ∆recA strains are. Currently, only the ampicillin-treated populations were reported. It is possible that the ∆recA strain has inherently higher mutagenesis than WT, with a larger subpopulation of resistant clones. Thus, ampicillin treatment might not, in fact, induce higher mutagenesis in ∆recA.

Summary of revised manuscript:

In their revisions, the authors have addressed my major concerns with additional experiments and changes to the text. Thank you!

Reviewer #3 (Public review):

Summary:

In the present work, Zhang et al investigate the involvement of the bacterial DNA damage repair SOS response in the evolution of beta-lactam drug resistance in Escherichia coli. Using a combination of microbiological, bacterial genetics, laboratory evolution, next-generation, and live-cell imaging approaches, the authors propose short-term (transient) drug resistance evolution can take place in RecA-deficient cells in an SOS response-independent manner. They propose the evolvability of drug resistance is alternatively driven by the oxidative stress imposed by accumulation of reactive oxygen species and compromised DNA repair. Overall, this is a nice study that addresses a growing and fundamental global health challenge (antimicrobial resistance).

Strengths:

The authors introduce new concepts to antimicrobial resistance evolution mechanisms. They show short-term exposure to beta-lactams can induce durably fixed antimicrobial resistance mutations. They propose this is due to compromised DNA repair and oxidative stress. Antibiotic resistance evolution under transient stress is poorly studied, so the authors' work is a nice mechanistic contribution to this field.

Weaknesses:

The authors revisions have significantly addressed weaknesses previously identified earlier in the review process.

Author response:

The following is the authors’ response to the previous reviews

Reviewer #1 (Public review):

Jin et al. investigated how the bacterial DNA damage (SOS) response and its regulator protein RecA affects the development of drug resistance under short-term exposure to beta-lactam antibiotics. Canonically, the SOS response is triggered by DNA damage, which results in the induction of error-prone DNA repair mechanisms. These error-prone repair pathways can increase mutagenesis in the cell, leading to the evolution of drug resistance. Thus, inhibiting the SOS regulator RecA has been proposed as means to delay the rise of resistance.

In this paper, the authors deleted the RecA protein from E. coli and exposed this ∆recA strain to selective levels of the beta-lactam antibiotic, ampicillin. After an 8h treatment, they washed the antibiotic away and allowed the surviving cells to recover in regular media. They then measured the minimum inhibitory concentration (MIC) of ampicillin against these treated strains. They note that after just 8 h treatment with ampicillin, the ∆recA had developed higher MICs towards ampicillin, while by contrast, wild-type cells exhibited unchanged MICs. This MIC increase was also observed subsequent generations of bacteria, suggesting that the phenotype is driven by a genetic change.

The authors then used whole genome sequencing (WGS) to identify mutations that accounted for the resistance phenotype. Within resistant populations, they discovered key mutations in the promoter region of the beta-lactamase gene, ampC; in the penicillin-binding protein PBP3 which is the target of ampicillin; and in the AcrB subunit of the AcrAB-TolC efflux machinery. Importantly, mutations in the efflux machinery can impact the resistances towards other antibiotics, not just beta-lactams. To test this, they repeated the MIC experiments with other classes of antibiotics, including kanamycin, chloramphenicol, and rifampicin. Interestingly, they observed that the ∆recA strains pre-treated with ampicillin showed higher MICs towards all other antibiotic tested. This suggests that the mutations conferring resistance to ampicillin are also increasing resistance to other antibiotics.

The authors then performed an impressive series of genetic, microscopy, and transcriptomic experiments to show that this increase in resistance is not driven by the SOS response, but by independent DNA repair and stress response pathways. Specifically, they show that deletion of the recA reduces the bacterium's ability to process reactive oxygen species (ROS) and repair its DNA. These factors drive accumulation of mutations that can confer resistance towards different classes of antibiotics. The conclusions are reasonably well-supported by the data, but some aspects of the data and the model need to be clarified and extended.

Strengths:

A major strength of the paper is the detailed bacterial genetics and transcriptomics that the authors performed to elucidate the molecular pathways responsible for this increased resistance. They systemically deleted or inactivated genes involved in the SOS response in E. coli. They then subjected these mutants the same MIC assays as described previously. Surprisingly, none of the other SOS gene deletions resulted an increase in drug resistance, suggesting that the SOS response is not involved in this phenotype. This led the authors to focus on the localization of DNA PolI, which also participates in DNA damage repair. Using microscopy, they discovered that in the RecA deletion background, PolI co-localizes with the bacterial chromosome at much lower rates than wild-type. This led the authors to conclude that deletion of RecA hinders PolI and DNA repair. Although the authors do not provide a mechanism, this observation is nonetheless valuable for the field and can stimulate further investigations in the future.

In order to understand how RecA deletion affects cellular physiology, the authors performed RNA-seq on ampicillin-treated strains. Crucially, they discovered that in the RecA deletion strain, genes associated with antioxidative activity (cysJ, cysI, cysH, soda, sufD) and Base Excision Repair repair (mutH, mutY, mutM), which repairs oxidized forms of guanine, were all downregulated. The authors conclude that down-regulation of these genes might result in elevated levels of reactive oxygen species in the cells, which in turn, might drive the rise of resistance. Experimentally, they further demonstrated that treating the ∆recA strain with an antioxidant GSH prevents the rise of MICs. These observations will be useful for more detailed mechanistic follow-ups in the future.

Weaknesses:

Throughout the paper, the authors use language suggesting that ampicillin treatment of the ∆recA strain induces higher levels of mutagenesis inside the cells, leading to the rapid rise of resistance mutations. However, as the authors note, the mutants enriched by ampicillin selection can play a role in efflux and can thus change a bacterium's sensitivity to a wide range of antibiotics, in what is known as cross-resistance. The current data is not clear on whether the elevated "mutagenesis" is driven ampicillin selection or by a bona fide increase in mutation rate.

Furthermore, on a technical level, the authors employed WGS to identify resistance mutations in the treated ampicillin-treated wild-type and ∆recA strains. However, the WGS methodology described in the paper is inconsistent. Notably, wild-type WGS samples were picked from non-selective plates, while ΔrecA WGS isolates were picked from selective plates with 50 μg/mL ampicillin. Such an approach biases the frequency and identity of the mutations seen in the WGS and cannot be used to support the idea that ampicillin treatment induces higher levels of mutagenesis.

Finally, it is important to establish what the basal mutation rates of both the WT and ∆recA strains are. Currently, only the ampicillin-treated populations were reported. It is possible that the ∆recA strain has inherently higher mutagenesis than WT, with a larger subpopulation of resistant clones. Thus, ampicillin treatment might not in fact induce higher mutagenesis in ∆recA.

Comments on revisions:

Thank you for responding to the concerns raised previously. The manuscript overall has improved.

We sincerely thank the reviewer for raising this important point. In our initial submission, we acknowledge that our mutation analysis was based on a limited number of replicates (n=6), which may not have been sufficient to robustly distinguish between mutation induction and selection. In response to this concern, we have substantially expanded our experimental dataset. Specifically, we redesigned the mutation rate validation experiment by increasing the number of biological replicates in each condition to 96 independent parallel cultures. This enabled us to systematically assess mutation frequency distributions under four conditions (WT, WT+ampicillin, ΔrecA, ΔrecA+ampicillin), using both maximum likelihood estimation (MLE) and distribution-based fluctuation analysis (new Figure 1F, 1G, and Figure S5).

These expanded datasets revealed that:

(1) While the estimated mutation rate was significantly elevated in ΔrecA+ampicillin compared to ΔrecA alone (Fig. 1G),

(2) The distribution of mutation frequencies in ΔrecA+ampicillin was highly skewed with evident jackpot cultures (Fig. 1F), and

(3) The observed pattern significantly deviated from Poisson expectations, which is inconsistent with uniform mutagenesis and instead supports clonal selection from an early-arising mutational pool (Fig. S5).

Importantly, these new results do not contradict our original conclusions but rather extend and refine them. The previous evidence for ROS-mediated mutagenesis remains valid and is supported by our GSH experiments, transcriptomic analysis of oxidative stress genes, and DNA repair pathway repression. However, the additional data now indicate that ROS-induced variants are not uniformly induced after antibiotic exposure but are instead generated stochastically under the stress-prone ΔrecA background and then selectively enriched upon ampicillin treatment.

Taken together, we now propose a two-step model of resistance evolution in ΔrecA cells (new Figure 5):

Step i: RecA deficiency creates a hypermutable state through impaired repair and elevated ROS, increasing the probability of resistance-conferring mutations.

Step ii: β-lactam exposure acts as a selective bottleneck, enriching early-arising mutants that confer resistance.

We have revised both the Results and Discussion sections to clearly articulate this complementary relationship between mutational supply and selection, and we believe this integrated model better explains the observed phenotypes and mechanistic outcomes.

Reviewer #2 (Public review):

This study aims to demonstrate that E. coli can acquire rapid antibiotic resistance mutations in the absence of a DNA damage response. The authors employed a modified Adaptive Laboratory Evolution (ALE) workflow to investigate this, initiating the process by diluting an overnight culture 50-fold into an ampicillin selection medium. They present evidence that a recA- strain develops ampicillin resistance mutations more rapidly than the wild-type, as indicated by the Minimum Inhibitory Concentration (MIC) and mutation frequency. Whole-genome sequencing of recA- colonies resistant to ampicillin showed predominant inactivation of genes involved in the multi-drug efflux pump system, contrasting with wild-type mutations that seem to activate the chromosomal ampC cryptic promoter. Further analysis of mutants, including a lexA3 mutant incapable of inducing the SOS response, led the authors to conclude that the rapid evolution of antibiotic resistance occurs via an SOS-independent mechanism in the absence of recA. RNA sequencing suggests that antioxidative response genes drive the rapid evolution of antibiotic resistance in the recA- strain. They assert that rapid evolution is facilitated by compromised DNA repair, transcriptional repression of antioxidative stress genes, and excessive ROS accumulation.

Strengths:

The experiments are well-executed and the data appear reliable. It is evident that the inactivation of recA promotes faster evolutionary responses, although the exact mechanisms driving this acceleration remain elusive and deserve further investigation.

Weaknesses:

Some conclusions are overstated. For instance, the conclusion regarding the LexA3 allele, indicating that rapid evolution occurs in an SOS-independent manner (line 217), contradicts the introductory statement that attributes evolution to compromised DNA repair.

We thank the reviewer for this insightful observation, which highlights a central conceptual advance of our study. Our data indeed indicate that resistance evolution in ΔrecA occurs independently of canonical SOS induction (as shown by the lack of resistance in lexA3, dpiBA, and translesion polymerase mutants), yet is clearly associated with impaired DNA repair capacity (e.g., downregulation of polA, mutH, mutY).

This apparent “contradiction” reflects the dual role of RecA: it functions both as the master activator of the SOS response and as a key factor in SOS-independent repair processes. Thus, the rapid resistance evolution in ΔrecA is not due to loss of SOS, but rather due to the broader suppression of DNA repair pathways that RecA coordinates, which elevates mutational load under stress (This point is discussed in further detail in our response to Reviewer 1).

The claim made in the discussion of Figure 3 that the hindrance of DNA repair in recA- is crucial for rapid evolution is at best suggestive, not demonstrative. Additionally, the interpretation of the PolI data implies its role, yet it remains speculative.

We appreciate this comment and would like to respectfully clarify that our conclusion regarding the role of DNA repair impairment is supported by several independent lines of mechanistic evidence.

First, our RNA-seq analysis revealed transcriptional suppression of multiple DNA repair genes in ΔrecA cells following ampicillin treatment, including polA (DNA Pol I) and the base excision repair genes mutH, mutY, and mutM (Fig. 4K). This indicates that multiple repair pathways, including those responsible for correcting oxidative DNA lesions, are downregulated under these conditions.

Second, we observed a significant reduction in DNA Pol I protein expression as well as reduced colocalization with chromosomal DNA in ΔrecA cells, suggesting impaired engagement of repair machinery (Fig. 3C-E). These phenotypes are not limited to transcriptional signatures but extend to functional protein localization.

Third, and most importantly, resistance evolution was fully suppressed in ΔrecA cells upon co-treatment with glutathione (GSH), which reduces ROS levels. As GSH did not affect ampicillin killing (Fig. 4J), these findings suggest that mutagenesis and thus the emergence of resistance requires both ROS accumulation and the absence of efficient repair.

Therefore, we believe these data go beyond correlation and demonstrate a mechanistic role for DNA repair impairment in driving stress-associated resistance evolution in ΔrecA. We have revised the Discussion to emphasize the strength of this evidence while avoiding overstatement.

In Figure 2A table, mutations in amp promoters are leading to amino acid changes.

We thank the reviewer for spotting this inconsistency. Indeed, the ampC promoter mutations we identified reside in non-coding regulatory regions and do not result in amino acid substitutions. We have corrected the annotation in Fig. 2A and clarified in the main text that these mutations likely affect gene expression through transcriptional regulation, rather than protein sequence alteration.

The authors' assertion that ampicillin significantly influences persistence pathways in the wild-type strain, affecting quorum sensing, flagellar assembly, biofilm formation, and bacterial chemotaxis, lacks empirical validation.

We thank the reviewer for pointing this out. In the original version, we acknowledged transcriptional enrichment of genes related to quorum sensing, flagellar assembly, and chemotaxis in the wild-type strain upon ampicillin treatment. However, as we did not directly assess persistence phenotypes (e.g., biofilm formation or persister levels), we agree that such functional inferences were not fully supported. We have revised the relevant statements to focus solely on transcriptomic changes and have removed language suggesting direct effects on persistence pathways.

Figure 1G suggests that recA cells treated with ampicillin exhibit a strong mutator phenotype; however, it remains unclear if this can be linked to the mutations identified in Figure 2's sequencing analysis.

We appreciate the reviewer’s comment. This point is discussed in further detail in our response to Reviewer 1.

Reviewer #3 (Public review):

In the present work, Zhang et al investigate involvement of the bacterial DNA damage repair SOS response in the evolution of beta-lactam drug resistance evolution in Escherichia coli. Using a combination of microbiological, bacterial genetics, laboratory evolution, next-generation, and live-cell imaging approaches, the authors propose short-term (transient) drug resistance evolution can take place in RecA-deficient cells in an SOS response-independent manner. They propose the evolvability of drug resistance is alternatively driven by the oxidative stress imposed by accumulation of reactive oxygen species and compromised DNA repair. Overall, this is a nice study that addresses a growing and fundamental global health challenge (antimicrobial resistance).

Strengths:

The authors introduce new concepts to antimicrobial resistance evolution mechanisms. They show short-term exposure to beta-lactams can induce durably fixed antimicrobial resistance mutations. They propose this is due to comprised DNA repair and oxidative stress. Antibiotic resistance evolution under transient stress is poorly studied, so the authors' work is a nice mechanistic contribution to this field.

Weaknesses:

The authors do not show any direct evidence of altered mutation rate or accumulated DNA damage in their model.

We appreciate the reviewer’s comment. This point is discussed in further detail in our response to Reviewer 1.

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

I would like to suggest two minor changes to the text.

(1) Re. WGS data.

The authors write in their response "We appreciate your concern regarding potential inconsistencies in the WGS methodology. However, we would like to clarify that the primary aim of the WGS experiment was to identify the types of mutations present in the wild type and ΔrecA strains after treatment of ampicillin, rather than to quantify or compare mutation frequencies. This purpose was explicitly stated in the manuscript.

I think the source of my confusion stemmed from this part in the text:

"In bacteria, resistance to most antibiotics requires the accumulation of drug resistance associated DNA mutations developed over time to provide high levels of resistance (29). To verify whether drug resistance associated DNA mutations have led to the rapid development of antibiotic resistance in recA mutant strain, we..."

I would change the phrase "verify whether drug resistance associated DNA mutations have led to the rapid development of antibiotic resistance in recA mutant strain" to "identify the types of mutations present in the wild type and ΔrecA strains after treatment of ampicillin." This would explicitly state what the sequencing was for (ie. ID-ing mutations). The current phrase can give the impression that WGS was used to validate rapid or high mutagenesis.

Thanks for this suggestion. We have revised this description to “In bacteria, resistance to most antibiotics requires the accumulation of drug resistance associated DNA mutations that can arise stochastically and, under stress conditions, become enriched through selection over time to confer high levels of resistance (33). Having observed a non-random and right-skewed distribution of mutation frequencies in ΔrecA isolates following ampicillin exposure, we next sought to determine whether specific resistance-conferring mutations were enriched in ΔrecA isolates following antibiotic exposure.”

(2) Re. whether the mutations are "induced" or "pre-existing."

The authors write:

"We appreciate your detailed feedback on the language used to describe our data. We understand the concern regarding the use of the term "induced" in relation to beta-lactam exposure. To clarify, we employed not only beta-lactam antibiotics but also other antibiotics, such as ciprofloxacin and chloramphenicol, in our experiments (data not shown). However, we observed that beta-lactam antibiotics specifically induced the emergence of resistance or altered the MIC in our bacterial populations. If resistance had pre-existed before antibiotic exposure, we would expect other antibiotics to exhibit a similar selective effect, particularly given the potential for cross-resistance to multiple antibiotics."

I think it is important to discuss the negative data for the other antibiotics (along with the other points made in your Reviewer response) in the main text.

This point is discussed in further detail in our response to Reviewer 1 (Public Review).

eLife Assessment

The authors provide a valuable contribution by documenting the role of microglia in pruning the axon terminals of AgRP neurons. The analysis of microglial axonal pruning is solid; however, the analysis of the effects inhibiting microglia on subsequent food consumption is not fully complete.

Reviewer #1 (Public review):

Summary:

This paper shows that maternal high-fat diet during lactation changes microglia morphology in the PVN, potentially to acquire a more active state. Further, the authors reveal that PVN microglia engulf AgRP terminals in the PVN during postnatal development, a previously unrecognized behavior. A notable finding of this paper is that pharmacological reduction of microglial cells can reverse weight gain and terminal loss in the offspring under maternal high fat diet conditions, even though an increase in microglial engulfment of AgRP+ terminals was not observed, suggesting an alternative mechanism. The data support these findings, although questions remain regarding the efficacy and timing of the pharmacological microglial knockdown.

Strengths

(1) The impact of microglia on hypothalamic synaptic pruning is poorly characterized, and thus, the findings herein are especially of interest.

Weaknesses

(1) Most minor concerns were addressed during revisions, including additional details in the methods and results sections that help interpret the data as presented.

(2) The AgRP staining is unclear. For example, in Figure 2, the figure legend says "labeled AgRP terminals (red)" (Fig 2A-D) but then concludes no difference in the number of "AgRP neurons" (Fig 2J). Is this quantification of AgRP+ neurons, terminals, or both?

(3) The PLX experiments are critical to their conclusion that during lactation, microglia in the PVN sculpt AgRP inputs; however, there is no demonstration that PLX treatment effectively eliminated microglia during this postnatal window. Microglia depletion was only assessed at P55, a month past the PLX treatment window making it unclear when and by what percentage the microglia were eliminated.

Reviewer #2 (Public review):

Hypothalamic neural circuits that control body weight develop during the lactation period in rodents. Exposure to maternal high-fat diet during this period (MHFD-L) program has lasting effects on their neuroanatomical organization and function. Microglia sense environmental signals and can sculpt developing circuits by promoting or pruning synaptic connections. Here, the authors examine the contribution of microglia to the effects of MHFD-L to reduce projections from AgRP neurons in the ARH to the PVH, a critical node in circuits regulating energy balance. Using detailed histomorphometric analyses of Iba-1+ cells in the three brain regions (ARH, PVH, and BNST) at two time points (P16 and P30), the authors show that microglial volume and complexity increase, while cell numbers decrease across this period. Exposure to MHFD-L is associated with a transient increase in microglial complexity/volume at P16 in the PVH but not in the other brain regions or time points assessed. Depleting microglia using a pharmacological approach reversed effects of MHD-L on AgRP outgrowth and body weight.

Strengths:

(1) The Introduction is well-written and provides a good overview of what is known about the roles of microglia in sculpting developing circuits in the hippocampus and cortex. This provides a strong rationale for the current investigations in the hypothalamus.

(2) High-quality imaging and detailed 3-D reconstructions of Iba-1 staining in microglia are used to perform unbiased analyses of microglial complexity and to quantify the spatial relationship between microglial processes and AgRP terminals.

Weaknesses:

(1) The central claim of the manuscript is that microglia in the PVH sculpt the density of AgRP inputs to the PVH in a temporally and spatially restricted manner. While the findings of the microglial ablation experiment are consistent with this hypothesis, they do not prove causality, since their manipulations were not limited to the PVH. Further studies are needed to exclude the possibility that increased outgrowth from AgRP neurons results from direct actions in the ARH or indirect consequences of changes in growth rates.

(2) Impacts of microglial depletion were only assessed in adulthood. Given the hypothesized importance of differences in microglia at P16 and not at P30, it would be helpful to demonstrate that PLX5622 does indeed affect microglia at P16, when the circuit is most sensitive to maternal influences.

Reviewer #3 (Public review):

Summary:

The authors interrogated the putative role of microglia in determining AgRP fiber maturation in offspring exposed to a maternal high-fat diet. They found that changes in specific parts of the hypothalamus (but not in others) occur in microglia and that the effect of microglia on AgRP fibers appears to be beyond synaptic pruning, a classical function of these brain-resident macrophages.

Strengths:

The work is very strong in neuroanatomy. The images are clear and nicely convey the anatomical differences. The microglia depletion study adds functional relevance to the paper; however, the pitfalls of the technology regarding functional relevance should be discussed.

Weaknesses:

There was no attempt to functionally interrogate microglia in different parts of the hypothalamus. Morphology alone does not reflect a potential for significant signaling alterations that may occur within and between these and other cell types.

Comments on revised submission: My advice is to change the title by removing "required" and state what is interrogated and found in the paper. A more accurate title would be (for example): Implication of Microglia for Developmental Specification of AgRP Innervation in the Hypothalamus of Offspring Exposed to Maternal High-Fat Diet During Lactation.

I suggest that the authors discuss the limitations of their approach and findings, and propose future directions to address them

Author response:

The following is the authors’ response to the original reviews

Public Reviews:

Reviewer #1 (Public reviews):

(1) A cartoon paradigm of the HFD treatment window would be a helpful addition to Figure 1. Relatedly, the authors might consider qualifying MHFD as 'lactational MHFD.' Readers might miss the fact that the exposure window starts at birth.

This is a good suggestion. The MHFD-L model has been used previously (e.g. Vogt et al. 2014). We have included a cartoon of the MHFD-L model and the PLX treatments to Figure 4, which we feel helps the readers and thank the reviewer for the suggestion.

(2) More details on the modeling pipeline are needed either in Figure 1 or text. Of the ~50 microglia that were counted (based on Figure 1J), were all 50 quantified for the morphological assessments? Were equal numbers used for the control and MHFD groups? Were the 3D models adjusted manually for accuracy? How much background was detected by IMARIS that was discarded? Was the user blind to the treatment group while using the pipeline? Were the microglia clustered or equally spread across the PVN?

In response to this suggestion, we have expanded the description of the image analysis routine in the methods. The analysis focused on detailed changes in microglial morphology as opposed to overall changes in microglia throughout the PVH as a whole. Accordingly, we applied anatomically matched ROIs to the PVH for the measurements. As described in the methods, the Imaris Filaments tool was used to visualize microglia fully contained within a tissue section and a mask derived from the 3D model for these cells was used to isolate them for further analysis, thereby separating these cells from interstitial labeling corresponding to parts of cell processes or other labeling not associated with selected cells. There was no formal “background subtraction.” This was an error in the previous version of the manuscript and we have revised the methods to reflect the process actually used. The images were segmented (to enhance signal to noise for 3D rendering), and then a Gaussian filter was applied to improve edge detection, which facilitates the morphological measurements.

(3) Suggest toning back some of the language. For example: "...consistent with enhanced activity and surveillance of their immediate microenvironment" (Line 195) could be "...perhaps consistent with...". Likewise, "profound" (Lines 194, 377) might be an overstatement.

Revisions have been made to both the Introduction and Discussion to modulate our representation of this controversial issue.

(4) Representative images for AgRP+ cells (quantified in Figure 2J) are missing. Why not a co-label of Iba1+/AgRP+ as per Figure 1, 3? Also, what was quantified in Figure 2J - soma? Total immunoreactivity?

Because the density of AgRP labeling does not change in the ARH we omitted the red channel image (AgRP labeling) to highlight the similarity of the microglial morphology. To address the reviewer’s concerns, in the revised figure we have reconstituted the figure with both the green (microglial) and red (AgRP) channels depicted.

Figure 2J displays the numbers of AgRP neurons counted in the ARH in selected R01s through the ARH. The Methods section has been revised to include the visualization procedure used for the cell counts.

(5) For the PLX experiment:

a) "...we depleted microglia during the lactation period" (Line 234). This statement suggests microglia decreased from the first injection at P4 and throughout lactation, which is inaccurate. PLX5622 effects take time, upwards of a week. Thus, if PLX5622 injections started at P4, it could be P11 before the decrease in microglia numbers is stable. Moreover, by the time microglia are entirely knocked down, the pups might be supplementing some chow for milk, making it unclear how much PLX5622 they were receiving from the dam, which could also impact the rate at which microglia repopulation commences in the fetal brain. Quantifying microglia across the P4-P21 treatment window would be helpful, especially at P16, since the PVN AgRP microglia phenotypes were demonstrated and roughly when pups might start eating some chow. b) I am surprised that ~70% of the microglia are present at P21. Does this number reflect that microglia are returning as the pups no longer receive PLX5622 from milk from the dam? Does it reflect the poor elimination of microglia in the first place?

This is an important point and have revised the first sentence in section 2.3 to clarify the PLX treatment logic and added a cartoon to Fig. 4 to show the treatment timeline. The PLX5622 was not administered to the dams but daily to the pups. We also agree with the interpretation that PLX5622 depleted numbers of microglia, as supported by the microglial cell counts, rather than effected a complete elimination and have made revisions to clarify this distinction. Although mice were weighed at weaning, cellular measurements were only made in mice perfused at P55.

(6) Was microglia morphology examined for all microglia across the PVN? It is possible that a focus on PVNmpd microglia would reveal a stronger phenotype? In Figure 4H, J, AgRP+ terminals are counted in PVN subregions - PVNmpd and PVNpml, with PVNmpd showing a decrease of ~300 AgRP+ terminals in MHFD/Veh (rescued in MHFD/PLX5622). In Figure 1K, AgRP+ terminals across what appears to be the entire PVN decrease by ~300, suggesting that PVNmpd is driving this phenotype. If true, then do microglia within the PVNmpd display this morphology phenotype?

We have revised the description of the analysis procedures to clarify these points. All measurements were made in user defined, matched regions of interest according to morphological features of the PVH. No measurements were made that included the entire PVH and we revised the Methods section to improve clarity.

(7) What chow did the pups receive as they started to consume solid food? Is this only a MHFD challenge, or could the pups be consuming HFD chow that fell into the cage?

The pups were weaned onto the same normal chow diet that the dams received prior to MHFD-L treatment. The cages were inspected daily and minimal HFD spillage was observed, although we cannot rule out with certainty any contribution of the pups directly consuming the HFD. We have edited Methods section 5.2 for clarity.

(8) Figure 5: Does internalized AgRP+ co-localize with CD68+ lysosomes? How was 'internalized' determined?

This important point has been clarified by revisions to the Methods section.

(9) Different sample sizes are used across experiments (e.g., Figure 4 NCD n=5, MHFD n=4). Does this impact statistical significance?

Sample size does impact power of ANOVA with larger samples reducing the chance of errors. ANOVA is generally robust in the face of moderate departures from the assumption of equal sample sizes and equal variance such as we experienced in the PLX5622 experiment. Here we used t-tests to detect differences in a single variable between two groups and two-way ANOVA to compare treatment by diet and treatment changes in the PLX5622 studies. Additional detail has been added to the Methods section to clarify this point.

Reviewer #2 (Public reviews):

(1) Under chow-fed conditions, there is a decrease in the number of microglia in the PVH and ARH between P16 and P30, accompanied by an increase in complexity/volume. With the exception of PVH microglia at P16, this maturation process is not affected by MHFD. This "transient" increase in microglial complexity could also reflect premature maturation of the circuit.

This is an interesting possibility that requires future investigation (see response to Recommended Suggestions, above).

(2) The key experiment in this paper, the ablation of microglia, was presumably designed to prevent microglial expansion/activation in the PVH of MHFD pups. However, it also likely accelerates and exaggerates the decrease in cell number during normal development regardless of maternal diet. Efforts to interpret these findings are further complicated because microglial and AgRP neuronal phenotypes were not assessed at earlier time points when the circuit is most sensitive to maternal influences.

We agree that evaluations of microglia and hypothalamic circuits at many more time points would indeed be informative (see comments above).

(3) Microglial loss was induced broadly in the forebrain. Enhanced AgRP outgrowth to the PVH could be caused by actions elsewhere, such as direct effects on AgRP neurons in the ARH or secondary effects of changes in growth rates.

A local effect of microglia in the ARH that affects growth of AgRP axons remains a distinct possibility that deserves a targeted examination (see response to Recommended Suggestions, above).

(4) Prior publications from the authors and other groups support the idea that the density of AgRP projections to the PVH is primarily driven by factors regulating outgrowth and not pruning. The failure to observe increased engulfment of AgRP fibers by PVH microglia is therefore not surprising. The possibility that synaptic connectivity is modulated by microglia was not explored.

Synaptic pruning and regulation of axon targeting are not mutually exclusive processes and microglia may participate in both. Here we evaluated innervation of the PVH, which is sensitive to MHFD-L exposure, and engulfment of AgRP terminals by microglia, which does appear to be altered by MHFD-L. Given previous observations of terminal engulfment by microglia in other brain regions in response to environmental changes (e.g. prolonged stress) it is not unreasonable to expect this outcome in the offspring of MHFD-L dams.  In future work it will be important to profile multiple cell types in the PVH for microglial dependent and MHFDL-sensitive changes in targeting of AgRP axons. Equally important is a full characterization of postsynaptic changes in PVH neurons.

Reviewer #3 (Public reviews):

There was no attempt to interrogate microglia in different parts of the hypothalamus functionally. Morphology alone does not reflect a potential for significant signaling alterations that may occur within and between these and other cell types.

The authors should discuss the limitations of their approach and findings and propose future directions to address them.

We agree that evaluations of microglia and hypothalamic circuits at many more time points that include analyses of multiple regions would indeed be informative. We have added statements to the manuscript that address the limitations of our experimental approach and suggest future studies that will extend understanding of underlying mechanisms beyond those investigated here.

Recommendations for the authors:

Reviewing Editors Comments:

(1) The Abstract is 405 words and should be shortened to less than 200 words.  

The abstract has been edited to 200 words.

(2) The authors might consider raising the question in the Introduction of whether reduced AgRP innervation of the PVN in MHFD-treated mice is due to decreased axonal growth, enhanced microglial-mediated pruning, or a combination of both. The potential effects on axonal growth should be given more consideration.

This is an important point that we agree deserves additional consideration in the manuscript. Our past work has focused on leptin’s ability to influence axonal targeting of PVH neurons by AgRP and PPG neurons through a cell-autonomous mechanism and our conclusion is that leptin primarily induces axon growth. Because in this study our design did not focus on changes in axon growth over time but on regional changes in microglia and their interactions with AgRP terminals we did not want to divert attention from our logic in the introduction by highlighting multiple mechanisms. However, we have added a brief mention in the Introduction and have expanded consideration of axonal growth effects to the Discussion. Distinguishing between microglia’s role in synaptic density or axon targeting in this pathway is an important goal of future work.

(3) Line 37, a high-fat diet should be defined here as HFD and used consistently thereafter. Note that "high-fat-diet exposure" requires two hyphens.

The suggested revisions have been made throughout the manuscript.

(4) Line 38 and elsewhere, MHFD does not adequately describe the treatment being limited to the lactation period, perhaps MLHFD would be better or just LHFD (because the pups can't lactate).

The suggested revisions have been made throughout the manuscript, and we have used MHFD-L to describe maternal consumption of a high-fat diet that is restricted to the lactation period.

(5) Line 110, leptin-deficient mice (add hyphen).

(6) Line 183, NCD should be defined.

The suggested revisions have been made throughout the manuscript.

(7) Lines 237- 238, it is not clear what is widespread in the rostral forebrain. Is it the loss of microglia? What is the dividing point between the rostral and caudal forebrain? Were microglia depleted in the caudal forebrain too?

We have revised this section of the manuscript to focus the description on the hypothalamus alone and specify that the reduction in microglial density is not restricted to the PVH.  

(8) Line 245, microglial-mediated effects (add hyphen).

(9) Line 247, vehicle-treated mice (add hyphen).

The suggested revisions have been made throughout the manuscript.

(10) Line 457, when referring to genes, the approved gene name should be used in italics, AgRP should be Agrp (italics).

The suggested revision has been made throughout the manuscript.

(11) Line 459, the name of the Syn-Tom mice in the Key Resource table, Methods, and Text should be consistent. It would be best to use the formal name of the Ai34 line of mice on the JAX website.

The suggested revisions have been made throughout the manuscript.

(12) Figure 1G H, and I um should have Greek micro; Fig. 1J and K, Replace # with Number. The same suggestions apply to all the other figures.

Both the manuscript and figures have been revised in accordance with this recommendation.

(13) Figures 4 G, H, I and J. and Figures 5 M and O. The font size is too small to see well.

Fonts have been changed in the figures to improve visibility.

Reviewer #1 (Recommendations for the authors):

(1) Figures are out of order in the text. For example, Figure 1A is followed next by the results for Figure 1J instead of Figure 1B.

We regret that the organization of figure panels makes for awkward matching for the reader as they proceed through the text. We designed the figures to facilitate comparisons between cellular responses and differences in labeling. After evaluating a reorganization, we decided to maintain the original panel configurations, but have revised the text to more closely follow the presentation of cellular features in the figures.

(2) Figure 1B.: All images lack scale bars.

(3) Line 433 - 'underlie' is spelled wrong.

(4) Rosin et al should be 2019 and not 2018.

These corrections have been implemented in the revised text and figures.

(5) The statement that "the effects of MHFD on microglial morphology in the PVH of offspring display both temporal and regional specificity, which correspond to a decrease in the density of AgRP inputs to the PVH" (Line 196) needs clarification, as the phrase "regional specificity" has not been substantiated in this section even though it is discussed later.

We agree with this comment and have revised section 2.1 to more closely match the data presented to this point in the manuscript.

Reviewer #2 (Recommendations for the authors):

(1) The claim of "spatial specificity" in the effects of MHFD on microglia is based on an increase in the complexity/volume of microglia at P16 in the PVH that was not seen in the ARH or BNST. The transient nature of the effect raises several questions: Does the effect on the PVH represent premature maturation?

This is an interesting suggestion. However, given how little is known about microglial maturation in the hypothalamus it is difficult to address. It is indeed possible that microglia mature at different rates in each AgRP target, and that MHFD-L exposure alters the rate of maturation in some regions but not others. This will require a great deal more analysis of both microglia and ARH projections to understand fully (see below).

(2) To support their central claim that microglia in the PVH "sculpt the density of AgRP inputs to the PVH" the authors report effects on Iba1+ cells in the PVH of chow-fed dams at P55, body weight at P21, and AgRP projections in the PVH at an unspecified age. It is hard to understand what is happening across "normal" development in chow-fed dams since the number of Iba1+ cells decreases from ~50 to ~25 between P16 and P30 (Figure 1), but then increases to >60 cells at P55 (Figure 4). Given the large fluctuations in microglial population across time, analyzing the same parameters (i.e. microglial number/morphology in the ARH and PVH, AgRP neuronal number in the ARH, and fiber density in the PVH, and body weight) across time points before, during and after the critical period in chow and MHFD conditions would be very helpful.

The time points we evaluated were chosen to be during and after the previously determined critical period for development of AgRP projections to the PVH, which were then compared with adults (which were all P55) to assess longevity of the effects. We have incorporated revisions to improve the clarity of when measurements were assessed, and treatments implemented. Defining the cellular dynamics of microglia across time remains a major challenge for the field and will certainly be informed by future studies with additional time points, as well as by in vivo imaging studies focused on regions identified here. Although such studies are beyond the scope of the present work, their completion would advance our current understanding of how microglia respond to nutritional changes during development of feeding circuits.

(3) As microglia are also ablated in the ARH, direct effects on AgRP neurons or indirect effects via changes in growth rates could also contribute to increased AgRP fiber density in the PVH. In support of the first possibility, postnatal microglial depletion increases the number of AgRP neurons (Sun, et al. 2023).

We agree with the suggestion, also raised by the Reviewing Editor, which has been addressed briefly in the Introduction, and in more detail by revisions to the Discussion section.

(4) The failure to assess alpha-MSH fibers in the same animals was a missed opportunity. They are also affected by MHFD but likely involve a distinct mechanism (Vogt, et al 2014).

Given the paired interest in POMC neurons and AgRP neurons I understand the reviewer’s comment. We chose to focus solely on AgRP neurons because we do not currently have a way to genetically target axonal labeling exclusively to POMC neurons due to the shared precursor origin of POMC neurons and a percentage of NPY neurons in the ARH, as shown by Lori Zeltser’s laboratory. Moreover, the elegant work by Vogt et al. focused on responses of POMC neurons in the MHFD-L model. However, it certainly remains possible that microglia in the PVH interact with terminals derived from POMC neurons, as well as with terminals derived from other afferent populations of neurons.

(5) All statistical analyses involved unpaired t-tests. Two-way ANOVAs should be used to assess the effects of age and HFD and interactions between these factors.

We used t-tests to detect differences in a single variable between two groups and two-way ANOVA to compare treatment by diet and treatment changes in the PLX5622 studies.  Additional detail has been added to the Methods section and information added to the figure legend for Fig. 4 to clarify this point.

Reviewer #3 (Recommendations for the authors):

I suggest exploring the deeper characterization of the microglia in various parts of the hypothalamus in different conditions. This could include cytokine assessment or spatial transcriptomic.

We agree that a great deal more work is needed to improve our understanding of how microglia impact hypothalamic development more broadly and to identify underlying molecular mechanisms. We are hopeful that the data presented here will motivate additional study of microglial dynamics in multiple hypothalamic regions, as well as detailed studies of cellular signaling events for factors derived from MHFD-L dams that impact neural development in the hypothalamus.

eLife Assessment

By performing a chemical screen of an FDA-approved library of small molecules against a leucine-dependent Mtb strain, this work discovered that semapimod inhibits Mtb growth by impairing leucine import. The work is useful because it connects leucine uptake with the cell wall lipids in Mtb; however, it remains incomplete as the evidence supporting semapimod's ability to target leucine uptake needs more substantial proof. The work requires significant experimental evidence to identify leucine transporter(s) and determine how PDIM participates in leucine uptake.

Reviewer #1 (Public review):

Summary:

In this manuscript, the authors used a leucine/pantothenate auxotrophic strain of Mtb to screen a library of FDA-approved compounds for their antimycobacterial activity and found significant antibacterial activity of the inhibitor semapimod. In addition to alterations in pathways, including amino acid and lipid metabolism and transcriptional machinery, the authors demonstrate that semapimod treatment targets leucine uptake in Mtb. The work presents an interesting connection between nutrient uptake and cell wall composition in mycobacteria.

Strengths:

(1) The link between the leucine uptake pathway and PDIM is interesting but has not been characterized mechanistically. The authors discuss that PDIM presents a barrier to the uptake of nutrients and shows binding of the drug with PpsB. However it is unclear why only the leucine uptake pathway was affected. We still do not know what PpsB actually does for amino acid uptake - is it a transporter? Does semapimod binding affect its activity? Does the auxotrophic Mtb have lower PDIM levels compared to wild-type Mtb?

(2) The authors show an interesting result where they observed antibacterial activity of semapimod against H37Rv only in vivo and not in vitro. Why do the authors think this is the basis of this observation? It is possible semapimod has an immunomodulatory effect on the host since leucine is an essential amino acid in mice. The authors could check pro-inflammatory cytokine levels in infected mouse lungs with and without drug treatment.

(3) The authors show that the semapimod-resistant auxotroph lacks PDIM. The conclusions would be further strengthened by including validations using PDIM mutants, including del-ppsB Mtb and other genes of the PDIM locus, whether in vivo this mutant would be more susceptible (or resistant) to semapimod treatment.

(4) Prolonged subculturing can introduce mutations in PDIM, which can be overcome by supplementing with propionate (Mullholland et al, Nat Microbiol, 2024). Did the authors also supplement their cultures with propionate? It would be interesting to see what mutations would result in Semr strains with propionate supplementation along with prolonged semapimod treatment.

Weaknesses:

I have summarized the limitations above in my comments. Overall, it would be helpful to provide more mechanistic details to study the connection between leucine uptake and PDIM.

Reviewer #2 (Public review):

Summary

This important study uncovers a novel mechanism for L-leucine uptake by M. tuberculosis and shows that targeting this pathway with 'Semapimod' interferes with bacterial metabolism and virulence. These results identify the leucine uptake pathway as a potential target to design new anti-tubercular therapy.

Strengths

The authors took numerous approaches to prove that L-leucine uptake of M. tuberculosis is an important physiological phenomenon and may be effectively targeted by 'Semapimod'. This study utilizes a series of experiments using a broad set of tools to justify how the leucine uptake pathway of M. tuberculosis may be targeted to design new anti-tubercular therapy.

Weaknesses

The study does not explain how L-leucine is taken up by M. tuberculosis, leaving the mechanism unclear. Even though 'Semapimod' binds to the PpsB protein, the relevant connection between changes in PDIM and amino acid transport remains incomplete. Also, the fact that the drug does not function on WT bacteria makes it a weak candidate to consider its usefulness for a therapeutic option.

Reviewer #3 (Public review):

Agarwal et al identified the small molecule semapimod from a chemical screen of repurposed drugs with specific antimycobacterial activity against a leucine-dependent strain of M. tuberculosis. To better understand the mechanism of action of this repurposed anti-inflammatory drug, the authors used RNA-seq to reveal a leucine-deficient transcriptomic signature from semapimod challenge. The authors then measured a decreased intracellular concentration of leucine after semapimod challenge, suggesting that semapimod disrupts leucine uptake as the primary mechanism of action. Unexpectedly, however, resistant mutants raised against semapimod had a mutation in the polyketide synthase gene ppsB that resulted in loss of PDIM synthesis. The authors believe growth inhibition is a consequence of decreased accumulation of leucine as a result of an impaired cell wall and a disrupted, unknown leucine transporter. This study highlights the importance of branched-chain amino acids for M. tuberculosis survival, and the chemical genetic interactions between semapimod and ppsB indicate that ppsB is a conditionally essential gene in a medium depleted of leucine.

The conclusions regarding the leucine and PDIM phenotypes are moderately supported by experimental data. The authors do not provide experimental evidence to support a specific link between leucine uptake and impaired PDIM production. Additional work is needed to support these claims and strengthen this mechanism of action.

(1) Since leucine uptake and PDIM synthesis are important concepts of the manuscript, experiments would benefit from exploring other BCAAs to know if the phenotypes observed are specific to leucine, and adding additional strains to the 2D TLC experiments to provide confidence in the absence of the PDIM band.

(2) The intriguing observation that wild-type H37Rv is resistant to semapimod but the leucine-auxotroph is sensitive should be further explored. If the authors are correct and semapimod does inhibit leucine uptake through a specific transporter or disrupted cell wall (PDIM synthesis), testing semapimod activity against the leucine-auxotroph in various concentrations of BCAAs could highlight the importance of intracellular leucine. H37Rv is still able to synthesize endogenous leucine and is able to circumvent the effect of semapimod.

Author response:

Public Reviews:

Reviewer #1 (Public review):

Summary:

n this manuscript, the authors used a leucine/pantothenate auxotrophic strain of Mtb to screen a library of FDA-approved compounds for their antimycobacterial activity and found significant antibacterial activity of the inhibitor semapimod. In addition to alterations in pathways, including amino acid and lipid metabolism and transcriptional machinery, the authors demonstrate that semapimod treatment targets leucine uptake in Mtb. The work presents an interesting connection between nutrient uptake and cell wall composition in mycobacteria.

Strengths:

The link between the leucine uptake pathway and PDIM is interesting but has not been characterized mechanistically. The authors discuss that PDIM presents a barrier to the uptake of nutrients and shows binding of the drug with PpsB. However it is unclear why only the leucine uptake pathway was affected.

We observe interference of L-leucine, but not of pantothenate, uptake in mc2 6206 strain upon semapimod treatment. At present, we do not have any clue whether PDIM presents a barrier exclusively to the uptake of L-leucine. Further studies may shed a light on underlying mechanism(s) by which L-leucine uptake is modulated by this small molecule.

We still do not know what PpsB actually does for amino acid uptake - is it a transporter?

By BLI-Octet we do not find any interaction between L-leucine and PpsB. Therefore, we doubt that PpsB is a transporter of L-leucine.

Does semapimod binding affect its activity?

Our study suggests that semapimod treatment alters PDIM architecture which becomes restrictive to L-leucine. However, at present the exact mechanism is not clear. Further studies are required to thoroughly examine the effect of semapimod on Mtb PpsB activity and alterations in PDIM by mass spectrometry.

Does the auxotrophic Mtb have lower PDIM levels compared to wild-type Mtb?

As per the published report by Mulholland et al, and by vancomycin susceptibility phenotype in our study, both the strains appear to have comparable PDIM levels.

The authors show an interesting result where they observed antibacterial activity of semapimod against H37Rv only in vivo and not in vitro. Why do the authors think this is the basis of this observation? It is possible semapimod has an immunomodulatory effect on the host since leucine is an essential amino acid in mice. The authors could check pro-inflammatory cytokine levels in infected mouse lungs with and without drug treatment.

Semapimod inhibits production of proinflammatory cytokines such as TNF-α, IL-1β, and IL-6, which would indeed help pathogen establish chronic infection. However, a significant reduction in bacterial loads in lungs and spleen upon semapimod treatment despite inhibition of proinflammatory cytokines clearly indicates bacterial dependence on host-derived exogenous leucine during intracellular growth.

The authors show that the semapimod-resistant auxotroph lacks PDIM. The conclusions would be further strengthened by including validations using PDIM mutants, including del-ppsB Mtb and other genes of the PDIM locus, whether in vivo this mutant would be more susceptible (or resistant) to semapimod treatment.

PDIM is a virulence factor, and plays an important role in the intracellular survival of the TB pathogen. Mtb strains lacking PDIM are expected to show attenuated growth during infection, even without semapimod treatment. In such a case, it might be difficult to draw any conclusions about the effect of semapimod against PDIM(-) strains in vivo.

Prolonged subculturing can introduce mutations in PDIM, which can be overcome by supplementing with propionate (Mullholland et al, Nat Microbiol, 2024). Did the authors also supplement their cultures with propionate? It would be interesting to see what mutations would result in Semr strains with propionate supplementation along with prolonged semapimod treatment.

Considering the fact that extensive subculturing may result in loss of PDIM, we avoided prolonged subculturing of bacteria. As presented in Fig. 6b, the WT bacteria retain PDIM. While performing the initial screening of drugs, we did not anticipate such phenotype, and hence bacteria were cultured in regular 7H9-OADS medium without propionate supplementation.

A comprehensive future study would help examining the effect of propionate on generation of semapimod resistant mutants in Mtb mc2 6206.

Weaknesses:

I have summarized the limitations above in my comments. Overall, it would be helpful to provide more mechanistic details to study the connection between leucine uptake and PDIM.

Reviewer #2 (Public review):

Summary

This important study uncovers a novel mechanism for L-leucine uptake by M. tuberculosis and shows that targeting this pathway with 'Semapimod' interferes with bacterial metabolism and virulence. These results identify the leucine uptake pathway as a potential target to design new anti-tubercular therapy.

Strengths

The authors took numerous approaches to prove that L-leucine uptake of M. tuberculosis is an important physiological phenomenon and may be effectively targeted by 'Semapimod'. This study utilizes a series of experiments using a broad set of tools to justify how the leucine uptake pathway of M. tuberculosis may be targeted to design new anti-tubercular therapy.

Weaknesses

The study does not explain how L-leucine is taken up by M. tuberculosis, leaving the mechanism unclear. Even though 'Semapimod' binds to the PpsB protein, the relevant connection between changes in PDIM and amino acid transport remains incomplete.

While Leucine uptake involves specific transporters in other bacteria, such transport system is not known in Mtb. By screening small molecule inhibitors, we came across a molecule, semapimod, which selectively kills the leucine auxotroph (mc2 6206), but not the WT Mtb. To understand the underlying mechanism of differential susceptibility of the WT and auxotrophic strains to this molecule, we evaluated the effect of restoration of leuCD and panCD expression on susceptibility of the auxotrophic strain to semapimod. Interestingly, our results demonstrated that upon endogenous expression of leuCD genes, mc2 6206 strain becomes resistant to killing by semapimod. In contrast, no effect of panCD expression was observed on semapimod susceptibility of mc2 6206. These findings were further substantiated by gene expression analysis of semapimod treated mc2 6206, which exhibits differential regulation of a set of genes that are altered upon leucine depletion in Mtb as well as in other bacteria. Overall results thus provide first evidence of perturbation of L-leucine uptake by semapimod treatment of the leucine auxotroph.

To further gain mechanistic insights into the effect of semapimod on leucine uptake in Mtb, we generated the semapimod resistant strain which exhibits point mutation in 4 genes including ppsB. Interestingly, overexpression of wild-type ppsB, but not of other genes, restored susceptibility of the resistant bacteria to semapimod. Our observations that semapimod interacts with PpsB, and semapimod resistant strain accumulates mutation in PpsB resulting in loss of PDIM together support the involvement of cell-wall PDIM in regulation of L-leucine transport in Mtb.

As mentioned above, we anticipate that semapimod treatment brings about certain modifications in PDIM which becomes more restrictive to L-leucine. A comprehensive future study will be helpful to examine the effect of semapimod on Mtb physiology.

Also, the fact that the drug does not function on WT bacteria makes it a weak candidate to consider its usefulness for a therapeutic option.

We agree that semapimod is not an appropriate drug candidate against TB owing to its inhibitory effect on production of proinflammatory cytokines such as TNF-α, IL-1β, and IL-6 that help pathogen establish chronic infection. However, a significant reduction in bacterial loads in lungs and spleen upon semapimod treatment despite inhibition of proinflammatory cytokines clearly indicates bacterial dependence on host-derived exogenous leucine during intracellular growth. Therefore targeting L-leucine uptake can be a novel therapeutic strategy against TB.

Reviewer #3 (Public review):

Agarwal et al identified the small molecule semapimod from a chemical screen of repurposed drugs with specific antimycobacterial activity against a leucine-dependent strain of M. tuberculosis. To better understand the mechanism of action of this repurposed anti-inflammatory drug, the authors used RNA-seq to reveal a leucine-deficient transcriptomic signature from semapimod challenge. The authors then measured a decreased intracellular concentration of leucine after semapimod challenge, suggesting that semapimod disrupts leucine uptake as the primary mechanism of action. Unexpectedly, however, resistant mutants raised against semapimod had a mutation in the polyketide synthase gene ppsB that resulted in loss of PDIM synthesis. The authors believe growth inhibition is a consequence of decreased accumulation of leucine as a result of an impaired cell wall and a disrupted, unknown leucine transporter. This study highlights the importance of branched-chain amino acids for M. tuberculosis survival, and the chemical genetic interactions between semapimod and ppsB indicate that ppsB is a conditionally essential gene in a medium depleted of leucine.

The conclusions regarding the leucine and PDIM phenotypes are moderately supported by experimental data. The authors do not provide experimental evidence to support a specific link between leucine uptake and impaired PDIM production. Additional work is needed to support these claims and strengthen this mechanism of action.

As mentioned above, overall results from this study provide first evidence of perturbation of L-leucine uptake by semapimod treatment of the leucine auxotroph. Our observations that semapimod interacts with PpsB, and semapimod resistant strain accumulates mutation in PpsB resulting in loss of PDIM together support the involvement of cell-wall PDIM in regulation of L-leucine transport in Mtb.

As hitherto mentioned, it appears that semapimod treatment brings about certain modifications in PDIM which becomes restrictive to L-leucine. Future studies are required to gain detailed mechanistic insights into the effect of semapimod on Mtb physiology.

Since leucine uptake and PDIM synthesis are important concepts of the manuscript, experiments would benefit from exploring other BCAAs to know if the phenotypes observed are specific to leucine, and adding additional strains to the 2D TLC experiments to provide confidence in the absence of the PDIM band.

We thank the peer reviewer for this suggestion. We would be happy to analyse the effect of semapimod on the level of other amino acids including BCAA by mass spectrometry.

The intriguing observation that wild-type H37Rv is resistant to semapimod but the leucine-auxotroph is sensitive should be further explored. If the authors are correct and semapimod does inhibit leucine uptake through a specific transporter or disrupted cell wall (PDIM synthesis), testing semapimod activity against the leucine-auxotroph in various concentrations of BCAAs could highlight the importance of intracellular leucine. H37Rv is still able to synthesize endogenous leucine and is able to circumvent the effect of semapimod.

We thank the peer reviewer for this suggestion. We would explore the possibility of analysing the effect of increasing concentrations of BCAAs on mc2 6206 susceptibility to semapimod.

eLife Assessment

This useful study shows a protective role of type 1 IFN during Mycobacterium tuberculosis infection. It shows that the type 1 IFN response in human skin TST inversely correlates with TB severity, suggesting its protective role. Considering that type I IFN is usually shown to be pro-pathogenic, the higher vulnerability of zebrafish larvae lacking stat2 to M marinum infection is a strong result. However, the conclusion that IFN-I is protective during mycobacterial infection remains indirect and incomplete; the study requires additional mechanistic insights and validation.

Reviewer #1 (Public review):

Summary:

The manuscript finds a negative relationship between tuberculin skin test-induced type I interferon activity with chest X-ray tuberculosis severity in humans. This evidence is between incomplete and solid. It needs a bioinfomatics/transcriptomics reviewer to make a more insightful judgement. The manuscript demonstrates a convincing role for Stat2 in controlling Mycobacterium marinum infection in zebrafish embryos, incomplete data are presented linking reduced leukocyte recruitment to the infection susceptibility phenotype.

Strengths:

(1) An interesting analysis of TST response correlated with chest X-ray pathology.

(2) Novel data on a protective role for Stat2 in a natural host-mycobacterial species infection pairing.

Weaknesses:

(1) The transcriptional modules are very large sets of genes that do not present a clear picture of what is actually being measured relative to other biological pathways.

(2) The link between infection-Stat2-leukocyte recruitment and containment of infection is plausible, but lacks a specific link to the first part of the manuscript.

Major concerns

(1) Line 158: The two transcriptional modules should be placed in the context of other DEG patterns. The macrophage type I interferon module, in particular, is quite large (361 genes). Can this be made more granular in terms of type I IFN ligands and STAT2-dependent genes?

(2) The ifnphi1 injection into mxa:mCherry stat2 crispants is a nice experiment to demonstrate loss of type I IFN responsiveness. Further data is required to demonstrate if important mycobacterial control pathways (IFNy, TNF, il6?, etc) are intact in stat2 crispants before being able to conclude that these phenotypes are specific to type I IFN.

Reviewer #2 (Public review):

Summary:

This study shows that type I interferon (IFN-I) signaling helps protect against mycobacterial infection. Using human gene expression data and a zebrafish model, the authors find that reduced IFN-I activity is linked to more severe disease. They also show that zebrafish lacking the IFN-I signaling gene stat2 are more vulnerable to infection due to poor macrophage migration. These results suggest a protective role for IFN-I in mycobacterial disease, challenging previous findings from other animal models.

Strengths:

Strengths of the manuscript include the use of human clinical samples to support relevance to disease, along with a genetically tractable zebrafish model that enables mechanistic insight.

Weaknesses:

(1) The manuscript presents intriguing human data showing an inverse correlation between IFN-I gene signatures and TB disease, but the findings remain correlative and may be cohort-specific. Given that the skin is not a primary site of TB and is relatively immunotolerant, the biological relevance of downregulated IFN-I-related genes in this tissue to systemic or pulmonary TB is unclear.

(2) The reliance on stat2 CRISPants in zebrafish offers a limited view of IFN-I signaling. Including additional crispant lines targeting other key regulators (e.g., ifnar1, tyk2, irf3, irf7) would strengthen the interpretation and clarify whether the observed effects reflect broader IFN-I pathway disruption.

(3) The conclusion that IFN-I is protective contrasts with established findings from murine and non-human primate models, where IFN-I is often detrimental. While the authors highlight species differences, the lack of functional human data and reliance on M. marinum in zebrafish limit the translational relevance. A more balanced discussion addressing these discrepancies would improve the manuscript.

(4) Quantification of bacterial burden using fluorescence intensity alone may not accurately reflect bacterial viability. Complementary methods, such as qPCR for bacterial DNA, would provide a more robust assessment of antimicrobial activity.

(5) Finally, the authors should clarify whether impaired macrophage recruitment in stat2 crispants results from defects in chemotaxis, differentiation, or survival, and address discrepancies between their human blood findings and prior studies.

Reviewer #3 (Public review):

Summary:

In this manuscript, the authors presented an interesting study providing an insight into the role of Type-I interferon responses in tuberculosis (TB) pathogenesis by combining transcriptome analysis of PBMCs and TST from tuberculosis patients. The zebrafish model was used to identify the changes in the innate immune cell population of macrophages and neutrophils. The findings suggested that Type-I interferon signatures inversely correlated with disease severity in the TST transcriptome data. The authors validated the observations by CRISPR-mediated disruption of stat2 (a critical transcription factor for type I interferon signaling) in zebrafish larvae, showing increased susceptibility to M. marinum infection. Traditionally, type-I interferon responses have been viewed as detrimental in mycobacterial infections, with studies suggesting enhanced susceptibility in certain mouse models. The study tried to identify and further characterize the understanding of the role of type-I interferons in TB.

Strengths:

Traditionally, type-I interferon responses have been viewed as detrimental in mycobacterial infections, with studies suggesting enhanced susceptibility in certain mouse models. The study tried to further understand the role of type-I interferons in TB pathogenesis.

Weaknesses:

Though the study showed an inverse correlation of Type-I interferon with radiological features of TB, the molecular mechanism is largely unexplored in the study, which is making it difficult to understand the basis of the results shown in the manuscript by the authors.

Author response:

Public Reviews:

Reviewer #1 (Public review):

Summary:

The manuscript finds a negative relationship between tuberculin skin test-induced type I interferon activity with chest X-ray tuberculosis severity in humans. This evidence is between incomplete and solid. It needs a bioinfomatics/transcriptomics reviewer to make a more insightful judgement. The manuscript demonstrates a convincing role for Stat2 in controlling Mycobacterium marinum infection in zebrafish embryos, incomplete data are presented linking reduced leukocyte recruitment to the infection susceptibility phenotype.

Strengths:

(1) An interesting analysis of TST response correlated with chest X-ray pathology.

(2) Novel data on a protective role for Stat2 in a natural host-mycobacterial species infection pairing.

We appreciate the reviewer’s positive comments.

Weaknesses:

(1) The transcriptional modules are very large sets of genes that do not present a clear picture of what is actually being measured relative to other biological pathways.

The transcriptional module analysis is a major strength of our approach. These gene signatures are derived from independent experiments, most of which have been previously published/validated [1,2]. To clarify, they represent co-regulated gene sets downstream of signalling pathways. Increased number of genes in these modules increases their combinatorial specificity for a given biological pathway. In the human data, they serve as orthogonal validation for the bioinformatic analysis showing enrichment of the type I IFN pathway among TST transcriptome genes that are negatively correlated with radiographic disease severity in pulmonary TB (see Figure 2). Importantly, our modules confirm the relationship with type I IFN signalling (see Figure 2E) by discriminating from type II IFN signalling, which is not statistically significantly correlated with radiographic TB severity (see Figure S6C-E).

(2) The link between infection-Stat2-leukocyte recruitment and containment of infection is plausible, but lacks a specific link to the first part of the manuscript.

For clarification, the first part of the study seeks to identify immune response pathways that relate to severity of human disease, leading to the identification of type I IFN signalling. Since the human data are limited to an observational analysis in which we cannot test causality, the second part of our study uses a genetically tractable experimental model to test the hypothesis that type I IFN signalling is host-protective and explore possible mechanisms for a beneficial effect. This leads to the observation that type I IFN responses contribute to early myeloid cell recruitment to the site of infection, that has previously been shown to be crucial for containment of mycobacterial infection in zebrafish larvae. We will further evaluate the introduction and results sections to ensure a clear link between the human and zebrafish work.

Major concerns

(1) Line 158: The two transcriptional modules should be placed in the context of other DEG patterns. The macrophage type I interferon module, in particular, is quite large (361 genes). Can this be made more granular in terms of type I IFN ligands and STAT2-dependent genes?

We respectfully disagree with this comment. For clarification, the 360 gene module reflects the zebrafish larval response to IFNphi1 protein [3]. Type I IFNs are known to induce hundreds of interferon stimulated genes [4]. As explained above, the size of the modules increases specificity for a given signalling pathway. In this case, we are most interested in discriminating type I and type II IFN signalling pathways that represent very different upstream biological processes. The discrimination we achieve with our modular approach is a major advance over previous reports of gene signatures in TB that do not discriminate between the two pathways. In this study, we did not discriminate between signalling downstream of type I IFN ligands and STAT2, consistent with existing literature showing that type I IFN signalling is STAT2 dependent [5,6].

(2) The ifnphi1 injection into mxa:mCherry stat2 crispants is a nice experiment to demonstrate loss of type I IFN responsiveness. Further data is required to demonstrate if important mycobacterial control pathways (IFNy, TNF, il6?, etc) are intact in stat2 crispants before being able to conclude that these phenotypes are specific to type I IFN.

Thank you for the positive comment. We acknowledge this point and will attempt to evaluate whether pro-inflammatory cytokine responses are intact in stat2 CRISPants by qPCR or bulk RNAseq. However, these experiments may prove inconclusive because of the limited sensitivity in this approach.

Reviewer #2 (Public review):

Summary:

This study shows that type I interferon (IFN-I) signaling helps protect against mycobacterial infection. Using human gene expression data and a zebrafish model, the authors find that reduced IFN-I activity is linked to more severe disease. They also show that zebrafish lacking the IFN-I signaling gene stat2 are more vulnerable to infection due to poor macrophage migration. These results suggest a protective role for IFN-I in mycobacterial disease, challenging previous findings from other animal models.

Strengths:

Strengths of the manuscript include the use of human clinical samples to support relevance to disease, along with a genetically tractable zebrafish model that enables mechanistic insight.

We welcome the reviewer’s positive summary of our study.

Weaknesses:

(1) The manuscript presents intriguing human data showing an inverse correlation between IFN-I gene signatures and TB disease, but the findings remain correlative and may be cohort-specific. Given that the skin is not a primary site of TB and is relatively immunotolerant, the biological relevance of downregulated IFN-I-related genes in this tissue to systemic or pulmonary TB is unclear.

We agree with the reviewer that the observational human data are correlative. That is precisely why we extend the study to undertake mechanistic studies in a genetically tractable animal model, using M. marinum infection of zebrafish larvae. In the introduction, we already provide a detailed rationale for the strengths of the TST model to study human immune responses to a standardised mycobacterial challenge. This approach mitigates against the confounding of heterogeneity in bacterial burden and sampling different stages of the natural history of infection in conventional observational human studies. Therefore, the application of the TST is a major strength of this study. We do not understand the context in which the reviewer suggests the skin is immunotolerant. In the present study and previous work we provide molecular level analysis of the TST as a robust cell mediated immune response that reflects molecular perturbation in granuloma from the site of pulmonary TB disease 1.

(2) The reliance on stat2 CRISPants in zebrafish offers a limited view of IFN-I signaling. Including additional crispant lines targeting other key regulators (e.g., ifnar1, tyk2, irf3, irf7) would strengthen the interpretation and clarify whether the observed effects reflect broader IFN-I pathway disruption.

We respectfully disagree with this comment. Our objective was to test the role of type I IFN signalling in M. marinum infection of zebrafish. We show that stat2 deletion effectively disrupts type I IFN signalling (Figure S8). Therefore, we do not see a compelling rationale to evaluate other molecules in the signalling pathway.

(3) The conclusion that IFN-I is protective contrasts with established findings from murine and non-human primate models, where IFN-I is often detrimental. While the authors highlight species differences, the lack of functional human data and reliance on M. marinum in zebrafish limit the translational relevance. A more balanced discussion addressing these discrepancies would improve the manuscript.

We acknowledge that our findings contrast with the prevailing view in published literature to date. We will further review the discussion to see how we can elaborate on the potential strengths and weaknesses of different experimental approaches, which may underpin these discrepancies.

(4) Quantification of bacterial burden using fluorescence intensity alone may not accurately reflect bacterial viability. Complementary methods, such as qPCR for bacterial DNA, would provide a more robust assessment of antimicrobial activity.

We and others have previously validated the use of the quantitative measures of fluorescence, used here as a measure of bacterial load [7,8]. Importantly, our measurements do not rely purely on the total fluorescence signal, but also measures of dissemination of infection, for which we see consistent findings. It is also widely recognised that DNA measurements do not necessarily correlate well with bacterial viability. Therefore, we respectfully disagree that a PCR-based approach will add substantial value to our existing analysis.

(5) Finally, the authors should clarify whether impaired macrophage recruitment in stat2 crispants results from defects in chemotaxis, differentiation, or survival, and address discrepancies between their human blood findings and prior studies.

We acknowledge that these are important questions. Our data show that stat2 disruption does not impact total macrophage numbers at baseline (Figure 4A,B) and therefore do not support any effect of Stat2 signalling on steady state macrophage survival or differentiation. The downregulation of macrophage mpeg1 expression in M. marinum infection precludes long-term follow-up of these cells in the context of infection [9]. Therefore, we cannot currently test the hypothesis that Stat2 signalling may influence death of macrophages recruited to the site of infection or make them more susceptible to the cytopathic effects of direct mycobacterial infection. We will attempt to confirm using short-term time-lapse imaging that cellular migration to the site of hindbrain M. marinum infection is reduced in stat2 deficient zebrafish. On the strength of what is possible to test and the established role of type I IFNs in induction of several chemokines [10,11], the most likely effect is that Stat2 signalling increases recruitment through chemokine production. We are exploring the possibility of testing changes to the chemokine profile in stat2 CRISPants by qPCR or bulk RNAseq, but these experiments may prove inconclusive because of the limitations of sensitivity in this approach.

We recognize that our finding of no relationship between peripheral blood type I IFN activity and severity of human TB contrasts with that of previous studies. As stated in the discussion, the most likely explanation for this is our use of transcriptional modules which reflect exclusive type I IFN responses. The signatures used in other studies include both type I and type II IFN inducible genes and therefore also reflect IFN gamma driven responses.

Reviewer #3 (Public review):

Summary:

In this manuscript, the authors presented an interesting study providing an insight into the role of Type-I interferon responses in tuberculosis (TB) pathogenesis by combining transcriptome analysis of PBMCs and TST from tuberculosis patients. The zebrafish model was used to identify the changes in the innate immune cell population of macrophages and neutrophils. The findings suggested that Type-I interferon signatures inversely correlated with disease severity in the TST transcriptome data. The authors validated the observations by CRISPR-mediated disruption of stat2 (a critical transcription factor for type I interferon signaling) in zebrafish larvae, showing increased susceptibility to M. marinum infection. Traditionally, type-I interferon responses have been viewed as detrimental in mycobacterial infections, with studies suggesting enhanced susceptibility in certain mouse models. The study tried to identify and further characterize the understanding of the role of type-I interferons in TB.

Strengths:

Traditionally, type-I interferon responses have been viewed as detrimental in mycobacterial infections, with studies suggesting enhanced susceptibility in certain mouse models. The study tried to further understand the role of type-I interferons in TB pathogenesis.

We thank the reviewer for their summary.

Weaknesses:

Though the study showed an inverse correlation of Type-I interferon with radiological features of TB, the molecular mechanism is largely unexplored in the study, which is making it difficult to understand the basis of the results shown in the manuscript by the authors.

We respectfully disagree with this comment. The observations in the human data lead to the hypothesis that type I IFN responses may be host-protective, which we then test specifically in the zebrafish model, and explore candidate mechanisms, focussing on myeloid cell recruitment to the site of infection.

References

(1) Bell, L.C.K., Pollara, G., Pascoe, M., Tomlinson, G.S., Lehloenya, R.J., Roe, J., Meldau, R., Miller, R.F., Ramsay, A., Chain, B.M., et al. (2016). In Vivo Molecular Dissection of the Effects of HIV-1 in Active Tuberculosis. PLoS Pathog. 12, e1005469. https://doi.org/10.1371/journal.ppat.1005469.

(2) Pollara, G., Turner, C.T., Rosenheim, J., Chandran, A., Bell, L.C.K., Khan, A., Patel, A., Peralta, L.F., Folino, A., Akarca, A., et al. (2021). Exaggerated IL-17A activity in human in vivo recall responses discriminates active tuberculosis from latent infection and cured disease. Sci. Transl. Med. 13, eabg7673. https://doi.org/10.1126/scitranslmed.abg7673.

(3) Levraud, J.-P., Jouneau, L., Briolat, V., Laghi, V., and Boudinot, P. (2019). IFN-Stimulated Genes in Zebrafish and Humans Define an Ancient Arsenal of Antiviral Immunity. J. Immunol. Baltim. Md 1950 203, 3361–3373. https://doi.org/10.4049/jimmunol.1900804.

(4) Schoggins, J.W. (2019). Interferon-Stimulated Genes: What Do They All Do? Annu. Rev. Virol. 6, 567–584. https://doi.org/10.1146/annurev-virology-092818-015756.

(5) Blaszczyk, K., Nowicka, H., Kostyrko, K., Antonczyk, A., Wesoly, J., and Bluyssen, H.A.R. (2016). The unique role of STAT2 in constitutive and IFN-induced transcription and antiviral responses. Cytokine Growth Factor Rev. 29, 71–81. https://doi.org/10.1016/j.cytogfr.2016.02.010.

(6) Begitt, A., Droescher, M., Meyer, T., Schmid, C.D., Baker, M., Antunes, F., Knobeloch, K.-P., Owen, M.R., Naumann, R., Decker, T., et al. (2014). STAT1-cooperative DNA binding distinguishes type 1 from type 2 interferon signaling. Nat. Immunol. 15, 168–176. https://doi.org/10.1038/ni.2794.

(7) Stirling, D.R., Suleyman, O., Gil, E., Elks, P.M., Torraca, V., Noursadeghi, M., and Tomlinson, G.S. (2020). Analysis tools to quantify dissemination of pathology in zebrafish larvae. Sci. Rep. 10, 3149. https://doi.org/10.1038/s41598-020-59932-1.

(8) Takaki, K., Davis, J.M., Winglee, K., and Ramakrishnan, L. (2013). Evaluation of the pathogenesis and treatment of Mycobacterium marinum infection in zebrafish. Nat. Protoc. 8, 1114–1124. https://doi.org/10.1038/nprot.2013.068.

(9) Benard, E.L., Racz, P.I., Rougeot, J., Nezhinsky, A.E., Verbeek, F.J., Spaink, H.P., and Meijer, A.H. (2015). Macrophage-expressed perforins mpeg1 and mpeg1.2 have an anti-bacterial function in zebrafish. J. Innate Immun. 7, 136–152. https://doi.org/10.1159/000366103.

(10) Lehmann, M.H., Torres-Domínguez, L.E., Price, P.J.R., Brandmüller, C., Kirschning, C.J., and Sutter, G. (2016). CCL2 expression is mediated by type I IFN receptor and recruits NK and T cells to the lung during MVA infection. J. Leukoc. Biol. 99, 1057–1064. https://doi.org/10.1189/jlb.4MA0815-376RR.

(11) Buttmann, M., Merzyn, C., and Rieckmann, P. (2004). Interferon-beta induces transient systemic IP-10/CXCL10 chemokine release in patients with multiple sclerosis. J. Neuroimmunol. 156, 195–203. https://doi.org/10.1016/j.jneuroim.2004.07.016.

eLife Assessment

This study presents an alternative to conventional and UV-based tetramers, which are easy to use and reliable for the identification of antigen-specific CD8 T cells. The authors demonstrate that tetramers for HLA alleles A0301, A1101, B0702, and C0702 can be subjected to specific temperatures that facilitate peptide exchange, whilst maintaining structural integrity. Whilst the strength of the evidence is currently incomplete, further development and validation of this approach is likely to provide a useful alternative to generating reagents for examining T cell specificities.

Reviewer #1 (Public review):

Summary:

A fundamental technique for the identification of peptide-specific CD8 T cells is the use of fluorophore-conjugated and peptide loaded MHC tetramers. Classically, refolding of specific peptides with MHC monomers can be labour intensive, and not optimal for screening large numbers of different peptides. Hence, UV-exchanged tetramers have been developed to upscale this, however, still has some associated challenges such as UV-mediated damage to peptide complexes. Here, Pothast, C.R. et al demonstrate the efficacy of using temperature exchanged tetramers for the prevalent alleles HLA-A*03:01, A*11:01, B*07:02, and C*07:02. Building upon their previous work with HLA-A*02:01, H-2Kb, and HLA-E. They first demonstrate the complex stability of tetramers with different affinity peptides at high temperature, showing complex destabilisation can be rescued with higher affinity peptides. This is followed by an optimisation of peptide exchange temperatures, tailored for each allele. The authors then demonstrate successful binding to clonal T cell lines, and then a step further with viral peptides against PBMCs from individuals with confirmed infection history. For the latter they compare to conventional tetramers and demonstrate comparable signal.<br /> Due to the prevalence of these 4 alleles, the ease-of-handling, and short time requirements, these tetramers are likely to show high utility.

Strengths:

The manuscript is well-written and the results are solid, although more detail may add clarity to some of the results, in particular Figures 1 and 2. Other than the points reported below, the study uses accurate controls to demonstrate the specificity of the tetramers, and the data are convincing.

Overall, the interpretation of the results is accurate, and the discussion is thorough. Additional comments may be included to cover potential tetramer batch variability and differences in the stability of different alleles. Specifically, whether certain alleles require higher-affinity peptides to be stable, compared to others.

Weaknesses:

The authors demonstrate the equivalence of temperature-exchanged tetramers to conventional ones, however, as they are an advancement on UV-exchange, it would be useful to show data on how their stability, exchange efficacy, and binding to T cell lines compare to UV-based tetramers. It would be supportive to show that temperature does not impact fluorophore intensity as well.

Reviewer #2 (Public review):

Summary:

The majority of CD8+ T cell responses rely on the proper presentation of antigens through stable MHC-I (but not requiring a stable immunological synapse). This work highlights a new approach to build an array of stable peptide MHC-I using temperature exchange, which can be used to identify antigen-specific CD8+ T cells.

Strengths:

In this work, the authors have proposed an alternative method to reload the peptide MHC-I molecule. Their temperature-exchange approach is distinct from current reloadable peptide MHC technologies involving photolabile peptide, empty MHC-I (Nat Commun 11, 1314 (2020). https://doi.org/10.1038/s41467-020-14862-4), tapasin/TAPBPR chaperone-assisted (eLife 7:e40126.), enzyme exchangeable (WO2020226570) and small alcohol (Curr Res Immunol. 2022 Aug 18;3:167-174. doi: 10.1016/j.crimmu.2022.08.002) approaches.

Weaknesses:

However, the proposed temperature-exchange approach does not substantially improve the quality of antigen-specific T cells that can be identified using the photolabile peptide MHC-I molecules.

The time saved using the temperature-exchange protocol may not be a pull factor as the photolabile peptide MHC-I approach is not unreasonably laborious.

Reviewer #3 (Public review):

Summary:

The study by Pothast and colleagues outlines an extension of their previously described temperature-based MHC-I peptide exchange method on 4 common HLA alleles, to enable the generation of peptide/MCH-I tetramers for characterization of antigen-specific T cells by flow cytometry.

Strengths:

This work outlines a protocol for generating MHC-I tetramers on 4 common HLA allotypes, which can then be applied to monitor T cell responses by flow cytometry studies. The work provides conditional ligands for exchange on each HLA and demonstrates proof of concept studies using clonotypic T cells and CD8+ PBMCs.

The results support that the temperature-exchanged tetramers can perform similarly to conventional tetramers in some settings.

Weaknesses:

Given that there are several proposed methodologies addressing the same task (including UV-mediated, disulfide-bond based stabilization of empty MHC-I conformers, and chaperone-based methods), the relevance of the proposed temperature-mediated technology is questionable.

More specifically, important limitations of the study include:

(1) A lack of quantification of exchanged molecules relative to molecules that retain the original placeholder peptides, or completely empty molecules present in the same sample.

(2) A lack of validation that peptide exchange has occurred in the absence of a reporter T cell line appears to be a significant limitation of the methodology for antigen / T cell discovery.

(3) The sub-optimal exchange efficiency relative to conventional prepared pMHC-I molecules, shown in Figure 4, is a significant limitation of the approach.

(4) There are no data to support that exchange proceeds through the generation of empty molecules during the temperature cycle, or by peptide binding on empty molecules that are already present in the sample. Understanding the mechanism of exchange is important for the necessary improvements to the methodology.

(5) It is possible that the temperature cycle causes protein aggregation or other irreversible changes to the sample - this should be explicitly quantified and addressed in the paper, since misfolded MHC-I molecules can lead to high levels of background staining.

(6) These potential limitations should limit detection of low-affinity/low-avidity interactions between TCRs and their cognate pMHC antigens - this should be addressed explicitly in a model antigen setting.

(7) The approach appears to be limited to the HLAs showing high thermal stability, which have been explored in this study. However, a large fraction of HLAs show sub-optimal thermal stabilities. It seems that explicit validation of peptide exchange would be required for any new HLA allele introduced into this process.

(8) Whether the approach can be used to load suboptimal peptides with lower thermal stabilities that are emerging immunotherapy targets is not addressed in the present study.

Because of these limitations, the present manuscript does not conclusively support the claim that temperature-based exchange can be used as a robust methodology to generate pMHC-I tetramers with desired peptide specificities.

As a result, the scope of applications using these suboptimal exchanged pHLA tetramers is limited, and should be addressed with further improvements of the methodology, including better characterization of exchange efficiency, demonstration of functionality across a broader range of HLA allotypes with varying thermal stability profiles, and validation with clinically relevant low-affinity peptides that would strengthen the potential utility of this approach in immunotherapy development and basic T cell biology research.

eLife Assessment

This study presents an important finding linking the bacterial metabolite trimethylamine and its receptor to circadian rhythms and olfaction. The current evidence supporting the claims of the authors is convincing, although further data and improvements to the presentation would further increase the impact of these results. This work will be of broad interest to researchers interested in nutrition, microbial metabolism, circadian rhythms, and host-microbiome interactions.

Reviewer #1 (Public review):

Summary:

This study focuses on the bacterial metabolite TMA, generated from dietary choline. These authors and others have previously generated foundational knowledge about the TMA metabolite TMAO, and its role in metabolic disease. This study extends those findings to test whether TMAO's precursor, TMA, and its receptor TAAR5 are also involved and necessary for some of these metabolic phenotypes. They find that mice lacking the host TMA receptor (Taar5-/-) have altered circadian rhythms in gene expression, metabolic hormones, gut microbiome composition, and olfactory and innate behavior. In parallel, mice lacking bacterial TMA production or host TMA oxidation have altered circadian rhythms.

Strengths:

These authors use state-of-the-art bacterial and murine genetics to dissect the roles of TMA, TMAO, and their receptor in various metabolic outcomes (primarily measuring plasma and tissue cytokine/gene expression). They also follow a unique and unexpected behavioral/olfactory phenotype. Statistics are impeccable.

Weaknesses:

Enthusiasm for the manuscript is dampened by some ambiguous writing and the presentation of ideas in the introduction, both of which could easily be improved upon revision.

Reviewer #2 (Public review):

Summary:

In the manuscript by Mahen et al., entitled "Gut Microbe-Derived Trimethylamine Shapes Circadian Rhythms Through the Host Receptor TAAR5," the authors investigate the interplay between a host G protein-coupled receptor (TAAR5), the gut microbiota-derived metabolite trimethylamine (TMA), and the host circadian system. Using a combination of genetically engineered mouse and bacterial models, the study demonstrates a link between microbial signaling and circadian regulation, particularly through effects observed in the olfactory system. Overall, this manuscript presents a novel and valuable contribution to our understanding of host-microbe interactions and circadian biology. However, several sections would benefit from improved clarity, organization, and mechanistic depth to fully support the authors' conclusions.

Strengths:

(1) The manuscript addresses an important and timely topic in host-microbe communication and circadian biology.

(2) The studies employ multiple complementary models, e.g., Taar5 knockout mice, microbial mutants, which enhance the depth of the investigation.

(3) The integration of behavioral, hormonal, microbial, and transcript-level data provides a multifaceted view of the observed phenotype.

(4) The identification of olfactory-linked circadian changes in the context of gut microbes adds a novel perspective to the field.

Weaknesses:

While the manuscript presents compelling data, several weaknesses limit the clarity and strength of the conclusions.

(1) The presentation of hormonal, cytokine, behavioral, and microbiome data would benefit from clearer organization, more detailed descriptions, and functional grouping to aid interpretation.

(2) Some transitions-particularly from behavioral to microbiome data-are abrupt and would benefit from better contextual framing.

(3) The microbial rhythmicity analyses lack detail on methods and visualization, and the sequencing metadata (e.g., sample type, sex, method) are not clearly stated.

(4) Several figures are difficult to interpret due to dense layouts or vague legends, and key metabolites and gene expression comparisons are either underexplained or not consistently assessed across models.

(5) Finally, while the authors suggest a causal role for TAAR5 and its ligand in circadian regulation, the current data remain correlative; mechanistic experiments or stronger disclaimers are needed to support these claims.

Reviewer #3 (Public review):

Summary:

Deletion of the TMA-sensor TAAR5 results in circadian alterations in gene expression, particularly in the olfactory bulb, plasma hormones, and neurobehaviors.

Strengths:

Genetic background was rigorously controlled.

Comprehensive characterization.

Weaknesses:

The weaknesses identified by this reviewer are minor.

Overall, the studies are very nicely done. However, despite careful experimentation, I note that even the controls vary considerably in their gene expression, etc, across time (eg, compare control graphs for Cry 1 in IB, 4B). It makes me wonder how inherently noisy these measurements are. While I think that the overall point that the Taar5 KO shows circadian changes is robust, future studies to dissect which changes are reproducible over the noise would be helpful.

Impact:

These data add to the growing literature pointing to a role for the TMA/TMAO pathway in olfaction and neurobehavioral.

eLife Assessment

Endothelial cell-specific loss of TGF-beta signaling in mice leads to CNS vascular defects, specifically impairing retinal development and promoting immune cell infiltration. The data are solid, showing that loss of TGF-beta signaling triggers vascular inflammation and attracts immune cells specific to CNS vasculature, but there are issues with the single-nucleus RNA sequencing of immune cells. These findings are valuable, highlighting TGF-beta's role in maintaining vascular-immune homeostasis and its therapeutic potential in neurovascular inflammatory diseases.

Reviewer #1 (Public review):

Summary:

This manuscript analyses primarily the effects of deleting the TgfbR1 and TgfbR2 receptors from endothelial cells at postnatal stages of vascular development and blood-retina barrier maturation in the retina. The authors find that deletion of these receptors affects vascular development in the retina, but importantly, it affects the infiltration of immune cells across the vessels in the retina. The findings demonstrate that Tgfb signaling through TgfbR1/R2 heterodimers regulates primarily the immune phenotypes of endothelial cells in addition to regulating vascular development. The data provided by the authors provide a solid support for their conclusions.

Strengths:

(1) The manuscript uses a variety of elegant genetic studies in mice to analyze the role of TgfbR1 and TgfbR2 receptors in endothelial cells at postnatal stages of vascular development and blood-retina barrier maturation in the retina.

(2) The authors provide a nice comparison of the vascular phenotypes in endothelial-specific knockout of TgfbR1 and TgfbR2 in the retina (and to a lesser degree in the brain) with those from Npd KO mice (loss of Ndp/Fzd signaling) or loss of VEGF-A signaling to dissect the specific roles of Tgf signaling for vascular development in the retina.

(3) The snRNAseq data of vessel segments from the brains of WT versus TgfbR1 -iECKO mice provides a nice analysis of pathways and transcripts that are regulated by Tgfb signaling in endothelial cells.

Weaknesses:

(1) The authors claim that choroidal neovascular tuft phenotypes are similar in TgfbrR1 KO and TgfbrR2 KO mice. However, the phenotypes look more severe in the TgfbrR1 KO rather than TgfbrR2 KO mice. Can the authors show a quantitative comparison of the number of choroidal neovascular tufts per whole eye cross-section in both genotypes?

(2) In the analysis of Sulfo-NHS-Biotin leakage in the retina to assess blood-retina barrier maturation. The authors claim that there is increased vascular leakage in the TgfbR1 KO mice. However, it does not seem like Sulfo-NHS-biotin is leaking outside the vessels. Therefore, it cannot be increased vascular permeability. Can the authors provide a detailed quantification of the leakage phenotype?

(3) The immune cell phenotyping by snRNAseq is premature, as the number of cells is very small. The authors should sort for CD45+ cells and perform single-cell RNA sequencing.

(4) The analysis of BBB leakage phenotype in TgfbR1 KO mice needs to be more detailed and include tracers as well as serum IgG leakage.

(5) A previous study (Zarkada et al., 2021, Developmental Cell) showed that EC-deletion of Alk5 affects the D tip cells. The phenotypes of those mice look very similar to those shown for TgfbrR1 KO mice. Are D-tip cells lost in these mutants by snRNAseq?

Reviewer #2 (Public review):

Summary:

The authors meticulously characterized EC-specific Tgfbr1, Tgfbr2, or double knockout in the retina, demonstrating through convincing immunostaining data that loss of TGF-β signaling disrupts retinal angiogenesis and choroidal neovascularization. Compared to other genetic models (Fzd4 KO, Ndp KO, VEGF KO), the Tgfbr1/2 KO retina exhibits the most severe immune cell infiltration. The authors proposed that TGF-β signaling loss triggers vascular inflammation, attracting immune cells - a phenotype specific to CNS vasculature, as non-CNS organs remain unaffected.

Strengths:

The immunostaining results presented are clear and robust. The authors performed well-controlled analyses against relevant mouse models. snRNA-seq corroborates immune cell leakage in the retina and vascular inflammation in the brain.

Weaknesses:

The causal link between TGF-β loss, vascular inflammation, and immune infiltration remains unresolved. The authors' model posits that EC-specific TGF-β loss directly causes inflammation, which recruits immune cells. However, an alternative explanation is plausible: Tgfbr1/2 KO-induced developmental defects (e.g., leaky vessels) permit immune extravasation, subsequently triggering inflammation. The observations that vein-specific upregulation of ICAM1 staining and the lack of immune infiltration phenotypes in the non-CNS tissues support the alternative model. Late-stage induction of Tgfbr1/2 KO (avoiding developmental confounders) could clarify TGF-β's role in retinal angiogenesis versus anti-inflammation.

eLife Assessment

This important study provides converging results from complementary neuroimaging and behavioral experiments to identify human brain regions involved in representing regular geometric shapes. Geometric shape concepts are universally present across diverse human cultures and possibly essential for unique human capabilities such as numerical cognition and symbolic reasoning, and identifying the brain networks involved in geometric shape representation is of broad interest to researchers studying human visual perception, reasoning, and cognitive development. The provided experimental evidence regarding the presence of geometric shape regularity representation in dorsal parietal and prefrontal cortex is solid, but the claimed link with mathematical reasoning, the influence of experimental tasks, and the role of experience in driving geometric shape representation in both humans and artificial vision models require further elucidation.

Reviewer #1 (Public review):

Summary:

This paper examines how geometric regularities in abstract shapes (e.g., parallelograms, kites) are perceived and processed in the human brain. The manuscript contains multimodal data (behavior, fMRI, MEG) from adults and additional fMRI data from 6-year-old children. The key findings show that (1) processing geometric shapes lead to reduced activity in ventral areas in comparison to complex stimuli and increased activity in intraparietal and inferior temporal regions, (2) the degree of geometric regularity modulates activity in intraparietal and inferior temporal regions, (3) similarity in neural representation of geometric shapes can be captured early by using CNN models and later by models of geometric regularity. In addition to these novel findings, the paper also includes a replication of behavioral data, showing that the perceptual similarity structure amongst the geometric stimuli used can be explained by a combination of visual similarities (as indexed by a feedforward CNN model of the ventral visual pathway) and geometric features.

Strengths:

(1) The study incorporates multi-modal data that uses more than one task and different populations of participants (adults and children).

(2) It replicates behavioral findings of an earlier study in a larger cohort.

(3) The paper comes with openly accessible code in a well-documented GitHub repository, and the data will be published with the paper on OpenNeuro.

Weaknesses:

I wonder how task difficulty and linguistic labels interact with the current findings. Based on the behavioral data, shapes with more geometric regularities are easier to detect when surrounded by other shapes. Do shape labels that are readily available (e.g., "square") help in making accurate and speedy decisions? Can the sensitivity to geometric regularity in intraparietal and inferior temporal regions be attributed to differences in task difficulty? Similarly, are the MEG oddball detection effects that are modulated by geometric regularity also affected by task difficulty?

Reviewer #2 (Public review):

Summary:

The current study seeks to understand the neural mechanisms underlying geometric reasoning. Using fMRI with both children and adults, the authors found that contrasting simple geometric shapes with naturalistic images (faces, tools, houses) led to responses in the dorsal visual stream, rather than ventral regions that are generally thought to represent shape properties. The authors followed up on this result using computational modeling and MEG to show that geometric properties explain distinct variance in the neural response beyond what is captured by a CNN.

Strengths:

These findings contribute much-needed neural and developmental data to the ongoing debate regarding shape processing in the brain and offer additional insights into why CNNs may have difficulty with shape processing. The motivation and discussion for the study are appropriately measured, and I appreciate the authors' use of multiple populations, neuroimaging modalities, and computational models to explore this question.

Weaknesses:

Given that the primary take away from this study is that geometric shape information is found in the dorsal stream, rather than the ventral stream there is very little there is very little discussion of prior work in this area (for reviews, see Freud et al., 2016; Orban, 2011; Xu, 2018). Indeed, there is extensive evidence of shape processing in the dorsal pathway in human adults (Freud, Culham, et al., 2017; Konen & Kastner, 2008; Romei et al., 2011), children (Freud et al., 2019), patients (Freud, Ganel, et al., 2017), and monkeys (Janssen et al., 2008; Sereno & Maunsell, 1998; Van Dromme et al., 2016), as well as the similarity between models and dorsal shape representations (Ayzenberg & Behrmann, 2022; Han & Sereno, 2022).

The presence of activation in aIPS led the authors to interpret their results to mean that geometric reasoning draws on the same processes as mathematical thinking. However, there is not enough evidence in the current study to support this claim.

Reviewer #3 (Public review):

Summary:

The authors report converging evidence from several brain-imaging techniques that geometric figures, notably quadrilaterals, are processed differently in visual (lower activation) and spatial (greater) areas of the human brain than representative figures. Comparison of mathematical models to fit activity for geometric figures shows the best fit for abstract geometric features like parallelism and symmetry. The brain areas active for geometric figures are also active in processing mathematical concepts, even in blind mathematicians, linking geometric shapes to abstract math concepts. The effects are stronger in adults than in 6-year-old Western children. Similar phenomena do not appear in great apes, suggesting that this is uniquely human and developmental.

Strengths:

Multiple converging techniques of brain imaging and testing of mathematical models. Careful reasoning at every step of research and presentation of research, anticipating and addressing possible reservations. Connecting these findings to other findings, brain, behavior, and historical/anthropological, to suggest broad and important fundamental connections between abstract visual-spatial forms and mathematical reasoning, further suggesting visual-spatial origins of mathematical reasoning.

Weaknesses:

Perhaps the manuscript could emphasize that the areas recruited by geometric figures but not objects are spatial, with reduced processing in visual areas. It also seems important to say that the images of real objects are interpreted as representations of 3D objects, as they activate the same visual areas as real objects. By contrast, the images of geometric forms are not interpreted as representations of real objects but rather perhaps as 2D abstractions. The authors use the term "symbolic." That use of that term could usefully be expanded here.

Pigeons have remarkable visual systems. According to my fallible memory, Herrnstein investigated visual categories in pigeons. They can recognize individual people from fragments of photos, among other feats. I believe pigeons failed at geometric figures and also at cartoon drawings of things they could recognize in photos. This suggests they did not interpret line drawings of objects as representations of objects.

Categories are established in part by contrast categories; are quadrilaterals, triangles, and circles different categories?

It would be instructive to investigate stimuli that are on a continuum from representational to geometric, e.g., table tops or cartons under various projections, or balls or buildings that are rectangular or triangular. Building parts, inside and out. like corners. Objects differ from geometric forms in many ways: 3D rather than 2D, more complicated shapes, and internal texture. The geometric figures used are flat, 2-D, but much geometry is 3-D (e. g. cubes) with similar abstract features. The feature space of geometry is more than parallelism and symmetry; angles are important, for example. Listing and testing features would be fascinating. Similarly, looking at younger or preferably non-Western children, as Western children are exposed to shapes in play at early ages.

What in human experience but not the experience of close primates would drive the abstraction of these geometric properties? It's easy to make a case for elaborate brain processes for recognizing and distinguishing things in the world, shared by many species, but the case for brain areas sensitive to processing geometric figures is harder. The fact that these areas are active in blind mathematicians and that they are parietal areas suggests that what is important is spatial far more than visual. Could these geometric figures and their abstract properties be connected in some way to behavior, perhaps with fabrication and construction as well as use? Or with other interactions with complex objects and environments where symmetry and parallelism (and angles and curvature--and weight and size) would be important? Manual dexterity and fabrication also distinguish humans from great apes (quantitatively, not qualitatively), and action drives both visual and spatial representations of objects and spaces in the brain. I certainly wouldn't expect the authors to add research to this already packed paper, but raising some of the conceptual issues would contribute to the significance of the paper.

eLife Assessment

This fundamental study combines in vitro reconstitution experiments and molecular dynamics simulations to elucidate how membrane lipids are transported from the outer to the inner membrane of mitochondria. The authors provide convincing evidence that a positive membrane curvature is critical for membrane lipid extraction. The work will be of broad interest to cell biologists and biochemists.

Reviewer #1 (Public review):

Lipid transfer proteins (LTPs) play a crucial role in the intramembrane lipid exchange within cells. However, the molecular mechanisms that govern this activity remain largely unclear. Specifically, the way in which LTPs surmount the energy barrier to extract a single lipid molecule from a lipid bilayer is not yet fully understood. This manuscript investigates the influence of membrane properties on the binding of Ups1 to the membrane and the transfer of phosphatidic acid (PA) by the LTP. The findings reveal that Ups1 shows a preference for binding to membranes with positive curvature. Moreover, coarse-grained molecular dynamics simulations indicate that positive curvature decreases the energy barrier associated with PA extraction from the membrane. Additionally, lipid transfer assays conducted with purified proteins and liposomes in vitro demonstrate that the size of the donor membrane significantly impacts lipid transfer efficiency by Ups1-Mdm35 complexes, with smaller liposomes (characterized by high positive curvature) promoting rapid lipid transfer.

This study offers significant new insights into the reaction cycle of phosphatidic acid (PA) transfer by Ups1 in mitochondria. Notably, the authors present compelling evidence that, alongside negatively charged phospholipids, positive membrane curvature enhances lipid transfer - an effect that is particularly relevant at the mitochondrial outer membrane. The experiments are technically robust, and my primary feedback pertains to the interpretation of specific results.

(1) The authors conclude from the lipid transfer assays (Figure 5) that lipid extraction is the rate-limiting step in the transfer cycle. While this conclusion seems plausible, it should be noted that the authors employed high concentrations of Ups1-Mdm35 along with less negatively charged phospholipids in these reactions. This combination may lead to binding becoming the rate-limiting factor. The authors should take this point into consideration. In this type of assay, it is challenging to clearly distinguish between binding, lipid extraction, and membrane dissociation as separate processes.

(2) The authors should discuss that variations in the size of liposomes will also affect the distance between them at a constant concentration, which may affect the rate of lipid transfer. Therefore, the authors should determine the average size and size distribution of liposomes after sonication (by DLS or nanoparticle analyzer, etc.).

(3) The authors use NBD-PA in the lipid transfer assays. Does the size of the donor liposomes affect the transfer of NBD-PA and DOPA similarly? Since NBD-labeled lipids are somewhat unstable within lipid bilayers (as shown by spontaneous desorption in Figure 5B), monitoring the transfer of unlabeled PA in at least one setting would strengthen the conclusion of the swap experiments.

(4) The present study suggests that membrane domains with positive curvature at the outer membrane may serve as starting points for lipid transport by Ups1-Mdm35. Is anything known about the mechanisms that form such structures? This should be discussed in the text.

Reviewer #2 (Public review):

Summary:

Lipid transfer between membranes is essential for lipid biosynthesis across different organelle membranes. Ups1-Mdm35 is one of the best-characterized lipid transfer proteins, responsible for transferring phosphatidic acid (PA) between the mitochondrial outer membrane (OM) and inner membrane (IM), a process critical for cardiolipin (CL) synthesis in the IM. Upon dissociation from Mdm35, Ups1 binds to the intermembrane space (IMS) surface of the OM, extracts a PA molecule, re-associates with Mdm35, and moves through the aqueous IMS to deliver PA to the IM. Here, the authors analyzed the early steps of this PA transfer - membrane binding and PA extraction - using a combination of in vitro biochemical assays with lipid liposomes and purified Ups1-Mdm35 to measure liposome binding, lipid transfer between liposomes, and lipid extraction from liposomes. The authors found that membrane curvature, a previously overlooked property of the membrane, significantly affects PA extraction but not PA insertion into liposomes. These findings were further supported by MD simulations.

Strengths:

The experiments are well-designed, and the data are logically interpreted. The present study provides an important basis for understanding the mechanism of lipid transfer between membranes.　

Weaknesses:

The physiological relevance of membrane curvature in lipid extraction and transfer still remains open.

Reviewer #3 (Public review):

The manuscript by Sadeqi et al. studies the interactions between the mitochondrial protein Ups1 and reconstituted membranes. The authors apply synthetic liposomal vesicles to investigate the role of pH, curvature, and charge on the binding of Ups1 to membranes and its ability to extract PA from them. The manuscript is well written and structured. With minor exceptions, the authors provide all relevant information (see minor points below) and reference the appropriate literature in their introduction. The underlying question of how the energy barrier for lipid extraction from membranes is overcome by Ups1 is interesting, and the data presented by the authors could offer a valuable new perspective on this process. It is also certainly a challenging in vitro reconstitution experiment, as the authors aim to disentangle individual membrane properties (e.g., curvature, charge, and packing density) to study protein adsorption and lipid transfer. I have one major suggestion and a few minor ones that the authors might want to consider to improve their manuscript and data interpretation:

Major Comments:

The experiments are performed with reconstituted vesicles, which are incubated with recombinant protein variants and quantitatively assessed in flotation and pelleting assays. According to the Materials and Methods section, the lipid concentration in these assays is kept constant at 5 µM. However, the authors change the size of the vesicles to tune their curvature. Using the same lipid concentration but varying vesicle sizes results in different total vesicle concentrations. Moreover, larger vesicles (produced by freeze-thawing and extrusion) tend to form a higher proportion of multilamellar vesicles, thus also altering the total membrane area available for binding. Could these differences in the experimental system account for the variation in binding? To address this, the authors would need to perform the experiments either under saturation (excess protein) conditions or find an experimental approach to normalize for these differences.

Author response:

Reviewer #1:

Lipid transfer proteins (LTPs) play a crucial role in the intramembrane lipid exchange within cells. However, the molecular mechanisms that govern this activity remain largely unclear. Specifically, the way in which LTPs surmount the energy barrier to extract a single lipid molecule from a lipid bilayer is not yet fully understood. This manuscript investigates the influence of membrane properties on the binding of Ups1 to the membrane and the transfer of phosphatidic acid (PA) by the LTP. The findings reveal that Ups1 shows a preference for binding to membranes with positive curvature. Moreover, coarse-grained molecular dynamics simulations indicate that positive curvature decreases the energy barrier associated with PA extraction from the membrane. Additionally, lipid transfer assays conducted with purified proteins and liposomes in vitro demonstrate that the size of the donor membrane significantly impacts lipid transfer efficiency by Ups1-Mdm35 complexes, with smaller liposomes (characterized by high positive curvature) promoting rapid lipid transfer.

This study offers significant new insights into the reaction cycle of phosphatidic acid (PA) transfer by Ups1 in mitochondria. Notably, the authors present compelling evidence that, alongside negatively charged phospholipids, positive membrane curvature enhances lipid transfer - an effect that is particularly relevant at the mitochondrial outer membrane. The experiments are technically robust, and my primary feedback pertains to the interpretation of specific results.

(1) The authors conclude from the lipid transfer assays (Figure 5) that lipid extraction is the rate-limiting step in the transfer cycle. While this conclusion seems plausible, it should be noted that the authors employed high concentrations of Ups1-Mdm35 along with less negatively charged phospholipids in these reactions. This combination may lead to binding becoming the rate-limiting factor. The authors should take this point into consideration. In this type of assay, it is challenging to clearly distinguish between binding, lipid extraction, and membrane dissociation as separate processes.

We thank the reviewer for the constructive and positive evaluation of our manuscript. We agree that, while our data support the interpretation that the rate-limiting step occurs at the donor membrane, it is difficult to dissect in our assay which of the individual steps at the donor membrane - such as binding of Ups1, lipid extraction into the binding pocket, or dissociation of Ups1 - is rate-limiting. Nevertheless, although we cannot exclude contributions from membrane binding or dissociation, several observations suggest that lipid extraction is a rate-limiting step under our experimental conditions.

The acceptor membrane has a similar lipid composition to the donor membrane (in tendency, the donor membrane is even a bit richer in binding-promoting lipids). If binding was ratelimiting, similar constraints would be expected at the acceptor membrane during lipid insertion. However, this is not observed.

Regarding dissociation, if this step were rate-limiting, one would expect similar constraints to be evident at the acceptor vesicles as well. Nevertheless, membrane dissociation might be mechanistically coupled to lipid extraction and thus difficult to evaluate as an independent step.

Based on our data and the considerations described above, we suggest that lipid extraction is the dominant rate-limiting step at the donor membrane under our conditions. However, we agree that a clear separation of these individual steps is not possible with the current experimental design. We will revise the corresponding passage to clarify that the rate-limiting step occurs at the donor membrane and, based on our observations, likely involves lipid extraction. Future studies aiming on dissecting these steps, will be important for elucidating the mechanism and regulation of Ups1-mediated lipid transfer both in vitro and in vivo.

(2) The authors should discuss that variations in the size of liposomes will also affect the distance between them at a constant concentration, which may affect the rate of lipid transfer. Therefore, the authors should determine the average size and size distribution of liposomes after sonication (by DLS or nanoparticle analyzer, etc.)

We agree that variations in liposome size will influence the average distance between vesicles at a given lipid concentration, which may in turn affect the rate of lipid transfer. As suggested, we will include DLS measurements to characterize the size distribution of our different liposome preparations.

Our setup was designed to keep the total membrane surface area comparable across conditions. This approach ensures a comparable overall binding capacity for Ups1 and enables the comparison of membrane binding and lipid extraction from different membranes. However, we agree that vesicle spacing, which is affected by liposome size at constant lipid concentration, could potentially influence certain steps in the transfer process, such as the time required for Ups1 to travel between donor and acceptor membranes. Whether this intermembrane travel time contributes to rate limitation is indeed an interesting question, and we will address this point through further discussion in the revised manuscript.

Investigating such effects in our current experimental system would require altering the vesicle concentration, which would in turn change the total membrane surface area and introduce additional variables. Nevertheless, exploring the influence of vesicle spacing and intermembrane distance on lipid transfer represents a promising direction for future studies aimed at dissecting the rate-limiting steps of the transfer cycle.

(3) The authors use NBD-PA in the lipid transfer assays. Does the size of the donor liposomes affect the transfer of NBD-PA and DOPA similarly? Since NBD-labeled lipids are somewhat unstable within lipid bilayers (as shown by spontaneous desorption in Figure 5B), monitoring the transfer of unlabeled PA in at least one setting would strengthen the conclusion of the swap experiments.

Ups1-mediated transfer of PA has been demonstrated both by mass spectrometry analysis of donor and acceptor vesicles (Connerth et al., 2012) and by NBD-fluorescence-based lipid transfer assays (Lu et al., 2020; Miliara et al., 2015; Miliara et al., 2019; Miliara et al., 2023; Potting et al., 2013; Watanabe et al., 2015). The fluorescence-based approach has been the most widely applied across multiple studies and has enabled detailed analysis of various aspects of lipid transfer by Ups1. It has been used to investigate mutants of key structural elements—such as the lipid-binding pocket and the α2–loop region. It has also been used to analyze fusion constructs between Ups1 and Mdm35, the influence of Mdm35 variants, and competition with excess Mdm35. Additionally, by comparing the transfer of NBD-labeled PA and NBD-labeled PS, this assay has provided insights into the determinants of the lipid specificity of Ups1. Hence, our experiments are based on the standard assay used to analyse lipid transfer in the field and thus can be corralated with the majority of published data.

Nevertheless, we agree that it is important to keep in mind that NBD labeling may alter the biophysical properties of lipids and, consequently, affect their transfer efficiency. Moreover, NBD-labeled lipids are not suitable for comparing the transfer efficiency of different PA species, as the label itself may mask differences in acyl chain composition. Therefore, it will be valuable to establish complementary methods that do not rely on NBD-labeled PA. We aim to develop these non-standard methods for possible inclusion in the present study, but even if not fully implemented at this stage, they will certainly form an important part of future investigations.

(4) The present study suggests that membrane domains with positive curvature at the outer membrane may serve as starting points for lipid transport by Ups1-Mdm35. Is anything known about the mechanisms that form such structures? This should be discussed in the text.

The origin of positively curved membrane domains is indeed highly relevant in the context of our findings, and while not the primary focus of this work, we will place more emphasis on discussing how such curvature may arise. Mechanisms include the action of curvature-generating proteins, asymmetric lipid composition and curvature induced at membrane contact sites. We have so far included examples of proteins in the outer mitochondrial membrane that are expected to influence curvature in their vicinity, and we will expand on this aspect and other contributing factors more thoroughly in the revised text.

Reviewer #2:

Summary:

Lipid transfer between membranes is essential for lipid biosynthesis across different organelle membranes. Ups1-Mdm35 is one of the best-characterized lipid transfer proteins, responsible for transferring phosphatidic acid (PA) between the mitochondrial outer membrane (OM) and inner membrane (IM), a process critical for cardiolipin (CL) synthesis in the IM. Upon dissociation from Mdm35, Ups1 binds to the intermembrane space (IMS) surface of the OM, extracts a PA molecule, re-associates with Mdm35, and moves through the aqueous IMS to deliver PA to the IM. Here, the authors analyzed the early steps of this PA transfer - membrane binding and PA extraction - using a combination of in vitro biochemical assays with lipid liposomes and purified Ups1-Mdm35 to measure liposome binding, lipid transfer between liposomes, and lipid extraction from liposomes. The authors found that membrane curvature, a previously overlooked property of the membrane, significantly affects PA extraction but not PA insertion into liposomes. These findings were further supported by MD simulations.

Strengths:

The experiments are well-designed, and the data are logically interpreted. The present study provides an important basis for understanding the mechanism of lipid transfer between membranes.  

Weaknesses:

The physiological relevance of membrane curvature in lipid extraction and transfer still remains open.

We thank the reviewer for the constructive feedback on our work. We agree that the physiological relevance of membrane curvature in lipid extraction and transfer remains an open question. Our data show that Ups1 binding to native-like OM membranes under physiological pH conditions is curvature-dependent, supporting the idea that this mechanism may optimize lipid transfer in vivo. While the intricate biophysical basis of this behaviour can only be dissected in vitro, these findings offer valuable insight into how curvature may functionally regulate Ups1 activity in the cellular context. To directly test this, it will be important in future studies to identify Ups1 mutants that lack curvature sensitivity and assess their performance in vivo, which will help clarify the physiological importance of this mechanism.

Reviewer #3:

The manuscript by Sadeqi et al. studies the interactions between the mitochondrial protein Ups1 and reconstituted membranes. The authors apply synthetic liposomal vesicles to investigate the role of pH, curvature, and charge on the binding of Ups1 to membranes and its ability to extract PA from them. The manuscript is well wrifen and structured. With minor exceptions, the authors provide all relevant information (see minor points below) and reference the appropriate literature in their introduction. The underlying question of how the energy barrier for lipid extraction from membranes is overcome by Ups1 is interesting, and the data presented by the authors could offer a valuable new perspective on this process. It is also certainly a challenging in vitro reconstitution experiment, as the authors aim to disentangle individual membrane properties (e.g., curvature, charge, and packing density) to study protein adsorption and lipid transfer. I have one major suggestion and a few minor ones that the authors might want to consider to improve their manuscript and data interpretation:

Major Comments:

The experiments are performed with reconstituted vesicles, which are incubated with recombinant protein variants and quantitatively assessed in flotation and pelleting assays. According to the Materials and Methods section, the lipid concentration in these assays is kept constant at 5 µM. However, the authors change the size of the vesicles to tune their curvature. Using the same lipid concentration but varying vesicle sizes results in different total vesicle concentrations. Moreover, larger vesicles (produced by freeze-thawing and extrusion) tend to form a higher proportion of multilamellar vesicles, thus also altering the total membrane area available for binding. Could these differences in the experimental system account for the variation in binding? To address this, the authors would need to perform the experiments either under saturation (excess protein) conditions or find an experimental approach to normalize for these differences.

We thank the reviewer for the constructive and positive comments. We agree that, since the total number of lipids was kept constant, the number of vesicles varied with vesicle size in our experiments. However, the setup was specifically designed to maintain a comparable total membrane surface area across conditions, ensuring a comparable number of available binding sites for Ups1. Because membrane surface area decreases with the square of the vesicle radius, keeping vesicle number constant would have led to a marked reduction in binding surface. Our approach was therefore aimed at preserving comparable binding capacity while varying membrane curvature.

With respect to multilamellarity, we thank the reviewer for addressing this important point. As described above, we aimed to maintain a constant total membrane surface area across all conditions to ensure an equal number of potential binding sites. We agree that multilamellarity in large liposomes could restrict accessibility to part of the membrane surface. However, we see in our experiments that even when the total membrane surface area of the small liposomes is reduced to one quarter of the standard amount, binding to the small liposomes remained stronger than to the larger liposomes at the higher concentration. This strongly indicates that restricted accessibility cannot account for the curvature-specific effect observed. Nonetheless, we will further address this aspect experimentally and in the discussion of the revised manuscript.

References

Connerth, M., Tatsuta, T., Haag, M., Klecker, T., Westermann, B., & Langer, T. (2012). Intramitochondrial transport of phosphatidic acid in yeast by a lipid transfer protein. Science, 338(6108), 815-818. https://doi.org/10.1126/science.1225625 

Lu, J., Chan, C., Yu, L., Fan, J., Sun, F., & Zhai, Y. (2020). Molecular mechanism of mitochondrial phosphatidate transfer by Ups1. Commun Biol, 3(1), 468. https://doi.org/10.1038/s42003-020-01121-x 

Miliara, X., Garnef, J. A., Tatsuta, T., Abid Ali, F., Baldie, H., Perez-Dorado, I., Simpson, P., Yague, E., Langer, T., & Mafhews, S. (2015). Structural insight into the TRIAP1/PRELI-like domain family of mitochondrial phospholipid transfer complexes. EMBO Rep, 16(7), 824-835. https://doi.org/10.15252/embr.201540229 

Miliara, X., Tatsuta, T., Berry, J. L., Rouse, S. L., Solak, K., Chorev, D. S., Wu, D., Robinson, C. V., Mafhews, S., & Langer, T. (2019). Structural determinants of lipid specificity within Ups/PRELI lipid transfer proteins. Nat Commun, 10(1), 1130. https://doi.org/10.1038/s41467-019-09089-x 

Miliara, X., Tatsuta, T., Eiyama, A., Langer, T., Rouse, S. L., & Mafhews, S. (2023). An intermolecular hydrogen bonded network in the PRELID-TRIAP protein family plays a role in lipid sensing. Biochim Biophys Acta Proteins Proteom, 1871(1), 140867. https://doi.org/10.1016/j.bbapap.2022.140867 

Posng, C., Tatsuta, T., Konig, T., Haag, M., Wai, T., Aaltonen, M. J., & Langer, T. (2013). TRIAP1/PRELI complexes prevent apoptosis by mediating intramitochondrial transport of phosphatidic acid. Cell Metab, 18(2), 287-295. https://doi.org/10.1016/j.cmet.2013.07.008 

Watanabe, Y., Tamura, Y., Kawano, S., & Endo, T. (2015). Structural and mechanistic insights into phospholipid transfer by Ups1-Mdm35 in mitochondria. Nat Commun, 6, 7922. https://doi.org/10.1038/ncomms8922

eLife Assessment

This is a very important study in which the authors have modified ChIP-seq and 4C-seq with a urea step, which drastically changes the pattern of chromatin interactions observed for SATB1, but not other proteins (including CTCF). The study highlights that the urea protocols provide a complementary view of protein-chromatin interactions for some proteins, which can uncover previously hidden, functionally significant layers of chromatin organization. If applied more widely, these protocols may significantly further our understanding of chromatin organization. The study's findings are supported by a wealth of controls, making the evidence compelling.

Reviewer #1 (Public review):

Summary:

The nuclear protein SATB1 was originally identified as a protein of the 'nuclear matrix', an aggregate of nuclear components that arose upon extracting nuclei with high salt. While the protein was assumed to have a global function in chromatin organization, it has subsequently been linked to a variety of pathological conditions, notably cancer. The mapping of the factor by conventional ChIP procedures showed strong enrichment in active, accessible chromatin, suggesting a direct role in gene regulation, perhaps in enhancer-promoter communication. These findings did not explain why SATB1-chromatin interaction resisted the 2 M salt extraction during early biochemical fractionation of nuclei.

The authors, who have studied SATB1 for many years, now developed an unusual variation of the ChIP procedure, in which they purify crosslinked chromatin by centrifugation through 8 M urea. Remarkably, while they lose all previously mapped signals for SATB1 in active chromatin, they now gain many binding events in silent regions of the genome, represented by lamin-associated domains (LADs).

SATB1 had previously been shown by the authors and others to bind to DNA with special properties, termed BUR for 'base-unpairing regions'). BURs are AT-rich and apparently enriched in equally AT-rich LADs. The 'urea-ChIP' pattern is essentially complementary to the classical ChIP pattern. The authors now speculate that the previously known SATB1 binding pattern determined by standard ChIP, which does not overlap BURs particularly well, is due to indirect chromatin binding, whereas they consider the urea-ChIP profile, which fits better to the BUR distribution on the chromosome, to be due to direct binding.

Building on the success with urea-ChIP the authors adapted the 4C-procedure of chromosome conformation mapping to work with urea-purified chromatin. The data suggest a model according to which BUR-bound SATB1 mediates long-distance interaction between active loci and some kind of scaffold structure formed by SATB1. Because cell type-specific differences are observed, they suggest that the SATB1 interactions are functionally relevant.

Strengths:

Given the unusual findings of essentially mutually exclusive 'standard ChIP' and 'urea-ChIP' profiles, the authors conducted many appropriate controls. They showed that all SATB1 peaks in urea-ChIP and 96% of peaks in standard-ChIP represent true signals, as they are not observed in a SATB1 knockout cell line. They also show that the urea-ChIP and standard ChIP yield similar profiles for CTCF and polycomb complex subunits. The data appear reproducible judged by at least two replicates and triangulation. The SATB1 KO cells provide a nice control for the specificity of signals, including those that arise from their elaborately modified 4C protocol.

In their revised manuscript the authors provide relevant background information concerning the effect of urea on the denaturation of macromolecules. Importantly, they argue convincingly that urea does not denature DNA under their conditions.

Weaknesses:

Despite the authors' efforts to explain their findings along with a lot of background information, some readers may be left confused due to the complexity of the system. BURs are found enriched in LADs, but are also present in active chromatin. SATB1 binds a subset of BURs, but the reason for discrimination remains unclear. SATB1 appears to bind chromatin in at least two modes with differing diffusion properties and exactly how this relates to the indirect and direct chromatin binding modes is mechanistically unclear.

The authors resort to the term 'SATB1-enriched subnuclear structure' to describe the profile gained through denaturing ChIP, thus avoiding strong statements about involvement of known nuclear structures (such as LADs or heterochromatin) and about functional implications.

The authors acknowledge a potential for RNA to be involved in modulating SATB1 interactions with chromatin, but leave this for future investigation.

Comment on revised version:

The authors revised their manuscript to my satisfaction.

Reviewer #2 (Public review):

Summary:

This study describes the key observation that SATB1 binds directly to so-called BUR elements. This is in contrast to several other reports describing SATB1 binding to promoters and enhancers. This discrepancy is explained by the authors to depend on the features of the ChIP technique being used. Urea-ChIP, innovated by the authors, strips off protein-protein interactions that compound conventional ChIP methods. The authors convincingly make the case that SATB1 and a key genome organiser, CTCF, largely bind different sites, as particularly evident in Figure 2A. In contrast, standard ChIP shows considerable overlap between their sites (Figure 2-figure supplement 1). The report documents convincingly that SATB1 partitions the genome independent of so-called TADs to influence expression patterns. SATB1 controls long-range interactions in thymocytes, and knock down of SATB1 does not affect the TAD patterns.

Strengths:

A new and innovative adaptation of ChIP-seq (urea ChIP-seq) has enabled the authors to successfully question existing data on the patterns of SATB1 binding to the genome. The authors provide a wealth of data to reinforce their claims. This report thus rectifies misconceptions about SATB1 function, which is particularly important given its role in metastasising cancer cells.

Weaknesses:

None

Author response:

The following is the authors’ response to the original reviews

Reviewer #1:

(1) 8 molar urea not only denatures proteins but also denatures DNA. Obviously, this does not affect the ChIP, since antibodies often recognize small linear epitopes and the proteins are crosslinked. However, under high urea conditions the BUR elements should be rendered single-stranded, and one wonders whether this has any effect on the procedure. The authors should alert the reader of these circumstances.

Thank you for raising this important question about the effects of 8M urea. We have added a brief paragraph explaining this point in the revised manuscript. Despite common misconceptions, 8M urea by itself does not actively convert double-stranded DNA to single-stranded DNA. For this conversion to occur, a heat denaturation step is required. Once DNA is heat-denatured to become single-stranded, urea can maintain this configuration. This is why the addition of 8M urea to acrylamide gel electrophoresis is a standard method for analyzing single-stranded oligonucleotides, but the DNA must first be denatured by heat (Summer et al., J. Vis. Exp. (32), e1485, DOI : 10.3791/1485). This is clearly described in published work comparing the status of DNA with and without heat treatment in an 8M urea-containing buffer (Hegedus et al., Nucl.Acids Res. 2009 (doi:10.1093/nar/gkp539).

We have additional evidence supporting this conclusion in the context of our urea ultracentrifugation experiment. Both crosslinked and un-crosslinked genomic DNA purified by 8M urea centrifugation can be digested with restriction enzymes, which indicates that the DNA remains double-stranded. For instance, we previously published SATB1 ChIP-3C results using Sau3A-digested DNA after urea purification. In the current paper, we used HindIII to digest urea-purified DNA for urea4C-seq. The BUR reference map can also be generated after restriction digestion of urea-purified DNA and isolating and sequencing SATB1-bound restriction fragments in vitro. If genomic DNA were denatured by 8M urea ultracentrifugation, we would not have been able to digest it with restriction enzymes to obtain these results.

We have now added a sentence noting that SATB1 is a double-stranded DNA-binding protein that does not bind to single-stranded DNA, as we have previously shown (Dickinson et al., 1992, Ref 32).

(2) An important conclusion is that urea-ChIP reveals direct DNA binding events, whereas standard ChIP shows indirect binding (which is stripped off by urea). I do not see any evidence for direct binding. At low resolution, predicted BUR elements are enriched in domains where SATB-1 is mapped by urea-ChIP. A statement like 'In a zoomed-in view, covering a 430 kb region, SATB1 sites identified from urea ChIP-seq precisely coincided with BUR peaks' is certainly not correct: most BUR peaks do not show significant SATB-1 binding. The randomly chosen regions shown in Figure 4 – Supplement 1 show how poor the overlap of SATB-1 and BURs is; indeed, they show that SATB-1 binds DNA mostly at non-BUR sites. I see Figure 2D, but such cumulative plots can be highly biased by very few cases. I suggest showing these data in heat maps instead.

We believe there may be some confusion regarding the interpretation of our figures. Looking at Track 3 (BUR reference map, RED peaks) and urea SATB1 Tracks 4 and 5 (replicas from two independent experiments) in Fig. 2B, the SATB1 peaks detected by urea ChIP-seq do indeed coincide with BUR peaks. In the revised manuscript, we have provided a further ‘zoomed-in’ view to better illustrate this point and also provided the underlying BUR sequence from one of these SATB1-bound regions (Figure 2—supplement figure 1).

It is true that many more BURs exist than SATB1-bound BURs, especially in gene-poor regions where BURs are clustered. However, from the perspective of SATB1-bound peaks, the majority of these coincide with BURs, as shown by both deepTools analyses and new heatmap, as suggested (Figure 2E, and Figure 7—supplement figure 3).

The results from our genome-wide quantitative analyses using deepTools to compare peaks from urea SATB1 ChIP-seq data and the BUR reference map shown in Supplementary Tables 1 and 2 are consistent with the heatmap analyses.

We must apologize for an error in the scaling of the y-axis in Figure 4-supplement figure 1 that likely contributed to some confusion. We have corrected our mistake in the revised manuscript. As we were preparing our figures, when placed in the figure and axes relabeled for legibility, the BUR reference peaks were mislabeled on their y-axis. In the figure the peaks were erroneously labeled on a scale of 0.1-1 read counts/million reads, but the data shown is actually scaled at 0.1 to 2 read counts per million reads. Unfortunately, we did not realize this error and, using the figure as a guide for scaling, provided urea SATB1 ChIP-seq peaks at a scale of 0.1-1 read counts/million reads to match the mislabeled BUR reference track. This had the effect of reducing the signal/noise in the SATB1 ChIP-seq data (Figure 1). We have now standardized the y-axis for fair comparison using a scaling of the y-axis at 0.1-2 for all tracks.  This will more clearly show that there are indeed more BUR peaks than SATB1-bound sites, consistent with our quantitative analysis.

We hope that these clarifications as well as the added heatmaps and binding site example allay the concerns about the specificity and overlap of SATB1 binding on BURS.

(3) In Figure 6C 'peaks' are compared. However, looking at Figure 4 - Supplement 1 again it is clear that peak calling can yield a misleading impression. Figure 6D suggests that there are more BURs than SATB-1 peaks but this is not true from looking at the browser.

We thank the reviewer for this observation. As noted in our response to point 2 above, the inconsistent y-axis scaling in Figure 4-supplement figure 1 created a misleading impression, which we have corrected in the revised manuscript. When properly displayed with consistent y-axis scaling, the browser view aligns with our quantitative data showing that there are indeed many more BURs than SATB1-bound sites. As mentioned under 2 above, we have performed genome-wide quantitative analysis by deepTools (Supplementary Tables 1 and 2) to confirm the results shown by bar graphs in Fig. 6C, 6D and Fig. 2D. 

In Figure 6C, the bars show the percentage of SATB1-bound peaks in each cell type (denominator) that overlap with confirmed BUR sites in the BUR reference map (numerator). In Figure 6D, we show the percentage of total BUR sites in the BUR reference map (denominator) that are bound by SATB1 from urea ChIP-seq (numerator). To avoid any confusion, we have added brief subtitles to Figures 6C and 6D in the revised manuscript.

(4) An important conclusion is that urea-ChIP reveals direct DNA binding events, whereas standard ChIP shows indirect binding (which is stripped off by urea). I do not yet see any evidence for direct binding. It cannot be excluded that the binding is RNA-mediated. The authors mention in passing that urea-ChIP material still contains (specific!) RNA. Given that this is a new procedure, the authors should document the RNA content of urea-ChIP and RNase-treat their samples prior to ChIP to monitor an RNA contribution.

Thank you for raising this important point. The direct binding of SATB1 to BURs is well-established in our previous work. Indeed, this was the main motivation to explore the reason for the lack of evidence for genome-wide SATB1 binding to BURs in the DNA-binding profile by standard ChIP-seq. This has been a major point of confusion for us for many years.

SATB1 was originally identified through a search for mammalian proteins that could recognize BURs specifically and not just any A+T-rich sequence. The Satb1 gene was originally cloned by an expression cDNA library and encoded SATB1 protein bound the BUR probe but not a mutated AT-rich BUR (control) probe.  Subsequent experiments confirmed that SATB1 specifically binds to many BURs without requiring additional factors. Furthermore, SATB1 recognizes BURs by binding in the minor groove of double-stranded DNA, presumably recognizing the altered phosphate backbone structure of BUR DNA, rather than accessing nucleotide bases (Dickinson et al, 1992).

We do agree with the reviewer, however, that there is a possibility that RNA can redirect SATB1 to different subsets of BURs and/or to interact indirectly with different regulatory regions depending on cell type or developmental stage. Although urea ultracentrifugation clearly separates most RNA (found in the middle region of the tube) from genomic DNA (pelleted at the bottom) (de Belle et al., 1998), upon crosslinking cells, a small quantity of RNA is found co-pelleted with DNA (our recent unpublished results). This RNA, tightly associated with crosslinked chromatin, may have some impact on SATB1 function.

Based on our preliminary data, we are currently planning to study the impact of RNA using RNase A as well as by targeting specific RNAs employing an anti-sense approach. We believe that thoroughly addressing the impact of RNA warrants a full paper, including the potential roles of specific non-coding RNAs in SATB1 function, and thus is beyond the scope of the current paper. However, we have now added discussion of this important point in the manuscript.

(5) An important aspect of the model is that SATB1 tethers active genes to inactive LADs. However, in the 4C experiment the BUR elements used to anchor the looping are both in the accessible, active chromatin domain. If the authors want to maintain their statement, they must show a 4C result that connects the 2 distinct domains and transverses A/B domain boundaries. Currently, the data only show a looping within accessible chromatin.

We appreciate REVIEWER 1 for bringing up the important point that our model could potentially be interpreted as “SATB1 tethers active genes to inactive LADs.” Since we describe that BURs are enriched in LADs and that SATB1 binds a subset of BURs, readers may assume that we aim to demonstrate, through urea 4C-seq, that SATB1 tethers active genes to transcriptionally-inactive LADs (via BURs). However, this is not our intention in the model (Figure 8). In the experiment we designed for our present study,  we selected BUR-1 and BUR-2 as viewpoints from a non-LAD gene-rich region (inter-LAD). Because these BURs are bound by SATB1, it indicates that these BURs are part of the “hard-to-access” SATB1-rich subnuclear structure, which resists extraction, in contrast to accessible chromatin. Thus, we illustrate in the model that BURs anchored to the SATB1-rich nuclear substructure make contact with accessible chromatin over long distances in a SATB1-dependent manner. Therefore, we do not intend to conclude that SATB1 mediates interactions between LADs and inter-LADs (accessible chromatin) from our current study: this would be a topic for future research. In the original model in the submitted manuscript, we used the terms “inaccessible” and “accessible.” In the revised version, we clarified this in the model by changing “inaccessible” to “SATB1-rich subnuclear structure” and carefully revised  the text in the Figure 8 legend to clarify the model. 

At this time, we do not know exactly how LADs and SATB1 nuclear architecture are related spatially and functionally. While LADs are mapped as genomic domains in proximity to Lamin B1 by LaminB1-DamID, BURs are mapped at ~300-500 bp resolution by urea ChIP-seq. To gain further insight into this important question, a large body of DNA-FISH and immunoDNA-FISH experiments will be required, comparing different cell types to see whether and how specific BURs move between LADs and SATB1 nuclear architecture. Such experiments may benefit from testing the Gabrg1 and Gabra2 loci, where many BURs are anchored to SATB1 in neurons but not in thymocytes, for instance.  This is included in Discussion in the revised manuscript.

Regarding the reviewer's second point about showing more extended domains for 4C interactions, we would like to highlight that Figure 5—supplement figure 3 in our submitted manuscript addresses this concern. This figure shows that BUR-interactions extend to multiple gene-rich regions across intervening gene-poor regions. Interestingly, BUR-1 and BUR-2 interactions skip a transcriptionally silent gene-rich region containing olfactory receptor genes but interact with subsequent gene-rich regions containing active genes. These data demonstrate that BUR-interactions do indeed traverse A- and B-compartment boundaries.  In the revised manuscript (in Figure 5—supplement figure 3), we newly added a Lamin B1-DamID (thymocyte) track.  Comparing with LADs, BUR-1 interactions occur mostly in non-LAD regions. Some minor overlap with LADs was detected in high resolution views (not shown). Future experiments testing BUR viewpoints that reside within LADs are required to assess whether SATB1 mediates interactions between B and A compartments.

(6) The description of the urea-co-immunoprecipitation experiment (Figure 3C) could be improved to make it unequivocally clear that co-binding to chromatin is tested, not protein-protein interaction (which is destroyed by urea).

Thank you for this helpful suggestion. We have revised the text in the manuscript by stating “Distinct from protein-protein co-immunoprecipitation (co-IP) using whole cell or nuclear extracts, we examined the direct co-binding status on chromatin in vivo of SATB1 and CTCF or cohesin by urea ChIP-Western”.

Reviewer #2:

(1) Since SATB1 has been described to interact with beta-catenin, I wonder if the authors have looked at TCF4/TCF7l2 binding patterns and their potential overlap with SATB1 binding patterns. This might appear a trivial request. However, uncontrolled WNT signalling is a major feature of cancer undergoing metastasis - a process that the authors have earlier associated with unscheduled SATB1 expression in triple-negative breast cancer.

We thank the reviewer for highlighting this important point about the potential relationship between SATB1 and TCF4/TCF7l2 binding patterns. Based on published observations with other factors (Rad21, CTCF, BRG1, RUNX) that show substantial overlap with SATB1 in standard ChIP-seq peaks(Kakugawa et al., Cell Rep 19, 1176-1188 (2017). DOI: 10.1016/j.celrep.2017.04.038. Poterlowicz et al., PLoS Genet, 2017 DOI: 10.1371/journal.pgen.1006966), we would anticipate that TCF4 might also show significant overlap with SATB1. An important question is whether the DNA binding profile of TCF4 depends on SATB1.

We have not yet generated ChIP-seq data for TCF4 in the presence and absence of SATB1, but we agree that such experiments could provide important insights into cancer progression as well as brain function. This represents an interesting direction for future work. We have added this point in our discussion based on your kind suggestion.

(2) The CTCF sizes indicated in the western blot analyses of Figures 3C and Figure 3 - supplement figure 2 do not display the normal size, which is around 130 kDa. Either the issue is erroneous marking or a so-called salt effect to slow the migration in the gel. Alternatively, it reflects a slower migrating form of CTCF generated by for example PARylation (by PARP1) that is known to approach 180 kDa. It would be useful if the authors could clarify this minor issue.

We appreciate the reviewer pointing out this discrepancy. As the reviewer correctly noted, CTCF can appear at a higher molecular weight due to post-translational modifications such as PARylation and O-GlcNAcylation, which alter its migration during electrophoresis.

Upon re-examination of our raw data for Figure 3—supplement figure 2A, we discovered that the marker lane for the CTCF panel was broken, and the 150kDa band was erroneously assigned. This led to the 150kDa marker being placed below the CTCF migration position, which is clearly an error. We thank the reviewer for bringing this to our attention.

We have checked our other data and consistently observe CTCF migrating below the 150kDa band, similar to the pattern shown on the Abcam website for the antibody we used (ab128873) (Figure 2). For Figure 3-supplement figure 2, we will use a marker lane from a parallel gel with identical composition and run time to correctly indicate the molecular weight. We havealso corrected the marker position in Figure 3C.

Reviewing Editor (Recommendations for the authors):

(1) The introduction states that urea ChIP-seq is "unbiased", which is difficult to unambiguously determine and therefore might be an overstatement. Maybe the authors could consider rephrasing.

We agree with the reviewer's assessment and have rephrased our description of the urea ChIP-seq method to avoid using the term "unbiased."

(2) The authors propose that in standard ChIP, most SATB1 is in the insoluble fraction. This seems easy to test and demonstrating it may help to further clarify the differences between the protocols.

We appreciate this suggestion and would like to clarify our description. What we stated in the manuscript was:

"We envision that SATB1 bound to inaccessible nuclear regions may be lost in the insoluble fraction."

This refers specifically to a subpopulation of SATB1 that is bound to the high-salt extraction-resistant nuclear substructure, not to the total SATB1 protein. We also noted elsewhere in the manuscript that:

"SATB1 proteins are found in high salt-resistant fraction as well as salt-extracted fraction (40). Thus, it is possible that soluble SATB1 may associate with open chromatin."

Our unpublished results show that SATB1 proteins exist in at least two distinct forms based on protein mobility: SATB1 with high mobility and another with very low or no mobility. While we have identified the SATB1 domain responsible for each of these distinct mobility patterns, we have not yet identified biochemical differences that would allow us to distinguish them conclusively. Therefore, an experiment to test the distribution of SATB1 in soluble versus insoluble fractions would show SATB1 in both fractions but would not necessarily provide information about the functional significance of these different populations. We believe this is an important area for future research and are working to develop tools to specifically distinguish and characterize SATB1 in the soluble versus insoluble fractions.

eLife Assessment

This useful modeling study shows how spatial representations, similar to those seen in experimental data, emerge in a recurrent neural network trained on a navigation task. The training required path integration and decodability but did not rely on grid cell inputs. The network modeling is solid, though the link to experimental data could be strengthened.

Reviewer #1 (Public review):

Summary:

This work studies representations in a network with one recurrent layer and one output layer that needs to path-integrate so that its position can be accurately decoded from its output. To formalise this problem, the authors define a cost function consisting of the decoding error and a regularisation term. They specify a decoding procedure that, at a given time, averages the output unit center locations, weighted by the activity of the unit at that time. The network is initialised without position information, and only receives a velocity signal (and a context signal to index the environment) at each timestep, so to achieve low decoding error it needs to infer its position and keep it updated with respect to its velocity by path integration.

The authors take the trained network and let it explore a series of environments with different geometries while collecting unit activities to probe learned representations. They find localised responses in the output units (resembling place fields) and border responses in the recurrent units. Across environments, the output units show global remapping and the recurrent units show rate remapping. Stretching the environment generally produces stretched responses in output and recurrent units. Ratemaps remain stable within environments and stabilise after noise injection. Low-dimensional projections of the recurrent population activity forms environment-specific clusters that reflect the environment's geometry, which suggests independent rather than generalised representations. Finally, the authors discover that the centers of the output unit ratemaps cluster together on a triangular lattice (like the receptive fields of a single grid cell), and find significant clustering of place cell centers in empirical data as well.

The model setup and simulations are clearly described, and are an interesting exploration of the consequences of a particular set of training requirements - here: path integration and decodability. But it is not obvious to what extent the modelling choices are a realistic reflection of how the brain solves navigation. Therefore, it is not clear whether the results generalize beyond the specifics of the setup here.

Strengths:

The authors introduce a very minimal set of model requirements, assumptions, and constraints. In that sense, the model can function as a useful 'baseline', that shows how spatial representations and remapping properties can emerge from the requirement of path integration and decodability alone. Moreover, the authors use the same formalism to relate their setup to existing spatial navigation models, which is informative.

The global remapping that the authors show is convincing and well-supported by their analyses. The geometric manipulations and the resulting stretching of place responses, without additional training, are interesting. They seem to suggest that the recurrent network may scale the velocity input by the environment dimensions so that the exact same path integrator-output mappings remain valid (but maybe there are other mechanisms too that achieve the same).

The simulations and analyses in the appendices serve as insightful controls for the main results.

The clustering of place cell peaks on a triangular lattice is intriguing, given there is no grid cell input. It could have something to do with the fact that a triangular lattice provides optimal coverage of 2d space? The included comparison with empirical data is valuable as a first exploration, showing a promising example, but doesn't robustly support the modelling results.

Weaknesses:

The navigation problem that needs to be solved by the model is a bit of an odd one. Without any initial position information, the network needs to figure out where it is, and then path-integrate with respect to a velocity signal. As the authors remark in Methods 4.2, without additional input, the only way to infer location is from border interactions. It is like navigating in absolute darkness. Therefore, it seems likely that the salient wall representations found in the recurrent units are just a consequence of the specific navigation task here; it is unclear if the same would apply in natural navigation. In natural navigation, there are many more sensory cues that help inferring location, most importantly vision, but also smell and whiskers/touch (which provides a more direct wall interaction; here, wall interactions are indirect by constraining velocity vectors). There is a similar but weaker concern about whether the (place cell like) localised firing fields of the output units are a direct consequence of the decoding procedure that only considers activity center locations.

The conclusion that 'representations are attractive' (heading of section 2) is not entirely supported. The authors show 'attractor-like behaviour' within a single context, but there could be alternative explanations for the recovery of stable ratemaps after noise injection. For example, the noise injection could scramble the network's currently inferred position, so that it would need to re-infer its position from boundary interactions along the trajectory. In that case the stabilisation would be driven by the input, not just internal attractor dynamics. Indeed, the useful control analysis in Appendix D suggests such a mechanism: without a velocity signal, only for small noise injections the network returns to a high correlation state. Correlated representations are recovered for larger noise injections due to the same mechanism that allow the network to determine its position upon from an uninformative initial hidden state upon entering a new environment, i.e. boundary interactions.

The authors report empirical data that shows clustering of place cell centers like they find for their output units. They report that 'there appears to be a tendency for the clusters to arrange in hexagonal fashion, similar to our computational findings'. This is an interesting observation on the distribution of place field centres which seems justified based on the example animal shown, but not across the population of animals included.

Reviewer #2 (Public review):

Summary:

The authors proposed a neural network model to explore the spatial representations of the hippocampal CA1 and entorhinal cortex (EC) and the remapping of these representations when multiple environments are learned. The model consists of a recurrent network and output units (a decoder) mimicking the EC and CA1, respectively. The major results of this study are: the EC network generates cells with their receptive fields tuned to a border of the arena; the decoder develops neuron clusters arranged in a hexagonal lattice. Thus, the model accounts for entrohinal border cells and CA1 place cells. It suggests that the remapping of place cells occurs between different environments through state transitions corresponding to unstable dynamical modes in the recurrent network.

Strengths:

The authors found a spatial arrangement of receptive fields similar to their model's prediction in experimental data recorded from CA1. Thus, the model proposes plausible mechanisms to generate hippocampal spatial representations without relying on grid cells. The model also suggests an interesting possibility that path integration is not the speciality of grid cells.

Weaknesses:

The role of grid cells in the proposed view, i.e., the boundary-to-place-to-grid model, remains elusive. The model can generate place cells without generating entorhinal grid cells. Moreover, the model can generate hexagonal grid patterns of place cells in a large arena. Whether and how the proposed model is integrated into the entire picture of the hippocampal-entorhinal meｍory processing remains elusive.

Reviewer #3 (Public review):

Summary:

The authors used recurrent neural network modelling of spatial navigation tasks to investigate border and place cell behaviour during remapping phenomena.

Strengths:

The neural network training seemed for the most part (see comments later) well-performed, and the analyses used to make the points were thorough.

The paper and ideas were well-explained.

Figure 4 contained some interesting and strong evidence for map-like generalisation as environmental geometry was warped.

Figure 7 was striking and potentially very interesting.

It was impressive that the RNN path-integration error stayed low for so long (Fig A1), given that normally networks that only work with dead-reckoning have errors that compound. I would have loved to know how the network was doing this, given that borders did not provide sensory input to the network. I could not think of many other plausible explanations... It would be even more impressive if it was preserved when the network was slightly noisy.

Update:

The analysis of how the RNN remapped, using a context signal to switch between largely independent maps, and the examination of the border like tuning in the recurrent units of the RNN, were both thorough and interesting. Further, in the updated response I appreciated the additional appendix E which helped substantiate the claim that the RNN neurons were border cells.

Weaknesses:

The remapping results were also puzzling. The authors present convincing evidence that the recurrent units effectively form 6 different maps of the 6 different environments (e.g. the sparsity of the code, or fig 6a), with the place cells remapping between environments. Yet, as the authors point out, in neural data the finding is that some cells generalise their co-firing patterns across environments (e.g. grid cells, border cells), while place cells remap, making it unclear what correspondence to make between the authors network and the brain. There are existing normative models that capture both entorhinal's consistent and hippocampus' less consistent neural remapping behaviour (Whittington et al. and probably others), what have we then learnt from this exercise?

Update: see summary below

I felt that the neural data analysis was unconvincing. Most notably, the statistical effect was found in only one of seven animals. Random noise is likely to pass statistical tests 1 in 20 times (at 0.05 p value), this seems like it could have been something similar? Further, the data was compared to a null model in which place cell fields were randomly distributed. The authors claim place cell fields have two properties that the random model doesn't (1) clustering to edges (as experimentally reported) and (2) much more provocatively, a hexagonal lattice arrangement. The test seems to collude the two; I think that nearby ball radii could be overrepresented, as in figure 7f, due to either effect. I would have liked to see a computation of the statistic for a null model in which place cells were random but with a bias towards to boundaries of the environment that matches the observed changing density, to distinguish these two hypotheses.

Update: the authors acknowledge these shortcomings and have appropriately tempered their data related claims.

Some smaller weaknesses:<br /> - Had the models trained to convergence? From the loss plot it seemed like not, and when including regularisors recent work (grokking phenomena, e.g. Nanda et al. 2023) has shown the importance of letting the regularisor minimise completely to see the resulting effect. Else you are interpreting representations that are likely still being learnt, a dangerous business.<br /> Update: I understand that practical limitations make testing this thoroughly impossible, which is fair enough.

- The claim that this work provided a mathematical formalism of the intuitive idea of a cognitive map seems strange, given that upwards of 10 of the works this paper cite also mathematically formalise a cognitive map into a similar integration loss for a neural network.<br /> Update: the introduction of these ideas hasn't changed, and my concerns above remain.

Aim Achieved? Impact/Utility/Context of Work

I think this is a thorough exploration of how this network with these losses is able to path-integrate its position and remap. This is useful, it is good to know how another neural network with slightly different constraints learns to perform these behaviours.

In the updated version of the manuscript I am happy to say that I think there are few claims that are unsubstantiated (see weakness section above that has been significantly updated). The link to neuroscience remains the biggest shortcoming of this work in my view. The authors point to two main results in this direction. First, the ability for interactions only between border-type and place cells to produce many observed place-cell results, providing a new hypothesis. Second, a connection between grid cells, place cells, and border cells, in the production of hexagonal arrangements of place cells.

Regarding the first, as the authors discuss, current evidence suggests border cells are invariant across environments whereas this work finds border cells for specific environments (they use the words rate-remapping boundary-type cells). It seems likely to me that there are many ways a neural network can path-integrate across different environments. In other models where the same base map is re-used (e.g. TEM) grid cells emerge, in this work where the maps for different environments are disjoint these border-like cells that do not match an observed cell type in their tuning to environment are involved. I find this a really interesting alternative (I think what an RNN does is interesting in its own right), but I don't see why I should think it is what the brain does, given that it appears to match observations less well (existence of grid cells, consistent firing patterns of border cells across environments). The smoking gun in favour of the author's hypothesis would be finding these sparse border like cells, or some other evidence of gating like interactions between border and place cells as they discuss. Finding such evidence sounds difficult (so not reasonable to ask for in a rebuttal), and to reiterate, I applaud the authors for clearly outlining an alternative, but I remain unconvinced.

Regarding the second point, while the grid-like placement of field centres was cool, and I applaud the authors for including real neural data comparisons, as the authors say, the data is preliminary, and further evidence would be required to fully substantiate this claim.

As such, in my mind it is an interesting alternative hypothesis. I look forward to seeing experimental predictions or comparisons that can tighten the link, substantiating the claim that what this particular RNN is doing reflects the algorithms at work in the brain.

Author response:

The following is the authors’ response to the original reviews

Public Reviews:

Reviewer #1 (Public Review):

Summary:

This work studies representations in a network with one recurrent layer and one output layer that needs to path-integrate so that its position can be accurately decoded from its output. To formalise this problem, the authors define a cost function consisting of the decoding error and a regularisation term. They specify a decoding procedure that at a given time averages the output unit center locations, weighted by the activity of the unit at that time. The network is initialised without position information, and only receives a velocity signal (and a context signal to index the environment) at each timestep, so to achieve low decoding error it needs to infer its position and keep it updated with respect to its velocity by path integration.

The authors take the trained network and let it explore a series of environments with different geometries while collecting unit activities to probe learned representations. They find localised responses in the output units (resembling place fields) and border responses in the recurrent units. Across environments, the output units show global remapping and the recurrent units show rate remapping. Stretching the environment generally produces stretched responses in output and recurrent units. Ratemaps remain stable within environments and stabilise after noise injection. Low-dimensional projections of the recurrent population activity forms environment-specific clusters that reflect the environment's geometry, which suggests independent rather than generalised representations. Finally, the authors discover that the centers of the output unit ratemaps cluster together on a triangular lattice (like the receptive fields of a single grid cell), and find significant clustering of place cell centers in empirical data as well.

The model setup and simulations are clearly described, and are an interesting exploration of the consequences of a particular set of training requirements - here: path integration and decodability. But it is not obvious to what extent the modelling choices are a realistic reflection of how the brain solves navigation. Therefore it is not clear whether the results generalize beyond the specifics of the setup here.

Strengths:

The authors introduce a very minimal set of model requirements, assumptions, and constraints. In that sense, the model can function as a useful 'baseline', that shows how spatial representations and remapping properties can emerge from the requirement of path integration and decodability alone. Moreover, the authors use the same formalism to relate their setup to existing spatial navigation models, which is informative.

The global remapping that the authors show is convincing and well-supported by their analyses. The geometric manipulations and the resulting stretching of place responses, without additional training, are interesting. They seem to suggest that the recurrent network may scale the velocity input by the environment dimensions so that the exact same path integrator-output mappings remain valid (but maybe there are other mechanisms too that achieve the same).

The clustering of place cell peaks on a triangular lattice is intriguing, given there is no grid cell input. It could have something to do with the fact that a triangular lattice provides optimal coverage of 2d space? The included comparison with empirical data is valuable, although the authors only show significant clustering - there is no analysis of its grid-like regularity.

First of all, we would like to thank the reviewer for their comprehensive feedback, and their insightful comments. Importantly, as you point out, our goal with this model was to build a minimal model of place cell representations, where representations were encouraged to be place-like, but free to vary in tuning and firing locations. By doing so, we could explore what upstream representations facilitate place-like representations, and even remapping (as it turned out) with minimal assumptions. However, we agree that our task does not capture some of the nuances of real-world navigation, such as sensory observations, which could be useful extensions in future work. Then again, the simplicity of our setup makes it easier to interpret the model, and makes it all the more surprising that it learns many behaviors exhibited by real world place cells.

As to the distribution of phases - we also agree that a hexagonal arrangement likely reflects some optimal configuration for decoding of location.

And we agree that the symmetry within the experimental data is important; we have revised analyses on experimental phase distributions, and included an analysis of ensemble grid score, to quantify any hexagonal symmetries within the data.

Weaknesses:

The navigation problem that needs to be solved by the model is a bit of an odd one. Without any initial position information, the network needs to figure out where it is, and then path-integrate with respect to a velocity signal. As the authors remark in Methods 4.2, without additional input, the only way to infer location is from border interactions. It is like navigating in absolute darkness. Therefore, it seems likely that the salient wall representations found in the recurrent units are just a consequence of the specific navigation task here; it is unclear if the same would apply in natural navigation. In natural navigation, there are many more sensory cues that help inferring location, most importantly vision, but also smell and whiskers/touch (which provides a more direct wall interaction; here, wall interactions are indirect by constraining velocity vectors). There is a similar but weaker concern about whether the (place cell like) localised firing fields of the output units are a direct consequence of the decoding procedure that only considers activity center locations.

Thank you for raising this point; we absolutely agree that the navigation task is somewhat niche. However, this was a conscious decision, to minimize any possible confounding from alternate input sources, such as observations. In part, this experimental design was inspired by the suggestion that grid cells support navigation/path integration in open-field environments with minimal sensory input (as they could, conceivably do so with no external input). This also pertains to your other point, that boundary interactions are necessary for navigation. In our model, using boundaries is one solution, but there is another way around this problem, which is conceivably better: to path integrate in an egocentric frame, starting from your initial position. Since the locations of place fields are inferred only after a trajectory has been traversed, the network is free to create a new or shifted representation every time, independently of the arena. In this case, one might have expected generalized solutions, such as grid cells to emerge. That this is not the case, seems to suggest that grid cells may somehow not be optimal for pure path integration, or at the very least, hard to learn (but may still play a part, as alluded to by place field locations). We have tried to make these points more evident in the revised manuscript.

As for the point that the decoding may lead to place-like representations, this is a fair point. Indeed, we did choose this form of decoding, inspired by the localized firing of place cells, in the hope that it would encourage minimally constrained, place-like solutions. However, compared to other works (Sorscher and Xu) hand tuning the functional form of their place cells, our (although biased towards centralized tuning curves) allows for flexible functional forms such as the position of the place cell centers, their tuning width, whether or not it is center-surround activity, and how they should tune to different environments/rooms. This allows us to study several features of the place cell system, such as remapping and field formation. We have revised to make this more clear in the model description.

The conclusion that 'contexts are attractive' (heading of section 2) is not well-supported. The authors show 'attractor-like behaviour' within a single context, but there could be alternative explanations for the recovery of stable ratemaps after noise injection. For example, the noise injection could scramble the network's currently inferred position, so that it would need to re-infer its position from boundary interactions along the trajectory. In that case the stabilisation would be driven by the input, not just internal attractor dynamics. Moreover, the authors show that different contexts occupy different regions in the space of low-dimensional projections of recurrent activity, but not that these regions are attractive.

We agree that boundary interactions could facilitate the convergence of representations after noise injection. We did try to moderate this claim by the wording “attractor-like”, but we agree that boundaries could confound this result. We have therefore performed a modified noise injection experiment, where we let the network run for an extended period of time, before noise injection (and no velocity signal), see Appendix Velocity Ablation in the revised text. Notably, representations converge to their pre-scrambled state after noise injection, even without a velocity signal. However, place-like representations do not converge for all noise levels in this case, possibly indicating that boundary interactions do serve an error-correcting function, also. Thank you for pointing this out.

As for the attractiveness of contexts, we agree that more analyses were required to demonstrate this. We have therefore conducted a supplementary analysis where we run the trained network with a mismatch in context/geometry, and demonstrate that the context signal fixes the representation, up to geometric distortions.

The authors report empirical data that shows clustering of place cell centers like they find for their output units. They report that 'there appears to be a tendency for the clusters to arrange in hexagonal fashion, similar to our computational findings'. They only quantify the clustering, but not the arrangement. Moreover, in Figure 7e they only plot data from a single animal, then plot all other animals in the supplementary. Does the analysis of Fig 7f include all animals, or just the one for which the data is plotted in 7e? If so, why that animal? As Appendix C mentions that the ratemap for the plotted animal 'has a hexagonal resemblance' whereas other have 'no clear pattern in their center arrangements', it feels like cherrypicking to only analyse one animal without further justification.

Thank you for pointing this out; we agree that this is not sufficiently explained and explored in the current version. We have therefore conducted a grid score analysis of the experimental place center distributions, to uncover possible hexagonal symmetries. The reason for choosing this particular animal was in part because it featured the largest number of included cells, while also demonstrating the most striking phase distribution, while including all distributions in the supplementary. Originally, this was only intended as a preliminary analysis, suggesting non-uniformity in experimental place field distributions, but we realize that these may all provide interesting insight into the distributional properties of place cells.

We have explained these choices in the revised text, and expanded analyses on all animals to showcase these results more clearly.

Reviewer #2 (Public Review):

Summary:

The authors proposed a neural network model to explore the spatial representations of the hippocampal CA1 and entorhinal cortex (EC) and the remapping of these representations when multiple environments are learned. The model consists of a recurrent network and output units (a decoder) mimicking the EC and CA1, respectively. The major results of this study are: the EC network generates cells with their receptive fields tuned to a border of the arena; decoder develops neuron clusters arranged in a hexagonal lattice. Thus, the model accounts for entorhinal border cells and CA1 place cells. The authors also suggested the remapping of place cells occurs between different environments through state transitions corresponding to unstable dynamical modes in the recurrent network.

Strengths:

The authors found a spatial arrangement of receptive fields similar to their model's prediction in experimental data recorded from CA1. Thus, the model proposes a plausible mechanisms to generate hippocampal spatial representations without relying on grid cells. This result is consistent with the observation that grid cells are unnecessary to generate CA1 place cells.

The suggestion about the remapping mechanism shows an interesting theoretical possibility.

We thank the reviewer for their kind feedback.

Weaknesses:

The explicit mechanisms of generating border cells and place cells and those underlying remapping were not clarified at a satisfactory level.

The model cannot generate entorhinal grid cells. Therefore, how the proposed model is integrated into the entire picture of the hippocampal mechanism of meｍory processing remains elusive.

We appreciate this point, and hope to clarify: From a purely architectural perspective, place-like representations are generated by linear combinations of recurrent unit representations, which, after training, appear border-like. During remapping, the network is simply evaluated/run in different geometries/contexts, which, it turns out, causes the network to exhibit different representations, likely as solutions to optimally encoding position in the different environments. We have attempted to revise the text to make some of these interpretations more clear. We have also conducted a supplementary analysis to demonstrate how representations are determined by the context signal directly, which helps to explain how recurrent and output units form their representations.

We also agree that our model does not capture the full complexity of the Hippocampal formation. However, we would argue that its simplicity (focusing on a single cell type and a pure path integration task), acts as a useful baseline for studying the role of place cells during spatial navigation. The fact that our model captures a range of place cell behaviors (field formation, remapping and geometric deformation) without grid cells also point to several interesting possibilities, such that grid cells may not be strictly necessary for place cell formation and remapping, or that border cells may account for many of the peculiar behaviors of place cells. However, we wholeheartedly agree that including e.g. sensory information and memory storage/retrieval tasks would prove a very interesting extension of our model to more naturalistic tasks and settings. In fact, our framework could easily accommodate this, e.g. by decoding contexts/observations/memories from the network state, alongside location.

Reviewer #3 (Public Review):

Summary:

The authors used recurrent neural network modelling of spatial navigation tasks to investigate border and place cell behaviour during remapping phenomena.

Strengths:

The neural network training seemed for the most part (see comments later) well-performed, and the analyses used to make the points were thorough.

The paper and ideas were well explained.

Figure 4 contained some interesting and strong evidence for map-like generalisation as environmental geometry was warped.

Figure 7 was striking, and potentially very interesting.

It was impressive that the RNN path-integration error stayed low for so long (Fig A1), given that normally networks that only work with dead-reckoning have errors that compound. I would have loved to know how the network was doing this, given that borders did not provide sensory input to the network. I could not think of many other plausible explanations... It would be even more impressive if it was preserved when the network was slightly noisy.

Thank you for your insightful comments! Regarding the low path integration error, there is a slight statistical signal from the boundaries, as trajectories tend to turn away from arena boundaries. However, we agree, that studying path integration performance in the face of noise would make for a very interesting future development.

Weaknesses:

I felt that the stated neuroscience interpretations were not well supported by the presented evidence, for a few reasons I'll now detail.

First, I was unconvinced by the interpretation of the reported recurrent cells as border cells. An equally likely hypothesis seemed to be that they were positions cells that are linearly encoding the x and y position, which when your environment only contains external linear boundaries, look the same. As in figure 4, in environments with internal boundaries the cells do not encode them, they encode (x,y) position. Further, if I'm not misunderstanding, there is, throughout, a confusing case of broken symmetry. The cells appear to code not for any random linear direction, but for either the x or y axis (i.e. there are x cells and y cells). These look like border cells in environments in which the boundaries are external only, and align with the axes (like square and rectangular ones), but the same also appears to be true in the rotationally symmetric circular environment, which strikes me as very odd. I can't think of a good reason why the cells in circular environments should care about the particular choice of (x,y) axes... unless the choice of position encoding scheme is leaking influence throughout. A good test of these would be differently oriented (45 degree rotated square) or more geometrically complicated (two diamonds connected) environments in which the difference between a pure (x,y) code and a border code are more obvious.

Thank you for pointing this out. This is an excellent point, that we agree could be addressed more rigorously. Note that there is no position encoding in our model; the initial state of the network is a vector of zeros, and the network must infer its location from boundary interactions and context information alone. So there is no way for positional information to leak through to the recurrent layer directly. However, one possible reason for the observed symmetry breaking, is the fact that the velocity input signal is aligned with the cardinal directions. To investigate this, we trained a new model, wherein input velocities are rotated 45 degrees relative to the horizontal, as you suggest. The results, shown and discussed in appendix E (Learned recurrent representations align with environment boundaries), do indicate that representations are tuned to environment boundaries, and not the cardinal directions, which hopefully improves upon this point.

Next, the decoding mechanism used seems to have forced the representation to learn place cells (no other cell type is going to be usefully decodable?). That is, in itself, not a problem. It just changes the interpretation of the results. To be a normative interpretation for place cells you need to show some evidence that this decoding mechanism is relevant for the brain, since this seems to be where they are coming from in this model. Instead, this is a model with place cells built into it, which can then be used for studying things like remapping, which is a reasonable stance.

This is a great point, and we agree. We do write that we perform this encoding to encourage minimally constrained place-like representations (to study their properties), but we have revised to make this more evident.

However, the remapping results were also puzzling. The authors present convincing evidence that the recurrent units effectively form 6 different maps of the 6 different environments (e.g. the sparsity of the code, or fig 6a), with the place cells remapping between environments. Yet, as the authors point out, in neural data the finding is that some cells generalise their co-firing patterns across environments (e.g. grid cells, border cells), while place cells remap, making it unclear what correspondence to make between the authors network and the brain. There are existing normative models that capture both entorhinal's consistent and hippocampus' less consistent neural remapping behaviour (Whittington et al. and probably others), what have we then learnt from this exercise?

Thanks for raising this point! We agree that this finding is surprising, but we hold that it actually shows something quite important: that border-type units are sufficient to create place-like representations, and learns several of the behaviors associated with place cells and remapping (including global remapping and field stretching). In other words, a single cell type known to exist upstream of place cells is sufficient to explain a surprising range of phenomena, demonstrating that other cell types are not strictly necessary. However, we agree that understanding why the boundary type units sometimes rate remap, and whether that can be true for some border type cells in the brain (either directly, or through gating mechanisms) would be important future developments. Related to this point, we also expanded upon the influence of the context signal for representation selection (appendix F)

Concerning the relationship to other models, we would argue that the simplicity of our model is one of its core strengths, making it possible to disentangle what different cell types are doing. While other models, including TEM, are highly important for understanding how different cell types and brain regions interact to solve complex problems, we believe there is a need for minimal, understandable models that allows us to investigate what each cell type is doing, and this is where we believe our work is important. As an example, our model not only highlights the sufficiency of boundary-type cells as generators of place cells, its lack of e.g. grid cells also suggest that grid cells may not be strictly necessary for e.g. open-field/sensory-deprived navigation, as is often claimed.

One striking result was figure 7, the hexagonal arrangement of place cell centres. I had one question that I couldn't find the answer to in the paper, which would change my interpretation. Are place cell centres within a single clusters of points in figure 7a, for example, from one cell across the 100 trajectories, or from many? If each cluster belongs to a different place cell then the interpretation seems like some kind of optimal packing/coding of 2D space by a set of place cells, an interesting prediction. If multiple place cells fall within a single cluster then that's a very puzzling suggestion about the grouping of place cells into these discrete clusters. From figure 7c I guess that the former is the likely interpretation, from the fact that clusters appear to maintain the same colour, and are unlikely to be co-remapping place cells, but I would like to know for sure!

This is a good point, and you are correct: one cluster tends to correspond to one unit. To make this more clear, we have revised Fig. 7, so that each decoded center is shaded by unit identity, which makes this more evident. And yes, this is, seemingly in line with some form of optimal packing/encoding of space, yes!

I felt that the neural data analysis was unconvincing. Most notably, the statistical effect was found in only one of seven animals. Random noise is likely to pass statistical tests 1 in 20 times (at 0.05 p value), this seems like it could have been something similar? Further, the data was compared to a null model in which place cell fields were randomly distributed. The authors claim place cell fields have two properties that the random model doesn't (1) clustering to edges (as experimentally reported) and (2) much more provocatively, a hexagonal lattice arrangement. The test seems to collude the two; I think that nearby ball radii could be overrepresented, as in figure 7f, due to either effect. I would have liked to see a computation of the statistic for a null model in which place cells were random but with a bias towards to boundaries of the environment that matches the observed changing density, to distinguish these two hypotheses.

Thanks for raising this point. We agree that we were not clear enough in our original manuscript. We included additional analyses in one animal, to showcase one preliminary case of non-uniform phases. To mitigate this, we have performed the same analyses for all animals, and included a longer discussion of these results (included in the supplementary material). We have also moderated the discussion on Ripley’s H to encompass only non-uniformity, and added a grid score analysis to showcase possible rotational symmetries in the data. We hope this gets our findings across more clearly

Some smaller weaknesses:

- Had the models trained to convergence? From the loss plot it seemed like not, and when including regularisors recent work (grokking phenomena, e.g. Nanda et al. 2023) has shown the importance of letting the regularisor minimise completely to see the resulting effect. Else you are interpreting representations that are likely still being learnt, a dangerous business.

Longer training time did not seem to affect representations. However, due to the long trajectories and statefulness involved, training was time-intensive and could become unstable for very long training. We therefore stopped training at the indicated time.

- Since RNNs are nonlinear it seems that eigenvalues larger than 1 doesn't necessarily mean unstable?

This is a good point; stability is not guaranteed. We have updated the text to reflect this.

- Why do you not include a bias in the networks? ReLU networks without bias are not universal function approximators, so it is a real change in architecture that doesn't seem to have any positives?

We found that bias tended to have a detrimental effect on training, possibly related to the identity initialization used (see e.g. Le et al. 2015), and found that training improved when biases were fixed to zero.

- The claim that this work provided a mathematical formalism of the intuitive idea of a cognitive map seems strange, given that upwards of 10 of the works this paper cite also mathematically formalise a cognitive map into a similar integration loss for a neural network.

We agree that other works also provide ways of formalizing this concepts. However, our goal by doing so was to elucidate common features across these seemingly disparate models. We also found that the concept of a learned and target map made it easier to come up with novel models, such as one wherein place cells are constructed to match a grid cell label.

Aim Achieved? Impact/Utility/Context of Work

Given the listed weaknesses, I think this was a thorough exploration of how this network with these losses is able to path-integrate its position and remap. This is useful, it is good to know how another neural network with slightly different constraints learns to perform these behaviours. That said, I do not think the link to neuroscience was convincing, and as such, it has not achieved its stated aim of explaining these phenomena in biology. The mechanism for remapping in the entorhinal module seemed fundamentally different to the brain's, instead using completely disjoint maps; the recurrent cell types described seemed to match no described cell type (no bad thing in itself, but it does limit the permissible neuroscience claims) either in tuning or remapping properties, with a potentially worrying link between an arbitrary encoding choice and the responses; and the striking place cell prediction was unconvincingly matched by neural data. Further, this is a busy field in which many remapping results have been shown before by similar models, limiting the impact of this work. For example, George et al. and Whittington et al. show remapping of place cells across environments; Whittington et al. study remapping of entorhinal codes; and Rajkumar Vasudeva et al. 2022 show similar place cell stretching results under environmental shifts. As such, this papers contribution is muddied significantly.

Thank you for this perspective; we agree that all of these are important works that arrive at complementary findings. We hold that the importance of our paper lies in its minimal nature, and its focus on place cells, via a purpose-built decoding that enables place-like representations. In doing so, we can point to possibly under explored relationships between cell types, in particular place cells and border cells, while challenging the necessity of other cell types for open-field navigation (i.e. grid cells). In addition, our work points to a novel connection between grid cells, place cells and even border cells, by way of the hexagonal arrangement of place unit centers. However, we agree that expanding our model to include more biologically plausible architectures and constraints would make for a very interesting extension in the future.

Thank you again for your time, as well as insightful comments.  

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

Even after reading Methods 5.3, I found it hard to understand how the ratemap population vectors that produce Fig 3e and Fig 5 are calculated. It's unclear to me how there can be a ratemap at a single timestep, because calculating a ratemap involves averaging the activity in each location, which would take a whole trajectory and not a single timestep. But I think I've understood from Methods 5.1 that instead the ratemap is calculated by running multiple 'simultaneous' trajectories, so that there are many visited locations at each timestep. That's a bit confusing because as far as I know it's not a common way to calculate ratemaps in rodent experiments (probably because it would be hard to repeat the same task 500 times, while the representations remain the same), so it might be worth explaining more in Methods 5.3.

We understand the confusion, and have attempted to make this more clear in the revised manuscript. We did indeed create ratemaps over many trajectories for time-dependent plots, for the reasons you mentioned. We also agree that this would be difficult to do experimentally, but found it an interesting way to observe convergence of representations in our simulated scenario.

Fig 3b-d shows multiple analyses to support output unit global remapping, but no analysis to support the claim that recurrent units remap by rate changes. The examples in Fig 3ai look pretty convincing, but it would be useful to also have a more quantitative result.

We agree, and only showed that units turn off/become silent using ratemaps. We have therefore added an explicit analysis, showcasing rate remapping in recurrent units (see appendix G; Recurrent units rate remap)

Reviewer #2 (Recommendations For The Authors):

Some parts of the current manuscript are hard to follow. Particularly, the model description is not transparent enough. See below for the details.

Major comments:

(1) Mathematical models should be explained more explicitly and carefully. I had to guess or desperately search for the definitions of parameters. For instance, define the loss function L in eq.(1). Though I can assume L represents the least square error (in A.8), I could not find the definition in Model & Objective. N should also be defined explicitly in equation (3). Is this the number of output cells?

Thank you for pointing this out, we have revised to make it more clear.

(2) In Fig. 1d, how were the velocity and context inputs given to individual neurons in the network? The information may be described in the Methods, but I could not identify it.

This was described in the methods section (Neural Network Architecture and Training), but we realize that we used confusing notation, when comparing with Fig. 1d. We have therefore changed the notation, and it should hopefully be clearer now. Thanks for pointing out this discrepancy.

(3) I took a while to understand equations (3) and (4) (for instance, t is not defined here). The manuscript would be easier to read if equations (5) and (6) are explained in the main text but not on page 18 (indeed, these equations are just copies of equations 3 and 4). Otherwise, the authors may replace equations (3) and (4) with verbal explanations similar to figure legend for Fig. 1b.

(4) Is there any experimental evidence for uniformly strong EC-to-CA1 projections assumed in the non-trainable decoder? This point should be briefly mentioned.

Thank you for raising this point. The decoding from EC (the RNN) to CA1 (the output layer) consists of a trainable weight matrix, and may thus be non-uniform in magnitude. The non-trainable decoding acts on the resulting “CA1” representation only. We hope that improvements to the model description also makes this more evident.  

(5) The explanation of Fig. 3 in the main text is difficult to follow because subpanels are explained in separate paragraphs, some of which are very short, as short as just a few lines.

This presentation style makes it difficult to follow the logical relationships between the subpanels. This writing style is obeyed throughout the manuscript but is not popular in neuroscience.

Thanks for pointing this out, we have revised to accommodate this.

(6) Why do field centers cluster near boundaries? No underlying mechanisms are discussed in the manuscript.

This is a good point; we have added a note on this; it likely reflects the border tuning of upstream units.

(7) In Fig. 4, the authors presented how cognitive maps may vary when the shape and size of open arenas are modified. The results would be more interesting if the authors explained the remapping mechanism. For instance, on page 8, the authors mentioned that output units exhibit global remapping between contexts, whereas recurrent units mainly rate remapping.

Why do such representational differences emerge?

We agree! Thanks for raising this point. We have therefore expanded upon this discussion in section 2.4.

(8) In the first paragraph of page 10, the authors stated ".. some output units display distinct field doubling (see both Fig. 4c), bottom right, and Fig. 4d), middle row)". I could not understand how Fig. 4d, middle row supports the argument. Similarly, they stated "..some output units reflect their main boundary input (with greater activity near one boundary)." I can neither understand what the authors mean to say nor which figures support the statement. Please clarify.

This is a good point, there was an identifier missing; we have updated to refer to the correct “magnification”. Thanks!

(9) The underlying mechanism of generating the hexagonal representation of output cells remains unclear. The decoder network uses a non-trainable decoding scheme based on localized firing patterns of output units. To what extent does the hexagonal representation depend on the particular decoding scheme? Similarly, how does the emergence of the hexagonal representation rely on the border representation in the upstream recurrent network? Showing several snapshots of the two place representations during learning may answer these questions.

This is an interesting point, and we have added some discussion on this matter. In particular, we speculate whether it’s an optimal configuration for position reconstruction, which is demanded by the task and thus highly likely dependent on the decoding scheme. We have not reached a conclusive method to determine the explicit dependence of the hexagonal arrangement on the choice of decoding scheme. Still, it seems this would require comparison with other schemes. In our framework, this would require changing the fundamental operation of the model, which we leave as inspiration for future work. We have also added additional discussion concerning the relationship between place units, border units, and remapping in our model. As for exploring different training snapshots, the model is randomly initialized, which suggests that earlier training steps should tend to reveal unorganized/uninformative phase arrangements, as phases are learned as a way of optimizing position reconstruction. However, we do call for more analysis of experimental data to determine whether this is true in animals, which would strongly support this observation. We also hope that our work inspires other models studying the formation and remapping of place cells, which could serve as a starting point for answering this question in the future.

(10) Figure 7 requires a title including the word "hexagonal" to make it easier to find the results demonstrating the hexagonal representations. In addition, please clarify which networks, p or g, gave the results shown here.

We agree, and have added it!

Minor comments:

(11) In many paragraphs, conclusions appear near their ends. Stating the conclusion at the beginning of each paragraph whenever possible will improve the readability.

We have made several rewrites to the manuscript, and hope this improves readability.

(12) Figure A4 is important as it shows evidence of the CA1 spatial representation predicted by the model. However, I could not find where the figure is cited in the manuscript. The authors can consider showing this figure in the main text.

We agree, and we have added more references to the experimental data analyses in the main text, as well as expanded this analysis.

(13) The main text cites figures in the following format: "... rate mapping of Fig. 3a), i), boundary ...." The parentheses make reading difficult.

We have removed the overly stringent use of double parentheses, thanks for letting us know.

(14) It would be nice if the authors briefly explained the concept of Ripley's H function on page 14.

Yes, we have added a brief descriptor.

eLife Assessment

The study investigated the effects of the peptide galanin on brain Ca2+ activity in zebrafish, which provides a useful model organism for whole-brain imaging because of its transparency. They found that galanin has distinct effects on hyperactivity and expression of galanin changes after activity increases. The strength of evidence was incomplete particularly for some of the conclusions regarding the use of convulsants and relevance to epilepsy because of limitations to the methods and interpretations of results.

Reviewer #1 (Public review):

Summary:

In this study, authors explored how galanin affects whole-brain activity in larval zebrafish using wide-field Ca2+ imaging, genetic modifications, and drugs that increase brain activity. The authors conclude that galanin has a sedative effect on the brain under normal conditions and during seizures, mainly through the galanin receptor 1a (galr1a). However, acute "stressors(?)" like pentylenetetrazole (PTZ) reduce galanin's effects, leading to increased brain activity and more seizures. Authors claim that galanin can reduce seizure severity while increasing seizure occurrence, speculated to occur through different receptor subtypes. This study confirms galanin's complex role in brain activity, supporting its potential impact on epilepsy.

Strengths:

The overall strength of the study lies primarily in its methodological approach using whole-brain Calcium imaging facilitated by the transparency of zebrafish larvae. Additionally, the use of transgenic zebrafish models is an advantage, as it enables genetic manipulations to investigate specific aspects of galanin signaling. This combination of advanced imaging and genetic tools allows for addressing galanin's role in regulating brain activity.

Weaknesses:

We have carefully reviewed the revised manuscript and the authors' responses. While the authors have attempted to address the points raised, I find that the revisions and rebuttals are insufficient and not entirely adequate. The authors seem not to have modified the manuscript in any way to take our comments into account.

In particular, many of the methodological and conceptual issues I initially raised remain unresolved. For example, the fundamental concern regarding the use of whole-brain calcium imaging - a method that may not effectively capture the localized and network-specific nature of seizure initiation and propagation - has not been adequately addressed. The authors acknowledge some limitations but do not sufficiently discuss how these affect the interpretation of their findings or propose mitigations. This could be added to the discussion section.

Additionally, the characterization of PTZ as a "stressor" remains problematic. Although the authors have retained this terminology, PTZ is widely understood to act primarily as a proconvulsant agent rather than a general stressor, and framing it otherwise continues to appear like a model-fitting rather than evidence-driven decision. The authors should consider changing the terminology throughout the manuscript and address these concerns when discussing their choice of PTZ as "stressor".

The discussion of the EAAT2 mutant model also remains incomplete. Although the authors mention preliminary transcriptome analyses, no new data were included, and it is stated that the evaluation is ongoing. Without thorough gene expression data, alternative explanations for the hypoactivity phenotype (such as changes in AMPA receptor or other critical neurotransmission-related genes) remain plausible and unaddressed. Moreover, the authors' acknowledgement that galanin upregulation is "at best one of a suite of regulatory mechanisms" further diminishes the centrality of their conclusions without sufficiently reworking the narrative of the study.

Finally, the finding that double knockout animals for EAAT2 and galanin showed little difference in seizure susceptibility compared to EAAT2 knockouts alone suggests that galanin upregulation may not play a dominant functional role, yet this important implication is not adequately reflected in the interpretation of the results.

Conclusion:

In summary, although the authors have made some efforts to respond to the critiques, I do not believe the manuscript has been substantially improved in response in R2, and I do not see reason to change my original assessment made after R1. The major conceptual and methodological concerns remain largely unaddressed, limiting the impact and validity of the study's conclusions. These concerns should be addressed not only in the rebuttal letter but also in the manuscript.

Reviewer #2 (Public review):

This revised paper describes an investigation of galanin and galanin receptor signaling on whole-brain activity in the context of recurrent seizure activity or under homeostatic basal conditions. The authors primarily use calcium imaging to observe whole-brain neuronal activity accompanied by galanin qPCR to determine how manipulations of galanin or the galr1a receptor affect the activity of the whole-brain under non-ictal conditions or when seizure activity occurs. The authors use their eaat2a-/- model (introduced in their Glia 2022 paper, PMID 34716961) that shows recurrent seizure activity as well as suppression of neuronal activity and locomotion interictally. It is compared to the well-known pentylenetetrazole (PTZ) pharmacological model of seizures in zebrafish. Given the literature cited in their Introduction, the authors hypothesize that galanin will exert a net inhibitory effect on brain activity in models of seizures/epilepsy. They were surprised to find that this hypothesis was only moderately supported in their eaat2a-/- model. In contrast, after PTZ, fish with galanin overexpression showed increased seizure number and reduced duration while fish with galanin KO showed reduced seizure number and increased duration.

Previous concerns about sex or developmental biological variables were addressed, as their model's seizure phenotype emerges rapidly and long prior to the establishment of zebrafish sexual maturity. However, it remains unclear whether all seizures detected via calcium imaging alone are also seizures that are detectable at the level of animal behavior. To confirm this, a validation of the threshold used for calcium imaging of "neuronal seizures" would be required to determine if this threshold detects only "neuronal seizures" that co-occur with behavioral seizures. Overall, this study is important and convincing, and carries clear value for understanding the multifaceted functions that neuronal galanin can perform under homeostatic and disease conditions.

Additional Concerns:

- The authors have validated their ability to measure behavioral seizures quantitatively in their 2022 Glia paper but the information provided on defining behavioral seizures as they map onto seizures detected via imaging alone was limited. The definition of behavioral seizure activity as it relates to calcium fluctuations is not expanded upon in this paper, but could provide detail about how the behavioral seizures relate to a seizure detected via calcium imaging alone.

Reviewer #3 (Public review):

Summary:

The neuropeptide galanin is primarily expressed in the hypothalamus and has been shown to play critical roles in homeostatic functions such as arousal, sleep, stress, and brain disorders such as epilepsy. Previous work in rodents using galanin analogs and receptor-specific knockout have provided convincing evidence for anti-convulsant effects of galanin.

In the present study, the authors sought to determine the relationship between galanin expression and whole-brain activity. The authors took advantage of the transparent nature of larval zebrafish to perform whole-brain neural activity measurements via widefield calcium imaging. Two models of seizures were used (eaat2a-/- and pentylenetetrazol; PTZ). In the eaat2a-/- model, spontaneous seizures occur and the authors found that galanin transcript levels were significantly increased and associated with reduced frequency of calcium events. Similarly, two hours after PTZ galanin transcript levels roughly doubled and the frequency and amplitude of calcium events were reduced, while the duration increased.

The authors also used a heat shock protein line (hsp70I:gal) where galanin transcripts levels are induced by activation of heat shock protein, but this line also shows higher basal transcript levels of galanin. Due to problems with whole-brain activity in wild-type larvae, the authors used the line without heat shock. They found higher level of galanin in hsp70I:gal larval zebrafish resulted in a reduction in the number of calcium events and amplitude. In contrast, galanin knockout (gal-/-) significantly increased calcium activity, indicated by an increased number of calcium events, but a reduction in amplitude and duration. Antibody staining confirmed the absence of galanin expression in gal-/- knockouts. Knockout of the galanin receptor subtype galr1a via crispants also increased the frequency of calcium events without influencing amplitude or duration.

In subsequent experiments in eaat2a-/- mutants were crossed with hsp70I:gal or gal-/- to modify galanin expression. These experiments showed modest effects, with eaat2a-/- x gal-/- knockouts showing an increased normalized area under the curve and seizure amplitude.

Lastly, the authors attempted to study the relationship between galanin and brain activity during a PTZ challenge. The hsp70I:gal larva showed increased number of seizures and reduced seizure duration during PTZ. In contrast, gal-/- mutants showed increased normalized area under the curve and a stark reduction in number of detected seizures, a reduction in seizure amplitude, but an increase in seizure duration. The authors then ruled out the role galanin a1 receptor in modulating this effect during PTZ, since the number of seizures was unaffected, whereas the amplitude and duration of seizures was increased in galr1a knockouts.

Strengths:

(1) The gain- and loss-of function galanin manipulations provided convincing evidence that galanin influences brain activity (via calcium imaging) during interictal and/or seizure-free periods. The relationship between galanin transcript levels and brain activity in figures 1 & 2 was convincing. Antibody staining also supports the absence of galanin in gal-/- mutants. Moreover, galanin transcript levels were unchanged in galr1ako brains, suggesting the lack of compensatory effects.

(2) The authors use two models of epilepsy (eaat2a-/- and PTZ).

(3) Supplementary video files for calcium imaging support the observations.

Weaknesses:

(1) I disagree with the idea that PTZ is a 'stressor'. This was raised in previous reviews and has not been acknowledged sufficiently.

(2) Although the relationship between galanin and brain activity during interictal or seizure-free periods was clear, the mechanisms that influence excitability during PTZ remain unclear. The authors show that galr1a does not mediate this effect, since seizure amplitude and duration were more severe in galr1a KO. Therefore, it remains unclear which galanin receptor is modulating this inhibitory effect.

(3) The manuscript is heavily reliant on calcium imaging for interpretation.<br /> Additional methods could strengthen the data, translational relevance, and interpretation (e.g., acute pharmacology using selective galanin agonists or antagonists, brain or cell recordings, biochemistry, etc).

Author response:

The following is the authors’ response to the previous reviews

Review 1:

Weaknesses:

The weaknesses of the study also stem from the methodological approach, particularly the use of whole-brain Calcium imaging as a measure of brain activity. While epilepsy and seizures involve network interactions, they typically do not originate across the entire brain simultaneously. Seizures often begin in specific regions or even within specific populations of neurons within those regions. Therefore, a whole-brain approach, especially with Calcium imaging with inherited limitations, may not fully capture the localized nature of seizure initiation and propagation, potentially limiting the understanding of Galanin's role in epilepsy.

We agree with the reviewers that the whole brain imaging approach is both a strength and a weakness. This manuscript and our previously published paper (Hotz et al., 2022) show indeed that the seizures have a initiation point and spread throughout the brain, interestingly affecting the telencephalon last. Localized seizure initiation was not the scope of this manuscript, however also here we would have to rely on imaging techniques. Using cell type specific drivers for specific neuronal subpopulation are an interesting approach, but outside of the scope of this study. An interesting approach would also include a more detailed analysis of glia in the context of epilepsy.

Furthermore, Galanin's effects may vary across different brain areas, likely influenced by the predominant receptor types expressed in those regions. Additionally, the use of PTZ as a "stressor" is questionable since PTZ induces seizures rather than conventional stress. Referring to seizures induced by PTZ as "stress" might be a misinterpretation intended to fit the proposed model of stress regulation by receptors other than Galanin receptor 1 (GalR1).

We also agree, that a more regional approach, after having more reliable information on the expression domains of the different galanin receptors, including more information on their respective role, is an important future research direction.

The description of the EAAT2 mutants is missing crucial details. EAAT2 plays a significant role in the uptake of glutamate from the synaptic cleft, thereby regulating excitatory neurotransmission and preventing excitotoxicity. Authors suggest that in EAAT2 knockout (KO) mice galanin expression is upregulated 15-fold compared to wild-type (WT) mice, which could be interpreted as galanin playing a role in the hypoactivity observed in these animals.

However, the study does not explore the misregulation of other genes that could be contributing to the observed phenotype. For instance, if AMPA receptors are significantly downregulated, or if there are alterations in other genes critical for brain activity, these changes could be more important than the upregulation of galanin. The lack of wider gene expression analysis leaves open the possibility that the observed hypoactivity could be due to factors other than, or in addition to, galanin upregulation.

We are in the process of preparing a manuscript describing a more detailed gene expression study of this and a chemically induced seizure model. Surprisingly we did not observe strong effects on glutamate receptor related genes. This does not preclude and indeed we deem it likely that additional factors play a role, e.g. other neuropeptides.

Moreover, the observation that in double KO mice for both EAAT2 and galanin there was little difference in seizure susceptibility compared to EAAT2 KO mice alone further supports the idea that galanin upregulation might not be the reason to the observed phenotype. This indicates that other regulatory mechanisms or gene expressions might be playing a more pivotal role in the manifestation of hypoactivity in EAAT2 mutants.

Yes, we agree that galanin is likely not the only player. This warrants further investigations.

These methodological shortcomings and conceptual inconsistencies undermine the perceived strengths of the study, and hinders understanding of Galanin's role in epilepsy and stress regulation.

Review 2:

Previous concerns about sex or developmental biological variables were addressed, as their model's seizure phenotype emerges rapidly and long prior to the establishment of zebrafish sexual maturity. However, in the course of re-review, some additional concerns (below) were detected that, if addressed, could further improve the manuscript. These concerns relate to how seizures were defined from the measurement of fluorescent calcium imaging data. Overall, this study is important and convincing, and carries clear value for understanding the multifaceted functions that neuronal galanin can perform under homeostatic and disease conditions.

We are pleased that we could dispel the initial concerns.

Additional Concerns:

- The authors have validated their ability to measure behavioral seizures quantitatively in their 2022 Glia paper but the information provided on defining behavioral seizures was limited. The definition of behavioral seizure activity is not expanded upon in this paper, but could provide detail about how the behavioral seizures relate to a seizure detected via calcium imaging.

In this paper we indeed do not address behavioral seizures but focus completely on neuronal seizures as defined in the material and methods section (“seizures were defined as calcium fluctuations reaching at least 100% of ΔF/F0 in the whole brain.”). Epileptic seizures in zebrafish, either evoked by pharmacological means or the result of genetic mutations, evoke stereotyped locomotor behavior in zebrafish as described in multiple publications (e.g. Baraban et al., 2005, Berghmans et al., 2007, Baxendale et al., 2012 and references therein).

- Related to the previous point, for the calcium imaging, the difference between an increase in fluorescence that the authors think reflects increased neuronal activity and the fluorescence that corresponds to seizures is not very clear. This detail is necessary because exactly when the term "seizure" describes a degree of increased activity can be difficult to distinguish objectively.

In our material and methods section, we describe our working definition of a seizure. Seizures are easily distinguished from increased activity by being synchronized.

- The supplementary movies that were added were very useful, but raised some questions. For example, what brain regions were pulsating? What areas seemed to constantly exhibit strong fluorescence and was this an artifact? It seemed that sometimes there was background fluorescence in the body. Perhaps an anatomical diagram could be provided for the readers. In addition, there were some movies with much greater fluorescence changes - are these the seizures? These are some reasons for our request for clarified definitions of the term "seizure".

The ”pulsating” (or “flickering”) brain activity is spontaneous neuronal activity. Some areas may appear to be more active, probably by a denser packing of neurons and intrinsically more spontaneous neuronal activity. However, since we only use normalized data, this does not affect our measurements.

- While it is not critical to change, I will again note the possible confusion that the use of the word "sedative" in this context may cause. However, I do understand this is a stylistic choice.

- Supplementary Figure 1B: the N values along the x-axis appear to have been duplicated and the duplications are offset and overlapping with one another by mistake.

Thank you for pointing this out. We have corrected the figure accordingly.

Review 3:

(1) Although the relationship between galanin and brain activity during interictal or seizure-free periods was clear, the revised manuscript still lacks mechanistic insight in the role of galanin during seizure-like activity induced by PTZ.

We agree that the mechanistic role of galanin still needs to be defined. The role is more complex that we expected, mainly due to its negative feedback properties. A complete mechanistic understanding will require a number of additional studies and is unfortunately outside of the scope of this manuscript.

(2) The revised manuscript continues to heavily rely on calcium imaging of different mutant lines. Confirmation of knockouts has been provided with immunostaining in a new supplementary figure. Additional methods could strengthen the data, translational relevance, and interpretation (e.g., acute pharmacology using galanin agonists or antagonists, brain or cell recordings, biochemistry, etc).

Cell recordings and biochemistry is challenging in the small larval zebrafish brain. We deem the genetic manipulations that we describe to be more informative than pharmacological experiments due to specificity issues.

eLife Assessment

The authors investigated KLF Transcription Factor 16 (KLF16) as an inhibitor of osteogenic differentiation, which plays a critical role in bone development, metabolism and repair. The results of the study are valuable as they could help to facilitate future research on the regulation of osteogenesis in vitro and in vivo. However, the evidence overall is incomplete, as validation by knockout mouse models would help to strengthen the conclusions.

Reviewer #1 (Public review):

Summary:

In this manuscript, Ru and colleagues investigated regulatory gene interactions during osteogenic differentiation. By profiling transcriptomic changes during mesenchymal stem cell differentiation, they identified KLF16 as a key transcription factor that inhibits osteogenic differentiation and mineralization. It was found that overexpression of KLF16 suppressed osteogenesis in vitro, while KLF16⁺/⁻ mice exhibited enhanced bone density, underscoring its regulatory role in bone formation.

Strengths:

(1) Bioinformatics is strong and comprehensive.

(2) Identification of KLF16 in osteoblast differentiation is exciting and innovative.

Weaknesses:

(1) The mechanism of KLF16 function is not studied.

(2) Studies of KLF16 in bone development, from both in vitro and in vivo perspectives, are descriptive.

(3) Findings in bioinformatics analysis are mostly redundant with previous studies in the field, and can be simplified.

Reviewer #2 (Public review):

In their manuscript with the title "Integrated transcriptomic analysis of human induced pluripotent stem cell (iPSC)-derived osteogenic differentiation reveals a regulatory role of KLF16", Ru et al. have analyzed the gene expression changes during the osteogenic differentiation of iPSC-derived mesenchymal stem/stromal cells into preosteoblasts and osteoblasts. As part of the computational analyses, they have investigated the transcription factor regulatory network mediating this differentiation process, which has also led to the identification of the transcription factor KLF16. Overexpression experiments in vitro and the analysis of heterozygous KLF16 knockout mice in vivo indicate that KLF16 is an inhibitor of osteogenic differentiation.

The integrated analysis of iPSC bulk transcriptomic data is a major strength of the study, and it is also great that the authors provide deeper functional characterization of the transcription factor KLF16, one of the newly identified candidate regulators of osteogenic differentiation.

However, characterization of KLF16 expression in the mouse and validation of the knockout model are currently lacking. Alternative explanations for the mutant phenotype should be considered to improve the strength of the conclusions.

If all issues can be addressed, the study would provide an important resource for the field that would facilitate future research on the regulation of osteogenesis in vitro and in vivo, with potential implications for preclinical and clinical research as well as bioengineering.

Author response

eLife Assessment

The authors investigated KLF Transcription Factor 16 (KLF16) as an inhibitor of osteogenic differentiation, which plays a critical role in bone development, metabolism and repair. The results of the study are valuable as they could help to facilitate future research on the regulation of osteogenesis in vitro and in vivo. However, the evidence overall is incomplete, as validation by knockout mouse models would help to strengthen the conclusions.

We appreciate the editors’ evaluation and recognition of the importance of our research. The primary goal and value of our study is to provide robust bioinformatics analyses of 20 independent iPSC lines, which can lead to the identification of novel genes involved in osteogenic differentiation. The identification of KLF16 serves to illustrate this goal. A thorough analysis of the function of any single gene both in vitro and in vivo is beyond the initial scope of this study. To validate KLF16’s inhibitory role in osteogenic differentiation, we provided evidence showing overexpression of Klf16 suppressed osteogenic differentiation in vitro, and Klf16^+/- mice exhibited enhanced bone mineral content and density in vivo.

Public Reviews:

Reviewer #1 (Public review):

Summary:

In this manuscript, Ru and colleagues investigated regulatory gene interactions during osteogenic differentiation. By profiling transcriptomic changes during mesenchymal stem cell differentiation, they identified KLF16 as a key transcription factor that inhibits osteogenic differentiation and mineralization. It was found that overexpression of KLF16 suppressed osteogenesis in vitro, while Klf16^+/- mice exhibited enhanced bone density, underscoring its regulatory role in bone formation.

Strengths:

(1) Bioinformatics is strong and comprehensive.

(2) Identification of KLF16 in osteoblast differentiation is exciting and innovative.

We appreciate the reviewer’s comments on our bioinformatic analyses of MSC osteogenic differentiation and the identification of KLF16 as a new osteogenesis regulator. The differentiation of iPSC-derived MSCs to OBs serves as a valuable model for investigating gene expression and regulatory networks in osteogenic differentiation. This study provides insights into the complex and dynamic regulation of the transcriptomic landscape in osteogenic differentiation and supplies a foundational resource for additional investigation into normal bone formation and the mechanisms underlying pathological conditions.

Weaknesses:

(1) The mechanism of KLF16 function is not studied.

(2) Studies of KLF16 in bone development, from both in vitro and in vivo perspectives, are descriptive.

Our study aims to apply rigorous bioinformatic analyses of 20 iPSC lines to identify novel genes involved in osteogenic differentiation. With this strategy, we successfully identified KLF16 as a regulator of osteogenic differentiation. We validated this with both in vitro and in vivo models even though we had limited availability of Klf16 knockout mice when the study was conducted. We demonstrated that overexpression of Klf16 suppressed osteogenesis in vitro, while Klf16^+/- mice exhibited increased bone mineral density, trabecular number, and cortical bone area, highlighting its role in bone formation. With these mice now available, further investigation into the mechanism of KLF16's function is possible.

(3) Findings in bioinformatics analysis are mostly redundant with previous studies in the field, and can be simplified.

We compared our bulk RNA-seq data with our previously published single-cell RNA-seq (scRNA-seq) data generated from iPSC-induced cells during osteogenic differentiation (Housman et al., 2022). The purpose is to corroborate the expression patterns of the genes we focused on during osteogenic differentiation. We found similar differential expression patterns in a pseudobulk analysis of the scRNA-seq data, even though there are significant differences between these two studies, including: cell culture conditions, sequencing approaches (bulk vs. single cell), goals of the studies (key TF drivers of osteoblast differentiation vs. mapping differentiation stages and inter-species gene programs in human and chimp), and findings (identification of TFs vs. identification of interspecific regulatory differences) .

Importantly, we performed network analyses to identify key transcription factors, which were not redundant with previous studies. We constructed a transcription factor regulatory network analysis during human osteogenic differentiation, and identified a network organized into five interactive modules. The most exciting finding was the identification of KLF16 as one of the strongest regulators in Module 5 (Figure 3), which previously was not demonstrated to be involved in bone formation. We also demonstrated known TF genes regulating osteogenic differentiation in these modules, and performed gene ontology (GO) and reactome pathway (RP) analyses for regulatory functions and pathways specific to each module. To clarify that our findings do not overlap with previous studies, we will revise the manuscript focusing on Module 5 and simplify the description of the bioinformatics analysis as the reviewer suggested.

Reviewer #2 (Public review):

In their manuscript with the title "Integrated transcriptomic analysis of human induced pluripotent stem cell (iPSC)-derived osteogenic differentiation reveals a regulatory role of KLF16", Ru et al. have analyzed the gene expression changes during the osteogenic differentiation of iPSC-derived mesenchymal stem/stromal cells into preosteoblasts and osteoblasts. As part of the computational analyses, they have investigated the transcription factor regulatory network mediating this differentiation process, which has also led to the identification of the transcription factor KLF16. Overexpression experiments in vitro and the analysis of heterozygous KLF16 knockout mice in vivo indicate that KLF16 is an inhibitor of osteogenic differentiation.

The integrated analysis of iPSC bulk transcriptomic data is a major strength of the study, and it is also great that the authors provide deeper functional characterization of the transcription factor KLF16, one of the newly identified candidate regulators of osteogenic differentiation.

We appreciate the reviewer’s summary and comments on the strength of our bioinformatic analyses of iPSC/MSC osteogenic differentiation and the deep functional characterization of the KLF16, as well as the novelty of our findings.

However, characterization of KLF16 expression in the mouse and validation of the knockout model are currently lacking. Alternative explanations for the mutant phenotype should be considered to improve the strength of the conclusions.

If all issues can be addressed, the study would provide an important resource for the field that would facilitate future research on the regulation of osteogenesis in vitro and in vivo, with potential implications for preclinical and clinical research as well as bioengineering.

We appreciate the reviewer’s valuable suggestions. Klf16 is highly expressed in mandibular, maxillary and tail mesenchyme at embryonic Day 12 (D'Souza et al., 2002), indicating its role in early bone development. We will further characterize the expression of Klf16 in mice, especially in the developing bones.

We identified Klf16 as a potential regulator of osteogenic differentiation, and then validated this with both in vitro and in vivo models. Overexpression of Klf16 suppressed osteogenesis in vitro, and Klf16^+/- mice showed increased bone mineral content and density, indicating its regulatory role in bone formation. We agree with the reviewer that the bone phenotypes of Klf16 knockout mice potentially can be affected by other factors in addition to osteogenic differentiation. As both bone formation and resorption are critical for bone development, we evaluated osteoclastogenesis in the Klf16^+/- mice by analyzing the expression of osteoclast marker CALCR and regulator RANKL in the femurs of the Klf16^+/- mice. Neither CALCR nor RANKL decreased in the bone of Klf16^+/- mice, indicating that osteoclastogenesis is not decreased; therefore, increased bone mineral content and density in the mutant mice is more likely attributed to enhanced bone formation rather than reduced resorption by osteoclasts. Additionally, we will discuss other alternative explanations for the bone phenotypes of Klf16 knockout mice as suggested by the reviewer.

References

D'Souza, U. M., Lammers, C.-H., Hwang, C. K., Yajima, S. and Mouradian, M. M. (2002). Developmental expression of the zinc finger transcription factor DRRF (dopamine receptor regulating factor). Mechanisms of Development 110, 197-201.

Housman, G., Briscoe, E. and Gilad, Y. (2022). Evolutionary insights into primate skeletal gene regulation using a comparative cell culture model. PLOS Genetics 18, e1010073-e1010073.

eLife Assessment

This useful study presents a virtual reality-based contextual fear conditioning paradigm for head-fixed mice. Solid evidence supports the claim that the reported methods provide a reliable paradigm for studying contextual fear conditioning in head-fixed mice. The approach provides a way to perform multiphoton imaging of neural circuits, and other techniques that are typically performed in head-fixed animals, during behaviors that have traditionally been studied in freely moving animals.

Reviewer #1 (Public review):

The authors have developed a contextual fear learning (CFC) paradigm in head-fixed mice that produces freezing as the conditioned response. Typically, lick suppression is the conditioned response in such designs, but this 1) introduces a potential confounding influence of reward learning on neural assessments of aversion learning and 2) does not easily allow comparison of head-fixed studies with extensive previous work in freely moving animals, which use freezing as the primary conditioned response. This report describes 3 versions of this virtual reality CFC paradigm, its validation using place-cell remapping, and provides suggestions for further refinement and application.

The first part of this study is a report on the development and outcomes of 3 variations of the CFC paradigm in a virtual reality environment. The fundamental design is strong, with head-fixed mice required to run down a linear virtual track to obtain a water reward. Once trained, the water reward is no longer necessary and mice will navigate virtual reality environments. There are rigorous performance criteria to ensure that mice that make it to the experimental stage show very low levels of inactivity prior to fear conditioning. These criteria do result in only 40% of the mice making it to the experimental stage, but high rates of activity in the VR environment is crucial for detecting learning-related freezing. It is possible that further adjustments to the procedure could improve attrition rates.

Paradigm versions 1 and 2 vary the familiarity of the control context while paradigm versions 2 and 3 vary the inter-shock interval. Version 1 is the most promising, showing the greatest increase in conditioned freezing (~40%) and good discrimination between contexts (delta ~15-20%). Version 2 showed no clear evidence of learning - average freezing at recall day 1 was not different than pre-shock freezing. First lap freezing showed a difference, but this single lap effect is not useful for many of the neural circuit questions for which this paradigm is meant to facilitate. Version 3 produces greater freezing and slower extinction than version 2. While the magnitude of the context discrimination is less than that in version 1, further optimization of the VR CFC is likely to produce robust learning and extinction. The authors discuss several options for further optimization.

The second part of the study is a validation of the head-fixed CFC VR protocol through demonstration that fear conditioning leads to remapping of dorsal CA1 place fields, similar to that observed in freely moving subjects. The results support this aim and largely replicate previous findings in freely moving subjects. One difference from previous work of note is that VR CFC led to remapping of the control environment, not just the conditioning context. The authors present several possible explanations for this lack of specificity to the shock context. While this experiment examined place cell remapping after fear conditioning, it did not attempt to link neural activity to the learned association or freezing behavior.

In summary, this is an important methodological innovation and this study sets the initial parameters and neuronal validation needed to further optimize a head-fixed CFC paradigm that produces freezing. In the discussion, the authors note the limitations of this study, suggest next steps in refinement, and point to several future directions using this protocol to significantly advance our understanding of the neural circuits of threat-related learning and behavior.

Comments on revisions:

The manuscript is much stronger with the additions and revisions the authors provided in their revised submission.

Reviewer #2 (Public review):

Summary:

In this manuscript, Krishnan et al devised three paradigms to perform contextual fear conditioning in head-fixed mice. Each of the paradigms relied on head-fixed mice running on a treadmill through virtual reality arenas. The authors tested the validity of three versions of the paradigms by using various parameters. The authors have addressed some of my initial concerns in their revised manuscript.

Strengths:

The authors have devised three new contextual fear conditioning paradigms in head-fixed mice. The authors tested a number of parameters towards optimization of this approach.

Weaknesses:

While some experimental parameters were tested in the manuscript, it appears that a large amount of additional testing and optimization will be required before reliable behavioral responses can be acquired and ultimately for the paradigm(s) to be useful for answering biological questions. One major factor will be optimizing parameters such that head-fixed mice in this paradigm can (largely) recapitulate what is observed in freely behaving mice. This may be challenging however, as they have previously published one of the three paradigms and the extensive additional testing they did in this current manuscript did not greatly improve the experimental setup. This may indicate limited immediate usefulness for the community as significant work likely remains for optimization.

Achievement of Aims:

The authors have put a significant amount of work in testing the paradigms, and as a result, progress has been made towards their usefulness in the field. However, a significant amount of optimization likely exists.

Impact on the field:

The development of a reliable paradigm for studying contextual fear in head-fixed animals would be a strong contribution to the field as it would enable sophisticated cell and circuit imaging analyses. This study is a good start towards this goal, but significant optimization is required for the paradigm(s) to fully benefit the field - especially to allow those who may have less experience in these approaches to use it in their own research.

Reviewer #3 (Public review):

Summary:

Krishnan et al. present a novel contextual fear conditioning (CFC) paradigm using a virtual reality (VR) apparatus to evaluate whether conditioned context-induced freezing can be elicited in head-fixed mice. By combining this approach with two-photon imaging, the authors aim to provide high-resolution insights into the neural mechanisms underlying learning, memory, and fear. Their experiments demonstrate that head-fixed mice can discriminate between threat and non-threat contexts, exhibit fear-related behavior in VR, and show context-dependent variability during extinction. Supplemental analyses further explore alternative behaviors and the influence of experimental parameters, while hippocampal neuron remapping is tracked throughout the experiments, showcasing the paradigm's potential for studying memory formation and extinction processes.

Strengths:

Methodological Innovation: The integration of a VR-based CFC paradigm with real-time two-photon imaging offers a powerful, high-resolution tool for investigating the neural circuits underlying fear, learning, and memory.

Versatility and Utility: The paradigm provides a controlled and reproducible environment for studying contextual fear learning, addressing challenges associated with freely moving paradigms.

Potential for Broader Applications: By demonstrating hippocampal neuron remapping during fear learning and extinction, the study highlights the paradigm's utility for exploring memory dynamics, providing a strong foundation for future studies in behavioral neuroscience.

Comprehensive Data Presentation: The inclusion of supplemental figures and behavioral analyses (e.g., licking behaviors and variability in extinction) strengthens the manuscript by addressing additional dimensions of the experimental outcomes.

Weaknesses:

Optimization: many parameters remain to be tested in the VR fear conditioning paradigm.

Extended training and attrition rate: the paradigm requires weeks of training and only 40% of mice reach criteria.

Author response:

The following is the authors’ response to the original reviews

We thank all the reviewers for their time and valuable feedback, which helped us improve our manuscript. Based on the comments, we have made several critical changes to the revised manuscript.

(1) We have changed our threshold for detecting freezing epochs from 1 cm/s to 0 cm/s in this revised manuscript. This change allows us to capture periods when animals are completely still on the treadmill, better matching the "true freezing" behavior seen in freely moving set-ups. We have added a new supplementary video (Supplementary Video 2) that better demonstrates the freezing response we observe. All results and figures in the revised manuscript reflect this updated threshold (Figure 2-6, Supplementary Figures 16, Tables 1-6). Our main findings remain robust, demonstrating that freezing serves as a reliable conditioned response in our paradigms, comparable to freely moving animals. Specifically, freezing behavior increased reliably in the fear-conditioned environment following CFC across all paradigms. We have also added data from a no-shock control group (Supplementary Figure 2) which, when compared to the conditioned group, shows that freezing responses in the conditioned group result from fear conditioning rather than immobility. We do observe other avoidance behaviors unique to our treadmill-based task— such as hesitation, backward movement, and slow crawls. These conditioned behaviors are captured through a separate metric: the time taken to complete a lap.

(2) As suggested by the reviewers, we have separately analyzed fear discrimination and extinction dynamics across recall days (Supplementary Figures 2, 5 and 6, Table 1-6). To assess fear discrimination, we use within-group comparisons to evaluate how well animals differentiate between the two VRs across days. For extinction, we use within-VR comparisons to examine freezing dynamics over time. Freezing across recall days is compared to baseline freezing (pre-conditioning) using a Linear Mixed Effects model (Tables 1-6), with recall days as fixed effects and mouse as a random effect, using baseline freezing as the reference.

(3) We have expanded the behavioral dataset in Paradigm 1 to investigate the effect of shock amplitude on the conditioned fear response (Supplementary Figure 2 C-E). Consistent with findings in freely moving animals, our data show that increasing shock intensity from 0.6 mA to 1.0 mA leads to stronger freezing. For the revised manuscript, we specifically increased the sample size in the 0.6 mA group (n = 8) in Paradigm 1, as this intensity is used in Paradigm 3. These additional data demonstrate that combining a lower shock amplitude with shorter inter-shock intervals and retaining the tail-coat during recall can enhance freezing, suggesting that these parameters help compensate for lower shock intensity.

(4) We have added more sample sizes to the imaging dataset (now n = 8, Figures 7-8).

Finally, we acknowledge that many aspects of this paradigm still require optimization. The headfixed CFC paradigm is in its early stages compared to the decades of research dedicated to understanding fear learning parameters in freely moving CFC paradigms. While there are numerous parameters that could be tested—both those identified through our own discussions and those raised by the reviewers—it is not feasible for a single lab to conduct a full evaluation of all the possible factors that could influence CFC in the head-fixed prep. A key limitation is that our approach requires robust navigation behavior in the VR without rewards, which requires weeks of training per mouse. It also necessitates larger sample sizes at the outset as not all animals will make it through our behavioral criteria required for CFC. Another important consideration is scalability. Unlike freely moving CFC paradigms, which allow parallel testing of many animals with minimal pre-training, the VR-CFC setup requires several weeks of behavior training and involves a more complex integration of hardware and software to accurately track behavior in virtual space. The number of VR rigs that can be operated simultaneously in a single lab is often limited, making high-throughput testing more challenging. These factors mean that the testing of a single parameter in a group of animals requires approximately 3–4 months to complete. Despite these constraints, we are committed to continue refining this paradigm over time. With this manuscript, our main aim was to provide a detailed framework, initial parameters, and evidence for conditioned behavior in the head-fixed preparation. By doing so, we hope to facilitate the adoption of this paradigm by researchers interested in studying the neural correlates of learning and memory using multiphoton imaging and stimulation techniques. This approach enables investigations that are not possible in freely moving animals, while the presence of freezing as a conditioned response allows for direct comparisons to the extensive body of work done in freely moving paradigms. Moving forward, we anticipate that optimizing this paradigm and identifying the key parameters that drive learning will be a collaborative, community-led effort.

Public Reviews:

Reviewer #1 (Public review):

The authors set out to develop a contextual fear learning (CFC) paradigm in head-fixed mice that would produce freezing as the conditioned response. Typically, lick suppression is the conditioned response in such designs, but this (1) introduces a potential confounding influence of reward learning on neural assessments of aversion learning and (2) does not easily allow comparison of head-fixed studies with extensive previous work in freely moving animals, which use freezing as the primary conditioned response.

The first part of this study is a report on the development and outcomes of 3 variations of the CFC paradigm in a virtual reality environment. The fundamental design is strong, with headfixed mice required to run down a linear virtual track to obtain a water reward. Once trained, the water reward is no longer necessary and mice will navigate virtual reality environments. There are rigorous performance criteria to ensure that mice that make it to the experimental stage show very low levels of inactivity prior to fear conditioning. These criteria do result in only 40% of the mice making it to the experimental stage, but high rates of activity in the VR environment are crucial for detecting learning-related freezing. It is possible that further adjustments to the procedure could improve attrition rates.

We acknowledge that further adjustments to the procedure could improve attrition rates, and we will continue to work on improving the paradigm.

Paradigm versions 1 and 2 vary the familiarity of the control context while paradigm versions 2 and 3 vary the inter-shock interval. Paradigm version 1 is the most promising, showing the greatest increase in conditioned freezing (~40%) and good discrimination between contexts (delta ~15-20%). Paradigm version 2 showed no clear evidence of learning - average freezing at recall day 1 was not different than pre-shock freezing. First-lap freezing showed a difference, but this single-lap effect is not useful for many of the neural circuit questions for which this paradigm is meant to facilitate. Also, the claim that mice extinguished first-lap freezing after 1 day is weak. Extinction is determined here by the loss of context discrimination, but this was not strong to begin with. First-lap freezing does not appear to be different between Recall Day 1 and 2, but this analysis was not done.

This is an important point. Following reviewer suggestions, we have replotted our figures for all paradigms to show within-VR freezing (see Supplementary Figures 2, 5 and 6) as the appropriate method for quantifying fear extinction across days. Using an LME model (Tables 16), we quantify freezing during recall days against baseline freezing levels measured before fear conditioning within each VR. In Paradigm 2, while some fear discrimination persists across days, extinction does occur rapidly. After the first lap in the CFC VR, we observed no significant differences in freezing compared to the baseline. These results are shown in the revised Supplementary Figure 5, and the revised text is in lines 393-399.

Paradigm version 3 has some promise, but the magnitude of the context discrimination is modest (~10% difference in freezing). Thus, further optimization of the VR CFC will be needed to achieve robust learning and extinction. This could include factors not thoroughly tested in this study, including context pre-exposure timing and duration and shock intensity and frequency.

We acknowledge that many aspects of this paradigm still need optimization, as virtual reality CFC is in its early stages, and we have not explored all of the parameter space. We describe above the reasoning for this. However, for this revised version of the paper we have added new behavioral data (Supplementary Figure 2 C-E) showing that increasing shock intensities from 0.6 mA to 1 mA enhances freezing, both in the first lap and on average. There are of course many other parameters that are likely important, like the ones pointed out here by the reviewer, but exploring the entire parameter space will take many years and will likely require many labs. The purpose of this paper is to show that VR-CFC fundamentally works and is a starting point from which the field can build on. We have now pointed out in the introduction (lines 54-58) and discussion (lines 730-737, 810-814) that there remains significant scope for improving this paradigm and optimizing parameters in the future.

The second part of the study is a validation of the head-fixed CFC VR protocol through the demonstration that fear conditioning leads to the remapping of dorsal CA1 place fields, similar to that observed in freely moving subjects. The results support this aim and largely replicate previous findings in freely moving subjects. One difference from previous work of note is that VR CFC led to the remapping of the control environment, not just the conditioning context. The authors present several possible explanations for this lack of specificity to the shock context, further underscoring the need for further refinement of the CFC protocol before it can be widely applied. While this experiment examined place cell remapping after fear conditioning, it did not attempt to link neural activity to the learned association or freezing behavior.

This is an interesting observation. We think that the remapping observed in the control context likely occurred due to the absence of reward in a previously rewarded environment. Our prior work has demonstrated that removal of reward causes increased remapping (Krishnan et al., 2022, Krishnan and Sheffield, 2023). In other words, the continued presence of reward within an environment stabilizes CA1 place fields. The Moita et al. (2004) paper, which showed remapping only in the fear conditioned context and not in the control context, provided rats with food pellets throughout the experimental session in both the control and conditioned context— likely to increase exploration necessary for identifying place cells. The presence of reward in the Moita et al experiment could explain the minimal remapping observed in their control context compared to our control context which lacked reward. Another possibility could lie in the differences in the intervals between place cell activity recordings in our study and that of Moita et al. While Moita et al. separated their recordings by just one hour, our recordings were separated by a full day, with a sleep period in between. The absence of sleep and the shorter time interval between conditioning and retrieval sessions in their study could explain the minimal remapping observed by Moita et al. compared to our findings. We have now addressed this discrepancy explicitly in lines 596-606.

Although we agree with the reviewer that it would be informative to perform analysis of how neural activity correlates with freezing responses, we think this warrants its own stand-alone manuscript as the neural dynamics and methods to appropriately analyze them are complicated. We are in the midst of analyzing this data further and will present these findings in a separate publication.

In summary, this is an important study that sets the initial parameters and neuronal validation needed to establish a head-fixed CFC paradigm that produces freezing behaviors. In the discussion, the authors note the limitations of this study, suggest the next steps in refinement, and point to several future directions using this protocol to significantly advance our understanding of the neural circuits of threat-related learning and behavior.

Reviewer #2 (Public review):

Summary:

In this manuscript, Krishnan et al devised three paradigms to perform contextual fear conditioning in head-fixed mice. Each of the paradigms relied on head-fixed mice running on a treadmill through virtual reality arenas. The authors tested the validity of three versions of the paradigms by using various parameters. As described below, I think there are several issues with the way the paradigms are designed and how the data are interpreted. Moreover, as Paradigm 3 was published previously in a study by the same group, it is unclear to me what this manuscript offers beyond the validations of parameters used for the previous publication. Below, I list my concerns point-by-point, which I believe need to be addressed to strengthen the manuscript.

Major comments

(1) In the analysis using the LME model (Tables 1 and 2), I am left wondering why the mice had increased freezing across recall days as well as increased generalization (increased freezing to the familiar context, where shock was never delivered). Would the authors expect freezing to decrease across recall days, since repeated exposure to the shock context should drive some extinction? This is complicated by the analysis showing that freeing was increased only on retrieval day 1 when analyzing data from the first lap only. Since reward (e.g., motivation to run) is removed during the conditioning and retrieval tests, I wonder if what the authors are observing is related to decreased motivation to perform the task (mice will just sit, immobile, not necessarily freezing per se). I think that these aspects need to be teased out.

This is an important point and we agree teasing out a lack of motivation versus fearful freezing would be useful. To address the possibility that reduced motivation to run without reward could contribute to the observed freezing behavior, we have now included a no-shock control group in the revised manuscript (n = 7; Supplementary Figure 2A-B, H–I). These control mice experienced the same protocol, including the wearing of a tail coat, but did not receive any shocks. We observed no increases in freezing across days in these controls, confirming that the increased freezing in the Familiar context of our experimental group stems from fear conditioning rather than the removal of reward from a previously rewarded context. If reduced motivation from reward removal were the primary driver, similar freezing patterns would have emerged in the no-shock controls. We have added lines 248-261 in the revised manuscript, discussing this point, and we thank the reviewer for motivating us to do this experiment and analysis.

That said, the precise mechanisms underlying the fear generalization observed in the nonconditioned context—particularly its emergence during later recall days—remain unclear. Studies in freely moving animals have shown that fear memories initially specific to the conditioned context can become generalized with repeated exposures, which may be occurring here (Biedenkapp & Rudy, 2007; Wiltgen & Silva, 2007). Alternatively, it is possible that the combination of fear conditioning and the removal of expected reward contributes to a delayed generalization effect. This may reflect a limitation of our approach, which relies on reward to motivate initial training. As noted by another reviewer, we have now addressed this potential drawback of reward-based training in the discussion (see lines 809-817). Clearly, unique factors specific to the head-fixed VR paradigm may contribute to this phenomenon. Understanding the mechanisms underlying fear generalization in the head-fixed VR CFC paradigm will be a valuable direction for future research.

(2) Related to point 1, the authors actually point out that these changes could be due to the loss of the water reward. So, in line 304, is it appropriate to call this freezing? I think it will be very important for the authors to exactly define and delineate what they consider as freezing in this task, versus mice just simply sitting around, immobile, and taking a break from performing the task when they realize there is no reward at the end.

As noted in point 1 above, we have added a no-shock control group (n = 7; Supplementary Figure 2A-B, H–I) to determine whether the observed freezing was driven by fear conditioning or by reduced motivation to run in the absence of reward. The absence of increased freezing in these controls supports the interpretation that the behavior in the conditioned group is fearrelated. In future studies, incorporating additional physiological measures—such as heart rate monitoring—could further help distinguish fear-related freezing from other forms of immobility.

(3) In the second paradigm, mice are exposed to both novel and (at the time before conditioning) neutral environments just before fear conditioning. There is a big chance that the mice are 'linking' the memories (Cai et al 2016) of the two contexts such that there is no difference in freezing in the shock context compared to the neutral context, which is what the authors observe (Lines 333-335). The experiment should be repeated such that exposure to the contexts does not occur on the conditioning day.

This is an interesting idea. However, if memory linking were driving the observed freezing patterns, we would expect to see similarly reduced fear discrimination across all three paradigms, as mice experience both contexts sequentially in each case. However, this effect appears to be specific to Paradigm 2, suggesting this may be due to other factors. We agree it would be informative to eliminate pre-conditioning exposure to both environments—to assess whether this improves fear discrimination and helps clarify the potential contribution of memory linking. This is something we plan to do in future studies that are beyond the scope of this initial paper on VR-CFC.

(4) On lines 360-361, the authors conclude that extinction happens rapidly, within the first lap of the VR trial. To my understanding, that would mean that extinction would happen within the first 5-10 seconds of the test (according to Figure S1E). That seems far too fast for extinction to occur, as this never occurs in freely behaving mice this quickly.

We agree with the reviewer that extinction in Paradigm 2 appears to occur relatively rapidly.

However, the average time to complete the first lap in the fear-conditioned context in Paradigm 2 is 25.68 ± 5.55 seconds (as stated in line 384), indicating that extinction occurs within approximately the first 30 seconds of context exposure—not within 5–10 seconds. This is specific to Paradigm 2 and does not happen in either of the other paradigms, as shown in Supplementary Figure 4. For clarification, Figure S1E pertains to baseline running in Paradigm 1 and does not apply to Paradigm 2.

As the reviewer points out, even at 30 seconds, extinction seems to be happening more quickly in Paradigm 2 than seen in freely moving setups. This may be due to a key structural difference in our setup. The VR-CFC task is organized into discrete trials, with mice being teleported back to the start after reaching the end of the virtual track. Completing a full lap without receiving a shock could serve as a clear signal that the threat is no longer present within the environment as the completion of a lap means that the animals have surveyed all locations within the environment. This structure could accelerate extinction compared to freely moving setups, where animals take longer to explore their complete environment due to the lack of discrete trials. Although this is true for all our paradigms, the accelerated extinction seen in paradigm 2 versus 1 and 3 may be driven by other factors. As noted by the reviewers, other task parameters—such as context pre-exposure timing, shock intensity, and conditioning duration— are likely to play a role in shaping extinction dynamics. These factors warrant further investigation, and we plan to explore them in future studies to better understand the conditions influencing extinction in the VR-CFC paradigm.

(5) Throughout the different paradigms, the authors are using different shock intensities. This can lead to differences in fear memory encoding as well as in levels of fear memory generalization. I don't think that comparisons can be made across the different paradigms as too many variables (including shock intensity - 0.5/0.6mA can be very different from 1.0 mA) are different. How can the authors pinpoint which works best? Indeed, they find Paradigm 3 'works' better than Paradigm 2 because mice discriminate better between the neutral and shock contexts. This can definitely be driven by decreased generalization from using a 0.6mA shock in Paradigm 3 compared to 1.0 mA shock in Paradigm 2.

The reviewer brings up important points here. We have now added new data evaluating 0.6 mA shocks in Paradigm 1 (Supplementary Figure 2A–E, n=8). These data show that 1.0 mA shocks produced stronger conditioned responses and greater fear discrimination compared to 0.6 mA. Our goal in Paradigm 3 was to begin with a lower shock intensity and assess whether additional modifications—specifically the shorter ISI and retention of the tail-coat during recall—could enhance fear conditioning. Surprisingly, despite the weaker shock intensity, Paradigm 3 resulted in improved discrimination and freezing behavior relative to Paradigm 2. We have now clarified this point in the manuscript (lines 466-470), and we interpret this outcome as evidence that the shorter ISIs and contextual cue continuity (tail-coat) likely play a more significant role in enhancing learning and recall. However, as noted in the text (lines 511-514), further testing is needed to determine the individual contributions of each parameter to successful VR-CFC. Fully optimizing the parameter settings will take additional time and resources, and we aim to continually refine the parameter space in the future, as has been done over the years for freely moving animals.

(6) There are some differences in the calcium imaging dataset compared to other studies, and the authors should perform additional testing to determine why. This will be integral to validating their head-fixed paradigm(s) and showing they are useful for modeling circuit dynamics/behaviors observed in freely behaving mice. Moreover, the sample size (number of mice) seems low.

The one notable difference between our imaging study and that done in freely moving animals is that we observed remapping of place cells in the control context. In contrast, Moita et al. (2004) reported more stable place fields in the control context. A key distinction is that their study included rewards in the control context, which may have contributed to the spatial stability. We now discuss this difference in the manuscript (lines 599-605).

It should be noted that there are many key distinctions among paradigms that study neural activity during fear conditioning in freely moving animals. These include varying exposure times to environments (1–6 days), the time interval between neural activity recordings, and the use of food rewards during the experiment stages in freely moving animals to encourage exploration for place cell identification. Although freely moving paradigms that investigate fear conditioning and place cells are heterogeneous, we were encouraged by the replication of several key findings. This validates VR-based CFC as a viable tool for neural circuit investigations. While future work will include more thorough analyses, our current findings demonstrate the paradigm's effectiveness for modeling circuit dynamics and behavior. We have now expanded our dataset, which includes four additional mice, further corroborating these original findings.

(7) It appears that the authors have already published a paper using Paradigm 3 (Ratigan et al 2023). If they already found a paradigm that is published and works, it is unclear to me what the current manuscript offers beyond that initial manuscript.

The reviewer is correct that we have published a paper using Paradigm 3. However, this manuscript goes beyond that one and provides a much more comprehensive description and fundamental analysis of the behavior and experimental parameters regarding VR-CFC, allowing the research community to adapt our paradigm reproducibly. While Ratigan et al. (2023) offered only a minimal description of behavior and included just Paradigm 3, we present two additional paradigms along with neuronal validation using hippocampal place cells. We have now explicitly stated this in the introduction (lines 50-55).

(8) As written, the manuscript is really difficult to follow with the averages and standard error reported throughout the text. This reporting in the text occurred heterogeneously throughout the text, as sometimes it was reported and other times it was not. Cleaning this reporting up throughout the paper would greatly improve the flow of the text and qualitative description of the results.

We completely agree with this point and have now cleaned up the text, leaving details only in a few places we felt were important.

Reviewer #3 (Public review):

Summary:

Krishnan et al. present a novel contextual fear conditioning (CFC) paradigm using a virtual reality (VR) apparatus to evaluate whether conditioned context-induced freezing can be elicited in head-fixed mice. By combining this approach with two-photon imaging, the authors aim to provide high-resolution insights into the neural mechanisms underlying learning, memory, and fear. Their experiments demonstrate that head-fixed mice can discriminate between threat and non-threat contexts, exhibit fear-related behavior in VR, and show context-dependent variability during extinction. Supplemental analyses further explore alternative behaviors and the influence of experimental parameters, while hippocampal neuron remapping is tracked throughout the experiments, showcasing the paradigm's potential for studying memory formation and extinction processes.

Strengths:

Methodological Innovation: The integration of a VR-based CFC paradigm with real-time twophoton imaging offers a powerful, high-resolution tool for investigating the neural circuits underlying fear, learning, and memory.

Versatility and Utility: The paradigm provides a controlled and reproducible environment for studying contextual fear learning, addressing challenges associated with freely moving paradigms.

Potential for Broader Applications: By demonstrating hippocampal neuron remapping during fear learning and extinction, the study highlights the paradigm's utility for exploring memory dynamics, providing a strong foundation for future studies in behavioral neuroscience.

Comprehensive Data Presentation: The inclusion of supplemental figures and behavioral analyses (e.g., licking behaviors and variability in extinction) strengthens the manuscript by addressing additional dimensions of the experimental outcomes.

Weaknesses:

Characterization of Freezing Behavior: The evidence supporting freezing behavior as the primary defensive response in VR is unclear. Supplementary videos suggest the observed behaviors may include avoidance-like actions (e.g., backing away or stopping locomotion) rather than true freezing. Additional physiological measurements, such as EMG or heart rate, are necessary to substantiate the claim that freezing is elicited in the paradigm.

To strengthen our claim that freezing is a conditioned response in this task, we have taken three key steps:

(1) We adjusted our freezing detection threshold from 1 cm/s to near 0 cm/s to capture only periods where the animal is virtually motionless on the treadmill. We validated this approach in Figure 2, particularly in the zoomed-in track position trace in Figure 2A, which clearly shows that the identified freezing epochs correspond to no change in track position. All analyses and figures have been updated to reflect this more stringent threshold.

(2) We have added a no-shock control group in the revised manuscript (n = 7; Supplementary Figure 2A-B, H–I) where mice experienced the same protocol, including wearing a tail-coat, but received no shocks. These mice showed no increases in freezing behavior, which further demonstrates that the increased freezing we observe is a result of fear conditioning.

(3) We have added a new supplementary video (Supplementary Video 2) that better illustrates the freezing behavior in our task.

That said, we fully agree with the reviewer that freezing is not the only defensive response observed. Other behaviors—such as hesitation, backward movement, and slowing down—also emerge that are unique to our treadmill-based paradigm. We chose to focus on freezing in this manuscript to align with convention in freely moving fear conditioning studies and to facilitate direct comparisons. We agree that additional physiological measurements (e.g., EMG or heart rate) would provide further validation and could help distinguish between different forms of defensive responses. We view this as an important future direction and plan to incorporate such measures in upcoming studies. We highlight this in the results section (lines 175-179, 262-268) and in the discussion (lines 739-750).

Analysis of Extinction: Extinction dynamics are only analyzed through between-group comparisons within each Recall day, without addressing within-group changes in behavior across days. Statistical comparisons within groups would provide a more robust demonstration of extinction processes.

This is an important distinction and we have now added figures (Supplementary Figures 2H-I, 5C-D, 6C-D) showing within-VR behavior across Recall days, along with statistical comparisons and a description of the extinction process based on these results.

Low Sample Sizes: Paradigm 1 includes conditions with very low sample sizes (N=1-3), limiting the reliability of statistical comparisons regarding the effects of shock number and intensity.

Increasing sample sizes or excluding data from mice that do not match the conditions used in Paradigms 2 and 3 would improve the rigor of the analysis.

While we included all conditions in Figure 2 for completeness, we have separated these conditions in Supplementary Figure 2 to ensure clarity. This allows researchers interested in this paradigm to see the approximate range of conditioned responses observed across different parameters. When comparing Paradigm 1 with Paradigms 2 and 3, we have only used data from 1mA, 6 shocks condition.

Potential Confound of Water Reward: The authors critique the use of reward in conjunction with fear conditioning in prior studies but do not fully address the potential confound introduced by using water reward during the training phase in their own paradigm.

We agree this is a point that needs discussion. We have now noted the limitation of using water rewards during training in the discussion section, particularly its effect on the animal’s motivation in the long term and on place cell activity (lines 814-820).

Recommendations for the authors

Reviewer #1 (Recommendations for the authors):

I suggest changing "3 paradigms" to "3 versions of a CFC paradigm," as the paradigm is fundamentally the same, but parameters were adjusted towards finding an optimal protocol.

We have changed this phrasing where applicable.

Figure S2: There appear to be different sets of shock parameters for different mice, most with an n of 1 or 2. This is not reliable for making a decision for optimal shock parameters and should not be discussed in that way until a full-powered comparison is completed. Also, the N adds up to 19, yet only 18 are described as being included in the study.

We thank the reviewer for this important point. We agree that the current study is not powered to definitively identify optimal parameter settings. We have been careful not to interpret it in that way in the text. Rather, we adopted a commonly used starting point from the freely moving literature—1 mA with six shocks—as our initial condition (lines 196-199). To provide context for others interested in pursuing this work, we have presented a range of conditioned responses from different parameter combinations to illustrate potential variability. In most cases, these data are intended for illustrative purposes only and are not meant to support firm conclusions. We agree that a systematic and fully powered investigation of each parameter would be highly valuable, and we plan to pursue this in future work (and hope other labs contribute to this goal, too), much like the iterative optimizations performed in freely moving paradigms over time.

We thank the reviewer for catching the sample size discrepancy and have now corrected it.

The number of animals for the no-shock condition should be included.

Thank you. We have now included this.

A possible explanation for the lower fear and poorer discrimination in versions 2 and 3 could be that 10 min pre-exposure to the CFC context on day -1 led to latent inhibition. Shorter (or eliminated) pre-exposure may improve outcomes.

We agree that the exposure time is a parameter that we should explore. We have highlighted this in the discussion (lines 729-736) as a parameter that is worth testing in the future.

For analysis of extinction, it is best to establish this within condition - is freezing to the CFC context significantly reduced compared with initial recall and similar to pre-training freezing? By using discrimination as your index of extinction, increases in control context freezing/inactivity can eliminate context discrimination without the conditioned response of freezing actually undergoing extinction.

This is a good point, and we have now included analysis and conclusions based on a within-VR comparison for the analysis of fear extinction (Supplementary Figures 2H-I, 5C-D, 6C-D).

Reviewer #3 (Recommendations for the authors):

Clarification of Treadmill Shape: The manuscript describes the treadmill as "spherical" throughout. However, based on representative images and videos, the treadmill appears cylindrical. This discrepancy should be clarified to ensure consistency between the text and visuals.

The reviewer is correct that the treadmill is cylindrical, and this was an error on our part. We have corrected it throughout.

Figure and Legend Labeling: To improve clarity, all figures and their legends should be explicitly labeled with the corresponding paradigm (1, 2, or 3) to facilitate interpretation.

We have now added a label on all figures that clarifies which Paradigm the figures are referring to. We have also explicitly added this to the figure legends.

Objective Language: Subjective language, such as "since we wanted animals to" (Line 850), should be revised to reflect an objective tone (e.g., "to allow animals to"). Similarly, phrases like "We believe" (Line 896) should be avoided to maintain an unbiased presentation.

We have removed subjective language from our text.

Placement of Future Directions: Speculations on future experimental plans, such as the use of sex as a biological variable (Lines 895-903), should be included in the Discussion section rather than the Methods. Additionally, remarks about the responsiveness of female mice to tail shocks should be moved to the main text for proper contextualization.

We have moved these lines as suggested by the reviewer.

eLife Assessment

This valuable study by Guo and colleagues reports the inhibitory activity of caffeic acid phenethyl ester (CAPE) against TcdB, a key toxin produced by Clostridioides difficile. C. difficile infections are a major public health concern, and this manuscript provides interesting data on toxin inhibition by CAPE, a potentially promising therapeutic alternative for this disease. The strength of the evidence to support the conclusions is solid, with some concerns about the moderate effects on the mouse infection model and direct binding assays of CAPE to the toxin.

Reviewer #1 (Public review):

Summary:

In this manuscript, Guo and colleagues used a cell rounding assay to screen a library of compounds for inhibition of TcdB, an important toxin produced by Clostridioides difficile. Caffeic acid and derivatives were identified as promising leads, and caffeic acid phenethyl ester (CAPE) was further investigated.

Strengths:

Considering the high morbidity rate associated with C. difficile infections (CDI), this manuscript presents valuable research in the investigation of novel therapeutics to combat this pressing issue. Given the rising antibiotic resistance in CDI, the significance of this work is particularly noteworthy. The authors employed a robust set of methods and confirmatory tests, which strengthen the validity of the findings. The explanations provided are clear, and the scientific rationale behind the results is well-articulated. The manuscript is extremely well written and organized. There is a clear flow in the description of the experiments performed. Also, the authors have investigated the effects of CAPE on TcdB in careful detail, and reported compelling evidence that this is a meaningful and potentially useful metabolite for further studies.

Weaknesses:

Although the authors have made changes to the manuscript to address some of my comments, many of the comments were not satisfactorily addressed. Many of the changes are still superficial, and some concerns still need to be addressed. Important details are still missing from the description of some experiments. Authors should carefully revise the manuscript to ascertain that all details that could affect interpretation of their results are presented clearly.

There is still very little discussion (none, really) in the manuscript about the fact that, because the authors observed a significant effect of CAPE on both bacterial growth and spore production, some of the phenotypes observed can no longer be attributed solely to toxin inhibition.

The details about mass spectrometry are still insufficient. It is still unclear whether metabolite identifications were always based on MS1 or MS2. Instead, several details that are really secondary were included. Authors should be unequivocally clear as to how metabolite identities were obtained. They should also indicate which mass spectrometer was used, and there should be a section in the Materials and Methods describing these experiments.

About the removal of carry-over compounds, the authors stated that ultrafiltration centrifugal partition was used. However, although the authors explained this in detail in their response to reviewers file, the details were omitted from the main text. Authors should clearly state in the manuscript text that "Due to the large molecular weight of TcdB, approximately 270 kDa, we selected a 100 kDa molecular weight cutoff ultrafiltration membrane. The centrifugation was performed at 4000 g for 5 min to eliminate the compounds that did not bind to TcdB."

These are important details which need to be included.

Reviewer #2 (Public review):

I appreciate the author's responses to my original review. This is a comprehensive analysis of CAPE on C. difficile activity. It seems like this compound effects all aspects of C. difficile, which could make it effective during infection but also make it difficult to understand the mechanism. Even considering the authors responses, I think it is critical for the authors to work on the conclusions regarding the infection model. There is some protection from disease by CAPE but some parameters are not substantially changed. For instance, weight loss is not significantly different in the C. difficile only group versus the C. difficile + CAPE group. Histology analysis still shows a substantial amount of pathology in the C. difficile + CAPE group. This should be discussed more thoroughly using precise language.

The authors did a good job addressing my concerns regarding the infection model by providing a more accurate descriptions in the Results section for histology. However, the weight loss improvement by CAPE does not look like a significant effect, although it is trending towards improvement. This should be more accurately described.

Another minor concern is that the current Abstract is overstating the amount of disease attenuation. I would replace "remarkably reduces the pathology" with "reduces some of the pathology"

Author response:

The following is the authors’ response to the previous reviews

Public Reviews:

Reviewer #1 (Public review):

Summary:

In this manuscript, Guo and colleagues used a cell rounding assay to screen a library of compounds for inhibition of TcdB, an important toxin produced by Clostridioides difficile. Caffeic acid and derivatives were identified as promising leads, and caffeic acid phenethyl ester (CAPE) was further investigated.

Strengths:

Considering the high morbidity rate associated with C. difficile infections (CDI), this manuscript presents valuable research in the investigation of novel therapeutics to combat this pressing issue. Given the rising antibiotic resistance in CDI, the significance of this work is particularly noteworthy. The authors employed a robust set of methods and confirmatory tests, which strengthen the validity of the findings. The explanations provided are clear, and the scientific rationale behind the results is well-articulated. The manuscript is extremely well written and organized. There is a clear flow in the description of the experiments performed. Also, the authors have investigated the effects of CAPE on TcdB in careful detail, and reported compelling evidence that this is a meaningful and potentially useful metabolite for further studies.

Weaknesses:

The authors have made some changes in the revised version. However, many of the changes were superficial, and some concerns still need to be addressed. Important details are still missing from the description of some experiments. Authors should carefully revise the manuscript to ascertain that all details that could affect interpretation of their results are presented clearly. For instance, authors still need to include details of how the metabolomics analyses were performed. Just stating that samples were "frozen for metabolomics analyses" is not enough. Was this mass-spec or NMR-based metabolomics. Assuming it was mass-spec, what kind? How was metabolite identity assigned, etc? These are important details, which need to be included. Even in cases where additional information was included, the authors did not discuss how the specific way in which certain experiments were performed could affect interpretation of their results. One example is the potential for compound carryover in their experiments. Another important one is the fact that CAPE affects bacterial growth and sporulation. Therefore, it is critical that authors acknowledge that they cannot discard the possibility that other factors besides compound interactions with the toxin are involved in their phenotypes. As stated previously, authors should also be careful when drawing conclusions from the analysis of microbiota composition data, and changes to the manuscript should be made to reflect this. Ascribing causality to correlational relationships is a recurring issue in the microbiome field. Again, I suggest authors carefully revise the manuscript and tone down some statements about the impact of CAPE treatment on the gut microbiota.

Thanks for your constructive suggestion. We have carefully revised the manuscript according to your suggestions.

Reviewer #2 (Public review):

I appreciate the author's responses to my original review. This is a comprehensive analysis of CAPE on C. difficile activity. It seems like this compound affects all aspects of C. difficile, which could make it effective during infection but also make it difficult to understand the mechanism. Even considering the authors responses, I think it is critical for the authors to work on the conclusions regarding the infection model. There is some protection from disease by CAPE but some parameters are not substantially changed. For instance, weight loss is not significantly different in the C. difficile only group versus the C. difficile + CAPE group. Histology analysis still shows a substantial amount of pathology in the C. difficile + CAPE group. This should be discussed more thoroughly using precise language.

Thanks for your constructive suggestion. We have carefully revised the manuscript according to your suggestions.

Reviewer #3 (Public review):

Summary:

The study is well written, and the results are solid and well demonstrated. It shows a field that can be explored for the treatment of CDI

Strengths:

Results are really good, and the CAPE shows a good and promising alternative for treating CDI.

Weaknesses:

Some references are too old or missing.

Comments on revisions:

I have read your study after comments made by all referees, and I noticed that all questions and suggestions addressed to the authors were answered and well explained. Some of the minor and major issues related to the article were also solved. I am satisfied with all the effort given by the authors to improve their manuscript.

Thanks again for your review.

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

The legend of Figure 3SB is incorrect. It should read "Growth curves of C. difficile BAA-1870 in the presence of varying concentrations of CAPE (0-64 µg/mL)". Also, there is something wrong with the symbols in this figure. I suspect what is happening is that the symbols for the concentrations of 32 and 64 µg/mL are superimposing, but this is a problem because the lower line looks like a closed circle, which is supposed to represent the condition where no CAPE was added. The authors should change the symbols to allow clear distinction between each of the conditions.

Thanks for your constructive suggestion. We have modified the panel and figure legend in Figure 3SB. The concentrations of 32 μg/mL and 64 μg/mL are quite similar, which makes it challenging to differentiate between the corresponding data points on the graph. To enhance clarity, we have utilized distinct colors to help distinguish these closely valued lines as effectively as possible.

Since the authors observed a significant effect of CAPE on both bacterial growth and spore production, their discussion and conclusions need to reflect the fact that the effects observed can no longer be attributed solely to toxin inhibition.

Thanks for your comments. We have modified the corresponding description according to your suggestions.

In lines 43-45, authors state that "CAPE treatment of C. difficile-challenged mice induces a remarkable increase in the diversity and composition of the gut microbiota (e.g., Bacteroides spp.)". It is still unclear to this reviewer why mention Bacteroides between parentheses. Does this mean that there was an increase in the abundance of Bacteroides? If that is the case this needs to be stated more clearly.

Thanks for your comments. Treatment with CAPE indeed significantly increased the abundance of Bacteroides spp. in the gut microbiota (Figure 7H-J). However, to avoid ambiguity in the abstract, we have chosen to delete the specific mention of Bacteroides spp. within the parentheses.

The modifications made to lines 132-135 still do not address my concern. Authors stated in the manuscript that "compounds that were not bound to TcdB were removed". But how was this done? This needs to be clearly explained in the manuscript. In the response to reviewers document, authors state that this was done through centrifugation. But given that the goal here is to separate excess of small molecule from a protein target, just stating that centrifugation was used is not enough. Did the authors use ultracentrifugation? What were the conditions employed. This is critical so that the reader can assess the degree of compound carryover that may have occurred. Also, authors need to clearly acknowledge the caveats of their experimental design by stating that they cannot rule out the contribution of compound carryover to their results.

Thanks for your comments. We employed ultrafiltration centrifugal partition to remove the unbound small molecule compounds. Due to the large molecular weight of TcdB, approximately 270 kDa, we selected a 100 kDa molecular weight cutoff ultrafiltration membrane. The centrifugation was performed at 4000 g for 5 min to eliminate the compounds that did not bind to TcdB. We have incorporated the relevant methods and discussed the potential impacts on the respective sections of the manuscript.

In line 142, authors added the molar concentration of caffeic acid, as requested. Although this helps, it is even more important that molar concentrations are added every time a compound concentration is mentioned. For instance, just 2 lines down there is another mention of a compound concentration. It would be informative if authors also added molar concentrations here and throughout the manuscript.

Thanks for your comments. In our initial test design, we have utilized the concentration unit of μg/mL. However, during the conversion to μM using the dilution method, some values do not result in neat, whole numbers. For instance, the conversion of 32 μg/mL of caffeic acid phenyl ethyl ester yields 112.55 μM, which appears somewhat irregular when expressed in this manner.

Line 277. For the sake of clarity, I would strongly suggest that authors use the term "control mice" instead of "model mice".

Thanks for your comments. We have modified “model mice” to “control mice” throughout the manuscript.

In line 302, the word taxa should not be capitalized. I capitalized it in my original comments simply to draw attention to it.

Thanks for your comments. We have modified this word.

In the section starting in line 318, authors still need to include details of how the metabolomics analyses were performed. Just stating that samples were "frozen for metabolomics analyses" is not enough. Was this mass-spec or NMR-based metabolomics. Assuming it was mass-spec, what kind? How was metabolite identity assigned? Etc, etc. These are important details, which need to be included.

Thanks for your comments. We have added some metabolomics methods in the corresponding section.

In line 338, the authors misunderstood my original comment. This sentence should read "...the final product of purine degradation, were markedly decreased in mice after...".

Thanks for your comments. We have modified this sentence.

Panels of figure 3 are still incorrectly labeled. The secondary structure predictions are shown in A and C, not A and B as is currently stated in the legend.

Thanks for your comments. We have modified the figure legend in Figure 3.

About Figure 5C, I think the authors for the clarification, but this explanation should be included in the figure legend.

Thanks for your comments. We have added the relevant information to the figure legend.

eLife Assessment

This manuscript provides an important biochemical analysis of p53 isoforms, highlighting their aggregation propensity, interaction with chaperones, and dominant-negative effects on p53 family members. The authors have substantially strengthened the original manuscript by incorporating new mass spectrometry data and clarifying isoform-specific oligomerization behavior. Although the use of high expression levels limits direct physiological interpretation, the work is carefully framed as an investigation of protein misfolding and stability. Overall, this study offers convincing insights into p53 isoform biophysics with broad implications for cancer biology.

Reviewer #1 (Public review):

Summary:

Brdar, Osterburg, Munick, et al. present an interesting cellular and biochemical investigation of different p53 isoforms. The authors investigate the impact of different isoforms on the in-vivo transcriptional activity, protein stability, induction of the stress response, and hetero-oligomerization with WT p53. The results are logically presented and clearly explained. Indeed, the large volume of data on different p53 isoforms will provide a rich resource for researchers in the field to begin to understand the biochemical effects of different truncations or sequence alterations.

Strengths:

The authors achieved their aims to better understand the impact/activity of different p53 is-forms, and their data well support their statements. Indeed, the major strengths of the paper lie in its comprehensive characterization of different p53 isoforms and the different assays that are measured. Notably, this includes p53 transcriptional activity, protein degradation, induction of the chaperone machinery, and hetero-oligomerization with wtp53. This will provide a valuable dataset where p53 researchers can evaluate the biological impact of different isoforms in different cell lines. The authors went to great lengths to control and test for the effect of (1) p53 expression level, (2) promotor type, and (3) cell type. I applaud their careful experiments in this regard.

Comments on revised version:

The authors have addressed all of my concerns convincingly, including with a new mass spectrometry experiment to quantify p53 peptides specifically.

eLife Assessment

This study presents an advance in efforts to use histone post-translational modification (PTM) data to model gene expression and to predict epigenetic editing activity. Such models are broadly useful to the research community, especially ones that can model and predict epigenetic editing activity, which is novel; additionally, the authors have nicely integrated datasets across cell types into their model. The work is mostly solid, but it would be strengthened by performing further comparisons to existing methods that predict gene expression from PTM data and from more comprehensive functional validation of model-predicted epigenome editing outcomes beyond dCas9-p300 based perturbations. This work will be of interest to the epigenetics and computational modeling communities.

Reviewer #1 (Public review):

Batra, Cabrera and Spence et al. present a model which integrates histone posttranslational modification (PTM) data across cell models to predict gene expression with the goal of using this model to better understand epigenetic editing. This gene expression prediction model approach is useful if a) it predicts gene expression in specific cell lines b) it predicts expression values rather than a rank or bin, c) if it helps us to better understand the biology of gene expression or d) it helps us to understand epigenome editing activity. Problematically for points a) and b) it is easier to directly measure gene expression than to measure multiple PTMs and so the real usefulness of this approach mostly relates to c) and d).

Other approaches have been published that use histone PTM to predict expression (e.g. PMID 27587684, 36588793). Is this model better in some way? No comparisons are made, although a claim is made that direct comparisons are difficult. I appreciate that the authors have not used the histone PTM data to predict gene expression levels of an "average cell" but rather that they are predicting expression within specific cell types or for unseen cell types. Approaches that predict expression levels are much more useful, whereas some previous approaches have only predicted expressed or not expressed or a rank order or bin-based ranking. The paper does not seem to have substantial novel insights into understanding the biology of gene expression.

The approach of using this model to predict epigenetic editor activity on transcription is interesting and to my knowledge novel although only examined in the context of a p300 editor. As the author point out the interpretation of the epigenetic editing data is convoluted by things like sgRNA activity scoring and to fully understand the results likely would require histone PTM profiling and maybe dCas9 ChIP-seq for each sgRNA which would be a substantial amount of work.

Furthermore from the model evaluation of H3K9me3 is seems the model is performing modestly for other forms of epigenetic or transcriptional editing- e.g. we know for the best studied transcriptional editor which is CRISPRi (dCas9-KRAB) that recruitment to a locus is associated with robust gene repression across the genome and is associated with H3K9me3 deposition by recruitment of KAP1/HP1/SETDB1 (PMID: 35688146, 31980609, 27980086, 26501517).

One concern overall with this approach is that dCas9-p300 has been observed to induce sgRNA independent off target H3K27Ac (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8349887/ see Figure S5D) which could convolute interpretation of this type of experiment for the model.

Comments on revisions: This resubmission adds a comparison to existing gene prediction methods, but add no new confirmation experiments with predicting epigenome editing efficiency and had only one minor text edit.

Reviewer #2 (Public review):

Summary:

The authors build a gene expression model based on histone post-translational modifications, and find that H3K27ac is correlated with gene expression. They compare to other gene prediction methods such as DeepChrome. They proceed to perturb H3K27ac at 13 gene promoters in two cell types, and measure gene expression changes to test their model.

Strengths:

The combination of multiple methods to model expression, along with utilizing 6 histone datasets in 13 cell types allowed the authors to build a model that correlates between 0.7-0.79 with gene expression.<br /> They compare three cells types to other prediction models, and this figure should be included in the main figures.<br /> They use dCas9-p300 fusions to perturb H3K27ac and monitor gene expression to test their model. Ranked correlations of the HEK293 data showed some support for the predictions after perturbation of H3K27ac.

Weaknesses:

The authors state in the latest submission that the primary use case of this work is related to predicting epigenome editing outcomes, not predicting gene expression from chromatin. However the first four figures all relate to gene expression prediction. The only main figure that shows epigenome editing prediction is panel 6E. If this authors wish to highlight the use case of this work they should redo figures, including moving panels from current supplemental figures to show this.

The perturbation of 5 genes in K562 with perturb-seq data shows a modest correlation of ~0.5 and is still only shown in supplemental figures, which is odd as this is the true test case of their model in my opinion. The authors are then left to speculate the reasons why the outcome of epigenome editing doesn't fit their predictions, which highlights the limited value in the current version of this method.<br /> As mentioned before, testing genes that were not expressed being most activated by dCas9-p300 weaken the correlations vs. looking at a broad range of different gene expression as the original model was trained on.

If the authors want this method to be used to predict outcomes of epigenome editing, expanding to dCas9-KRAB and other CRISPRa methods (SAM and VPR) would be useful. Those datasets are published and could be analyzed for this manuscript and show how the model holds up across cell types and epigenome editing methods.

The utility of this method as described here, to predict gRNA outcomes seems modest and limited. It is fairly trivial to test 10 or more gRNAs for a single gene to find the best one, and the authors show limited prediction and occasionally no benefit. For example, with CHD8 and CD79 the gRNA with the highest prediction had the lowest actual impact on gene expression of the gRNAs tested. For many other genes the gRNA's prediction and gene expression outcome show no correlation.

Author response:

The following is the authors’ response to the previous reviews

Reviewer #1 (Public Review):

Batra, Cabrera and Spence et al. present a model which integrates histone posttranslational modification (PTM) data across cell models to predict gene expression with the goal of using this model to better understand epigenetic editing. This gene expression prediction model approach is useful if a) it predicts gene expression in specific cell lines b) it predicts expression values rather than a rank or bin, c) if it helps us to better understand the biology of gene expression or d) it helps us to understand epigenome editing activity. Problematically for point a) and b) it is easier to directly measure gene expression than to measure multiple PTMs and so the real usefulness of this approach mostly relates to c) and d).

We appreciate this point from Reviewer #1 and the instructive comments and helpful feedback on our study. We designed our approach keeping in mind that the primary use case is to understand how epigenome editing would affect gene expression.

Other approaches have been published that use histone PTM to predict expression (e.g. PMID 27587684, 36588793). Is this model better in some way? No comparisons are made although a claim is made that direct comparisons are difficult. I appreciate that the authors have not used the histone PTM data to predict gene expression levels of an "average cell" but rather that they are predicting expression within specific cell types or for unseen cell types. Approaches that predict expression levels are much more useful whereas some previous approaches have only predicted expressed or not expressed or a rank order or bin-based ranking. The paper does not seem to have substantial novel insights into understanding the biology of gene expression.

We thank Reviewer #1 again for this insightful comment. We have included citations for a series of papers (PMIDs: 27587684, 30147283, 36588793) that performed gene expression prediction using histone PTM data. However, each of these methods performs classification of gene expression as opposed to predicting the actual gene expression value via regression. Additionally, the referenced studies all work with Roadmap Epigenomics read-depth data as opposed to p-values obtained from the ENCODE pipelines, making it difficult to make direct comparisons. We outline in the Discussion section that by creating a comprehensive dataset of epigenome editing outcomes, which include quantification of histone PTMs before and after in situ 1 perturbations, will improve our understanding of the effects of dCas9-p300 on gene expression and assist in the design of gRNAs for achieving fine-tuned control over gene expression levels. In this revised version of our study, we have also added new data (Figure 3 – figure supplement 3) to further benchmark our model against others.

The approach of using this model to predict epigenetic editor activity on transcription is interesting and to my knowledge novel although only examined in the context of a p300 editor. As the author point out the interpretation of the epigenetic editing data is convoluted by things like sgRNA activity scoring and to fully understand the results likely would require histone PTM profiling and maybe dCas9 ChIP-seq for each sgRNA which would be a substantial amount of work.

We agree with the Reviewer and view these experiments as important components of future studies.

Furthermore from the model evaluation of H3K9me3 is seems the model is performing modestly for other forms of epigenetic or transcriptional editing- e.g. we know for the best studied transcriptional editor which is CRISPRi (dCas9-KRAB) that recruitment to a locus is associated with robust gene repression across the genome and is associated with H3K9me3 deposition by recruitment of KAP1/HP1/SETDB1 (PMID: 35688146, 31980609, 27980086, 26501517).

This is an interesting point. We have included new data (Figure 4 – figure supplement 1), that quantifies how sensitive the trained gene expression model is to perturbations in H3K9me3. Indeed our data suggests that the model predictions are sensitive to perturbations in H3K9me3. For instance, there is a clear decrease and a gradual increase as the position where the perturbation is performed moves from upstream to downstream of the TSS. Additionally, the magnitude of the predicted fold-change is a function of how much the H3K9me3 is perturbed and hence the magnitude of change would be even higher if the perturbation magnitude is increased. However, this precise magnitude is hard to estimate In the absence of experimental perturbation data for H3K9me3. Leveraging our model in combination with KRAB-based CRISPRi is an exciting and important aspect of future studies.

One concern overall with this approach is that dCas9-p300 has been observed to induce sgRNA independent off target H3K27Ac (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8349887/ see Figure S5D) which could convolute interpretation of this type of experiment for the model.

This remains an excellent point and indeed, we and others have observed that dCas9-p300 can result in off-target H3K27ac levels (both increased and suppressed) across the genome. Our study focused on p300, because the molecule is one of the few known proteins that can catalyze H3K27ac in the human genome, and H3K27ac remains a proxy for active genomic regulatory elements. Nevertheless, any off target activity of dCas9-p300 could certainly convolute our analyses. We have included language to address this caveat in our discussion.

Reviewer #2 (Public review):

Summary:

The authors build a gene expression model based on histone post-translational modifications, and find that H3K27ac is correlated with gene expression. They proceed to perturb H3K27ac at 13 gene promoters in two cell types, and measure gene expression changes to test their model.

We remain appreciative of the constructive feedback and input from Reviewer #2 on our manuscript.

Strengths:

The combination of multiple methods to model expression, along with utilizing 6 histone datasets in 13 cell types allowed the authors to build a model that correlates between 0.7-0.79 with gene expression. They use dCas9-p300 fusions to perturb H3K27ac and monitor gene expression to test their model. Ranked correlations of the HEK293 data showed some support for the predictions after perturbation of H3K27ac.

Weaknesses:

The perturbation of 5 genes in K562 with perturb-seq data shows a modest correlation of ~0.5 and isn't included in the main figures. The authors are then left to speculate reasons why the outcome of epigenome editing doesn't fit their predictions, which highlights the limited value in the current version of this method.

We agree with the reviewer’s suggestion and highlight in our conclusion that generating epigenome editing data across a variety of cell types and across many genes will help uncover the underlying mechanisms of gene expression modulation.

As mentioned before, testing genes that were not expressed being most activated by dCas9-p300 weaken the correlations vs. looking at a broad range of different gene expression as the original model was trained on.

We appreciate this comment from Reviewer #2. We note that the data generated from this dCas9-p300 perturb-seq experiment used gRNAs from a pre-existing library published previously (PMID: 37034704). While this library enabled deeper interrogation of dCas9-p300 driven effects compared to our previous revision, the gRNAs in this library were designed against genes associated with haploinsufficiency in neuronal cell types, and which were generally lowly-expressed in K562 cells. Further, we restricted our analysis here to promoter-proximal gRNAs (as opposed to enhancer-targeted gRNAs in the library), focusing our scope even more so. Thus the genes ultimately used for analysis are enriched for low expression.

If the authors want this method to be used to predict outcomes of epigenome editing, expanding to dCas9-KRAB and other CRISPRa methods (SAM and VPR) would be useful. Those datasets are published and could be analyzed for this manuscript.

This is an exciting suggestion from Reviewer #2. We agree, and view this as a component of future work in this area.

The authors don't compare their method to other prediction methods.

In this revised version of our study, we have also added new data (Figure 3 – figure supplement 3) to further benchmark our model against others. These data demonstrate that our CNN model outperforms existing approaches across multiple cell types.

Recommendations for the authors:

Reviewer #2 (Recommendations for the authors):

Looking at the individual genes in K562 shows a random looking range of predictions and observed, with the exception of Bcl11A which is one of two genes in this set of 5 that are not expressed. I will repeat my earlier comment, that epigenome editing and CRISPRa methods generally show the most upregulation with the lowest expressed genes. I speculate that plotting endogenous expression vs. outcome (assuming using all gRNAs within a reasonable and similar distance to TSS) would produce a correlation of -0.5 or greater and be as useful as this method.

We agree, and believe that this demonstrates more work is needed in this emerging research area.

The methods describe Perturb-seq analysis but not the bench experiments.

We have added the bench methods related to our Perturb-seq experiments to our revised manuscript under the Experimental Methods section in the Appendix.

I don't understand why the authors can't compare to other methods as that is fairly standard in new prediction papers. I get that others used REMC vs. ENCODE, and were rank or binary based, but the authors could use REMC data and/or convert their data to ranked or binary and still compare. Lacking that it's hard to judge this manuscript.

We have added benchmarking against existing methods as Figure 3 – figure supplement 3.

eLife Assessment

This study, which includes additional experiments in response to the reviewer comments, presents valuable findings illustrating the role of PI3Kα in heterotopic ossification in FOP model mice. The methods, data, and analyses are solid and generally support the claims although as noted by one of the reviewers, there is no data demonstrating the effect of BYL79 on cell growth, and it remains unclear whether BYL79 also inhibits the Smad2/3 pathway. While this study provides new insights into the role of the PI3Kα pathway as a therapeutic target for FOP, questions about the mechanism of BYL79 still exist.

Reviewer #1 (Public review):

Summary:

In the present study, the authors examined the possibility of using phosphatidyl-inositol kinase 3-kinase alpha (PI3Ka) inhibitors for heterotopic ossification in fibrodysplasia ossificans progressiva. Administration of BYL719, a chemical inhibitor of PI3Ka, prevented heterotopic ossification in a mouse model of FOP that expressed a mutated ACVR1 receptor. Genetic ablation of PI3Ka also suppressed heterotopic ossification in mice. BYL719 blocked osteo/chondroprogenitor specification and reduced inflammatory responses by reducing the number of fibro-adipogenic progenitors (FAPs) and promoting muscle fibre regeneration in vivo. The authors claimed that inhibition of PI3Ka is a safe and effective therapeutic strategy for heterotopic ossification.

Strengths:

Taking together previous reports on the specificity of BY718 in PI3K, it was suggested that BYL719 inhibits heterotopic ossification by reducing FAPs and promoting muscle regeneration through the PI3K pathway in vivo.

Weaknesses:

In the original manuscript, there was the possibility that BYL719 inhibited heterotopic ossification through non-specific and toxic effects rather than the PI3k pathway.

However, the authors added new data and explanations in the revision to solve the possibility. The findings of the authors would be useful and would provide an additional direction to develop a therapeutic strategy for heterotopic ossification.

Reviewer #2 (Public review):

Summary:

Authors in this study previously reported that BYL719, an inhibitor of PI3Kα, suppressed heterotopic ossification in mice model of a human genetic disease, fibrodysplasia ossificans progressive, which is caused by the activation of mutant ACVR1/R206H by Activin A. The aim of this study is to identify the mechanism of BYL719 for the inhibition of heterotopic ossification. They found that BYL719 suppressed heterotopic ossification in two ways: one is to inhibit the specification of precursor cells for chondrogenic and osteogenic differentiation and the other is to suppress the activation of inflammatory cells.

Strengths:

This study is based on authors' previous reports and the experimental procedures including the animal model are established. In addition, to confirm the role of PI3Kα, authors used the conditional knock-out mice of the subunit of PI3Kα. They clearly demonstrated the evidence indicating that the targets of PI3Kα is not members of TGFBR by a newly established experimental method.

Weaknesses:

Overall, the presented data were closely related to those previously published by authors' group or others and there were very few new findings. The molecular mechanisms through which BYL719 inhibits HO remain unclear, even in the revised manuscript.<br /> Heterotopic ossification in mice model was not stable and inappropriate for the scientific evaluation.<br /> The method for chondrogenic differentiation was not appropriate, and the scientific evidence of successful differentiation was lacking.<br /> The design of gene expression profile comparison was not appropriate and failed to obtain the data for the main aim of this study.<br /> The experiments of inflammatory cells were performed cell lines without ACVR1/R206H mutation, and therefore the obtained data were not precisely related to the inflammation in FOP.

Comments on revisions:

In the R2 version, the authors performed additional experiments using mice with inducible human R206H ACVR1A. BM-MSCs isolated from these mice were used to investigate the effect of Activin-A. The results again suggested that BYL79 inhibited the chondrogenic differentiation of BM-MSCs. However, there are still no data demonstrating the effect of BYL79 on cell growth in these in vitro experiments. In Figures 7A-D, 10 μM BYL79 strongly inhibited the proliferation of inflammatory cells, suggesting that growth inhibition may have contributed to the results shown in Figure 5.

The main point of discussion concerns the significance of the comparisons made. The fundamental disagreement arises from the role of Activin-A in R206H cells and its effect on chondrogenic differentiation. The authors' rebuttal regarding my comments on the RNA-seq analyses should be reconsidered. The core issue lies in the interpretation of Activin-A's role in R206H cells and the distinction between chondrogenic differentiation and ossification.

A key feature of R206H mutant cells is that they respond to Activin-A by activating Smad1/5 signaling-comparable in quality to the signaling induced by BMP6 in WT cells. Another important point, as also acknowledged by the authors, is that Activin-A can transduce Smad2/3 signaling via its canonical receptor, ACVR1B. These dual signaling pathways synergistically contribute to chondrogenic differentiation in precursor cells such as FAPs. Several reports have demonstrated that the combined activation of TGF-β and BMP signaling promotes chondrogenesis more strongly than either pathway alone.

Since the PI3Kα inhibition effect on HO is already known, a critical question in this study is whether BYL79 also inhibits the Smad2/3 pathway. A straightforward experiment would be to compare WT cells treated with Activin-A alone versus Activin-A plus BYL79, and to perform GO term enrichment analyses related specifically to chondrogenic differentiation, not ossification. Additionally, comparing R206H cells treated with Activin-A/BYL79 and WT cells treated with BMP6/BYL79 could help identify gene sets inhibited by BYL79 via Smad2/3 signaling. If these comparisons reveal no specific effect on genes related to chondrogenesis, the effect of BYL79 may be limited to suppression of BMP-mediated osteogenesis. Unfortunately, the authors appear to show little interest in addressing this issue.

Regarding Figure 7, the authors' rebuttal should also be reconsidered. Since the R2 version employed FOP model mice, it would have been possible to evaluate the effects of BYL79 on inflammatory cells harboring the R206H mutation. This could have enabled a more precise assessment of BYL79's influence on inflammatory signaling. While the authors repeatedly claim that BYL79's effect is not specific to any particular ligand or the presence of the FOP mutation, the role of TGF-β signaling in the development of endochondral heterotopic ossification is well recognized. Therefore, the mechanism of BYL79 should be clarified before considering its therapeutic application

Author response:

The following is the authors’ response to the previous reviews

Our revised manuscript thoroughly addresses all comments and suggestions raised by the reviewers, as detailed in our point-by-point response. To strengthen our findings, we have conducted additional in vivo experiments to evaluate the presence of fibro-adipogenic progenitors (FAPs) at different time points during HO formation in control and BYL719-treated mice. Our results indicate that BYL719 reduces the accumulation of FAPs and promotes muscle fiber regeneration in vivo. We have also expanded our discussion on BYL719’s effects on mTOR signaling, further clarifying key points raised by Reviewer #1, and have addressed all minor comments.

Additionally, in response to Reviewer #2, we have employed an orthogonal and complementary approach using a new model. We conducted chondrogenic differentiation experiments with murine MSCs expressing either ACVR1wt or ACVR1^R206H. qPCR analysis of chondrogenic gene markers (Sox9, Acan, Col2a1) demonstrates that Activin A enhances their expression in ACVR1^R206H cells, whereas BYL719 strongly suppresses their expression, regardless of ACVR1 mutational status. These new data further confirm that BYL719 effectively inhibits genes involved in ossification and osteoblast differentiation, independent of the ACVR1 mutation. We have also expanded our discussion to further clarify points raised by Reviewer #2 and have addressed all remaining minor comments.

Below, we provide a detailed point-by-point response to the reviewers’ comments:

Rreviewer #1:

Point 1: In this revised manuscript, the authors clearly showed that BYL719 suppressed the proliferation and differentiation of murine myoblasts, C2C12 cells, in addition to human MSCs in vitro. Furthermore, BYL719 decreased migratory activity in vitro in monocytes and macrophages without suppressing proliferation. Overall, these data suggested that BYL719 is not a specific chemical compound for cell types or signaling pathways as mentioned in the manuscript by the authors themselves. Therefore, it was still unclear how to explain the molecular mechanisms in inhibition of HO by the compound in a specific signaling pathway in a specific cell type, MSCs, contradicting many other possibilities. The authors should add logical explanations in the manuscript.

Regarding its selectivity, BYL719 is a potent and highly selective inhibitor of PI3Kα. It has been demonstrated in multiple studies and in several in vitro kinase assay panels (Furet et al. PMID: 23726034, Fritsch et al. PMID: 24608574). The IC50 or Kd values for BYL719 against PI3Kα were at least 50 times lower than for most of other kinases tested. Moreover, BYL719 is also highly selective for PI3Kα (IC50 = 4.6 nmol/L) compared to other class I PI3K (PI3Kβ (IC50 = 1,156 nmol/L), PI3Kδ (IC50 = 290 nmol/L), PI3Kγ (IC50 = 250 nmol/L)) (Fritsch et al). Consistent with these data, we show that, at the concentrations tested, BYL719 does not have a direct effect on any kinase receptor within the TGF-b superfamily, including ACVR1 or ACVR1^R206H.

Rather than blocking ACVR1 kinase activity, in our manuscript we provide evidence that BYL719 has the potential to inhibit osteochondroprogenitor specification and prevent an exacerbated inflammatory response in vivo (Valer et al., 2019a PMID: 31373426, and this manuscript) through different mechanisms, such as (i) increasing SMAD1/5 degradation, (ii) reducing transcriptional responsiveness to BMPs and Activin, (iii) blocking non-canonical ACVR1 responses such as the activation of AKT/mTOR. All these defined molecular mechanisms contribute to suppress HO in vitro and in vivo, as we report and explain throughout the manuscript. Selective PI3Kα inhibition is at the core of the different molecular pathways described. As such, PI3Kα blockade inhibits the phosphorylation of GSK3 and compromises SMAD1 protein stability, thereby altering canonical responsiveness and osteochondroprogenitor specification (Gamez et al PMID: 26896753; Valer et al PMID: 31373426). Moreover, PI3Kα blockade downregulates Akt/mTOR signalling, which is critical for FOP and non‐genetic (trauma induced) HO in preclinical models (Hino et al, 2017 PMID: 28758906; Hino et al. PMID: 30392977). Finally, PI3Kα inhibition hampers a number of proinflammatory pathways, thereby limiting the expression of pro-inflammatory cytokines, reducing the proliferation of monocytes, macrophages and mast cells, and partially blocking the migration of monocytes. As we suggest in the discussion of the manuscript, this effect likely causes a poor recruitment of monocytes and macrophages at injury sites and throughout the in vivo ossification process.

Noteworthy, in our manuscript we do not refer to a “specific chemical compound for cell types”. Rather, in the Discussion we write “the administration of BYL719 prevented an exacerbated inflammatory response in vivo, possibly due to specific effects observed on immune cell populations.” This sentence did not intend to imply that BYL719 only affects these specific cell types, but aimed to emphasize the effects observed on those cell populations, even though systemic BYL719 may affect all populations. We rephrased it to “the administration of BYL719 prevented an exacerbated inflammatory response in vivo, possibly due to the effects observed on immune cell populations.” to provide a clearer message as suggested by the reviewer. We thank the reviewer for these questions and hope that these explanations and changes in the text improve the clarity of the message.

Mesenchymal stem/stromal cells (MSCs) are osteochondroprogenitor cells that can follow distinct differentiation paths. In this study, we use these cells as an in vitro model for the study of osteochondrogenitor specification. MSCs, and induced MSCs (iMSCs), have been widely used as in vitro cellular models of osteochondroprogenitor specification for the analysis of markers, signaling, modulation, and differentiation potential or capacity. Their use as models for this purpose has been extensively studied in wild type MSCs, and in the presence of FOP mutations (Boeuf and Richter PMID: 20959030; Schwartzl et al. PMID: 37923731).

Point 2: Related to comment #1, the effects of BYL719 on the proliferation and differentiation of fibro-adipogenic cells in skeletal muscle, which are potential progenitor cells of HO, should be important to support the claim of the authors.

We have performed additional in vivo experiments to assess the presence of fibro-adipogenic precursors (FAPs) at different time-points during HO formation in control and BYL719-treated in the mouse model of heterotopic ossification. We analyzed the number of fibro-adipogenic progenitor (FAPs) during the progression of the HO. These data are shown in the new Figure3-Figure Supplement 1. We demonstrate that BYL719 reduces the number of PDGFRA+ cells (FAPs, red) throughout the ossification process in vivo. Moreover, now we also show an enlargement of the diameter of myofibers (labelled with wheat germ agglutinin, green) when animals were treated with BYL719, indicating improved muscle regeneration and further validating the data reported as supplementary figures that were added in the first revision of this manuscript.

Point 3: BYL719 inhibited signaling through not only ACVR1-R206H and ACVR1-Q207D but also wild type ACVR1 and suppressed the chondrogenic differentiation of parental MSCs regardless of the expression of wild type or mutant ACVR1. Again, these findings suggest that BYL719 inhibits HO through a multiple and nonspecific pathway in multiple types of cells in vivo. The authors are encouraged to explain logically the use of bone marrow-derived MSCs to examine the effects of BYL719.

As detailed in main point 1, we consider that the main target, molecular mechanisms and inhibited pathways by BYL719 are specific and well characterised in other research articles and further defined in this manuscript, including the generation of PI3Ka deficient mice in an FOP background, that undoubtedly demonstrates an essential role for PI3Ka in ACVR1-driven heterotopic ossification in vivo. Altogether, we are confident that BYL719 inhibits HO through multiple and specific pathways that arise from the PI3Kα inhibition. As a systemically administrated drug, BYL719 affects the multiple types of cells in vivo that express PI3Kα. It is well known that PI3Kα is exquisitely required for chondrogenesis and osteogenesis (Zuscik et al. PMID; Gamez et al PMID: 26896753 1824619). Accordingly, throughout the manuscript we refrain from suggesting a specific effect on ACVR1-R206H cells but instead an inhibitory effect on cell number and differentiation regardless on the ACVR1 form expressed.

Similarly, as detailed in main point 1, MSCs and hiPSCs have been extensible used as in vitro cellular models of osteochondroprogenitor specification for the analysis of markers, signaling, modulation, and differentiation potential or capacity (Barruet et al., PMID: 28716551; Kan et al., PMID: 39308190).

Point 4: BYL719 clearly inhibits an mTOR pathway. Is there a possibility that BYL719 suppresses HO by inhibiting mTOR rather than PI3K? The authors are encouraged to show the unique role of PI3K in BYL719-suppressed HO formation.

As clarified above, BYL719 is a potent and selective inhibitor of PI3Kα, with minimal off-target inhibition against other kinases, as it has been demonstrated in multiple studies and in several in vitro kinase assay panels. In the same study, while IC50 of BYL719 against PI3Kα was (IC50 = 4.6 nmol/L), IC50 against mTOR was (IC50= >9,100 nmol/L), indicating that it was not directly inhibited. mTOR is one of the well-known pathways that are activated downstream of PI3K. Therefore, there is no surprise that blocking PI3Kα will block mTOR signalling. This potential effect was already demonstrated in previous publications (Valer et al., 2019a PMID: 31373426) and discussed throughout the first revision. We consider that the additive effect of mTOR inhibition and other molecular mechanisms downstream of PI3Kα, including reduced SMAD1/5 protein levels, contribute to the in vivo HO inhibition by BYL719.

Reviewer #2:

Point 1: It is also important to note that, in most of the data, there is no significant difference between cells with wild-type ACVR1 and those with the R206H mutation. The authors demonstrated that ACVR1 is not a target of BYL719 based on NanoBRET assay data, suggesting that BYL719's effect is not specific to FOP cells, even though they used an FOP mouse model to show in vivo effects.

The main effect of R206H mutation is the gain of function in response to Activin A. For most of the responses to other ACVR1 ligands (e.g. BMP6/7), we observe a slightly increased response in the presence of the mutation (which is consistent with previous research, usually labelling RH as a “weak activating mutant” unless Activin A is added (Song et al., PMID: 20463014)). Therefore, as expected, most of the differences between WT and RH mutant cells can be observed mostly upon Activin A addition, as observed, for example, in Figure 3 of our manuscript.

We agree with the reviewer that, at the concentrations used, BYL719 does not specifically target FOP cells. However, we believe that it targets downstream pathways of PI3Kα inhibition that are essential for osteochondrogenic specification, regardless of mutation status. This therapeutic strategy aligns with other experimental drugs, including Palovarotene (validated for FOP) and Garetosmab and Saracatinib (in advanced clinical trials), which target Activin A function, ACVR1 activity, or osteochondrogenic differentiation irrespective of the mutant allele. Unlike these molecules, BYL719 has been chronically administered to patients (including children) without major side effects (Gallagher et al.; PMID: 38297009), further supporting its potential for safe long-term use.

The authors should consider that the effect of Activin A on R206H cells is not identical to that of BMP6 on WT cells. If the authors aim to identify the target of BYL719 in FOP cells, they should compare R206H cells treated with Activin A/BYL719 to WT cells treated with BMP6/BYL719.

We use Activin A and BMP6, both high-affinity ACVR1 ligands, to demonstrate, as observed in figure 6, that PI3Kα inhibition can inhibit the expression of genes within GO terms ossification and osteoblast differentiation. It is important to note, however, that Activin A canonical signaling receptor is ACVR1B. Since BYL719 blocks the induction of a heterotopic ossification gene expression signature common to Activin A and BMP6, in the context of the FOP mutation R206H, our results indicate that BYL719 inhibition affects a signaling pathway downstream of ACVR1, activated by either BMP6 (wild type receptor, relevant for non-genetic heterotopic ossifications) or Activin (R206H mutant receptor, relevant for FOP).

We consider that the comparison (RH ACTA BYL vs WT BMP6 BYL) would provide confounding results raised from intrinsic model differences in basal expression programs (WT vs RH), and differences in the quantitative level of signaling of the different ligands at these specific doses. First, if we only consider SMAD1/5 signaling, Activin A and BMP6 won’t have identical signaling, and differences will arise from the strength of that signaling. Secondly, in the suggested comparison we would find, mostly, all the differential gene expression promoted by Activin A canonical signaling through type I receptors ACVR1B/ALK4 in complex with ACVR2A or ACVR2B, promoting SMAD2/3 activation (in addition to the altered signaling that ACVR1-R206H could promote). Examples of differential response in pSMAD1/5 in ACVR1-WT or RH with BMP ligands and R206H with Activin A ligand, and examples of pSMAD2/3 canonical signaling in R206H cells have been described in Ramachandran et al, PMID: 34003511; Hatsell et al., PMID: 26333933).

Point 2: The interpretation of the data in the new Figure 5 is inappropriate. Based on the expression levels of SOX9, COL2A1, and ACAN, it is unclear whether the effect of BYL719 is due to the inhibition of differentiation or proliferation. The addition of Activin A showed no difference between ACVR1/WT and ACVR1/R206H cells, suggesting that these cells did not accurately replicate the FOP condition.

To gain consistency in our manuscript, we decided to use an orthogonal and complementary approach in a completely new model. We performed new experiments of chondrogenic differentiation using murine MSCs from UBC-Cre-ERT2/ACVR1^R206H knock-in mice. These cells, when treated with 4OH-tamoxifen, express the intracellular exons of human ACVR1^R206H in the murine Acvr1 locus. Therefore, we can compare differentiation of wild type and R206H MSCs isolated form the same mice. We initiated the chondrogenic differentiation assay from confluent cells to minimize changes in cell proliferation throughout the process. These new results are shown in the new Figure 5F. Mutant (RH) cells display an enhanced chondrogenic response to activin A compared to wild type cells. The treatment with BYL719 decreased the expression of chondrogenic markers irrespective of the mutational status of ACVR1 in the cells, further supporting our previous results in this manuscript and published article (Valer et al., 2019a PMID: 31373426).

Point 3: The additional investigation of RNA-seq data provided useful information but was insufficient to fully address the purpose of this study. The authors should identify downregulated genes by comparing WT cells treated with Activin A/BYL719 and Activin A alone and then compare these identified genes with those shown in Figure 5E. Additionally, they should compare R206H cells treated with Activin A/BYL719 to WT cells treated with BMP6/BYL719. These comparisons will clarify whether there are FOP-specific BYL719-regulated genes.

We thank the reviewer for considering that RNAseq data provides useful information. As already discussed in our answer above, our results indicate that regardless of the ligand (Activin A or BMP6) and regardless of the ACVR1 mutation (WT, relevant for non-genetic heterotopic ossifications or RH, relevant for FOP), BYL719 can inhibit the expression of the genes relevant to endochondral ossification. In our opinion, this is a very relevant conclusion of this study.

We have deeply considered the strategy proposed by the reviewer, comparing “WT cells treated with Activin A/BYL719 and Activin A alone and then compare these identified genes with those shown in Figure 5E” and/or comparing “R206H cells treated with Activin A/BYL719 to WT cells treated with BMP6/BYL719”. While we have discussed why we do not consider appropriate the first comparison proposed, there are a number of reasons why we are not confident that the second comparison would provide a straightforward conclusion.

Regarding the second suggested comparison already in Main point 1, we consider that it would provide confounding results due to all the arguments detailed in Main point 1. Regarding the first suggested comparison, we also consider that it would provide confounding results. There are several reasons why we do not consider that the genes only found in the RH comparison can be confidently considered genes that are only affected by BYL719 in RH cells.

First, the effect of BYL719 in an osteogenic-prone sample (for example, RH-ActA) is higher than the effect that we can observe in absence of this activation (for example, WT-ActA), as observed in the higher number of significantly downregulated genes in RH ActA BYL vs RH ActA comparison, compared to WT ActA BYL vs WT ActA. Similar results are observed in figure 3C, where the expressions of the genes are significantly inhibited in RH ActA compared to RH ActA BYL. This inhibition is not significantly observed in in WT ActA compared to WT ActA BYL because the osteogenic expression of these genes is already very weak in the absence of ACVR1 R206H. This weak signaling of pSMAD1/5 in the absence of osteogenic signaling (RH without ligand or, especially, WT with Activin A) has already been described (Ramachandran et al. MID: 34003511). Therefore, even though the inhibition is present in both comparisons, as observed in figure 6C, the extent of the observed effect is different. Second, we are comparing a different number of DEGs for each comparison between them. If we compare the 67 downregulated genes from one comparison and 38 downregulated genes from the other comparison, the unequal list size may inflate the number of unique genes in the group with more downregulated genes. To prove these concerns, we performed the comparison that the reviewer suggested and we found, for example, that amongst the 38 differentially downregulated ossification genes in (WT_ActA_BYL vs WT_ActA) and 67 differentially downregulated ossification genes in (RH_ActA_BYL vs RH_ActA), 39 genes were only found in the RH comparison, while 10 were only found in the WT comparison, and 28 were found in both.

These effects are present, for example, when studying the ID genes, well-known downstream mediators of BMP signaling. In this case, ID1 is downregulated in both comparisons, while ID2, ID3, and ID4, are downregulated only in the RH-group, despite the fact that all ID1, ID2, ID3, and ID4 are similarly regulated and increase their expression with similar time curves upon BMP signaling activation (Yang et al., PMID: 23771884). Therefore, we consider that the comparisons proposed will not help us to identify specific BYL719-regulated genes relevant for FOP and/or ACVR1 R206H signaling. Again, we consider that BYL719 effect is not specific of FOP cells. Our results show that regardless of the ligand (Activin A or BMP6) and regardless of the ACVR1 mutation (WT, relevant for non-genetic heterotopic ossifications or RH, relevant for FOP), BYL719 can inhibit the expression of the genes linked to ossification and osteoblast differentiation, which could be important for the treatment of FOP and non-genetic heterotopic ossifications.

Point 4: The data in Figure 7 are not relevant to the aim of this study because the cell lines used in these experiments did not have ACVR1/R206H mutations. The authors mentioned that BMP6 is a ligand for ACVR1 and, therefore, these experiments reflect the situation of inflammatory cells in FOP. This is inappropriate and not rational. As mentioned above, the effect of Activin A on FOP cells is not identical to the effect of BMP6 in wild-type cells. The data in Figure 7 indicated that the effect of BYL719 is unrelated to the presence of BMP6, clearly demonstrating that these experiments are not related to the activation of ACVR1. In the gene expression analyses, almost all genes showed no changes with the addition of BMP6. Only TGF and CCL2 showed upregulation in THP1 cells, and the treatment with BYL719 failed to inhibit the effect of BMP6, suggesting that these experiments merely demonstrate the effect of BYL719 on inflammatory cells irrespective of the presence of the HO signal.

We consider that Figure 7 is relevant to the aim of this study. As shown in Fig. 8, treatment of FOP mice with BYL719 led to a decreased recruitment of immune cells within the FOP lesions, suggesting a direct effect of BYL719 in immune cells. This is very relevant for the FOP pathology, since flare-ups have been linked with inflammatory episodes since the very early characterization of the disease (Mejias-Rivera et al., PMID: 38672135). Given the technical difficulties to transduce THP1, RAW264 and HMC1 cell lines with lentiviral particles carrying ACVR1 R206H, we decided to partially recapitulate ACVR1 R206H activation with recombinant BMP6 and to test the effect of BYL719 in these conditions. In these models, we found that BYL719 inhibited the expression of key genes driving immune cell activation, in a cell-type and ligand independent manner. To clarify this rationale, we have swapped Figures 7 and 8 and adjusted our conclusions accordingly. We have softened our interpretations, emphasizing the absence of the ACVR1 R206H mutant receptor in these experiments.

eLife Assessment

Using a unique cerebellar disruption approach in non-human primates, this study provides valuable new insight into how cerebellar inputs to the motor cortex contribute to reaching. The findings convincingly demonstrate that reaching movements following cerebellar disruption slow down because of both an acute deficit in producing muscle activity as well as a progressive decline in compensating for limb dynamics. This work will be of interest to neuroscientists and clinicians interested in cerebellar function and pathology.

Reviewer #1 (Public review):

Summary:

In a previous work Prut and colleagues had shown that during reaching, high frequency stimulation of the cerebellar outputs resulted in reduced reach velocity. Moreover, they showed that the stimulation produced reaches that deviated from a straight line, with the shoulder and elbow movements becoming less coordinated. In this report they extend their previous work by addition of modeling results that investigate the relationship between the kinematic changes and torques produced at the joints. The results show that the slowing is not due to reductions in interaction torques alone, as the reductions in velocity occur even for movements that are single joint. More interestingly, the experiment revealed evidence for decomposition of the reaching movement, as well as an increase in the variance of the trajectory.

Strengths:

This is a rare experiment in a non-human primate that assessed the importance of cerebellar input to the motor cortex during reaching.

Weaknesses:

None

Reviewer #2 (Public review):

This manuscript asks an interesting and important question: what part of 'cerebellar' motor dysfunction is an acute control problem vs a compensatory strategy to the acute control issue? The authors use a cerebellar 'blockade' protocol, consisting of high frequency stimuli applied to the cerebellar peduncle which is thought to interfere with outflow signals. This protocol was applied in monkeys performing center out reaching movements and has been published from this laboratory in several preceding studies. I found the take-home-message broadly convincing and clarifying - that cerebellar block reduces muscle activation acutely particularly in movements that involve multiple joints and therefore invoke interaction torques, and that movements progressively slow down to in effect 'compensate' for these acute tone deficits. The manuscript was generally well written, data were clear, convincing and novel. The key strengths are differentiating acute from sub-acute (within session but not immediate) kinematic consequences of cerebellar block.

Reviewer #3 (Public review):

Summary:

In their revised manuscript, Sinha and colleagues aim to identify distinct causes of motor impairments seen when perturbing cerebellar circuits. This goal is an important one, given the diversity of movement related phenotypes in patients with cerebellar lesion or injury, which are especially difficult to dissect given the chronic nature of the circuit damage. To address this goal, the authors use high-frequency stimulation (HFS) of the superior cerebellar peduncle in monkeys performing reaching movements. HFS provides an attractive approach for transiently disrupting cerebellar function previously published by this group. First, they find a reduction in hand velocities during reaching, which was more pronounced for outward versus inward movements. By modeling inverse dynamics, they find evidence that shoulder muscle torques are especially affected. Next, the authors examine the temporal evolution of movement phenotypes over successive blocks of HFS trials. Using this analysis, they find that in addition to the acute, specific effects on torques in early HFS trials, there was an additional progressive reduction in velocity during later trials, which they interpret as an adaptive response to the inability to effectively compensate for interaction torques during cerebellar block. Finally, the authors examine movement decomposition and trajectory, finding that even when low velocity reaches are matched to controls, HFS produces abnormally decomposed movements and higher than expected variability in trajectory.

Strengths:

Overall, this work provides important insight into how perturbation of cerebellar circuits can elicit diverse effects on movement across multiple timescales.

The HFS approach provides temporal resolution and enables analysis that would be hard to perform in the context of chronic lesions or slow pharmacological interventions. Thus, this study describes an important advance over prior methods of circuit disruption in the monkey, and their approach can be used as a framework for future studies that delve deeper into how additional aspects of sensorimotor control are disrupted (e.g., response to limb perturbations).

In addition, the authors use well-designed behavioral approaches and analysis methods to distinguish immediate from longer-term adaptive effects of HFS on behavior. Moreover, inverse dynamics modeling provides important insight into how movements with different kinematics and muscle dynamics might be differentially disrupted by cerebellar perturbation.

In this revised version of the manuscript, the authors have provided additional analyses and clarification that address several of the comments from the original submission.

Remaining comments:

The argument that there are acute and adaptive effects to perturbing cerebellar circuits is compelling, but there seems to be a lost opportunity to leverage the fast and reversible nature of the perturbations to further test this idea and strengthen the interpretation. Specifically, the authors could have bolstered this argument by looking at the effects of terminating HFS - one might hypothesize that the acute impacts on joint torques would quickly return to baseline in the absence of HFS, whereas the longer-term adaptive component would persist in the form of aftereffects during the 'washout' period. As is, the reversible nature of the perturbation seems underutilized in testing the authors' ideas. While this experimental design was not implemented here, it seems like a good opportunity for future work using these approaches.

The analysis showing that there is a gradual reduction in velocity during what the authors call an adaptive phase is convincing. While it is still not entirely clear why disruption of movement during the adaptive phase is not seen for inward targets, despite the fact that many of the inward movements also exhibit large interaction torques, the authors do raise potential explanations in the Discussion.

The text in the Introduction and in the prior work developing the HFS approach overstates the selectivity of the perturbations. First, there is an emphasis on signals transmitted to the neocortex. As the authors state several times in the Discussion, there are many subcortical targets of the cerebellar nuclei as well, and thus it is difficult to disentangle target-specific behavioral effects using this approach. Second, the superior cerebellar peduncle contains both cerebellar outputs and inputs (e.g., spinocerebellar). Therefore, the selectivity in perturbing cerebellar output feels overstated. Readers would benefit from a more agnostic claim that HFS affects cerebellar communication with the rest of the nervous system, which would not affect the major findings of the study. In the revised manuscript, the authors do provide additional anatomical and evolutionary context and discuss potential limitations in the selectivity of HFS in the Materials and Methods. However, I feel that at least a brief mention of these caveats in the Introduction, where it is stated, "we then reversibly blocked cerebellar output to the motor cortex", would benefit the reader.

eLife Assessment

By using sparse Cre-dependent deletion of GluN1 subunit, in vitro quadruple patch clamp recordings, and pharmacological interventions, the authors show that spike timing dependent plasticity at between L5 synapses in the mouse visual cortex is: (i) dependent on presynaptic NMDA receptors; (ii) mediated by non-ionotropic NMDA receptor signaling, and (iii) reliant on presynaptic JNK2/Syntaxin-1a interactions. These fundamental findings advance our understanding of the molecular mechanisms underlying spike time dependent plasticity. The data are compelling and are supported by the elegant application of sophisticated experimental approaches.

Reviewer #2 (Public review):

Summary:

The study characterized the dependence of spike timing-dependent long-term depression (tLTD) on presynaptic NMDA receptors and the intracellular cascade after NMDAR activation possibly involved in the observed decrease in glutamate probability release at L5-L5 synapses of the visual cortex in mouse brain slices.

Strengths:

The genetic and electrophysiological experiments are thorough. The experiments are well reported and mainly support the conclusions. This study confirms and extends current knowledge by elucidating additional plasticity mechanisms at cortical synapses, complementing existing literature.

Weaknesses:

No direct testing for ions passing trough standard NMDAR, mainly sodium and calcium is shown.

Reviewer #3 (Public review):

Summary:

In this manuscript, "Neocortical Layer-5 tLTD Relies on Non-Ionotropic Presynaptic NMDA Receptor Signaling", Thomazeau et al. seek to determine the role of presynaptic NMDA receptors and the mechanism by which they mediate expression of frequency-independent timing-dependent long-term depression (tLTD) between layer-5 (L5) pyramidal cells (PCs) in the developing mouse visual cortex. By utilizing sophisticated methods, including sparse Cre-dependent deletion of GluN1 subunit via neonatal iCre-encoding viral injection, in vitro quadruple patch clamp recordings, and pharmacological interventions, the authors elegantly show that L5 PC->PC tLTD is 1) dependent on presynaptic NMDA receptors, 2) mediated by non-ionotropic NMDA receptor signaling, and 3) is reliant on JNK2/Syntaxin-1a (STX1a) interaction (but not RIM1αβ) in the presynaptic neuron. The study elegantly and pointedly addresses a long-standing conundrum regarding the lack of frequency dependence of tLTD.

Strengths:

The authors did a commendable job presenting a very polished piece of work with high-quality data that this Reviewer feels enthusiastic about. The manuscript has several notable strengths. Firstly, the methodological approach used in the study is highly sophisticated and technically challenging, and successfully produced high-quality data that were easily accessible to a broader audience. Secondly, the pharmacological interventions used in the study targeted specific players and their mechanistic roles, unveiling the mechanism in question step-by-step. Lastly, the manuscript is written in a well-organized manner that is easy to follow. Overall, the study provides a series of compelling evidence that leads to a clear illustration of mechanistic understanding.

Weakness:

No major weaknesses were noted.

Author response:

The following is the authors’ response to the original reviews

Reviewer 1 (Public review)

Summary

The results offer compelling evidence that L5-L5 tLTD depends on presynaptic NMDARs, a concept that has previously been somewhat controversial. It documents the novel finding that presynaptic NMDARs facilitate tLTD through their metabotropic signaling mechanism.

We thank Reviewer 1 for their kind words and thoughtful feedback!

Strengths

The experimental design is clever and clean. The approach of comparing the results in cell pairs where NMDA is deleted either presynaptically or postsynaptically is technically insightful and yields decisive data. The MK801 experiments are also compelling.

We are very grateful for this kind feedback!

Weaknesses

No major weaknesses were noted by this reviewer.

We were happy to see that Reviewer 1 had no concerns in the Public Review. We address their Recommendations here below.

Reviewer #1 (Recommendations for the authors):

There is one minor issue that the authors might want to address. In Figure 6C, the average time course of the controls (blue symbols) shows a clear decline in the baseline. The rate of this decline appears to be similar to the initial decline rate observed after inducing tLTD.

Sorry, the x-axis was truncated so the first data points were not visible. We fixed Fig 6C as well as 6G, which suffered from the same problem.

Reviewer 2 (Public review)

Summary

The study characterized the dependence of spike-timing-dependent long-term depression (tLTD) on presynaptic NMDA receptors and the intracellular cascade after NMDAR activation possibly involved in the observed decrease in glutamate probability release at L5-L5 synapses of the visual cortex in mouse brain slices.

We are grateful for Reviewer 2’s thoughtful and detailed feedback!

Strengths

The genetic and electrophysiological experiments are thorough. The experiments are well-reported and mainly support the conclusions. This study confirms and extends current knowledge by elucidating additional plasticity mechanisms at cortical synapses, complementing existing literature.

We were thrilled to see that the reviewer thinks our experiments are “thorough”, “well-reported” and they “mainly support the conclusions”!

Weaknesses

While one of the main conclusions (preNMDARs mediating presynaptic LTD) is resolved in a very convincing genetic approach, the second main conclusion of the manuscript (non-ionotropic preNMDARs) relies on the use of a high concentration of extracellular blockers (MK801, 2 mM; 7-clorokinurenic acid: 100 microM), but no controls for the specific actions of these compounds are shown.

We thank the reviewer for calling our genetic approach “very convincing”!

Regarding the pharmacological controls: for MK-801, we deliberately used a high extracellular concentration in the mM-range to match the intracellular concentrations used both in our own experiments and in prior studies (Berretta and Jones, 1996; Brasier and Feldman, 2008; Buchanan et al., 2012; Corlew et al., 2007; Humeau et al., 2003; Larsen et al., 2011; Rodríguez-Moreno et al., 2011; Rodríguez-Moreno and Paulsen, 2008). Our goal was to isolate the variable of application site (internal vs. external) while keeping concentration constant. If we had used the lower, more conventional µM-range extracellular concentrations (e.g., Huettner and Bean, 1988; Kemp et al., 1988; Tovar and Westbrook, 1999), differences in outcome might have reflected differences in drug efficacy rather than localization — particularly since failure to observe an effect at low concentrations would be hard to interpret.

We now clarify this rationale in the revised manuscript (lines 578-585).

As for 7-chlorokynurenic acid (7-CK), the 100 µM concentration we used is standard for effectively blocking the glycine-binding site of NMDARs (e.g., Nabavi et al., 2013).

We also added two supplementary figures to show the effects of washing in MK-801 and 7-CK. In MK-801, responses are stable at low frequency (clarified in the manuscript lines 155-157 and Supp Fig 1 caption text). However, 7-CK suppresses responses appreciably, which takes time to stabilize. We clarify in the revised manuscript that in 7-CK experiments, we waited for this stabilization before inducing tLTD (lines 167-172 and Supp Fig 2 caption text). This additional suppression is consistent with 7-CK also acting as a potent competitive inhibitor of L-glutamate transport into synaptic vesicles (Bartlett et al., 1998).

In addition, no direct testing for ions passing through preNMDAR has been performed.

Sorry for being unclear, we have previously tested directly for ions passing through preNMDARs. For example, we showed blockade with Mg²⁺ before (Abrahamsson et al., 2017; Wong et al., 2024), and we showed preNMDAR Ca²⁺ supralinearities before (Abrahamsson et al., 2017; Buchanan et al., 2012). To improve the manuscript, we clarified the text accordingly (lines 140-141).

It is not known if the results can be extrapolated to adult brain as the data were obtained from 11-18 days-old mice slices, a period during which synapses are still maturing and the cortex is highly plastic.

Thank you, this is a good point. We address this point in the revised manuscript (lines 428-432). While our study focuses on the early postnatal period (P11–P18), when plasticity mechanisms are prominent and synaptic maturation is ongoing, we agree that extrapolation to the adult brain should be made with caution.

Reviewer #2 (Recommendations for the authors):

Points 1-3 were also found in the Public Review so are not addressed again here.

(4) Results seem to be obtained in the absence of inhibition blocking and the role of inhibition in tLTD is not described. It should be indicated whether present results are obtained with or without the functional inhibitory synapse activation. If GABAergic synapses are not blocked authors need to show what happens when this inhibition is blocked.

We agree that extracellular stimulation can inadvertently recruit inhibitory circuits. However, in our paired whole-cell recordings, synaptic responses are always subthreshold and exclusively reflect the direct connection between the two recorded neurons (Chou et al., 2024; Song et al., 2005). Under these conditions, inhibitory synapses are not activated, and we therefore did not apply GABAergic blockers. We thank the reviewer for raising this, which is now clarified in the Methods (lines 539-541) of the revised manuscript.

(5) In some figures, the number of experiments seems to be low, and this number of experiments might be increased (Figures 1C, 3C, 4B).

We acknowledge that the number of experiments in these figures is modest, but these recordings are technically demanding, and the data are carefully curated. Importantly, the observed effects were statistically significant, indicating that the sample sizes were sufficient. We also note that concerns about statistical power are typically more critical in the case of negative or null results, whereas our findings were positive.

(6) The discussion is detailed but it is not clear that the activation of JNK2 needs to be achieved by a non-ionotropic action of NMDAR as activation after ionotropic NMDAR activation has been described in the literature. This point needs to be clarified and expanded.

Sorry that we were unclear on this point. We clarified this on lines 371-372 of the manuscript.

(7) Adding a cartoon/schematic summarizing the proposed mechanism for tLTD would help the reading of the manuscript.

We appreciate this suggestion and agree that a schematic would be helpful. However, we prefer to hold off on including one at this stage, as aspects of the underlying mechanism — particularly the role of CB1 receptors in presynaptic pyramidal cells (Sjöström et al., 2003) — are currently under active investigation in a separate project. To avoid potentially misleading oversimplifications, we would prefer to revisit a summary schematic once these uncertainties have been resolved.

Minor:

(1) Concentration of compounds is recommended to be included in the figures or in the text. This would make it easy to follow the results.

We appreciate the suggestion. However, we avoid repeating concentrations to emphasize that conditions are consistent unless otherwise stated. All compound concentrations are clearly listed in the Methods and remain unchanged across experiments. We believe this streamlined approach avoids redundancy while keeping the results clear.

(2) In some figures, failures in synaptic transmission can be observed (and changes after tLTD). The authors may analyse changes in a number of failures in synaptic transmission after tLTD as an additional indication of a presynaptic expression of this form of tLTD. PPR may also be included in all figures.

While failures in synaptic transmission are occasionally visible, we chose to focus on CV analysis, which is mathematically equivalent to failure rate analysis, as both rely on the same underlying variability in synaptic responses (Brock et al., 2020). Provided failures are reliably extracted (which requires sufficient signal-to-noise), CV and failure rate analyses should yield consistent conclusions.

In contrast, PPR analysis is not mathematically equivalent to CV analysis and may offer complementary insights into presynaptic mechanisms. However, the presence of preNMDARs complicates the use of paired-pulse stimulation during baseline: preNMDARs enhance release during high-frequency activity (Abrahamsson et al., 2017; Sjöström et al., 2003; Wong et al., 2024), so repeated stimulation can suppress synaptic responses when preNMDARs are blocked, potentially confounding interpretation. For this reason, we limited PPR analysis to Figures 5 and 6, where conditions were appropriate.

Admittedly, our manuscript was previously not clear on when we did paired-pulse stimulation and when we did not. We have clarified this in the revised manuscript (lines 548- 551 and lines 569-574).

(3) Discussion: Line 363-64, hippocampal (SC-CA1 synapses) results exist where postsynaptic MK801 blocks presynaptic tLTD, this may be added here and in the references.

While we acknowledge that postsynaptic MK-801 has been shown to block presynaptic tLTD at hippocampal SC–CA1 synapses, we note that the hippocampus is part of the archicortex, whereas our study focuses on neocortical circuits, as highlighted in the manuscript title. Given the substantial anatomical and functional differences between these regions, we prefer to keep our discussion focused on the neocortex to maintain conceptual coherence.

(4) Discussion: While authors indicate "non-ionotropic" they do not discuss whether this action can be named properly "metabotropic" and whether G-proteins may be in fact needed for this action. The authors may briefly discuss this point.

We previously referred to non-ionotropic NMDAR signaling as “metabotropic,” but reconsidered after discussions with colleagues, including Juan Lerma, who pointed out that the term typically implies G-protein coupling, which has not been definitively shown in this context. While the term “metabotropic” is used inconsistently in the literature (Heuss and Gerber, 2000; Heuss et al., 1999) — sometimes broadly to indicate non-ion flow signaling — we prefer to avoid potential confusion and therefore use “non-ionotropic” unless and until G-protein involvement is clearly demonstrated. We clarified this on lines 423-427 of the Discussion.

(5) Page 19, line 451 NMDR needs to be corrected to NMDAR.

Thanks! This was corrected.

Reviewer 3 (Public review)

Summary

In this manuscript, "Neocortical Layer-5 tLTD Relies on Non-Ionotropic Presynaptic NMDA Receptor Signaling", Thomazeau et al. seek to determine the role of presynaptic NMDA receptors and the mechanism by which they mediate expression of frequency-independent timing-dependent long-term depression (tLTD) between layer-5 (L5) pyramidal cells (PCs) in the developing mouse visual cortex. By utilizing sophisticated methods, including sparse Cre-dependent deletion of GluN1 subunit via neonatal iCre-encoding viral injection, in vitro quadruple patch clamp recordings, and pharmacological interventions, the authors elegantly show that L5 PC->PC tLTD is (1) dependent on presynaptic NMDA receptors, (2) mediated by non-ionotropic NMDA receptor signaling, and (3) is reliant on JNK2/Syntaxin-1a (STX1a) interaction (but not RIM1αβ) in the presynaptic neuron. The study elegantly and pointedly addresses a long-standing conundrum regarding the lack of frequency dependence of tLTD.

We thank the reviewer for calling our methods “sophisticated” and our study “elegant”! We appreciate the kind feedback!

Strengths

The authors did a commendable job presenting a very polished piece of work with high-quality data that this Reviewer feels enthusiastic about. The manuscript has several notable strengths. Firstly, the methodological approach used in the study is highly sophisticated and technically challenging and successfully produced high-quality data that were easily accessible to a broader audience. Secondly, the pharmacological interventions used in the study targeted specific players and their mechanistic roles, unveiling the mechanism in question step-by-step. Lastly, the manuscript is written in a well-organized manner that is easy to follow. Overall, the study provides a series of compelling evidence that leads to a clear illustration of mechanistic understanding.

We are elated that the reviewer described our study with words such as “polished”, “high-quality”, “sophisticated”, and “compelling”!

Minor comments

(1) For the broad readership, a brief description of JNK2-mediated signaling cascade underlying tLTD, including its intersection with CB1 receptor signaling may be desired.

Thank you, this is a great suggestion for improving clarity. We briefly address this point in the revised manuscript (lines 360-363).

(2) The authors used juvenile mice, P11 to P18 of age. It is a typical age range used for plasticity experiments, but it is also true that this age range spans before and after eye-opening in mice (~P13) and is a few days before the onset of the classical critical period for ocular dominance plasticity in the visual cortex. Given the mechanistic novelty reported in the study, can authors comment on whether this signaling pathway may be age-dependent?

Thanks, Reviewer 2 also raised this point. In the revised manuscript, we discuss this point (lines 428-432).

Reviewer #3 (Recommendations for the authors):

(1) Minor typos: page 4 line 101: sensitivity -> sensitive.

We fixed this typo.

(2) Page 15 line 333: sensitivity -> sensitive.

We fixed this typo.

(3) Minor aesthetic suggestion: On the scale bars for all examples, LTP and LTD data are easily confused with the letter L. I'd suggest flipping them left to right.

We thank the reviewer for the suggestion. We flipped the scale bars in all figures.

References

Abrahamsson, T., Chou, C.Y.C., Li, S.Y., Mancino, A., Costa, R.P., Brock, J.A., Nuro, E., Buchanan, K.A., Elgar, D., Blackman, A.V., et al. 2017. Differential Regulation of Evoked and Spontaneous Release by Presynaptic NMDA Receptors. Neuron 96: 839-855 e835

Bartlett, R.D., Esslinger, C.S., Thompson, C.M., and Bridges, R.J. 1998. Substituted quinolines as inhibitors of L-glutamate transport into synaptic vesicles. Neuropharmacology 37: 839-846

Berretta, N., and Jones, R.S. 1996. Tonic facilitation of glutamate release by presynaptic N-methyl-D-aspartate autoreceptors in the entorhinal cortex. Neuroscience 75: 339-344.

Brasier, D.J., and Feldman, D.E. 2008. Synapse-specific expression of functional presynaptic NMDA receptors in rat somatosensory cortex. J Neurosci 28: 2199-2211

Brock, J.A., Thomazeau, A., Watanabe, A., Li, S.S.Y., and Sjöström, P.J. 2020. A Practical Guide to Using CV Analysis for Determining the Locus of Synaptic Plasticity. Frontiers in Synaptic Neuroscience 12:11 10.3389/fnsyn.2020.00011

Buchanan, K.A., Blackman, A.V., Moreau, A.W., Elgar, D., Costa, R.P., Lalanne, T., Tudor Jones, A.A., Oyrer, J., and Sjöström, P.J. 2012. Target-Specific Expression of Presynaptic NMDA Receptors in Neocortical Microcircuits. Neuron 75: 451-466

Chou, C.Y.C., Wong, H.H.W., Guo, C., Boukoulou, K.E., Huang, C., Jannat, J., Klimenko, T., Li, V.Y., Liang, T.A., Wu, V.C., and Sjöström, P.J. 2024. Principles of visual cortex excitatory microcircuit organization. The Innovation 6: 1-11

Corlew, R., Wang, Y., Ghermazien, H., Erisir, A., and Philpot, B.D. 2007. Developmental switch in the contribution of presynaptic and postsynaptic NMDA receptors to long-term depression. J Neurosci 27: 9835-9845

Heuss, C., and Gerber, U. 2000. G-protein-independent signaling by G-protein-coupled receptors. Trends in Neurosciences 23: 469-475

Heuss, C., Scanziani, M., Gähwiler, B.H., and Gerber, U. 1999. G-protein-independent signaling mediated by metabotropic glutamate receptors. Nature Neuroscience 2: 1070-1077

Huettner, J.E., and Bean, B.P. 1988. Block of N-methyl-D-aspartate-activated current by the anticonvulsant MK-801: selective binding to open channels. PNAS 85: 1307-1311.

Humeau, Y., Shaban, H., Bissière, S., and Lüthi, A. 2003. Presynaptic induction of heterosynaptic associative plasticity in the mammalian brain. Nature 426: 841-845

Kemp, J.A., Foster, A.C., Leeson, P.D., Priestley, T., Tridgett, R., Iversen, L.L., and Woodruff, G.N. 1988. 7-Chlorokynurenic acid is a selective antagonist at the glycine modulatory site of the N-methyl-D-aspartate receptor complex. PNAS 85: 6547-6550

Larsen, R.S., Corlew, R.J., Henson, M.A., Roberts, A.C., Mishina, M., Watanabe, M., Lipton, S.A., Nakanishi, N., Perez-Otano, I., Weinberg, R.J., and Philpot, B.D. 2011. NR3A-containing NMDARs promote neurotransmitter release and spike timing-dependent plasticity. Nat Neurosci 14: 338-344

Nabavi, S., Kessels, H.W., Alfonso, S., Aow, J., Fox, R., and Malinow, R. 2013. Metabotropic NMDA receptor function is required for NMDA receptor-dependent long-term depression. PNAS 110: 4027-4032

Rodríguez-Moreno, A., Kohl, M.M., Reeve, J.E., Eaton, T.R., Collins, H.A., Anderson, H.L., and Paulsen, O. 2011. Presynaptic induction and expression of timing-dependent long-term depression demonstrated by compartment-specific photorelease of a use-dependent NMDA receptor antagonist. J Neurosci 31: 8564-8569

Rodríguez-Moreno, A., and Paulsen, O. 2008. Spike timing-dependent long-term depression requires presynaptic NMDA receptors. Nat Neurosci 11: 744-745

Sjöström, P.J., Turrigiano, G.G., and Nelson, S.B. 2003. Neocortical LTD via coincident activation of presynaptic NMDA and cannabinoid receptors. Neuron 39: 641-654

Song, S., Sjöström, P.J., Reigl, M., Nelson, S., and Chklovskii, D.B. 2005. Highly nonrandom features of synaptic connectivity in local cortical circuits. PLoS biology 3: e68

Tovar, K.R., and Westbrook, G.L. 1999. The incorporation of NMDA receptors with a distinct subunit composition at nascent hippocampal synapses in vitro. J Neurosci 19: 4180-4188

Wong, H.H., Watt, A.J., and Sjöström, P.J. 2024. Synapse-specific burst coding sustained by local axonal translation. Neuron 112: 264-276 e266

eLife Assessment

This study presents valuable findings that advance our understanding of mural cell dynamics and vascular pathology in a zebrafish model of cerebral small vessel disease. The authors provide compelling evidence that partial loss of foxf2 function leads to progressive, cell-intrinsic defects in pericytes and associated endothelial abnormalities across the lifespan, leveraging powerful in vivo imaging and genetic tools. The strength of evidence could be further improved by additional mechanistic insight and quantitative or lineage-tracing analyses to clarify how pericyte number and identity are affected in the mutant model.

Reviewer #1 (Public review):

Summary:

The paper by Graff et al. investigates the function of foxf2 in zebrafish to understand the progression of cerebral small vessel disease. The authors use a partial loss of foxf2 (zebrafish possess two foxf2 genes, foxf2a and foxf2b, and the authors mainly analyze homozygous mutants in foxf2a) to investigate the role of foxf2 signaling in regulating pericyte biology. They find that the number of pericytes is reduced in foxf2a mutants and that the remaining pericytes display alterations in their morphologies. The authors further find that mutant animals can develop to adulthood, but that in adult animals, both endothelial and pericyte morphologies are affected. They also show that mutant pericytes can partially repopulate the brain after genetic ablation.

Strengths:

The paper is well written and easy to follow.

Weaknesses:

The results are mainly descriptive, and it is not clear how they will advance the field at their current state, given that a publication on mice has already examined the loss of foxf2 phenotype on pericyte biology (Reyahi, 2015, Dev. Cell).

(1) Reyahi et al. showed that loss of foxf2 in mice leads to a marked downregulation of pdgfrb expression in perivascular cells. In contrast to expectation, perivascular cell numbers were higher in mutant animals, but these cells did not differentiate properly. The authors use a transgenic driver line expressing gal4 under the control of the pdgfrb promoter and observe a reduction in pericyte (pdgfrb-expressing) cells in foxf2a mutants. In light of the mouse data, this result might be due to a similar downregulation of pdgfrb expression in fish, which would lead to a downregulation of gal4 expression and hence reduced labelling of pericytes. The authors show a reduction of pdgfrb expression also in zebrafish in foxf2b mutants (Chauhan et al., The Lancet Neurology 2016). It would be important to clarify whether, also in zebrafish, foxf2a/foxf2b mutants have reduced or augmented numbers of perivascular cells and how this compares to the data in the mouse. The authors should perform additional characterization of perivascular cells using marker gene expression (for a list of markers, see e.g., Shih et al. Development 2021) and/or genetic lineage tracing.

(2) The authors motivate using foxf2a mutants as a model of reduced foxf2 dosage, "similar to human heterozygous loss of FOXF2". However, it is not clear how the different foxf2 genes in zebrafish interact with each other transcriptionally. Is there upregulation of foxf2b in foxf2a mutants and vice versa? This is important to consider, as Reyahi et al. showed that foxf2 gene dosage in mice appears to be important, with an increase in foxf2 gene dosage (through transgene expression) leading to a reduction in perivascular cell numbers.

(3) Figures 3 and 4 lack data quantification. The authors describe the existence of vascular defects in adult fish, but no quantifiable parameters or quantifications are provided. This needs to be added.

(4) The analysis of pericyte phenotypes and morphologies is not clear. On page 6, the authors state: "In the wildtype brain, adult pericytes have a clear oblong cell body with long, slender primary processes that extend from the cytoplasm with secondary processes that wrap around the circumference of the blood vessel." Further down on the same page, the authors note: "In wildtype adult brains, we identified three subtypes of pericytes, ensheathing, mesh and thin-strand, previously characterized in murine models." In conclusion, not all pericytes have long, slender primary processes, but there are at least three different sub-types? Did the authors analyze how they might be distributed along different branch orders of the vasculature, as they are in the mouse? Which type of pericyte is affected in foxf2a mutant animals? Can the authors identify the branch order of the vasculature for both wildtype and mutant animals and compare which subtype of pericyte might be most affected? Are all subtypes of pericytes similarly affected in mutant animals? There also seems to be a reduction in smooth muscle cell coverage.

(5) Regarding pericyte regeneration data (Figure 7): Are the values in Figure 7D not significantly different from each other (no significance given)?

(6) In the discussion, the authors state that "pericyte processes have not been studied in zebrafish". Ando et al. (Development 2016) studied pericyte processes in early zebrafish embryos, and Leonard et al. (Development 2022) studied zebrafish pericytes and their processes in the developing fin.

Reviewer #2 (Public review):

Summary:

This study investigates the developmental and lifelong consequences of reduced foxf2 dosage in zebrafish, a gene associated with human stroke risk and cerebral small vessel disease (CSVD). The authors show that a ~50% reduction in foxf2 function through homozygous loss of foxf2a leads to a significant decrease in brain pericyte number, along with striking abnormalities in pericyte morphology-including enlarged soma and extended processes-during larval stages. These defects are not corrected over time but instead persist and worsen with age, ultimately affecting the surrounding endothelium. The study also makes an important contribution by characterizing pericyte behavior in wild-type zebrafish using a clever pericyte-specific Brainbow approach, revealing novel interactions such as pericyte process overlap not previously reported in mammals.

Strengths:

This work provides mechanistic insight into how subtle, developmental changes in mural cell biology and coverage of the vasculature can drive long-term vascular pathology. The authors make strong use of zebrafish imaging tools, including longitudinal analysis in transgenic lines to follow pericyte number and morphology over larval development, and then applied tissue clearing and whole brain imaging at 3 and 11 months to further dissect the longitudinal effects of foxf2a loss. The ability to track individual pericytes in vivo reveals cell-intrinsic defects and process degeneration with high spatiotemporal resolution. Their use of a pericyte-specific Zebrabow line also allows, for the first time, detailed visualization of pericyte-pericyte interactions in the developing brain, highlighting structural features and behaviors that challenge existing models based on mouse studies. Together, these findings make the zebrafish a valuable model for studying the cellular dynamics of CSVD.

Weaknesses:

While the findings are compelling, several aspects could be strengthened. First, quantifying pericyte coverage across distinct brain regions (forebrain, midbrain, hindbrain) would clarify whether foxf2a loss differentially impacts specific pericyte lineages, given known regional differences in developmental origin, with forebrain pericytes being neural crest-derived and hindbrain pericytes being mesoderm-derived. Second, measuring foxf2b expression in foxf2a mutants would better support the interpretation that total FOXF2 dosage is reduced in a graded fashion in heterozygote and homozygote foxf2a mutants. Finally, quantifying vascular density in adult mutants would help determine whether observed endothelial changes are a downstream consequence of prolonged pericyte loss. Correlating these vascular changes with local pericyte depletion would also help clarify causality.

Reviewer #3 (Public review):

Summary:

The goal of the work by Graff et al. is to model CSVD in the zebrafish using foxf2a mutants. The mutants show loss of cerebral pericyte coverage that persists through adulthood, but it seems foxf2a does not regulate the regenerative capacity of these cells. The findings are interesting and build on previous work from the group. Limitations of the work include little mechanistic insight into how foxf2a alters pericyte recruitment/differentiation/survival/proliferation in this context, and the overlap of these studies with previous work in fox2a/b double mutants. However, the data analysis is clean and compelling, and the findings will contribute to the field.

eLife Assessment

This study identifies 53BP1 as an interaction partner of GMCL1 (a likely CUL3 substrate receptor). The study seeks to link this finding to regulation of the mitotic surveillance pathway and paclitaxel resistance in cancer. The evidence for these claims is currently inadequate; concerns include the use of cell lines that have been reported to lack the mitotic surveillance pathway, insufficient consideration of paclitaxel mechanisms of action, and an overinterpretation of correlative results.

Reviewer #1 (Public review):

In this manuscript, Pagano and colleagues test the idea that the protein GMCL1 functions as a substrate receptor for a Cullin RING 3 E3 ubiquitin ligase (CUL3) complex. Using a pulldown approach, they identify GMCL1 binding proteins, including the DNA damage scaffolding protein 53BP1. They then focus on the idea that GMCL1 recruits 53BP1 for CUL3-dependent ubiquitination, triggering subsequent proteasomal degradation of ubiquitinated 53BP1.

In addition to its DNA damage signalling function, in mitosis, 53BP1 is reported to form a stopwatch complex with the deubiquitinating enzyme USP28 and the transcription factor p53 (PMID: 38547292). These 53BP1-stopwatch complexes generated in mitosis are inherited by G1 daughter cells and help promote p53-dependent cell cycle arrest independent from DNA damage (PMID: 38547292). Several studies show that knockout of 53BP1 overcomes G1 cell cycle arrest after mitotic delays caused by anti-mitotic drugs or centrosome ablation (PMID: 27432897, 27432896). In this model, it is crucial that 53BP1 remains stable in mitosis and more stopwatch complex is formed after delayed mitosis.

Pagano and coworkers suggest that 53BP1 levels can sometimes be suppressed in mitosis if the cells overexpress GMCL1. They carry out a bioinformatic analysis of available public data for p53 wild-type cancer cell lines resistant to the anti-mitotic drug paclitaxel and related compounds. Stratifying GMCL1 into low and high expression groups reveals a weak (p = 0.05 or ns) correlation with sensitivity to taxanes. It is unclear on what basis the authors claim paclitaxel-resistant and p53 wild-type cancer cell lines bypass the mitotic surveillance/timer pathway. They have not tested this. Figure 3 is a correlation assembled from public databases but has no experimental tests. Figure 4 looks at proliferation but not cell cycle progression or the length of mitosis. The main conclusions relating to cell cycle progression and specifically the link to mitotic delays are therefore not supported by experimental data. There is no imaging of the cell cycle or cell fate after mitotic delays, or analysis of where the cells arrest in the cell cycle. Most of the cell lines used have been reported to lack a functional mitotic surveillance pathway in the recent work by Meitinger. To support these conclusions, the stability of endogenous 53BP1 under different conditions in cells known to have a functional mitotic surveillance pathway needs to be examined. A key suggestion in the work is that the level of GMCL1 expression correlates with resistance to taxanes. For the mitotic surveillance pathway, the type of drug (nocodazole, taxol, etc) used to induce a delay isn't thought to be relevant, only the length of the delay. Do GMCL1-overexpressing cells show resistance to anti-mitotics in general?

Importantly, if GMCL1 specifically degrades 53BP1 during prolonged mitotic arrests, the authors should show what happens during normal cell divisions without any delays or drug treatments. How much 53BP1 is destroyed in mitosis under those conditions? Does 53BP1 destruction depend on the length of mitosis, drug treatment, or does 53BP1 get degraded every mitosis regardless of length? Testing the contribution of key mitotic E3 ligase activities on mitotic 53BP1 stability, such as the anaphase-promoting complex/cyclosome (APC/C) is important in this regard. One previous study reported an analysis of putative APC/C KEN-box degron motifs in 53BP1 and concluded these play a role in 53BP1 stability in anaphase (PMID: 28228263).

There is no direct test of the proposed mechanism, and it is therefore unclear if 53BP1 is ubiquitinated by a GMCL1-CUL3 ligase in cells, and how efficient this process would be at different cell cycle stages. A key issue is the lack of experimental data explaining why the proposed mechanism would be restricted to mitosis. Indirect effects, such as loss of 53BP1 from the chromatin fraction during M phase upon GMCL1 overexpression, do not necessarily mean that 53BP1 is degraded. PLK1-dependent chromatin-cytoplasmic shuttling of 53BP1 during mitotic delays has been described previously (PMID: 38547292, 37888778). These papers are cited in the text, but the main conclusions of those papers on 53BP1 incorporation into a stopwatch complex during mitotic delays have been ignored. Are the authors sure that 53BP1 is destroyed in mitosis and not simply re-localised between chromatin and non-chromatin fractions? At the very least, these reported findings should be discussed in the text.

The authors use a variety of cancer cell line models throughout their study, most of which have been reported to lack a functional mitotic surveillance pathway. U2OS and HCT116 cells do not respond normally to mitotic delays, despite being annotated as p53 WT. Other studies have used p53 wild-type hTERT RPE-1 cells to study the mitotic surveillance pathway. If the model is correct, then over-expressing GMCL1 in hTERT-RPE1 cells should suppress cell cycle arrest after mitotic delays, and GMCL1 KO should make the cells more sensitive to delays. These experiments are needed to provide an adequate test of the proposed model.

To conclude, while the authors propose a potentially interesting model on how GMCL1 overexpression could regulate 53BP1 stability to limit p53-dependent cell cycle arrest, it is unclear what triggers this pathway or when it is relevant. 53BP1 is known to function in DNA damage signalling, and GMCL1 might be relevant in that context. The manuscript contains the initial description of GMCL1-53BP1 interaction but lacks a proper analysis of the function of this interaction and is therefore a preliminary report.

Reviewer #2 (Public review):

Summary:

This study investigates the role of GMCL1 in regulating the mitotic surveillance pathway (MSP), a protective mechanism that activates p53 following prolonged mitosis. The authors identify a physical interaction between 53BP1 and GMCL1, but not with GMCL2. They propose that the ubiquitin ligase complex CRL3-GMCL1 targets 53BP1 for degradation during mitosis, thereby preventing the formation of the "mitotic stopwatch" complex (53BP1-USP28-p53) and subsequent p53 activation. The authors show that high GMCL1 expression correlates with resistance to paclitaxel in cancer cell lines that express wild-type p53. Importantly, loss of GMCL1 restores paclitaxel sensitivity in these cells, but not in p53-deficient lines. They propose that GMCL1 overexpression enables cancer cells to bypass MSP-mediated p53 activation, promoting survival despite mitotic stress. Targeting GMCL1 may thus represent a therapeutic strategy to re-sensitize resistant tumors to taxane-based chemotherapy.

Strengths:

This manuscript presents potentially interesting observations. The major strength of this article is the identification of GMCL1 as a 53BP1 interaction partner. The authors identified relevant domains and showed that GMCL1 controls 53BP1 stability. The authors further show a potentially interesting link between GMCL1 status and sensitivity to Taxol.

Weaknesses:

However, the manuscript is significantly weakened by unsubstantiated mechanistic claims, overreliance on a non-functional model system (U2OS), and overinterpretation of correlative data. To support the conclusions of the manuscript, the authors must show that the GMCL1-dependent sensitivity to Taxol depends on the mitotic surveillance pathway.

Reviewer #3 (Public review):

Summary:

In this study, Kito et al follow up on previous work that identified Drosophila GCL as a mitotic substrate recognition subunit of a CUL3-RING ubiquitin ligase (CRL3) complex.

Here they characterize mutants of the human ortholog of GCL, GMCL1, that disrupt the interaction with CUL3 (GMCL1E142K) and that lack the substrate interaction domain (GMCL1 BBO). Immunoprecipitation followed by mass spectrometry identified 9 proteins that interacted with wild-type FLAG-GMCL1 and GMCL1 EK but not GMCL1 BBO. These proteins included 53BP1, which plays a well-characterized role in double-strand break repair but also functions in a USP28-p53-53BP1 "mitotic stopwatch" complex that arrests the cell cycle after a substantially prolonged mitosis. Consistent with the IP-MS results, FLAG-GMCL1 immunoprecipitated 53BP1. Depletion of GMCL1 during mitotic arrest increased protein levels of 53BP1, and this could be rescued by wild-type GMCL1 but not the E142K mutant or a R433A mutant that failed to immunoprecipitate 53BP1.

Using a publicly available dataset, the authors identified a relatively small subset of cell lines with high levels of GMCL1 mRNA that were resistant to the taxanes paclitaxel, cabazitaxel, and docetaxel. This type of analysis is confounded by the fact that paclitaxel and other microtubule poisons accumulate to substantially different levels in various cell lines (DOI: 10.1073/pnas.90.20.9552 , DOI: 10.1091/mbc.10.4.947 ), so careful follow-up experiments are required to validate results. The correlation between increased GMCL1 mRNA and taxane resistance was not observed in lung cancer cell lines. The authors propose this was because nearly half of lung cancers harbor p53 mutations, and lung cancer cell lines with wild-type but not mutant p53 showed the correlation between increased GMCL1 mRNA and taxane resistance. However, the other cancer cell types in which they report increased GMCL1 expression correlates with taxane sensitivity also have high rates of p53 mutation. Furthermore, p53 status does not predict taxane response in patients (DOI: 10.1002/1097-0142(20000815)89:4<769::aid-cncr8>3.0.co;2-6 , DOI: 10.1002/(SICI)1097-0142(19960915)78:6<1203::AID-CNCR6>3.0.CO;2-A , PMID: 10955790).

The authors then depleted GMCL1 and reported that it increased apoptosis in two cell lines with wild-type p53 (MCF7 and U2OS) due to activation of the mitotic stopwatch. This is surprising because the mitotic stopwatch paper they cite (DOI: 10.1126/science.add9528 ) reported that U2OS cells have an inactive stopwatch and that activation of the stopwatch results in cell cycle arrest rather than apoptosis in most cell types, including MCF7. Beyond this, it has recently been shown that the level of taxanes and other microtubule poisons achieved in patient tumors is too low to induce mitotic arrest (DOI: 10.1126/scitranslmed.3007965 , DOI: 10.1126/scitranslmed.abd4811 , DOI: 10.1371/journal.pbio.3002339 ), raising concerns about the relevance of prolonged mitosis to paclitaxel response in cancer. The findings here demonstrating that GMCL1 mediates degradation of 53BP1 during mitotic arrest are solid and of interest to cell biologists, but it is unclear that these findings are relevant to paclitaxel response in patients.

Strengths:

This study identified 53BP1 as a target of CRL3GMCL1-mediated degradation during mitotic arrest. AlphaFold3 predictions of the binding interface, followed by mutational analysis, identified mutants of each protein (GMCL1 R433A and 53BP1 IEDI1422-1425AAAA) that disrupted their interaction. Knock-in of a FLAG tag into the C-terminus of GMCL1 in HCT116 cells, followed by FLAG immunoprecipitation, confirmed that endogenous GMCL1 interacts with endogenous CUL3 and 53BP1 during mitotic arrest.

Weaknesses:

The clinical relevance of the study is overinterpreted. The authors have not taken relevant data about the clinical mechanism of taxanes into account. Supraphysiologic doses of microtubule poisons cause mitotic arrest and can activate the mitotic stopwatch. However, in physiologic concentrations of clinically useful microtubule poisons, cells proceed through mitosis and divide their chromosomes on mitotic spindles that are at least transiently multipolar. Though these low concentrations may result in a brief mitotic delay, it is substantially shorter than the arrest caused by high concentrations of microtubule poisons, and the one mimicked here by 16 hours of 0.4 mg/mL nocodazole, which is not used clinically and does not induce multipolar spindles. Resistance to mitotic arrest occurs through different mechanisms than resistance to multipolar spindles. No evidence is presented in the current version of the manuscript that GMCL1 affects cellular response to clinically relevant doses of paclitaxel.

eLife Assessment

This valuable study expands the inventory of polyadenylated RNAs cleaved by the double-stranded RNA endonuclease Rnt1 in budding yeast, using solid methodology based on high-throughput sequencing. Previous studies had anecdotally discovered mRNA substrates, and this global characterization is comprehensive with multiple complementary controls. However, the study would be stronger with a deeper investigation into the biological function of Rnt1, as well as experiments directly probing the interaction between Rnt1 and its putative substrates.

Reviewer #1 (Public review):

Strengths:

Sarpaning et al. provide a thorough characterization of putative Rnt1 cleavage of mRNA in S. cerevisiae. Previous studies have discovered Rnt1 mRNA substrates anecdotally, and this global characterization expands the known collection of putative Rnt1 cleavage sites. The study is comprehensive, with several types of controls to show that Rnt1 is required for several of these cleavages.

Weaknesses:

Formally speaking, the authors do not show a direct role of Rnt1 in mRNA cleavage - no studies were done (e.g., CLIP-seq or similar) to define direct binding sites. Is the mutant Rnt1 expected to trap substrates? Without direct binding studies, the authors rely on genetics and structure predictions for their argument, and it remains possible that a subset of these sites is an indirect consequence of rnt1. This aspect should be addressed in the discussion.

The comprehensive list of putative Rnt1 mRNA cleavage sites is interesting insofar as it expands the repertoire of Rnt1 on mRNAs, but the functional relevance of the majority of these sites remains unknown. Along these lines, the authors should present a more thorough characterization of putative Rnt1 sites recovered from in vitro Rnt1 cleavage.

The authors need to corroborate the rRNA 3'-ETS tetraloop mutations with a northern analysis of 3'-ETS processing to confirm an ETS processing defect (which might need to be done in decay mutants to stabilize the liberated ETS fragment). They state that the tetraloop mutation does not yield a growth defect and use this as the basis for concluding that rRNA cleavage is not the major role of Rnt1 in vivo, which is a surprising finding. But it remains possible that tetraloop mutations did not have the expected disruptive effect in vivo; if the ETS is processed normally in the presence of tetraloop mutations, it would undermine this interpretation. This needs to be more carefully examined.

To support the assertion that YDR514C cleavage is required for normal "homeostasis," and more specifically that it is the major contributor to the rnt1∆ growth defect, the authors should express the YDR514C-G220S mutant in the rDNA∆ strains with mutations in the 3'-ETS (assuming they disrupt ETS processing, see above). This simple experiment should provide a relative sense of "importance" for one or the other cleavage being responsible for the rnt1∆ defect. Given the accepted role of Rnt1 cleavage in rRNA processing and a dogmatic view that this is the reason for the rnt1∆ growth defect, such a result would be surprising and elevate the functional relevance and significance of Rnt1 mRNA cleavage.

Given that some Rnt1 mRNA cleavage is likely nuclear, it is possible that some of these targets are nascent mRNA transcripts, as opposed to mature but unexported mRNA transcripts, as proposed in the manuscript. A role for Rnt1 in co-transcriptional mRNA cleavage would be conceptually similar to Rnt1 cleavage of the rRNA 3'-ETS to enable RNA Pol I "torpedo" termination by Rat1, described by Proudfoot et al (PMID 20972219). To further delineate this point, the authors could e.g., examine the poly-A tails on abundant Rnt1 targets to establish whether they are mature, polyadenylated mRNAs (e.g., northern analysis of oligo-dT purified material). A more direct test would be PARE analysis of oligo-dT enriched or depleted material to determine the poly-A status of the cleavage products. Alternatively, their association with chromatin could be examined.

While laboratory strains of budding yeast have a single RNase III ortholog Rnt1, several other budding yeast have a functional RNAi system with Dcr and Ago (PMID 19745116), and laboratory yeast strains are a derived state due to pressure from the killer virus to lose the RNAi system (PMID 21921191). The current study could provide new insight into the relative substrate preferences of Rnt1 and budding yeast Dicer, which could be experimentally confirmed by expressing Dcr in RNT1 and rnt1∆ strains. In lieu of experiments, discussion of the relevance of Rnt1 cleavage compared to yeast RNAi should be included in the discussion before the "human implications" section.

For SNR84 in Figure S3D, it appears that the TSS may be upstream of the annotated gene model. Does RNA-seq coverage (from external datasets) extend upstream to these additional mapped cleavages? The assertion that the mRNA is uncapped is concerning; an alternative explanation is that the nascent mRNA has a cap initially but is subsequently cleaved by Rnt1. This point should be clarified or reworded for accuracy.

Reviewer #2 (Public review):

The yeast double-stranded RNA endonuclease Rnt1, a homolog of bacterial RNAse III, mediates the processing of pre-rRNA, pre-snRNA, and pre-snoRNA molecules. Cells lacking Rnt1 exhibit pronounced growth defects, particularly at lower temperatures. In this manuscript, Notice-Sarpaning examines whether these growth defects can be attributed at least in part to a function of Rnt1 in mRNA degradation. To test this, the authors apply parallel analysis of RNA ends (PARE), which they developed in previous work, to identify polyA+ fragments with 5' monophosphates in RNT1 yeast that are absent in rnt1Δ cells. Because such RNAs are substrates for 5' to 3' exonucleolytic decay by Rat1 in the nucleus or Xrn1 in the cytoplasm, these analyses were performed in a rat1-ts xrn1Δ background. The data recapitulate known Rtn1 cleavage sites in rRNA, snRNAs, and snoRNAs, and identify 122 putative novel substrates, approximately half of which are mRNAs. Of these, two-thirds are predicted to contain double-stranded stem loop structures with A/UGNN tetraloops, which serve as a major determinant of Rnt1 substrate recognition. Rtn1 resides in the nucleus, and it likely cleaves mRNAs there, but cleavage products seem to be degraded after export to the cytoplasm, as analysis of published PARE data shows that some of them accumulate in xrn1Δ cells. The authors then leverage the slow growth of rnt1Δ cells for experimental evolution. Sequencing analysis of thirteen faster-growing strains identifies mutations predominantly mapping to genes encoding nuclear exosome co-factors. Some of the strains have mutations in genes encoding a larat-debranching enzyme, a ribosomal protein nuclear import factor, poly(A) polymerase 1, and the RNA-binding protein Puf4. In one of the puf4 mutant strains, a second mutation is also present in YDR514C, which the authors identify as an mRNA substrate cleaved by Rnt1. Deletion of either puf4 or ydr514C marginally improves the growth of rnt1Δ cells, which the authors interpret as evidence that mRNA cleavage by Rnt1 plays a role in maintaining cellular homeostasis by controlling mRNA turnover.

While the PARE data and their subsequent in vitro validation convincingly demonstrate Rnt1-mediated cleavage of a small subset of yeast mRNAs, the data supporting the biological significance of these cleavage events is substantially less compelling. This makes it difficult to establish whether Rnt1-mediated mRNA cleavage is biologically meaningful or simply "collateral damage" due to a coincidental presence of its target motif in these transcripts.

(1) A major argument in support of the claim that "several mRNAs rely heavily on Rnt1 for turnover" comes from comparing number of PARE reads at the transcript start site (as a proxy for fraction of decapped transcripts) and at the Rnt1 cleavage site (as a proxy for fraction of Rnt1-cleaved transcripts). The argument for this is that "the major mRNA degradation pathway is through decapping". However, polyA tail shortening usually precedes decapping, and transcripts with short polyA tails would be strongly underrepresented in PARE sequencing libraries, which were constructed after two rounds of polyA+ RNA selection. This will likely underestimate the fraction of decapped transcripts for each mRNA. There is a wide range of well-established methods that can be used to directly measure differences in the half-life of Rnt1 mRNA targets in RNT1 vs rnt1Δ cells. Because the PARE data rely on the presence of a 5' phosphate to generate sequencing reads, they also cannot be used to estimate what fraction of a given mRNA transcript is actually cleaved by Rnt1.

(2) Rnt1 is almost exclusively nuclear, and the authors make a compelling case that its concentration in the cytoplasm would likely be too low to result in mRNA cleavage. The model for Rnt1-mediated mRNA turnover would therefore require mRNAs to be cleaved prior to their nuclear export in a manner that would be difficult to control. Alternatively, the Rnt1 targets would need to re-enter prior to cleavage, followed by export of the cleaved fragments for cytoplasmic decay. These processes would need to be able to compete with canonical 5' to 3' and 3' to 5' exonucleolytic decay to influence mRNA fate in a biologically meaningful way.

(3) The experimental evolution clearly demonstrates that mutations in nuclear exosome factors are the most frequent suppressors of the growth defects caused by Rnt1 loss. This can be rationalized by stabilization of nuclear exosome substrates such as misprocessed snRNAs or snoRNAs, which are the major targets of Rnt1. The rescue mutations in other pathways linked to ribosomal proteins (splicing, ribosomal protein import, ribosomal mRNA binding) support this interpretation. By contrast, the potential suppressor mutation in YDR514C does not occur on its own but only in combination with a puf4 mutation; it is also unclear whether it is located within the Rnt1 cleavage motif or if it impacts Rnt1 cleavage at all. This can easily be tested by engineering the mutation into the endogenous YDR514C locus with CRISPR/Cas9 or expressing wild-type and mutant YDR514C from a plasmid, along with assaying for Rnt1 cleavage by northern blot. Notably, the growth defect complementation of YDR514C deletion in rnt1Δ cells is substantially less pronounced than the growth advantage afforded by nuclear exosome mutations (Figure S9, evolved strains 1 to 5). These data rather argue for a primary role of Rnt1 in promoting cell growth by ensuring efficient ribosome biogenesis through pre-snRNA/pre-snoRNA processing.

eLife Assessment

Complex traits are influenced by genes and the environment, but especially the latter is difficult to pin down. This important study uses C. elegans to demonstrate that non-genetic differences in gene expression, partly influenced by the environment, correlate with individual differences in two reproductive traits. This supports the use of gene expression data as a key intermediate for understanding complex traits. The clever study design makes for compelling evidence, which is further strengthened by experimental confirmation that identified differentially expressed genes indeed influence these traits.

Reviewer #1 (Public review):

Summary:

Genome-wide association studies have been an important approach to identifying the genetic basis of human traits and diseases. Despite their successes, for many traits, a substantial amount of variation cannot be explained by genetic factors, indicating that environmental variation and individual 'noise' (stochastic differences as well as unaccounted for environmental variation) also play important roles. The authors' goal was to address whether gene expression variation in genetically identical individuals, driven by historical environmental differences and 'noise', could be used to predict reproductive trait differences.

Strengths:

To address this question, the authors took advantage of genetically identical C. elegans individuals to transcriptionally profile 180 adult hermaphrodite individuals that were also measured for two reproductive traits. A major strength of the paper is its experimental design. While experimenters aim to control the environment that each worm experiences, it is known that there are small differences that each worm experiences even when they are grown together on the same agar plate - e.g. the age of their mother, their temperature, the amount of food they eat, and the oxygen and carbon dioxide levels depending on where they roam on the plate. Instead of neglecting this unknown variation, the authors design the experiment up front to create two differences in the historical environment experienced by each worm: 1) the age of its mother and 2) 8 8-hour temperature difference, either 20 or 25 {degree sign}C. This helped the authors interpret the gene expression differences and trait expression differences that they observed.

Using two statistical models, the authors measured the association of gene expression for 8824 genes with the two reproductive traits, considering both the level of expression and the historical environment experienced by each worm. Their data supports several conclusions. They convincingly show that gene expression differences are useful for predicting reproductive trait differences, predicting ~25-50% of the trait differences depending on the trait. Using RNAi, they also show that the genes they identify play a causal role in trait differences. Finally, they demonstrate an association with trait variation and the H3K27 trimethylation mark, suggesting that chromatin structure can be an important causal determinant of gene expression and trait variation.

Overall, this work supports the use of gene expression data as an important intermediate for understanding complex traits. This approach is also useful as a starting point for other labs in studying their trait of interest.

Weaknesses:

There are no major weaknesses that I have noted. Some important limitations of the work (that I believe the authors would agree with) are worth highlighting, however:

(1) A large remaining question in the field of complex traits remains in splitting the role of non-genetic factors between environmental variation and stochastic noise. It is still an open question which role each of these factors plays in controlling the gene expression differences they measured between the individual worms.

(2) The ability of the authors to use gene expression to predict trait variation was strikingly different between the two traits they measured. For the early brood trait, 448 genes were statistically linked to the trait difference, while for egg-laying onset, only 11 genes were found. Similarly, the total R2 in the test set was ~50% vs. 25%. It is unclear why the differences occur, but this somewhat limits the generalizability of this approach to other traits.

(3) For technical reasons, this approach was limited to whole worm transcription. The role of tissue and cell-type expression differences is important to the field, so this limitation is important.

Reviewer #2 (Public review):

Summary:

This paper measures associations between RNA transcript levels and important reproductive traits in the model organism C. elegans. The authors go beyond determining which gene expression differences underlie reproductive traits, but also (1) build a model that predicts these traits based on gene expression and (2) perform experiments to confirm that some transcript levels indeed affect reproductive traits. The clever study design allows the authors to determine which transcript levels impact reproductive traits, and also which transcriptional differences are driven by stochastic vs environmental differences. In sum, this is a rather comprehensive study that highlights the power of gene expression as a driver of phenotype, and also teases apart the various factors that affect the expression levels of important genes.

Strengths:

Overall, this study has many strengths, is very clearly communicated, and has no substantial weaknesses that I can point to. One question that emerges for me is about the extent to which these findings apply broadly. In other words, I wonder whether gene expression levels are predictive of other phenotypes in other organisms. I think this question has largely been explored in microbes, where some studies (PMID: 17959824) but not others (PMID: 38895328) find that differences in gene expression are predictive of phenotypes like growth rate. Microbes are not the primary focus here, and instead, the discussion is mainly focused on using gene expression to predict health and disease phenotypes in humans. This feels a little complicated since humans have so many different tissues. Perhaps an area where this approach might be useful is in examining infectious single-cell populations (bacteria, tumors, fungi). But I suppose this idea might still work in humans, assuming the authors are thinking about targeting specific tissues for RNAseq.

In sum, this is a great paper that really got me thinking about the predictive power of gene expression and where/when it could inform about (health-related) phenotypes.

Reviewer #3 (Public review):

Summary:

Webster et al. sought to understand if phenotypic variation in the absence of genetic variation can be predicted by variation in gene expression. To this end they quantified two reproductive traits, the onset of egg laying and early brood size in cohorts of genetically identical nematodes exposed to alternative ancestral (two maternal ages) and same generation life histories (either constant 20C temperature or 8-hour temperature shift to 25C upon hatching) in a two-factor design; then they profiled genome-wide gene expression in each individual.

Using multiple statistical and machine learning approaches, they showed that, at least for early brood size, phenotypic variation can be quite well predicted by molecular variation, beyond what can be predicted by life history alone.

Moreover, they provide some evidence that expression variation in some genes might be causally linked to phenotypic variation.

Strengths:

(1) Cleverly designed and carefully performed experiments that provide high-quality datasets useful for the community.

(2) Good evidence that phenotypic variation can be predicted by molecular variation.

Weaknesses:

What drives the molecular variation that impacts phenotypic variation remains unknown. While the authors show that variation in expression of some genes might indeed be causal, it is still not clear how much of the molecular variation is a cause rather than a consequence of phenotypic variation.

eLife Assessment

This useful study presents the potentially interesting concept that LRRK2 regulates cellular BMP levels and their release via extracellular vesicles, with GCase activity further modulating this process in mutant LRRK2-expressing cells. However, the evidence supporting the conclusions remains incomplete, and certain statistical analyses are inadequate. This work would be of interest to cell biologists working on Parkinson's disease.

Reviewer #1 (Public review):

Summary:

Even though mutations in LRRK2 and GBA1 (which encodes the protein GCase) increase the risk of developing Parkinson's disease (PD), the specific mechanisms driving neurodegeneration remain unclear. Given their known roles in lysosomal function, the authors investigate how LRRK2 and GCase activity influence the exocytosis of the lysosomal lipid BMP via extracellular vesicles (EVs). They use fibroblasts carrying the PD-associated LRRK2-R1441G mutation and pharmacologically modulate LRRK2 and GCase activity.

Strengths:

The authors examine both proteins at endogenous levels, using MEFs instead of cancer cells. The study's scope is potentially interesting and could yield relevant insights into PD disease mechanisms.

Weaknesses:

Many of the authors' conclusions are overstated and not sufficiently supported by the data. Several statistical errors undermine their claims. Pharmacological treatment is very long, leading to potential off-target effects. Additionally, the authors should be more rigorous when using EV markers.

Reviewer #2 (Public review):

Summary:

In this paper, the authors used MEFs expressing the R1441G mutant of leucine-rich repeat kinase 2 (LRRK2), a mutant associated with the early onset of Parkinson's disease. They report that in these cells LAMP2 fluorescence is higher but BMP fluorescence is lower, MVE size is reduced, and that MVEs contain less ILVs. They also report that LAMP2-positive EVs are increased in mutant cells in a process sensitive to LRRK2 kinase inhibition but are further increased by glucocerebrosidase (GCase) inhibition, and that total di-22:6-BMP and total di-18:1-BMP are increased in mutant LRRK2 MEFs compared to WT cells by mass spectrometry. They also report that LRRK2 kinase inhibition partially restores cellular BMP levels, and that GCase inhibition further increases BMP levels, and that in EVs from the LRRK2 mutant, LRRK2 inhibition decreases BMP while GCase inhibition has the opposite effect. Moreover, they report that the BMP increase is not due to increased BMP synthesis, although the authors observe that CLN5 is increased in LRRK2 mutant cells. Finally, they report that GW4869 decreases EV release and exosomal BMP, while bafilomycin A1 increases EV release. They conclude that LRRK2 regulates BMP levels (in cells) and release (via EVs). They also conclude that the process is modulated by GCase in LRRK2 mutant cells, and that these studies may contribute to the use of BMP-positive EVs as a biomarker for Parkinson's disease and associated treatments.

Strengths:

This is an interesting paper, which provides novel insights into the biogenesis of exosomes with exciting biomedical potential. However, I have comments that authors need to address to clarify some aspects of their study.

Weaknesses:

(1) The intensity of LAMP2 staining is increased significantly in cells expressing the R1441G mutant of LRRK2 when compared to WT cells (Figure 1C). Yet mutant cells contain significantly smaller MVEs with fewer ILVs, and the MVE surface area is reduced (Figure 1D-F). This is quite surprising since LAMP2 is a major component of the limiting membrane of late endosomes. Are other proteins of endo-lysosomes (eg, LAMP1, CD63, RAB7) or markers (lysotracker) also decreased (see also below)?

(2) LRRK2 has been reported to interact with endolysosomal membranes. Does the R1441G mutant bind LAMP2- and/or BMP-positive membranes? Does the mutant affect endolysosomes?

(3) Immunofluorescence data indicate that BMP is decreased in mutant LRRK2-expressing cells compared to WT (Figure 1A-B), but mass spec data indicate that di-22:6-BMP and di-18:1-BMP are increased (Figure 3). Authors conclude that the BMP pool detected by mass spec in mutant cells is less antibody-accessible than that present in wt cells, or that the anti-BMP antibody is less specific and that it detects other analytes. This is an awkward conclusion, since the IF signal with the antibody is lower (not higher): why would the antibody be less specific? Could it be that the antibody does not see all BMP isoforms equally well? Moreover, the observations that mutant cells contain smaller MVEs (Figure 1D-F) with fewer ILVs are consistent with the IF data and reduced BMP amounts. This needs to be clarified.

Mass spectrometry data are only shown for two BMP species (di-22:6, di-18:1). What are the major BMP isoforms in WT cells? The authors should show the complete analysis for all BMP species if they wish to draw quantitative conclusions about the amounts of BMP in wt and mutant cells. Finally, BMP and PG are isobaric lipids. Fragmentation of BMPs or PGs results in characteristic fingerprints, but the presence of each daughter ion is not absolutely specific for either lipid. This should be clarified, e.g., were BMP and PG separated before mass spec analysis? Was PG affected? The authors should also compare the BMP data with mass spec data obtained with a control lipid, e.g., PC.

(4) It is quite surprising that the amounts of labeled BMP continue to increase for up to 24h after a short 25min pulse with heavy BMP precursors (Figure 4B).

(5) It is argued that upregulation of CLN5 may be due to an overall upregulation of lysosomal enzymes, as LAMP2 levels were also increased (Figure 2A, C, E). Again, this is not consistent with the observed decrease in MVE size and number (Figure 1D-F). As mentioned above, other independent markers of endo-lysosomes should be analyzed (eg, LAMP1, CD63, RAB7), and/or other lysosomal enzymes (e.g. cathepsin. D).

(6) The authors report that the increase in BMP is not due to an increase in BMP synthesis (Figure 4), although they observe a significant increase in CLN5 (Figure 5A) in LRRK2 mutant cells. Some clarification is needed.

(7) Authors observe that both LAMP2 and BMP are decreased in EVs by GW4869 and increased by bafilomycin (Figure 6). Given my comments above on Figure 1, it would also be nice to illustrate/quantify the effects of these compounds on cells by immunofluorescence.

eLife Assessment

This fundamental study advances our understanding of the role that energy metabolism, specifically anaerobic glycolysis, plays during retinal development. Convincing in vitro genetic and pharmacological evidence demonstrates that glycolytic flux controls retinal progenitor cell proliferation rates and the timing of photoreceptor maturation. Interesting evidence suggests potential downstream roles for intracellular pH and Wnt/β-catenin signaling; however, more direct evidence is needed to show they are the key mechanisms through which glycolytic flux regulates retinogenesis in vivo. This work is expected to stimulate broad interest and possible future studies investigating the link between metabolism and development in other tissue systems.

[Editors’ note: Primary data for this manuscript are not available due to a corrupted hard drive that occurred during the course of peer review. However, preprocessed data are available.]

Reviewer #1 (Public review):

Summary:

This paper seeks to understand the upstream regulation and downstream effectors of glycolysis in retinal progenitor cells, using mouse retinal explants as the main model system. The paper presents evidence that high glycolysis in retinal progenitor cells is required for their proliferation and timely differentiation into photoreceptors. Retinal glycolysis increases after deletion of Pten. The authors suggest that high glycolysis controls cell proliferation and differentiation by promoting intracellular alkalinization, beta-catenin acetylation and stabilization and consequent activation of the canonical Wnt pathway.

Strengths:

- The experiments showing that PFKFB3 overexpression is sufficient to increase proliferation of retinal progenitors (which are already highly dividing cells) and photoreceptor differentiation are striking and the result unanticipated. It suggests that glycolytic flux is normally limiting for proliferation in embryos.<br /> - Likewise the result that an increase in pH from 7.4 to 8.0 is sufficient to increase proliferation implies that pH regulation may have instructive roles in setting the tempo of retinal development and embryonic cell proliferation. Similarly for the results showing that acetate supplementation increases proliferation (I think this result should be moved to the main figures).

Weaknesses:

- Epistatic experiments to test if changes in pH mediate the effects of glycolysis on photoreceptor differentiation, or if Wnt activation is the main downstream effector of glycolysis in controlling differentiation are not presented.<br /> - It is likely that metabolism changes ex vivo vs in vivo, and therefore stable isotope tracing experiments in the explants may not reflect in vivo metabolism.<br /> - The retina at P0 is composed of both progenitors and differentiated cells. It is not clear if the results of the RNA-seq and metabolic analysis reflect changes in the metabolism of progenitors, or of mature cells, or changes in cell type composition rather than direct metabolic changes in a specific cell type.<br /> - The biochemical links between elevated glycolysis and pH and beta-catenin stability are unclear. White et al found that higher pH decreased beta-catenin stability (JCB 217: 3965) in contrast to the results here. Oginuma et al found that inhibition of glycolysis or beta-catenin acetylation does not affect beta-catenin stability (Nature 584:98), again in contrast to these results. Another paper showed that acidification inhibits Wnt signaling by promoting the expression of a transcriptional repressor and not via beta-catenin stability (Cell Discovery 4:37). There are also additional papers showing increased pH can promote cell proliferation via other mechanisms (e.g. Nat Metab 2:1212). It is possible that there is organ-specificity in these signaling pathways however some clarification of these divergent results is warranted.<br /> - The gene expression analysis is not completely convincing. E.g. expression of additional glycolytic genes should be shown in Fig. 1. It is not clear why Hk1 and Pgk1 are specifically shown, and conclusions about changes in glycolysis are difficult to draw from expression of these two genes. The increase in glycolytic gene expression in the Pten-deficient retina is generally small.<br /> - Is it possible that glycolytic inhibition with 2DG slows down development and production of most new differentiated cells rather than specifically affecting photoreceptor differentiation?<br /> - Are the prematurely-born cells caused by PFKFB3 overexpression photoreceptors as assessed by morphology or markers (in addition to position)?

Reviewer #3 (Public review):

Summary:

This study examines the metabolic regulation of progenitor proliferation and differentiation in the developing retina. The authors observe dynamic changes in glycolytic gene expression in retinal progenitors and use various strategies to test the role of glycolysis. They find that elevated glycolysis in Pten-cKO retinas results in alteration of RPC fate, while inhibition of glycolysis has converse effects. They specifically test the role of elevated glycolysis using dominant active cytoPFKB3, which demonstrates the selective effects of elevated glycolysis on progenitor proliferation and rod differentiation. They then show that elevated glycolysis modulates both pHi and Wnt signaling, and provide evidence that these pathways impact proliferation and differentiation of progenitors, particularly affecting rod photoreceptor differentiation.

Strengths:

This is a compelling and rigorous study that provides an important advance in our understanding of metabolic regulation of retina development, addressing a major gap in knowledge. A key strength is that the study utilizes multiple genetic and pharmacological approaches to address how both increased or decreased glycolytic flux affect retinal progenitor proliferation and differentiation. They discover elevated Wnt signaling pathway genes in Pten cKO retina, revealing a potential link between glycolysis and Wnt pathway activation. Altogether the study is comprehensive and adds to the growing body of evidence that regulation of glycolysis plays a key role in tissue development.

Weaknesses:

(1) Following expression of cytoPFKB3, which results in increased glycolytic flux, BrDU labeling was performed at e12.5 and increased labeled cells were detected in the outer nuclear layer. But whether these are cones or rods is not established. The rest of the analysis is focused on the precocious maturation of rhodopsin-labelled outer segments, and the major conclusions emphasize rod photoreceptor differentiation. Therefore it is unclear whether there is an effect on cone differentiation for either Pten cKO or cytoPFKB3 transgenic retina. It is also not established whether rods are born precociously. Presumably this would be best detected by BrDU labeling at later embryonic stages.

(2) The authors find that there is upregulation of multiple Wnt pathway components in Pten cKO retina. They further show that inhibiting Wnt signaling phenocopies the effects of reducing glycolysis. However, they do not test whether pharmacological inhibition of Wnt signaling reverses the effects of high glycolytic activity in Pten cKO retinas. Thus the argument that Wnt is a key downstream effector pathway regulating rod photoreceptor differentiation is weak.

(3) The use of sodium acetate to force protein acetylation is quite non-specific and will have effects beyond beta-catenin acetylation (which the authors acknowledge). Thus it is a stretch to state that "forced activation of beta-catenin acetylation" mimics the impact of Pten<br /> loss/high glycolytic activity in RPCs since the effects could be due to acetylation of other proteins.

Author response:

The following is the authors’ response to the original reviews

Reviewer 1 (Public review):

(1) “It is likely that metabolism changes ex vivo vs in vivo, and therefore stable isotope tracing experiments in the explants may not reflect in vivo metabolism.”

We agree with the reviewer that metabolic changes may differ ex vivo versus in vivo. We now state: “Lastly, an important caveat to our study is that metabolism changes ex vivo versus in vivo, and thus, in the future, in vivo studies can be performed to assess metabolic changes.” (lines 591-593).

(2) “The retina at P0 is composed of both progenitors and differentiated cells. It is not clear if the results of the RNA-seq and metabolic analysis reflect changes in the metabolism of progenitors, or of mature cells, or changes in cell type composition rather than direct metabolic changes in a specific cell type.”

We have clarified that the metabolic changes may be in RPCs or in other retinal cell types on lines 149-152: “Since these measurements were performed in bulk, and the ratio of RPCs to differentiated cells declines as development proceeds, it is not clear whether glycolytic activity is temporally regulated within RPCs or in other retinal cell types.”

However, since we mined a single cell (sc) RNA-seq dataset, we are able to attribute gene expression specifically within RPCs (Figure 1).

(3) “The biochemical links between elevated glycolysis and pH and beta-catenin stability are unclear. White et al found that higher pH decreased beta-catenin stability (JCB 217: 3965) in contrast to the results here. Oginuma et al found that inhibition of glycolysis or beta-catenin acetylation does not affect beta-catenin stability (Nature 584:98), again in contrast to these results. Another paper showed that acidification inhibits Wnt signaling by promoting the expression of a transcriptional repressor and not via beta-catenin stability (Cell Discovery 4:37). There are also additional papers showing increased pH can promote cell proliferation via other mechanisms (e.g. Nat Metab 2:1212). It is possible that there is organ-specificity in these signaling pathways however some clarification of these divergent results is warranted.”

We have added the information and references brought up by the reviewer in our discussion (lines 529-549 and 570-574). We have also suggested future experiments to further analyse our system in line with the studies now referenced (lines 580-589).

(4) The gene expression analysis is not completely convincing. E.g. the expression of additional glycolytic genes should be shown in Figure 1. It is not clear why Hk1 and Pgk1 are specifically shown, and conclusions about changes in glycolysis are difficult to draw from the expression of these two genes. The increase in glycolytic gene expression in the Pten-deficient retina is generally small.

We have expanded the list of glycolytic genes analysed, in modified Figure 1B, and expanded the description of these results on lines 156-166.

(5) Is it possible that glycolytic inhibition with 2DG slows down the development and production of most newly differentiated cells rather than specifically affecting photoreceptor differentiation?

We added a comment to this effect to the discussion: “It is possible that glycolytic inhibition with 2DG slows down the development and production of most newly differentiated cells rather than specifically affecting photoreceptor differentiation, which we could assess in the future.“ (lines 600-603).

(6) “Likewise the result that an increase in pH from 7.4 to 8.0 is sufficient to increase proliferation implies that pH regulation may have instructive roles in setting the tempo of retinal development and embryonic cell proliferation. Similarly, the results show that acetate supplementation increases proliferation (I think this result should be moved to the main figures).”

We have added the acetate data to main Figure 7E.

We added a supplemental data table that was inadvertently not included in our last submission. Figure 2– Data supplement 1.

Reviewer #2 (Recommendations for the authors):

Major points

(1) Assuming that increased glycolysis gets RPCs to exit from the proliferative stage earlier, the total number of retinal cells, notably that of the rod photoreceptors, should be reduced since the pool of proliferating cells is depleted earlier. Is that really the case for a mature retina? To address this question, the authors should perform quantifications of photoreceptors at a stage where most developmental cell death has concluded (i.e. at P14 or later; Young, J. Comp. Neurol. 229:362-373, 1984) and check whether or not there are more or less photoreceptors present.

We have previously quantified numbers of each cell type in Pten RPC-cKO retinas, and as suggested by the reviewer, there are fewer rod photoreceptors at P7 (Tachibana et al. 2016. J Neurosci 36 (36) 9454-9471) and P21 (Hanna et al. 2025. IOVS. Mar 3;66(3):45). We have edited the following sentence: “Using cellular birthdating, we previously showed that Pten-cKO RPCs are hyperproliferative and differentiate on an accelerated schedule between E12.5 and E18.5, yet fewer rod photoreceptors are ultimately present in P7 (Tachibana et al., 2016) and P21 (Hanna et al., 2025) retinas, suggestive of a developmental defect. (lines 184-187).

(2) Figure 1B, 1H: On what data are these two figures based? The plots suggest that a high-density time series of gene expression and rod photoreceptor birth was performed, yet it is not clear where and how this was done. The authors should provide the data, plot individual data points, and, if applicable perform a statistical analysis to support their idea that glycolytic gene expression (as a surrogate for glycolysis) overlaps in time with rod photoreceptor birth (Figure 1B) and that in Pten KO the glycolytic gene expression is shifted forward in time (Figure 1H). If the data required to construct these plots (min. 5 data points, min 3 repeats each) does not exist or cannot be generated (e.g. from reanalysis of previously published datasets), then these graphs should be removed.

We have removed the previous Figure 1B and Figure 1H.

(3) Figure 2E: Which PKM isozyme was analyzed here? Does the genetic analysis allow us to distinguish between PKM1 and PKM2? Since PKM governs the key rate-limiting step of glycolysis but was not significantly upregulated, does this not contradict the authors' main hypothesis? If PKM at some point was inhibited (see also below comment to Figure 5) one would expect an accumulation of glycolytic intermediates, including phosphoenolpyruvate. Was such an effect observed?

The data in Figure 2E is bulk RNA-seq data. Since there is only a single Pkm gene that is alternatively spliced, the RNA-sequencing data cannot distinguish between the four PK isozymes that arise from alternative splicing. Specifically, we used Illumina NextSeq 500 for sequencing of 75bp Single-End reads that will sequence transcripts for alternatively spliced Pkm1 and Pkm2 mRNAs, which carry a common 3’end. We added a statement to this effect: “However, since we employed 75 bp single-end sequencing, we could not distinguish between alternatively spliced Pkm1 and Pkm2 mRNAs.“ (lines 215-216).

We have not performed metabolic analyses of glycolytic intermediates, but we have proposed such a strategy as an important avenue of investigation for future studies in the Discussion: “Lastly, an important caveat to our study is that metabolism changes ex vivo versus in vivo, and thus, in the future, in vivo studies can be performed to assess metabolic changes.” (lines 591-593).

(4) Figure 3 and materials & methods: For the retinal explant cultures, was the RPE included in the cultured explants? If so, how can the authors distinguish drug effects on neuroretina and RPE? If the RPE was not included, then the authors should discuss how the missing RPE - neuroretina interaction could have influenced their results.

We remove the RPE from the retinal explants, as indicated in the Methods section. The RPE is a metabolic hub that allows transport of nutrients for the retina, so in the absence of the RPE, there is not an immediate source of energy, such as glucose, to the retina. However, the media (DMEM) contains 25 mM glucose to replace the RPE as an energy source, and we now show that RPCs express GLUT1, which allows uptake of glucose (see new Figure 3A).

We added the following sentence “P0 explants were mounted on Nucleopore membranes and cultured on top of retinal explant media, providing a source of nutrients, growth factors and glucose. “(lines 241-243).

(5) Figure 3: It seems rather odd that, if glycolysis was so important for retinal proliferation, differentiation, and metabolism in general, the inhibition of glycolysis with 2DG should not produce a strong degeneration. However, since 2DG competes with glucose, and must be used at nearly equimolar concentration to block glycolysis in a meaningful way, it is possible that the 2DG concentration used simply was not high enough to substantially inhibit glycolysis. Since the inhibitory effect of 2DG depends on the glucose concentration, the authors should measure and provide the concentration of glucose in the explant culture medium. This value should be given either in results or materials and methods.

We recently published a manuscript showing that 2DG treatments at the same concentrations employed in this study are effective at reducing lactate production in the developing retina in vivo, which is the expected effect of reduced glycolysis (Hanna et al. 2025. IOVS). However, in this study, we did not observe an impact on cell survival.

We do not agree that it is necessary to measure glucose in the media since the anti-proliferative effect of 2DG is well known, and we are working in the effective range established by multiple groups. We have clarified that we are in the effective range by adding the following sentences: “2DG is typically used in the range of 5-10 mM in cell culture studies and in general, has anti-proliferative effects. To test whether 2DG treatment was in the effective range, explants were exposed to BrdU, which is incorporated into S-phase cells, for 30 minutes prior to harvesting. 2DG treatment resulted in a dose-dependent inhibition of RPC proliferation as evidenced by a reduction in BrdU⁺ cells (Figure 3D), indicating that our treatment was in the effective range.” (lines 246-251).

(6) Figure 3F: The authors use immunostaining for cleaved, activated caspase-3 to assess the amount of apoptotic cell death. However, there are many different possible mechanisms for neuronal cells to die, the majority of which are caspase-independent. To assess the amount of cell death occurring, the authors should perform a TUNEL assay (which labels apoptotic and non-apoptotic forms of cell death; Grasl-Kraupp et al., Hepatology 21:1465-8, 1995), quantify the numbers of TUNEL-positive cells in the retina, and compare this to the numbers of cells positive for activated caspase-3.

We agree with the reviewer that there are more ways for a cell to die than just apoptosis, and TUNEL would pick up dying cells that may undergo apoptosis or necrosis, for example, our data with cleaved caspase-3, an executioner protease for apoptosis, provides us with clear evidence of cell death in our different conditions. Since this manuscript is not focused on cell death pathways, we have not performed the additional TUNEL assay.

(7) Figure 4F and 4I: At post-natal day P7 the rod outer segments (OSs) only just start to grow out and the characteristic, rhodopsin-filled disk stacks are not yet formed. To test whether the PFKB3 gain-of function or the Pten KO has a marked effect on OS formation and length, the authors should perform the same tests on older, more mature retina at a time when rod OS show their characteristic disk structures (e.g. somewhere between P14 to P30). The same applies to the 2DG inhibition on the Pten KO retina.

The precocious differentiation of rod outer segments observed in P7 Pten-cKO retinas does not persist in adulthood, and instead reflects a developmental acceleration. Indeed, we found that in Pten cKO retinas at 3-, 6- and 12-months of age, rod and cone photoreceptors degenerate, and cone outer segments are shorter (Hanna et al., 2025; Tachibana et al., 2016). These data demonstrate that Pten is required to support rod and cone survival.

(8) Figure 5: Lowering media pH is a rather coarse and untargeted intervention that will have multiple metabolic consequences independent of PKM2. It is thus hardly possible to attribute the effects of pH manipulation to any specific enzyme. To assess this and possibly confirm the results obtained with low pH, the authors should perform a targeted inhibition experiment, for instance using Shikonin (Chen et al., Oncogene 30:4297-306, 2011), to selectively inhibit PKM2. If the retinal explant cultures contained the RPE, an additional question would be how the changes in RPE would alter lactate flux and metabolization between RPE and neuroretina (see also question 4 above).

We have reframed the rationale for the pH manipulation experiments, highlighting the importance of pH in cell fate specification, and indicating that the aggregation of PKM2 is only one possible effect of lower pH.

We wrote: “Given that altered glycolysis influences intracellular pH, which in turn controls cell fate decisions, we set out to assess the impact of manipulating pH on cell fate selection in the retina. One of the expected impacts of lowering pH was the aggregation of PKM2, a rate-limiting enzyme for glycolysis, which aggregates in reversible, inactive amyloids (Cereghetti et al., 2024).” (lines 362-366). 

We have also added a discussion point “Whether pH manipulations also impact the stability of other retinal proteins, such as PKM2, can be further investigated in the future using specific PKM2 inhibitors, such as Shikonin (Chen et al., 2011). (lines 545-547).

(9) Figure 5G: As for Figure 3F, the authors should perform TUNEL assays to assess the number of cells dying independent of caspase-3.

Please see response to point 6.

(10) Figure 7E: In the figure legend "K" should read "E". From the figure and the legend, it is not clear to which cell type this diagram should refer. This must be specified. Importantly, the insulin-dependent glucose-transporter 4 (GLUT4) highlighted in Figure 7E, while expressed on inner retinal vasculature endothelial cells, is not expressed in retinal neurons. What GLUTs exactly are expressed in what retinal neurons may still be to some extent contentious (cf. Chen et al., elife, https://doi.org/10.7554/eLife.91141.3 ; and reviewer comments therein), yet RPE cells clearly express GLUT1, photoreceptors likely express GLUT3, Müller glia cells may express GLUT1, while horizontal cells likely express GLUT2 (Yang et al., J Neurochem. 160:283-296, 2022).’

We have removed this summary schematic for simplicity.

(11) Materials and methods: The retinal explant culture system must be described in more detail. Important questions concern the use of medium and serum for which the providers, order numbers, and batch/lot numbers (whichever is applicable) must be given. The glucose concentration in the medium (including the serum content) should be measured. A key concern is whether the explants were cultivated submerged into the medium - this would prevent sufficient oxygenation and drive metabolism towards glycolysis (i.e. the Pasteur effect) - or whether they were cultivated on top of the liquid medium, at the interface between air and liquid (i.e. a situation that would favor OXPHOS).

We have added further detail to the methods section for the explant assay (lines 686-689). We cultured the retinal explants on membranes on top of the media, which is the standard methodology in the field and in our laboratory (Cantrup et al., 2012; Tachibana et al., 2016; Touahri et al., 2024). Typically, RPCs undergo aerobic glycolysis, meaning that even in the presence of oxygen, they still prefer glycolysis rather than OXPHOS. We demonstrated that 2DG blocks RPC proliferation when treated with 2DG, indicating that RPCs are indeed favoring glycolysis in our assay system.

(12) A point the authors may want to discuss additionally is the potential relevance of their data for the pathogenesis of human diseases, especially early developmental defects such as they occur in oxygen-induced retinopathy of prematurity.

We would like to thank the reviewer for their valuable comment. Given that retinopathy of prematurity (ROP) is primarily vascular in nature, and we have not investigated vascular defects in this study, we have elected not to add a discussion of ROP to our manuscript.

Minor points

(1) Please add a label indicating the ages of the retina to images showing the entire retina (i.e. "P7"; e.g. in Figures 1F, 3, 4D, 5, etc.).

Figure 1:

1D: E18.5 indicated at the bottom of the two panels

1F – P0 is indicated at the bottom of the two panels.

Figure 3C-H: P0 explant stage and days of culture indicated

Figure 4D: E12.5 BrdU and P7 harvest date indicated

Figure 5C-H: P0 explant stage and days of culture indicated

Figure 7A-E: P0 explant stage and days of culture indicated

(2) The term Ctnnb1 should be introduced also in the abstract.

We now state that Ctnnb1 encodes for b-catenin in the abstract.

(3) Line 249: "...remaining..." should probably read "...remained...".

Changed (now line 260).

(4) Line 381: The sentence "...correlating with the propensity of some RPCs to continue to proliferate while others to differentiate.", should probably be rewritten to something like "...correlating with the propensity of some RPCs to continue to proliferate while others differentiate.".

We have corrected this sentence.

(5) The structure of the discussion might benefit from the introduction of subheadings.

We have introduced subheadings.

Reviewer #3 (Recommendations for the authors):

(1) Figure 1H shows the kinetics of rod photoreceptor production as accelerated, but does not represent the fact that fewer rods are ultimately produced, which appears to be the case from the data. If so, the Pten cKO curve should probably be lower than WT to reflect that difference.

We have removed this graph (as per Reviewer #2, point 2).

(2) KEGG analysis also showed that the HIF-1 signaling pathway is altered in the Pten cKO retina. What is the significance of that, and is it related to metabolic dysregulation? It has been shown that lactate can promote vessel growth, which initiates at birth in the mouse retina.

We have added some information on HIF-1 to the Discussion. “The increased glycolytic gene expression in Pten-cKO retinas is likely tied to the increased expression of hypoxia-induced-factor-1-alpha (Hif1a), a known target of mTOR signaling that transcriptionally activates Slc1a3 (GLUT1) and glycolytic genes (Hanna et al., 2022). Indeed, mTOR signaling is hyperactive in Pten-cKO retinas (Cantrup et al., 2012; Tachibana et al., 2016; Tachibana et al., 2018; Touahri et al., 2024), and likewise, in Tsc1-cKO retinas, which also increase glycolysis via HIF-1A (Lim et al., 2021).” (lines 489-494).

Cantrup, R., Dixit, R., Palmesino, E., Bonfield, S., Shaker, T., Tachibana, N., Zinyk, D., Dalesman, S., Yamakawa, K., Stell, W. K., Wong, R. O., Reese, B. E., Kania, A., Sauve, Y., & Schuurmans, C. (2012). Cell-type specific roles for PTEN in establishing a functional retinal architecture. PLoS One, 7(3), e32795. https://doi.org/10.1371/journal.pone.0032795

Cereghetti, G., Kissling, V. M., Koch, L. M., Arm, A., Schmidt, C. C., Thüringer, Y., Zamboni, N., Afanasyev, P., Linsenmeier, M., Eichmann, C., Kroschwald, S., Zhou, J., Cao, Y., Pfizenmaier, D. M., Wiegand, T., Cadalbert, R., Gupta, G., Boehringer, D., Knowles, T. P. J., Mezzenga, R., Arosio, P., Riek, R., & Peter, M. (2024). An evolutionarily conserved mechanism controls reversible amyloids of pyruvate kinase via pH-sensing regions. Dev Cell. https://doi.org/10.1016/j.devcel.2024.04.018

Chen, J., Xie, J., Jiang, Z., Wang, B., Wang, Y., & Hu, X. (2011). Shikonin and its analogs inhibit cancer cell glycolysis by targeting tumor pyruvate kinase-M2. Oncogene, 30(42), 4297-4306. https://doi.org/10.1038/onc.2011.137

Hanna, J., Touahri, Y., Pak, A., David, L. A., van Oosten, E., Dixit, R., Vecchio, L. M., Mehta, D. N., Minamisono, R., Aubert, I., & Schuurmans, C. (2025). Pten Loss Triggers Progressive Photoreceptor Degeneration in an mTORC1-Independent Manner. Invest Ophthalmol Vis Sci, 66(3), 45. https://doi.org/10.1167/iovs.66.3.45

Tachibana, N., Cantrup, R., Dixit, R., Touahri, Y., Kaushik, G., Zinyk, D., Daftarian, N., Biernaskie, J., McFarlane, S., & Schuurmans, C. (2016). Pten Regulates Retinal Amacrine Cell Number by Modulating Akt, Tgfbeta, and Erk Signaling. J Neurosci, 36(36), 9454-9471. https://doi.org/10.1523/JNEUROSCI.0936-16.2016

Touahri, Y., Hanna, J., Tachibana, N., Okawa, S., Liu, H., David, L. A., Olender, T., Vasan, L., Pak, A., Mehta, D. N., Chinchalongporn, V., Balakrishnan, A., Cantrup, R., Dixit, R., Mattar, P., Saleh, F., Ilnytskyy, Y., Murshed, M., Mains, P. E., Kovalchuk, I., Lefebvre, J. L., Leong, H. S., Cayouette, M., Wang, C., Sol, A. D., Brand, M., Reese, B. E., & Schuurmans, C. (2024). Pten regulates endocytic trafficking of cell adhesion and Wnt signaling molecules to pattern the retina. Cell Rep, 43(4), 114005. https://doi.org/10.1016/j.celrep.2024.114005

eLife Assessment

This landmark study describes the structure of the human RAD51 filament with a recombination intermediate called the displacement loop (D-loop). Using cryogenic structural, biochemical, and single-molecule analyses, the authors provide compelling evidence on how the RAD51 filament promotes strand exchange between single-stranded and double-stranded DNAs. The work will be of interest to the community of homologous recombination and DNA repair, as well as genome stability more generally.

Reviewer #1 (Public review):

Summary:

The paper describes the cryoEM structure of RAD51 filament on the recombination intermediate. In the RAD51 filament, the insertion of a DNA-binding loop called the L2 loop stabilizes the separation of the complementary strand for the base-pairing with an incoming ssDNA and the non-complementary strand, which is captured by the second DNA-binding channel called the site II. The molecular structure of the RAD51 filament with a recombination intermediate provides a new insight into the mechanism of homology search and strand exchange between ssDNA and dsDNA.

Strengths:

This is the first human RAD51 filament structure with a recombination intermediate called the D-loop. The work has been done with great care, and the results shown in the paper are compelling based on cryo-EM and biochemical analyses. The paper is really nice and important for researchers in the field of homologous recombination, which gives a new view on the molecular mechanism of RAD51-mediated homology search and strand exchange.

Weaknesses:

The authors need more careful text writing. Without page and line numbers, it is hard to give comments.

Reviewer #2 (Public review):

Summary:

Homologous recombination (HR) is a critical pathway for repairing double-strand DNA breaks and ensuring genomic stability. At the core of HR is the RAD51-mediated strand-exchange process, in which the RAD51-ssDNA filament binds to homologous double-stranded DNA (dsDNA) to form a characteristic D-loop structure. While decades of biochemical, genetic, and single-molecule studies have elucidated many aspects of this mechanism, the atomic-level details of the strand-exchange process remained unresolved due to a lack of atomic-resolution structure of RAD51 D-loop complex.<br /> In this study, the authors achieved this by reconstituting a RAD51 mini-filament, allowing them to solve the RAD51 D-loop complex at 2.64 Å resolution using a single particle approach. The atomic resolution structure reveals how specific residues of RAD51 facilitate the strand exchange reaction. Ultimately, this work provides unprecedented structural insight into the eukaryotic HR process and deepens the understanding of RAD51 function at the atomic level, advancing the broader knowledge of DNA repair mechanisms.

Strengths:

The authors overcame the challenge of RAD51's helical symmetry by designing a minifilament system suitable for single-particle cryo-EM, enabling them to resolve the RAD51 D-loop structure at 2.64 Å without imposed symmetry. This high resolution revealed precise roles of key residues, including F279 in Loop 2, which facilitates strand separation, and basic residues on site II that capture the displaced strand. Their findings were supported by mutagenesis, strand exchange assays, and single-molecule analysis, providing strong validation of the structural insights.

Weaknesses:

Despite the detailed structural data, some structure-based mutagenesis data interpretation lacks clarity. Additionally, the proposed 3′-to-5′ polarity of strand exchange relies on assumptions from static structural features, such as stronger binding of the 5′-arm-which are not directly supported by other experiments. This makes the directional model compelling but contradicts several well-established biochemical studies that support a 5'-to-3' polarity relative to the complementary strand (e.g., Cell 1995, PMID: 7634335; JBC 1996, PMID: 8910403; Nature 2008, PMID: 18256600).

Overall:

The 2.6 Å resolution cryoEM structure of the RAD51 D-loop complex provides remarkably detailed insights into the residues involved in D-loop formation. The high-quality cryoEM density enables precise placement of each nucleotide, which is essential for interpreting the molecular interactions between RAD51 and DNA. Particularly, the structural analysis highlights specific roles for key domains, such as the N-terminal domain (NTD), in engaging the donor DNA duplex.

This structural interpretation is further substantiated by single-molecule fluorescence experiments using the KK39,40AA NTD mutant. The data clearly show a significant reduction in D-loop formation by the mutant compared to wild-type, supporting the proposed functional role of the NTD observed in the cryoEM model.

However, the strand exchange activity interpretation presented in Figure 5B could benefit from a more rigorous experimental design. The current assay measures an increase in fluorescence intensity, which depends heavily on the formation of RAD51-ssDNA filaments. As shown in Figure S6A, several mutants exhibit reduced ability to form such filaments, which could confound the interpretation of strand exchange efficiency. To address this, the assay should either: (1) normalize for equivalent levels of RAD51-ssDNA filaments across samples, or (2) compare the initial rates of fluorescence increase (i.e., the slope of the reaction curve), rather than endpoint fluorescence, to better isolate the strand exchange activity itself.

Based on the structural features of the D-loop, the authors propose that strand pairing and exchange initiate at the 3'-end of the complementary strand in the donor DNA and proceed with a 3'-to-5' polarity. This conclusion, drawn from static structural observations, contrasts with several well-established biochemical studies that support a 5'-to-3' polarity relative to the complementary strand (e.g., Cell 1995, PMID: 7634335; JBC 1996, PMID: 8910403; Nature 2008, PMID: 18256600). While the structural model is compelling and methodologically robust, this discrepancy underscores the need for further experiments.

Reviewer #3 (Public review):

Summary:

Built on their previous pioneer expertise in studying RAD51 biology, in this paper, the authors aim to capture and investigate the structural mechanism of human RAD51 filament bound with a displacement loop (D-loop), which occurs during the dynamic synaptic state of the homologous recombination (HR) strand-exchange step. As the structures of both pre- and post-synaptic RAD51 filaments were previously determined, a complex structure of RAD51 filaments during strand exchange is one of the key missing pieces of information for a complete understanding of how RAD51 functions in the HR pathway. This paper aims to determine the high-resolution cryo-EM structure of RAD51 filament bound with the D-loop. Combined with mutagenesis analysis and biophysical assays, the authors aim to investigate the D-loop DNA structure, RAD51-mediated strand separation and polarity, and a working model of RAD51 during HR strand invasion in comparison with RecA.

Strengths:

(1) The structural work and associated biophysical assays in this paper are solid, elegantly designed, and interpreted.  These results provide novel insights into RAD51's function in HR.

(2) The DNA substrate used was well designed, taking into consideration the nucleotide number requirement of RAD51 for stable capture of donor DNA. This DNA substrate choice lays the foundation for successfully determining the structure of the RAD51 filament on D-loop DNA using single-particle cryo-EM.

(3) The authors utilised their previous expertise in capping DNA ends using monomeric streptavidin and combined their careful data collection and processing to determine the cryo-EM structure of full-length human RAD51 bound at the D-loop in high resolution. This interesting structure forms the core part of this work and allows detailed mapping of DNA-DNA and DNA-protein interaction among RAD51, invading strands, and donor DNA arms (Figures 1, 2, 3, 4). The geometric analysis of D-loop DNA bound with RAD51 and EM density for homologous DNA pairing is also impressive (Figure S5). The previously disordered RAD51's L2-loop is now ordered and traceable in the density map and functions as a physical spacer when bound with D-loop DNA. Interestingly, the authors identified that the side chain position of F279 in the L2_loop of RAD51_H differs from other F279 residues in L2-loops of E, F, and G protomers. This asymmetric binding of L2 loops and RAD51_NTD binding with donor DNA arms forms the basis of the proposed working model about the polarity of csDNA during RAD51-mediated strand exchange.

(4) This work also includes mutagenesis analysis and biophysical experiments, especially EMSA, single-molecule fluorescence imaging using an optical tweezer, and DNA strand exchange assay, which are all suitable methods to study the key residues of RAD51 for strand exchange and D-loop formation (Figure 5).

Weaknesses:

(1) The proposed model for the 3'-5' polarity of RAD51-mediated strand invasion is based on the structural observations in the cryo-EM structure. This study lacks follow-up biochemical/biophysical experiments to validate the proposed model compared to RecA or developing methods to capture structures of any intermediate states with different polarity models.

(2) The functional impact of key mutants designed based on structure has not been tested in cells to evaluate how these mutants impact the HR pathway.

The significance of the work for the DNA repair field and beyond:

Homologous recombination (HR) is a key pathway for repairing DNA double-strand breaks and involves multiple steps. RAD51 forms nucleoprotein filaments first with 3' overhang single-strand DNA (ssDNA), followed by a search and exchange with a homologous strand. This function serves as the basis of an accurate template-based DNA repair during HR. This research addressed a long-standing challenge of capturing RAD51 bound with the dynamic synaptic DNA and provided the first structural insight into how RAD51 performs this function. The significance of this work extends beyond the discovery of biology for the DNA repair field, into its medical relevance. RAD51 is a potential drug target for inhibiting DNA repair in cancer cells to overcome drug resistance. This work offers a structural understanding of RAD51's function with the D-loop and provides new strategies for targeting RAD51 to improve cancer therapies.

Author response:

Public Reviews:

Reviewer #1 (Public review):

Summary:

The paper describes the cryoEM structure of RAD51 filament on the recombination intermediate. In the RAD51 filament, the insertion of a DNA-binding loop called the L2 loop stabilizes the separation of the complementary strand for the base-pairing with an incoming ssDNA and the non-complementary strand, which is captured by the second DNA-binding channel called the site II. The molecular structure of the RAD51 filament with a recombination intermediate provides a new insight into the mechanism of homology search and strand exchange between ssDNA and dsDNA.

Strengths:

This is the first human RAD51 filament structure with a recombination intermediate called the D-loop. The work has been done with great care, and the results shown in the paper are compelling based on cryo-EM and biochemical analyses. The paper is really nice and important for researchers in the field of homologous recombination, which gives a new view on the molecular mechanism of RAD51-mediated homology search and strand exchange.

Weaknesses:

The authors need more careful text writing. Without page and line numbers, it is hard to give comments.

We would like to thank the reviewer for their kind words of appreciation of our work.

Reviewer #2 (Public review):

Summary:

Homologous recombination (HR) is a critical pathway for repairing double-strand DNA breaks and ensuring genomic stability. At the core of HR is the RAD51-mediated strand-exchange process, in which the RAD51-ssDNA filament binds to homologous double-stranded DNA (dsDNA) to form a characteristic D-loop structure. While decades of biochemical, genetic, and single-molecule studies have elucidated many aspects of this mechanism, the atomic-level details of the strand-exchange process remained unresolved due to a lack of atomic-resolution structure of RAD51 D-loop complex.

In this study, the authors achieved this by reconstituting a RAD51 mini-filament, allowing them to solve the RAD51 D-loop complex at 2.64 Å resolution using a single particle approach. The atomic resolution structure reveals how specific residues of RAD51 facilitate the strand exchange reaction. Ultimately, this work provides unprecedented structural insight into the eukaryotic HR process and deepens the understanding of RAD51 function at the atomic level, advancing the broader knowledge of DNA repair mechanisms.

Strengths:

The authors overcame the challenge of RAD51's helical symmetry by designing a minifilament system suitable for single-particle cryo-EM, enabling them to resolve the RAD51 D-loop structure at 2.64 Å without imposed symmetry. This high resolution revealed precise roles of key residues, including F279 in Loop 2, which facilitates strand separation, and basic residues on site II that capture the displaced strand. Their findings were supported by mutagenesis, strand exchange assays, and single-molecule analysis, providing strong validation of the structural insights.

Weaknesses:

Despite the detailed structural data, some structure-based mutagenesis data interpretation lacks clarity. Additionally, the proposed 3′-to-5′ polarity of strand exchange relies on assumptions from static structural features, such as stronger binding of the 5′-arm-which are not directly supported by other experiments. This makes the directional model compelling but contradicts several well-established biochemical studies that support a 5'-to-3' polarity relative to the complementary strand (e.g., Cell 1995, PMID: 7634335; JBC 1996, PMID: 8910403; Nature 2008, PMID: 18256600).

Overall:

The 2.6 Å resolution cryoEM structure of the RAD51 D-loop complex provides remarkably detailed insights into the residues involved in D-loop formation. The high-quality cryoEM density enables precise placement of each nucleotide, which is essential for interpreting the molecular interactions between RAD51 and DNA. Particularly, the structural analysis highlights specific roles for key domains, such as the N-terminal domain (NTD), in engaging the donor DNA duplex.

This structural interpretation is further substantiated by single-molecule fluorescence experiments using the KK39,40AA NTD mutant. The data clearly show a significant reduction in D-loop formation by the mutant compared to wild-type, supporting the proposed functional role of the NTD observed in the cryoEM model.

However, the strand exchange activity interpretation presented in Figure 5B could benefit from a more rigorous experimental design. The current assay measures an increase in fluorescence intensity, which depends heavily on the formation of RAD51-ssDNA filaments. As shown in Figure S6A, several mutants exhibit reduced ability to form such filaments, which could confound the interpretation of strand exchange efficiency. To address this, the assay should either: (1) normalize for equivalent levels of RAD51-ssDNA filaments across samples, or (2) compare the initial rates of fluorescence increase (i.e., the slope of the reaction curve), rather than endpoint fluorescence, to better isolate the strand exchange activity itself.

Based on the structural features of the D-loop, the authors propose that strand pairing and exchange initiate at the 3'-end of the complementary strand in the donor DNA and proceed with a 3'-to-5' polarity. This conclusion, drawn from static structural observations, contrasts with several well-established biochemical studies that support a 5'-to-3' polarity relative to the complementary strand (e.g., Cell 1995, PMID: 7634335; JBC 1996, PMID: 8910403; Nature 2008, PMID: 18256600). While the structural model is compelling and methodologically robust, this discrepancy underscores the need for further experiments.

We would like to thank the reviewer for highlighting the importance of our findings to our understanding of the mechanism of homologous recombination.

We agree with the reviewer that the reduced filament-forming ability of some of the RAD51 mutants complicates a straightforward interpretation of their strand-exchange assay. Interestingly, the RAD51 mutants that appear most impaired are the esDNA-capture mutants that do not contact the ssDNA in the structure of the pre-synaptic filament. However, the RAD51 NTD mutants, that display the most severe defect in strand-exchange, have a near-WT filament forming ability.

The reviewer correctly points out that the polarity of strand exchange by RecA and RAD51 is an extensively researched topic that has been characterised in several authoritative studies. In our paper, we simply describe the mechanistic insights obtained from the structural D-loop models of RAD51 (our work) and RecA (Yang et al, PMID: 33057191).The structures illustrate a very similar mechanism of D-loop formation that proceeds with opposite polarity of strand exchange for RAD51 and RecA. Comparison of the D-loop structures for RecA and RAD51 provides an attractive explanation for the opposite polarity, as caused by the different positions of their dsDNA-binding domains in the filament structure. We agree with the reviewer that further investigation will be needed for an adequate rationalisation of the available evidence. We will mention the relevant literature in the revised version of the manuscript.

Reviewer #3 (Public review):

Summary:

Built on their previous pioneer expertise in studying RAD51 biology, in this paper, the authors aim to capture and investigate the structural mechanism of human RAD51 filament bound with a displacement loop (D-loop), which occurs during the dynamic synaptic state of the homologous recombination (HR) strand-exchange step. As the structures of both pre- and post-synaptic RAD51 filaments were previously determined, a complex structure of RAD51 filaments during strand exchange is one of the key missing pieces of information for a complete understanding of how RAD51 functions in the HR pathway. This paper aims to determine the high-resolution cryo-EM structure of RAD51 filament bound with the D-loop. Combined with mutagenesis analysis and biophysical assays, the authors aim to investigate the D-loop DNA structure, RAD51-mediated strand separation and polarity, and a working model of RAD51 during HR strand invasion in comparison with RecA.

Strengths:

(1) The structural work and associated biophysical assays in this paper are solid, elegantly designed, and interpreted.  These results provide novel insights into RAD51's function in HR.

(2) The DNA substrate used was well designed, taking into consideration the nucleotide number requirement of RAD51 for stable capture of donor DNA. This DNA substrate choice lays the foundation for successfully determining the structure of the RAD51 filament on D-loop DNA using single-particle cryo-EM.

(3) The authors utilised their previous expertise in capping DNA ends using monomeric streptavidin and combined their careful data collection and processing to determine the cryo-EM structure of full-length human RAD51 bound at the D-loop in high resolution. This interesting structure forms the core part of this work and allows detailed mapping of DNA-DNA and DNA-protein interaction among RAD51, invading strands, and donor DNA arms (Figures 1, 2, 3, 4). The geometric analysis of D-loop DNA bound with RAD51 and EM density for homologous DNA pairing is also impressive (Figure S5). The previously disordered RAD51's L2-loop is now ordered and traceable in the density map and functions as a physical spacer when bound with D-loop DNA. Interestingly, the authors identified that the side chain position of F279 in the L2_loop of RAD51_H differs from other F279 residues in L2-loops of E, F, and G protomers. This asymmetric binding of L2 loops and RAD51_NTD binding with donor DNA arms forms the basis of the proposed working model about the polarity of csDNA during RAD51-mediated strand exchange.

(4) This work also includes mutagenesis analysis and biophysical experiments, especially EMSA, single-molecule fluorescence imaging using an optical tweezer, and DNA strand exchange assay, which are all suitable methods to study the key residues of RAD51 for strand exchange and D-loop formation (Figure 5).

Weaknesses:

(1) The proposed model for the 3'-5' polarity of RAD51-mediated strand invasion is based on the structural observations in the cryo-EM structure. This study lacks follow-up biochemical/biophysical experiments to validate the proposed model compared to RecA or developing methods to capture structures of any intermediate states with different polarity models.

(2) The functional impact of key mutants designed based on structure has not been tested in cells to evaluate how these mutants impact the HR pathway.

The significance of the work for the DNA repair field and beyond:

Homologous recombination (HR) is a key pathway for repairing DNA double-strand breaks and involves multiple steps. RAD51 forms nucleoprotein filaments first with 3' overhang single-strand DNA (ssDNA), followed by a search and exchange with a homologous strand. This function serves as the basis of an accurate template-based DNA repair during HR. This research addressed a long-standing challenge of capturing RAD51 bound with the dynamic synaptic DNA and provided the first structural insight into how RAD51 performs this function. The significance of this work extends beyond the discovery of biology for the DNA repair field, into its medical relevance. RAD51 is a potential drug target for inhibiting DNA repair in cancer cells to overcome drug resistance. This work offers a structural understanding of RAD51's function with the D-loop and provides new strategies for targeting RAD51 to improve cancer therapies.

We thank the reviewer for their positive comments on the significance of our work. Concerning the proposed polarity of strand exchange based on our structural finding, please see our reply to the previous reviewer; we agree with the reviewer that further experimentation will be needed to reach a settled view on this.

Testing the functional effects of the RAD51 mutants on HR in cells was not an aim of the current work but we agree that it would be a very interesting experiment, which would likely provide further important insights into the mechanism of strand exchange at the core of the HR reaction.

eLife Assessment

This important study by Lee et al. investigates the heterogeneous response of non-growing bacteria to the antimicrobial peptide (AMP) tachyplesin. In this response, a subpopulation of bacteria limits the accumulation of a fluorescent analog of the AMP, avoiding lethal damage. The study provides compelling evidence of the reduced susceptibility to the antimicrobial peptide antibiotic tachyplesin in a subpopulation of cells characterized by reduced drug accumulation. The evidence on the underlying molecular mechanisms is solid.

Reviewer #1 (Public review):

Summary:

This work contributes several important and interesting observations regarding the heterotolerance of non-growing Escherichia coli and Pseudomonas aeruginosa to the antimicrobial peptide tachyplesin. The primary mechanism of action of tachyplesin is thought to be disruption of the bacterial cell envelope, leading to leakage of cellular contents after a threshold level of accumulation. Although the MIC for tachyplesin in exponentially growing E. coli is just 1 ug/ml, the authors observe that a substantial fraction of a stationary phase population of bacteria survives much higher concentrations, up to 64 ug/ml. By using a fluorescently labelled analogue of tachyplesin, the authors show that the amount of per-cell intracellular accumulation of tachyplesin displays a bimodal distribution, and that the fraction of "low accumulators" correlates with the fraction of survivors. Using a microfluidic device, they show that low accumulators exclude propidium iodide, suggesting that their cell envelopes remain largely intact, while high accumulators of tachyplesin also stain with propidium iodide. They show that this phenomenon holds for several clinical isolates of E. coli with different genetic determinants of antibiotic resistance, and for a strain of Pseudomonas aeruginosa. However, the bimodal distribution does not occur in these organisms for several other antimicrobial peptides, or for tachyplesin in Klebsiella pneumoniae or Staphylococcus aureus, indicating some degree of specificity in the interaction between AMP and bacterial cell envelope. They next explore the dynamics of the fluorescent tachyplesin accumulation and show interestingly that a high degree of accumulation is initially seen in all cells, but that the "low accumulator" subpopulation manages to decrease the amount of intracellular fluorescence over time, while the "high accumulator"subpopulation continues to increase its intracellular fluorescence. Focusing on increased efflux as a hypothesised mechanism for the "low accumulator" phenotype, based on transcriptomic analysis of the two subpopulations, the authors screen putative efflux inhibitors to see if they can block the formation of the low accumulator subpopulation. They find that both the protonophore CCCP and the SSRI sertraline can block the formation of this subpopulation and that a combination of sertraline plus tachyplesin kills a greater fraction of the stationary phase cells than either agent alone, similar to the killing observed when growing cells are treated with tachyplesin.

Strengths:

This study provides new insight into the heterogeneous behaviours of non-growing bacteria when exposed to an antimicrobial peptide, and into the dynamics of their response. The single-cell analysis by FACS and microscopy is compelling. The results provide a much-needed single cell perspective on the phenomenon of tolerance to AMPs and a good starting point for further exploration.

Weaknesses:

The authors have substantially improved the clarity of the manuscript and have added additional experiments to probe further the location of the AMP relative to low and high accumulators, and the physiological states of these sub-populations. These experiments strengthen the assertion that low accumulators keep the AMP at the cell surface while high accumulators permit intracellular access to the AMP.

However, many questions still remain about the physiological characterisation of the "low accumulator" cells. While the evidence presented does support an induced response that removes the AMP from the interior of the cell, no clear mechanism for this is favoured by the experiments presented.

A double deletion of acrA and tolC (two out of the three components of the major constitutive RND efflux pump) reduces the appearance of the low accumulator phenotype, but interestingly, the single deletions have no effect, and a well-characterised inhibitor of RND efflux pumps also has no effect. The authors identify a two-component system, qseCB, that appears necessary for the appearance of low accumulators, but this system has pleiotropic effects on many cellular systems, with only tenuous connections to efflux. The selected pharmacological agents that could prevent the appearance of low accumulators do not offer clear insight into the mechanism by which low accumulators arise, because they have diverse modes of action.

The transcriptomics data collected for low and high accumulator sub-populations are interesting, but in my opinion, the conclusions that can be drawn from these data remain overstated. It is not possible to make any claims about the total amount of "protein synthesis, energy production, and gene expression" on the basis of RNA-Seq data. The reads from each sample are normalised, so there is no information about the total amount of transcript. Many elements of total cellular activity are post-transcriptionally regulated, so it is impossible to assess from transcriptomics alone. Finally, the transcriptomic data are analysed in aggregated clusters of genes that are enriched for biological processes, for example: "Cluster 2 included processes involved in protein synthesis, energy production, and gene expression that were downregulated to a greater extent in low accumulators than high accumulators". However, this obscures the fact that these clusters include genes that are generally inhibitory of the process named, as well as genes that facilitate the process.

The authors have added an experiment to attempt to assess overall metabolic activity in the low accumulator and high accumulator populations, which is a welcome addition. They apply the redox dye resazurin and observe lower resorufin (reduced form) fluorescence in the low accumulator population, which they take to indicate a lower respiration rate. This seems possible, however, an important caveat is that they have shown the low accumulator population to retain substantially lower amounts of multiple different fluorescent molecules (tachyplesin-NBD, propidium iodide, ethidium bromide) intracellularly compared to the high accumulator population. It seems possible that the low accumulator population is also capable of removing resazurin or resorufin from the intracellular space, regardless of metabolic rate. Indeed, it has previously been shown that efflux by RND efflux pumps influences resazurin reduction to resorufin in both P. aeruginosa and E. coli. By measuring only the retained redox dye using flow cytometry, the results may be confounded by the demonstrated ability of the low accumulator population to remove various fluorescent dyes. More work is needed to strongly support broad conclusions about the physiological states of the low and high accumulator populations.

The phenomenon of the emergence of low accumulators, which are phenotypically tolerant to the antimicrobial peptide tachyplesin, is interesting and important even if there is still work to be done to understand the mechanism by which it occurs.

Reviewer #2 (Public review):

Summary:

This study reports on the existence of subpopulations of isogenic E. coli and P. aeruginosa cells that are tolerant to the antimicrobial peptide tachyplesin and are characterized by accumulation of low levels of a fluorescent tachyplesin-NBD conjugate. The authors then set out to address the molecular mechanisms, providing interesting insights even though the mechanism remains incompletely defined: The work demonstrates that increased efflux may cause this phenotype, putatively together with other changes in membrane lipid composition. The authors further demonstrate that pharmacological manipulation can prevent generation of tolerance. The authors are cautious in their interpretation and the claims made are largely justified by the data.

Strengths:

Going beyond the commonly used bulk techniques for studying susceptibility to AMPs , Lee et al. used of fluorescent antibiotic conjugates in combination with flow cytometry analysis to study variability in drug accumulation at the single cell level. This powerful approach enabled the authors to expose bimodal drug accumulation pattern that were condition dependent, but conserved across a variety of E. coli clinical isolates. Using cell sorting in combination with colony-forming unit assays as well as quantitative fluorescence microscopic analysis in a microfludics-setup the authors compellingly demonstrate that low accumulators (where fluorescence signal is mostly restricted to the membrane), can survive antibiotic treatment, whereas high accumulators (with high intracellular fluorescence) were killed.

The relevance of efflux for the ´low accumulator´ phenotype and its survival is convincingly demonstrated by the following lines of evidence: i) A time-course experiment on tachyplesin-NBD pre-loaded cells revealed that all cells initially were high accumulators, before a subpopulation of cells subsequently managed to reduce signal intensity, demonstrating that the ´low accumulator´ phenotype is an induced response and not a pre-existing property. Ii) Double-mutants deficient in the delta acrA delta tolC double-KO, which showed reduced levels of low accumulators´. Interestingly, ´low accumulator´populations were nearly abrogated in bacteria deficient in the qse quorum sensing system, suggesting its centrality for the tachyplesin response. Even though this system may control acrA, the strength of the phenotype may suggest that it may control additional as-of-yet unidenitified factors relevant in the response to tachyplesin. Iii) treatment with efflux pump inhibitor sertraline and verapamil (even though some caution needs to be taken since it is not perfectly selective, see weakness) prevents generation of low accumulators. The observation that sertraline enhances tachyplesin-based killing is an important basis for developing combination therapies.

The study convincingly illustrates how susceptibility to tachyplesin adaptively changes in a heterogeneous way dependent on the growth phases and nutrient availability. This is highly relevant also beyond the presented example of tachyplesin and similar subpopulation-based adaptive changes to the susceptibility towards antimicrobial peptides or other drugs may occur during infections in vivo and they would likely be missed by standardized in vitro susceptibility testing.

Weaknesses:

Some mechanistic questions regarding tachyplesin-accumulation and survival remain. One general shortcoming of the setup of the transcriptomics experiment is that the tachyplesin-NBD probe itself has antibiotic efficacy and induces phenotypes (and eventually cell death) in the ´high accumulator´ cells. As the authors state themselves, this makes it challenging to interpret whether any differences seen between the two groups are causative for the observed accumulation pattern of if they are a consequence of differential accumulation and downstream phenotypic effects.

I have a few minor concerns regarding new data that was added during the revision:

- The statement ´ Moreover, we found that the fluorescence of low accumulators decreased over time when bacteria were treated with 20 μg mL´ is, in my opinion, not supported by the data shown in Figure S4C. That figure shows that the abundance of ´low accumulator´ cells decreases over time. Following the rationale that protease K treatment may cleave surface-associated/extracellular tachyplesin-NDB, this should lead to a shift of ´low accumulator´population to the left, indicating reduced fluorescence intensity per cell. This is not so case, but the population just disappears. However, after 120 min of treatment more cells appear in the ´high accumulator´ state. This result is somewhat puzzling.

- The authors used the metabolic dye resazurin to measure the metabolic activity of low vs. high accumulators. I am not entirely convinced that the lower fluorescence resorufin-fluorescence in tachyplesin-NBD accumulators really indicates lower metabolic activity, since a cell's fluorescence levels would also be affected by the cellular uptake and efflux. It appears plausible that the lower resorufin-fluorescence may result from reduced accumulation/increased efflux in the´low-tachyplesin NBD´ population.

Comment on revisions: All my previous comments have been satisfactorily addressed by the authors.

Reviewer #3 (Public review):

Summary:

This important study shows that stationary phase bacteria survive antimicrobial peptide treatment by switching on efflux pumps, generating low accumulating subpopulations that evade killing-a finding with clear implications for the design of peptide based antibiotics and for researchers studying antimicrobial resistance. The evidence is solid and frequently convincing, as diverse single cell assays, genetics and chemical inhibition coherently link reduced intracellular peptide to survival, even though a few mechanistic details warrant further exploration.

Strengths:

The authors investigate how Escherichia coli (and, to a lesser extent, Pseudomonas aeruginosa) survive exposure to the antimicrobial peptide (AMP) tachyplesin. Because resistance to AMPs is thought to rely heavily on non genetic adaptations rather than on classical mutation based mechanisms, the study focuses on phenotypic heterogeneity and seeks to pinpoint the cellular processes that protect a subset of cells. Using fluorescently labelled tachyplesin, single cell imaging, flow cytometry, transcriptomics, targeted genetics, and chemical perturbations, the authors report that stationary phase cultures harbor two phenotypic states: high accumulating cells that die and low accumulating cells that survive. They further propose and show that inducible efflux activity is the primary driver of survival and show that either efflux inhibition (sertraline, verapamil) or nutrient supplementation prevents the emergence of low accumulators and boosts killing.<br /> The experiments unambiguously reveal that the cells respond to stress heterogeneously, with two distinct subpopulations - one with better survival than the other. This primary phenotype is convincingly shown across various E. coli strains, including clinical isolates. The authors probed the underlying mechanism from several angles, with important additional experiments in the revised version that strengthens the original conclusions in several ways. Newly added efflux assays with ethidium bromide, together with proteinase treatment experiments and ΔacrAΔtolC and ΔqseB/qseC mutant data, illustrate that the low accumulating subpopulation can actively export intracellular compounds. The authors took great care to temper their language to acknowledge other potential alternatives that could explain some of the data such as altered influx, vesicle release or proteolysis, metabolic activity of the cells, indirect effects of sertraline treatment, etc. Additional metabolic dye measurements confirm that low accumulators are less metabolically active, and a new data on nutrient supplementation shows that forcing growth increases peptide uptake and lethality. The authors clarify the crucial point of where antimicrobial peptides actually bind on the cell within the broader survival mechanism and present their conclusions, along with potential caveats, with commendable clarity.

Weaknesses:

Despite these advances, the contribution of efflux may require more direct evidence to further dissect whether efflux is necessary, sufficient, or contributory. The facts that the key low-efflux mutant still retains a small fraction of survivors and that the inhibitors used may cause other physiological changes leading to higher efflux are still unaccounted for. The lipidomic and vesicle findings, while intriguing, remain descriptive, and direct tests of their functional relevance would further solidify the mechanistic models.

Conclusion:

Even with these limitations, the study provides valuable insight into non genetic resistance mechanisms to AMPs and highlights inducible heterogeneity as a critical obstacle to peptide therapeutics. In a much broader context, this study also underscores the importance of efflux physiology even for those antimicrobials that seemingly would not have intracellular targets.

Author response:

The following is the authors’ response to the original reviews

Reviewer 1:

(1) The initial high accumulation by all cells followed by the emergence of a sub-population that has reduced its intracellular levels of tachyplesin is a key observation and I agree with the authors' conclusion that this suggests an induced response to the AMP is important in facilitating the bimodal distribution. However, I think the conclusion that upregulated efflux is driving the reduction in signal in the "low accumulator" subpopulation is not fully supported. Steady-state amounts of intracellular fluorescent AMP are determined by the relative rates of influx and efflux and a decrease could be caused by decreasing influx (while efflux remained unchanged), increasing efflux (while influx remained unchanged), or both decreasing influx and increasing efflux. Given the transcriptomic data suggest possible changes in the expression of enzymes that could affect outer membrane permeability and outer membrane vesicle formation as well as efflux, it seems very possible that changes to both influx and efflux are important. The "efflux inhibitors" shown to block the formation of the low accumulator subpopulation have highly pleiotropic or incompletely characterised mechanisms of action so they also do not exclusively support a hypothesis of increased efflux.

We agree with the reviewer that the emergence of low accumulators after 30 min in the presence of extracellular tachyplesin-NBD (Figure 4A) could be due to either decreased influx while efflux remained unchanged, increased efflux while influx remained unchanged, or both decreasing influx and increasing efflux. Increased proteolytic activity or increased secretion of OMVs could also play a role.

We have now acknowledged that “Reduced intracellular accumulation of tachyplesin-NBD in the presence of extracellular tachyplesin-NBD could be due to decreased drug influx, increased drug efflux, increased proteolytic activity or increased secretion of OMVs.” (lines 313-315).

However, the emergence of low accumulators after 60 min in the absence of extracellular tachyplesin-NBD in our efflux assays (Figure 4C) cannot be due to decreased influx while efflux remained unchanged because of the absence of extracellular tachyplesin-NBD. We acknowledge that in our original manuscript we did not explicitly state that the efflux assays reported in Figure 4C-D were performed in the absence of tachyplesin-NBD in the extracellular environment. We have now clarified this point in our manuscript, we have added illustrations in Figure 4A, 4C-D and we have also carried out efflux assays using ethidium bromide (EtBr) to further support our conclusions about the primary role played by efflux in reducing tachyplesin accumulation in low accumulators. We have added the following paragraphs to our revised manuscript:

“Next, we performed efflux assays using ethidium bromide (EtBr) by adapting a previously described protocol [62]. Briefly, we preloaded stationary phase E. coli with EtBr by incubating cells at a concentration of 254 µM EtBr in M9 medium for 90 min. Cells were then pelleted and resuspended in M9 to remove extracellular EtBr. Single-cell EtBr fluorescence was measured at regular time points in the absence of extracellular EtBr using flow cytometry. This analysis revealed a progressive homogeneous decrease of EtBr fluorescence due to efflux from all cells within the stationary phase E. coli population (Figure S13A). In contrast, when we performed efflux assays by preloading cells with tachyplesin-NBD (46 μg mL^-1 or 18.2 μM), followed by pelleting and resuspension in M9 to remove extracellular tachyplesin-NBD, we observed a heterogeneous decrease in tachyplesin-NBD fluorescence in the absence of extracellular tachyplesin-NBD: a subpopulation retained high tachyplesin-NBD fluorescence, i.e. high accumulators; whereas another subpopulation displayed decreased tachyplesin-NBD fluorescence, 60 min after the removal of extracellular tachyplesin-NBD (Figure 4B). Since these assays were performed in the absence of extracellular tachyplesin-NBD, decreased tachyplesin-NBD fluorescence could not be ascribed to decreased drug influx or increased secretion of OMVs in low accumulators, but could be due to either enhanced efflux or proteolytic activity in low accumulators.

Next, we repeated efflux assays using EtBr in the presence of 46 μg mL^-1 (or 20.3 µM) extracellular tachyplesin-1. We observed a heterogeneous decrease of EtBr fluorescence with a subpopulation retaining high EtBr fluorescence (i.e. high tachyplesin accumulators) and another population displaying reduced EtBr fluorescence (i.e. low tachyplesin accumulators, Figure S14B) when extracellular tachyplesin-1 was present. Moreover, we repeated tachyplesin-NBD efflux assays in the presence of M9 containing 50 μg mL^-1 (244 μM) carbonyl cyanide m-chlorophenyl hydrazone (CCCP), an ionophore that disrupts the proton motive force (PMF) and is commonly employed to abolish efflux and found that all cells retained tachyplesin-NBD fluorescence (Figure S15B). However, it is important to note that CCCP does not only abolish efflux but also other respiration-associated and energy-driven processes [63].

Taken together, our data demonstrate that in the absence of extracellular tachyplesin, stationary phase E. coli homogeneously efflux EtBr, whereas only low accumulators are capable of performing efflux of intracellular tachyplesin after initial tachyplesin accumulation. In the presence of extracellular tachyplesin, only low accumulators can perform efflux of both intracellular tachyplesin and intracellular EtBr. However, it is also conceivable that besides enhanced efflux, low accumulators employ proteolytic activity, OMV secretion, and variations to their bacterial membrane to hinder further uptake and intracellular accumulation of tachyplesin in the presence of extracellular tachyplesin.”

These amendments can be found on lines 316-350 and in the new Figure S13 and Figure 4. We have also carried out more tachyplesin-NBD accumulation assays using single and double gene-deletion mutants lacking efflux components, please see Response 3 to reviewer 2 and the data reported in Figure 4B.

(2) A conclusion of the transcriptomic analysis is that the lower accumulating subpopulation was exhibiting "a less translationally and metabolically active state" based on less upregulation of a cluster of genes including those involved in transcription and translation. This conclusion seems to borrow from well-described relationships referred to as bacterial growth laws in which the expression of genes involved in ribosome production and translation is directly related to the bacterial growth (and metabolic) rate. However, the assumptions that allow the formulation of the bacterial growth laws (balanced, steady state, exponential growth) do not hold in growth arrest. A non-growing cell could express no genes at all or could express ribosomal genes at a very low level, or efflux pumps at a high level. The distribution of transcripts among the functional classes of genes does not reveal anything about metabolic rates within the context of growth arrest - it only allows insight into metabolic rates when the constraint of exponential growth can be assumed. Efflux pumps can be highly metabolically costly; for example, Tn-Seq experiments have repeatedly shown that mutants for efflux pump gene transcriptional repressors have strong fitness disadvantages in energy-limited conditions. There are no data presented here to disprove a hypothesis that the low accumulators have high metabolic rates but allocate all of their metabolic resources to fortifying their outer membranes and upregulating efflux. This could be an important distinction for understanding the vulnerabilities of this subpopulation. Metabolic rates can be more directly estimated for single cells using respiratory dyes or pulsed metabolic labelling, for example, and these data could allow deeper insight into the metabolic rates of the two subpopulations. My main recommendation for additional experiments to strengthen the conclusions of the paper would be to attempt to directly measure metabolic or translational activity in the high- and low-accumulating populations. I do not think that the transcriptomic data are sufficient to draw conclusions about this but it would be interesting to directly measure activity. Otherwise, it might be reasonable to simply soften the language describing the two populations as having different activity levels. They do seem to have different transcriptional profiles, and this is already an interesting observation.

We agree with the reviewer that it might be misleading to draw conclusions on bacterial metabolic states solely based on transcriptomic data. We have therefore removed the statement “low accumulators displayed a less translationally and metabolically active state”. We have instead stated the following: “Our transcriptomics analysis showed that low tachyplesin accumulators downregulated protein synthesis, energy production, and gene expression processes compared to high accumulators”. Moreover, we have employed the membrane-permeable redox-sensitive dye C₁₂-resazurin, which is reduced to the fluorescent C₁₂-resorufin in metabolically active cells, to obtain a more direct estimate of the metabolic state of low and high accumulators of tachyplesin. We have added the following paragraph reporting our new data:

“Our transcriptomics analysis also showed that low tachyplesin accumulators downregulated protein synthesis, energy production, and gene expression compared to high accumulators. To gain further insight on the metabolic state of low tachyplesin accumulators, we employed the membrane-permeable redox-sensitive dye, resazurin, which is reduced to the highly fluorescent resorufin in metabolically active cells. We first treated stationary phase E. coli with 46 μg mL^-1 (18.2 μM) tachyplesin-NBD for 60 min, then washed the cells, and then incubated them in 1 μM resazurin for 15 min and measured single-cell fluorescence of resorufin and tachyplesin-NBD simultaneously via flow cytometry. We found that low tachyplesin-NBD accumulators also displayed low fluorescence of resorufin, whereas high tachyplesin-NBD accumulators also displayed high fluorescence of resorufin (Figure S16), suggesting lower metabolic activity in low tachyplesin-NBD accumulators.”

These amendments can be found on lines 398-408 and in Figure S16.

(3) The observation that adding nutrients to the stationary phase cultures pushes most of the cells to the "high accumulator" state is presented as support of the hypothesis that the high accumulator state is a higher metabolism/higher translational activity state. However, it is important to note that adding nutrients will cause most or all of the cells in the population to start to grow, thus re-entering the familiar regime in which bacterial growth laws apply. This is evident in the slightly larger cell sizes seen in the nutrient-amended condition. In contrast to stationary phase cells, growing cells largely do not exhibit the bimodal distribution, and they are much more sensitive to tachyplesin, as demonstrated clearly in the supplement. Growing cells are not necessarily the same as the high-accumulating subpopulation of non-growing cells.

Following the reviewer’s suggestion, we are no longer using the nutrient supplementation data to support the hypothesis that high accumulators possess higher metabolism or translational activity.

The nutrient supplementation data is now only used to investigate whether tachyplesin-NBD accumulation and efficacy can be increased, and not to show that high tachyplesin-NBD accumulators are more metabolically or translationally active.

Furthermore, our previous statement “Our data suggests that such slower-growing subpopulations might display lower antibiotic accumulation and thus enhanced survival to antibiotic treatment.” has now been removed from the discussion.

(4) It might also be worth adding some additional context around the potential to employ efflux inhibitors as therapeutics. It is very clear that obtaining sufficient antimicrobial drug accumulation within Gram-negative bacteria is a substantial barrier to effective treatments, and large concerted efforts to find and develop therapeutic efflux pump inhibitors have been undertaken repeatedly over the last 25 years. Sufficiently selective inhibitors of bacterial efflux pumps with appropriate drug-like properties have been challenging to find and none have entered clinical trials. Multiple psychoactive drugs have been shown to impact efflux in bacteria but usually using concentrations in the 10-100 uM range (as here). Meanwhile, the Ki values for their human targets are usually in the sub- to low-nanomolar range. The authors rightly note that the concentration of sertraline they have used is higher than that achieved in patients, but this is by many orders of magnitude, and it might be worth expanding a bit on the substantial challenge of finding efflux inhibitors that would be specific and non-toxic enough to be used therapeutically. Many advances in structural biology, molecular dynamics, and medicinal chemistry may make the quest for therapeutic efflux inhibitors more fruitful than it has been in the past but it is likely to remain a substantial challenge.

We agree with this comment and we have now added the following statement:

“This limitation underscores the broader challenge of identifying EPIs that are both effective and minimally toxic within clinically achievable concentrations, while also meeting key therapeutic criteria such as broad-spectrum efficacy against diverse efflux pumps, high specificity for bacterial targets, and non-inducers of AMR [117]. However, advances in biochemical, computational, and structural methodologies hold the potential to guide rational drug design, making the search for effective EPIs more promising [118]. Therefore, more investigation should be carried out to further optimise the use of sertraline or other EPIs in combination with tachyplesin and other AMPs.”

This amendment can be found on lines 535-542.

(5) My second recommendation is that the transcriptomic data should be made available in full and in a format that is easier for other researchers to explore. The raw data should also be uploaded to a sequence repository, such as the NCBI Geo database or the EMBL ENA. The most useful format for sharing transcriptomic data is a table (such as an excel spreadsheet) of transcripts per million counts for each gene for each sample. This allows other researchers to do their own analyses and compare expression levels to observations from other datasets. When only fold change data are supplied, data cannot be compared to other datasets at all, because they are relative to levels in an untreated control which are not known. The cluster analysis is one way of gaining insight into biological function revealed by transcriptional profile, but it can hide interesting additional complexities. For example, rpoS is named as one of the transcription-associated genes that are higher in the high accumulator subpopulation and evidence of generally increased activity. But RpoS is the stress sigma factor that drives much lower levels of expression generally than the housekeeping sigma factor RpoD, even though it recognises many of the same promoters (and some additional stress-specific promoters). Therefore, increased RpoS occupancy of RNAP would be expected to result in overall lower levels of transcription. However, it is also true that the transcript level for the rpoS gene is a particularly poor indicator of expression - rpoS is largely post-transcriptionally regulated. More generally, annotations are always evolving and key functional insights related to each gene might change in the future, so the results are a more durable resource if they are presented in a less analysed form as well as showing the analysis steps. It can also be important to know which genes were robustly expressed but did not change, versus genes that were not detected.

Sequencing data associated with this study have now been uploaded and linked under NCBI BioProject accession number PRJNA1096674 (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA1096674).

We have added this link to the methods under subheading “Accession Numbers” on lines 858-860. Additionally, transcripts per million counts for each gene for each sample have been added to the Figure 3 - Source Data file as requested by the reviewer.

(6) In the introduction, the susceptibility of AMP efficacy to resistance mechanisms is discussed:

"However, compared to small molecule antimicrobials, AMP resistance genes typically confer smaller increases in resistance, with polymyxin-B being a notable exception 7, 8. Moreover, mobile resistance genes against AMPs are relatively rare, and horizontal acquisition of AMP resistance is hindered by phylogenetic barriers owing to functional incompatibility with the new host bacteria9, again with plasmid-transmitted polymyxin resistance being a notable exception."

It seems worth pointing out that polymixins are the only AMPs that can reasonably be compared with small molecule antibiotics in terms of resistance acquisition since they are the only AMPs that have been widely used as drugs and therefore had similar chances to select for resistance among diverse global microbial populations.

We have now clarified that we are referring to laboratory evolutionary analyses of resistance towards small molecule antibiotics and AMPs (Spohn et al., 2019) and that polymyxins are the only AMPs that have been used in antibiotic treatment to date.

We have added the following statement to address this point:

“Bacteria have developed genetic resistance to AMPs, including proteolysis by proteases, modifications in membrane charge and fluidity to reduce affinity, and extrusion by AMP transporters. However, compared to small molecule antimicrobials, AMP resistance genes typically confer smaller increases in resistance in experimental evolution analyses, with polymyxin-B and CAP18 being notable exceptions [8]. Moreover, mobile resistance genes against AMPs are relatively rare and horizontal acquisition of AMP resistance is hindered by phylogenetic barriers owing to functional incompatibility with the new host bacteria [9]. Plasmid-transmitted polymyxin resistance constitutes a notable exception [10], possibly because polymyxins are the only AMPs that have been in clinical use to date [9].”

This amendment can be found on lines 57-65.

(7) In the description of Figure 4, " tachyplesin monotherapy" is mentioned. It is not really appropriate to describe the treatment of a planktonic culture of bacteria in a test tube as a therapy since there is no host that is benefitting.

We have now replaced “tachyplesin monotherapy” with “tachyplesin treatment”.

(8) In the discussion, it is stated that " tachyplesin accumulates intracellularly only in bacteria that do not survive tachyplesin exposure" but this is clearly not true. All bacteria accumulate tachyplesin intracellularly initially, but if the bacteria are non-growing during the exposure, some of them are able to reduce their intracellular levels. The fraction of survivors is roughly correlated with the fraction of bacteria that do not maintain high intracellular levels of tachyplesin and that do not stain with propidium iodide, but for any given cell it seems that there is no clear point at which a high intracellular level of tachyplesin means that it will definitely not survive.

We have now clarified this statement as follows: “We show that after an initial homogeneous tachyplesin accumulation within a stationary phase E. coli population, tachyplesin is retained intracellularly by bacteria that do not survive tachyplesin exposure, whereas tachyplesin is retained only in the membrane of bacteria that survive tachyplesin exposure.”

This amendment can be found on lines 443-446.

(9) Also in the discussion: " Our data suggests that such slower-growing subpopulations might display lower antibiotic accumulation and thus enchanced [sic] survival to antibiotic treatment." This does not really relate to the results here because the bimodal distributions were primarily studied in the absence of growth. In the LB/exponential growth situations where the population was growing but a very small subpopulation of low accumulators was observed, no measurements were made to indicate subpopulation growth rates.

We have now removed this statement from the manuscript.

(10) In discussion, L-Ara4N appears to be referred to as both positively charged and negatively charged; this should be clarified.

We have now clarified that L-Ara4N is positively charged.

This amendment can be found on line 496.

(11) Discussion of TF analysis seems to overstate what is supported by the evidence. The correlation of up- and downregulated genes with previously described TF regulons (probably measured in very different conditions) does not really demonstrate TF activity. This could be measured directly with additional experiments but in the absence of those experiments claims about detecting TF activity should probably be avoided. The attempts to directly demonstrate the importance of those transcription factors to the observed accumulation activity were not successful.

We have now removed from the discussion the previous paragraph related to the TF analysis. We have also modified the results section reported the TF analysis as follows: “Next, we sought to infer transcription factor (TF) activities via differential expression of their known regulatory targets [61]. A total of 126 TFs were inferred to exhibit differential activity between low and high accumulators (Data Set S4). Among the top ten TFs displaying higher inferred activity in low accumulators compared to high accumulators, four regulate transport systems, i.e. Nac, EvgA, Cra, and NtrC (Figure S12). However, further experiments should be carried out to directly measure the activity of these TFs.”

Finally, we have also moved the TFs’ data from Figure 3 to Figure S12 in the Supplementary information.

These amendments can be found on lines 288-293.

(12) When discussing the possibility of nutrient supplementation versus efflux inhibition as a potential therapeutic strategy, it could be noted that nutrient supplementation cannot be done in many infection contexts. The host immune system and host/bacterial cell density control nutrient access.

We have now added the following statement: “Moreover, nutrient supplementation as a therapeutic strategy may not be viable in many infection contexts, as host density and the immune system often regulate access to nutrients [3]”.

These amendments can be found on lines 553-555.

Reviewer 2:

(1) Some questions regarding the mechanism remain. One shortcoming of the setup of the transcriptomics experiment is that the tachyplesin-NBD probe itself has antibiotic efficacy and induces phenotypes (and eventually cell death) in the ´high accumulator´cells. This makes it challenging to interpret whether any differences seen between the two groups are causative for the observed accumulation pattern or if they are a consequence of differential accumulation and downstream phenotypic effects.

We agree with the reviewer and we have now acknowledged that “tachyplesin-NBD has antibiotic efficacy (see Figure 2) and has an impact on the E. coli transcriptome (Figure 3). Therefore, we cannot conclude whether the transcriptomic differences reported between low and high accumulators of tachyplesin-NBD are causative for the distinct accumulation patterns or if they are a consequence of differential accumulation and downstream phenotypic effects.”

These amendments can be found on lines 283-287.

(2) It would be relevant to test and report the MIC of sertraline for the strain tested, particularly since in Figure 4G an initial reduction in CFUs is observed for sertraline treatment, which suggests the existence of biological effects in addition to efflux inhibition.

We have now measured the MIC of sertraline against E. coli BW25113 finding the MIC value to be 128 μg mL^-1 (418 µM). This value is more than four times higher compared to the sertraline concentration employed in our study, i.e. 30 μg mL^-1 (98 μM).

These amendments can be found on lines 389-391 and data has been added to Figure 4 – Source Data.

(3) The role of efflux systems is further supported by the finding that efflux pump inhibitors sensitize E. coli to tachyplesin and prevent the occurrence of the tolerant ´low accumulator´ subpopulations. In principle, this is a great way of validating the role of efflux pumps, but the limited selectivity of these inhibitors (CCCP is an uncoupling agent, and for sertraline direct antimicrobial effects on E. coli have been reported by Bohnert et al.) leaves some ambiguity as to whether the synergistic effect is truly mediated via efflux pump inhibition. To strengthen the mechanistic angle of the work analysis of tachyplesin-NBD accumulation in mutants of the identified efflux components would be interesting.

We have now performed tachyplesin-NBD accumulation assays using 28 single and 4 double E. coli BW25113 gene-deletion mutants of efflux components and transcription factors regulating efflux. While for the majority of the mutants we recorded bimodal distributions of tachyplesin-NBD accumulation similar to the distribution recorded for the E. coli BW25113 parental strain (Figure 4B and Figure S13), we found unimodal distributions of tachyplesin-NBD accumulation constituted only of high accumulators for both DqseB and DqseBDqseC mutants as well as reduced numbers of low accumulators for the DacrADtolC mutant (Figure 4B). Considering that the AcrAB-TolC tripartite RND efflux system is known to confer genetic resistance against AMPs like protamine and polymyxin-B [29,30] and that the quorum sensing regulators qseBC might control the expression of acrA [64] , these data further corroborate the hypothesis that low accumulators can efflux tachyplesin and survive treatment with this AMP.

These amendments can be found on lines 351-361, in the new Figure 4B and in the new Figure S14.

Moreover, we have also carried out further efflux assays with both ethidium bromide and tachyplesin-NBD to further demonstrate the role of efflux in reduced accumulation of tachyplesin as well as acknowledging that other mechanisms (i.e reduced influx, increased protease activity or increased secretion of OMVs) could play an important role, please see Response 1 to Reviewer 1.

(4) The authors imply that protease could contribute to the low accumulator mechanism. Proteases could certainly cleave and thus inactivate AMPs/tachyplesin, but would this effect really lead to a reduction in fluorescence levels since the fluorophore itself would not be affected by proteolytic cleavage?

We agree with the reviewer that nitrobenzoxadiazole (NBD) might not be cleaved by proteases that inactivate tachyplesin and other AMPs. Therefore, inactivation of tachyplesin by proteases might not affect cellular fluorescence levels unless efflux of NBD is possible following the cleavage of tachyplesin-NBD. We have therefore removed the statement “Conversely, should efflux or proteolytic activities by proteases underpin the functioning of low accumulators, we should observe high initial tachyplesin-NBD fluorescence in the intracellular space of low accumulators followed by a decrease in fluorescence due to efflux or proteolytic degradation.” We have now stated the following: “Low accumulators displayed an upregulation of peptidases and proteases compared to high accumulators, suggesting a potential mechanism for degrading tachyplesin (Table S1 and Data Set S3).”

These amendments can be found on lines 280-282.

(5) To facilitate comparison with other literature (e.g. papers on sertraline) it would be helpful to state compound concentrations also as molar concentrations.

We have now added the molar concentrations alongside all instances where concentrations are stated in μg mL^-1.

(6) The authors tested a series of efflux pump inhibitors and found that CCCP and sertraline prevented the generation of the low accumulator subpopulation, whereas other inhibitors did not. An overview and discussion of the known molecular targets and mode of action of the different selected inhibitors could reveal additional insights into the molecular mechanism underlying the synergy with tachyplesin.

We have now added molecular targets and mode of action of the different inhibitors where known. “Moreover, we repeated tachyplesin-NBD efflux assays in the presence of M9 containing 50 μg mL^-1 (244 μM) carbonyl cyanide m-chlorophenyl hydrazone (CCCP), an ionophore that disrupts the proton motive force (PMF) and is commonly employed to abolish efflux and found that all cells retained tachyplesin-NBD fluorescence (Figure S15B). However, it is important to note that CCCP does not only abolish efflux but also other respiration-associated and energy-driven processes [63].” And “Interestingly, M9 containing 30 µg mL^-1 (98 μM) sertraline (Figure 4D and S15C), an antidepressant which inhibits efflux activity of RND pumps, potentially through direct binding to efflux pumps [65] and decreasing the PMF [66], or 50 µg mL^-1 (110 μM) verapamil (Figure S15D), a calcium channel blocker that inhibits MATE transporters [67] by a generally accepted mechanism of PMF generation interference [68,69], was able to prevent the emergence of low accumulators. Furthermore, tachyplesin-NBD cotreatment with sertraline simultaneously increased tachyplesin-NBD accumulation and PI fluorescence levels in individual cells (Figure 4E and F, p-value < 0.0001 and 0.05, respectively). The use of berberine, a natural isoquinoline alkaloid that inhibits MFS transporters [70] and RND pumps [71], potentially by inhibiting conformational changes required for efflux activity [70], and baicalein, a natural flavonoid compound that inhibits ABC [72] and MFS [73,74] transporters, potentially through PMF dissipation [75], prevented the formation of a bimodal distribution of tachyplesin accumulation, however displayed reduction in fluorescence of the whole population (Figure S15E and F). Phenylalanine-arginine beta-naphthylamide (PAbN), a synthetic peptidomimetic compound that inhibits RND pumps [76] through competitive inhibition [77], reserpine, an indole alkaloid that inhibits ABC and MFS transporters, and RND pumps [78], by altering the generation of the PMF [69], and 1-(1-naphthylmethyl)piperazine (NMP), a synthetic piperazine derivative that inhibits RND pumps [79], through non-competitive inhibition [80], did not prevent the emergence of low accumulators (Figure S15G-I).”

These amendments can be found on lines 337-342 and 367-385.

(7) Page 8. The term ´medium accumulators´ for a 1:1 mix of low and high accumulators is misleading.

We have now replaced the term “medium accumulators” with “a 1:1 (v/v) mixture of low and high accumulators”.

These amendments to the description can be found on lines 238-239.

(8) Figure 3. It may be more appropriate to rephrase the title of the figure to ´biological processes associated with low tachyplesin accumulation´ (rather than ´facilitate accumulation´). The same applies to the section title on page 8.

We have amended the title of Figure 3 as requested by the reviewer.

(9) The fact that the low accumulation phenotype depends on the growth media and conditions and can be prevented by nutrients is highly relevant. I would encourage the authors to consider showing the corresponding data in the main manuscript rather than in the SI.

We have created a new Figure 5, displaying the impact of the nutritional environment and bacterial growth phase on both tachyplesin-NBD accumulation and efficacy.

(10) In the discussion the authors state´ Heterogeneous expression of efflux pumps within isogenic bacterial populations has been reported 29,32,33,67-69. However, recent reports have suggested that efflux is not the primary mechanism of antimicrobial resistance within stationary-phase bacteria 31,70.´. In light of the authors´ findings that the response to tachyplesin is induced by exposure and is not pre-selected, could they speculate on why this specific response can be induced in stationary, but not exponential cells? Could there be a combination of pre-existing traits and induced responses at play? Could e.g. the reduced growth rate/metabolism in these cells render these cells less susceptible to the intracellular effects of tachyplesin and slow down the antibiotic efficacy, giving the cells enough time to mount additional protective responses that then lead to the low accumulation phenotype?

We have now acknowledged that it is conceivable that other pre-existing traits of low accumulators also contribute to reduced tachyplesin accumulation. For example, reduced protein synthesis, energy production and gene expression in low accumulators could slow down tachyplesin efficacy, giving low accumulators more time to mount efflux as an additional protective response.

“As our accumulation assay did not require the prior selection for phenotypic variants, we have demonstrated that low accumulators emerge subsequent to the initial high accumulation of tachyplesin-NBD, suggesting enhanced efflux as an induced response. However, it is conceivable that other pre-existing traits of low accumulators also contribute to reduced tachyplesin accumulation. For example, reduced protein synthesis, energy production, and gene expression in low accumulators could slow down tachyplesin efficacy, giving low accumulators more time to mount efflux as an additional protective response.”

This amendment can be found on lines 482-489.

(11) In the abstract: Is it true that low accumulators ´sequester´ the drug in their membrane? In my understanding ´sequestering´ would imply that low accumulators would bind higher levels of tachyplesin-NBD in their membrane compared to high accumulators (and thereby preventing it from entering the cells). According to Figure 1 J, K, it rather seems that the fluorescent signal around the membrane is also stronger in high accumulators.

We have now removed the sentence “low accumulators sequester the drug in their membrane” from the abstract. We have instead stated: “These phenotypic variants display enhanced efflux activity to limit intracellular peptide accumulation.”

These amendments can be found on lines 34-35.

Reviewer 3:

(1) The authors' claims about high efflux being the main mechanism of survival are unconvincing, given the current data. There can be several alternative hypotheses that could explain their results, such as lower binding of the AMP, lower rate of internalization, metabolic inactivity, etc. It is unclear how efflux can be important for survival against a peptide that the authors claim binds externally to the cell. The addition of efflux assays would be beneficial for clear interpretations. Given the current data, the authors' claims about efflux being the major mechanism in this resistance are unconvincing (in my humble opinion). Some direct evidence is necessary to confirm the involvement of efflux. The data with CCCP in Figure 4C can only indicate accumulation, not efflux. The authors are encouraged to perform direct efflux assays using known methods (e.g., PMIDs 20606071, 30981730, etc.). Figure 4A: The data does not support the broad claims about efflux. First, if the peptide is accumulated on the outside of the outer membrane, how will efflux help in survival? The dynamics shown in 4A may be due to lower binding, lower entry, or lower efflux. These mechanisms are not dissected here. Second, the heterogeneity can be preexisting or a result of the response to this stress. Either way, whether active efflux or dynamic transcriptomic changes are responsible for these patterns is not clear. Direct efflux assays are crucial to conclude that efflux is a major factor here.

This important comment is similar in scope to the first comment of reviewer 1 and it is partly due to the fact that we had not clearly explained our efflux assays reported in Figure 4 in the original manuscript. We kindly refer this reviewer to our extensive response 1 to reviewer 1 and corresponding amendments on lines 316-350 and in the new Figure S13 and Figure 4 (reported in the response 1 to reviewer 1 above), where we have now fully addressed this reviewer’s and reviewer 1 concerns, as well as performing new experiments following their important suggestions and the methods described in PMIDs 20606071 suggested by this reviewer.

(2) The fluorescent imaging experiments can be conducted in the presence of externally added proteases, such as proteinase K, which has multiple cleavage sites on tachyplesin. This would ensure that all the external peptides (both free and bound) are removed. If the signal is still present, it can be concluded that the peptide is present internally. If the peptide is primarily external, the authors need to explain how efflux could help with externally bound peptides. Figure 1J-K: How are the authors sure about the location of the intensity? The peptide can be inside or outside and still give the same signal. To prove that the peptide is inside or outside, a proteolytic cleavage experiment is necessary (proteinase K, Arg-C proteinase, clostripain, etc.).

We thank the reviewer for this important suggestion.

We have now performed experiments where stationary phase E. coli was incubated in 46 μg mL^-1 (18.2 μM) tachyplesin-NBD in M9 for 60 min. Next, cells were pelleted and washed to remove extracellular tachyplesin-NBD and then incubated in either M9 or 20 μg mL^-1 (0.7 μΜ) proteinase K in M9 for 120 min. We found that the fluorescence of low accumulators decreased over time in the presence of proteinase K; in contrast, the fluorescence of high accumulators did not decrease over time in the presence of proteinase K. These data therefore suggest that tachyplesin-NBD is present only on the cell membrane of low accumulators and both on the membrane and intracellularly in high accumulators.

Moreover, confocal microscopy using tachyplesin-NBD along with the membrane dye FM™ 4-64FX further confirmed that tachyplesin-NBD is present only on the cell membrane of low accumulators and both on the membrane and intracellularly in high accumulators.

These amendments can be found on lines 173-179, lines 188-192 and in the new Figures S4 and S6.

(3) Further genetic experiments are necessary to test whether efflux genes are involved at all. The genetic data presented by the authors in Figure S11 is crucial and should be further extended. The problem with fitting this data to the current hypothesis is as follows: If specific efflux pumps are involved in the resistance mechanism, then single deletions would cause some changes to the resistance phenotype, and the data in Figure S11 would look different. If there is redundancy (as is the case in many efflux phenotypes), the authors may consider performing double deletions on the major RND regulators (for example, evgA and marA). Additionally, the deletion of pump components such as TolC (one of the few OM components) and adaptors (such as acrA/D) might also provide insights. If the peptide is present in the periplasm, then deletions involving outer components would become important.

This important comment is similar in scope to the third comment of reviewer 2. We have now performed tachyplesin-NBD accumulation assays using 28 single and 4 double E. coli BW25113 gene-deletion mutants of efflux components and transcription factors regulating efflux. While for the majority of the mutants we recorded bimodal distributions of tachyplesin-NBD accumulation similar to the distribution recorded for the E. coli BW25113 parental strain (Figure 4B and Figure S13), we found unimodal distributions of tachyplesin-NBD accumulation constituted only of high accumulators for both DqseB and DqseBDqseC mutants as well as reduced numbers of low accumulators for the DacrADtolC mutant.

These amendments can be found on lines 351-361, in the new Figure 4B and in the new Figure S14, please also see our response to comment 3 of reviewer 2.

(4) Line numbers would have been really helpful. Please mention the size of the peptide (length and spatial) for readers.

We have now added line numbers to the revised manuscript. The length and molecular weight of tachyplesin-1 have now been added on lines 75.

(5) Figure S4 is unclear. How were the low accumulators collected? What prompted the low-temperature experiment? The conclusion that it accumulates at the outer membrane is unjustified. Where is the data for high accumulators?

We have now corrected the results section to state that tachyplesin-NBD accumulates on the cell membranes, rather than at the outer membrane of E. coli cells.

These amendments can be found on lines 178 and 190.

We would like to clarify that in Figure S4 we compare the distribution of tachyplesin-NBD single-cell fluorescence at low temperature versus 37 °C across the whole stationary phase E. coli population, we did not collect low accumulators only.

The low-temperature experiment was prompted by a previous publication paper (Zhou Y et al. 2015: doi: 10.1021/ac504880r. Epub 2015 Mar 24. PMID: 25753586) that showed non-specific adherence of antimicrobials to the bacterial surface occurs at low temperatures and that passive and active transport of antimicrobials across the membrane is significantly diminished. Additionally, there are previous reports that suggest low temperatures inhibit post-binding peptide-lipid interactions, but not the primary binding step (PMID: 16569868; PMCID: PMC1426969; PMID: 3891625; PMCID: PMC262080).

Therefore, the low-temperature experiment was performed to quantify the fluorescence of cells due to non-specific binding. This quantification allowed us to deduce that fluorescence levels of high accumulators are above the measured non-specific binding fluorescence (measured in the low-temperature experiment for the whole stationary phase E. coli population) is the result of intracellular tachyplesin-NBD accumulation. In contrast, the comparable fluorescence levels between all the cells in the low-temperature experiment and the low accumulator subpopulation at 37 °C suggest that tachyplesin-NBD is predominantly accumulated on the cell membranes of low accumulators instead of intracellularly.

Please also see our response to comment 2 above for further evidence supporting that tachyplesin-NBD accumulates only on the cell membranes of low accumulators and both on the cell membranes and intracellularly in low accumulators.

(6) Figure S5: Describe the microfluidic setup briefly. Why did the distribution pattern change (compared to Figure 1A)? Now, there are more high accumulators. Does the peptide get equally distributed between daughter cells?

We have now added a brief description of the microfluidic setup on lines 182-184.

The difference in the abundance of low and high accumulators between the microfluidics and flow cytometry measurements is likely due to differences in cell density, i.e. a few cells per channel vs millions of cells in a tube. A second major difference is that tachyplesin-NBD is continuously supplied in the microfluidic device for the entire duration of the experiment, therefore, the extracellular concentration of tachyplesin-NBD does not decrease over time. In contrast, tachyplesin-NBD is added to the tube only at the beginning of the experiment, therefore, the extracellular concentration of tachyplesin-NBD likely decreases in time as it is accumulated by the bacteria. The relative abundance of low and high accumulators changes with the extracellular concentration of tachyplesin-NBD as shown in Figure 1A.

We have added a sentence to acknowledge this discrepancy on lines 186-187.

No instances of cell division were observed in stationary phase E. coli in the absence of nutrients in all microfluidics assays. Therefore, we cannot comment on the distribution of tachyplesin-NBD across daughter cells.

(7) How did the authors conclude this: "tachyplesin accumulation on the bacterial membrane may not be sufficient for bacterial eradication"? It is completely unclear to this reviewer.

We presented this hypothesis at the end of the section “Tachyplesin accumulates primarily in the membranes of low accumulators” as a link to the following section “Tachyplesin accumulation on the bacterial membranes is insufficient for bacterial eradication” where we test this hypothesis. For clarity, we have now moved this sentence to the beginning of the section “Tachyplesin accumulation on the bacterial membranes is insufficient for bacterial eradication”.

(8) What is meant by membrane accumulation? Outside, inside, periplasm? Where? Figure 2H conclusions are unjustified. Bacterial killing with many antibiotics is associated with membrane damage, which is an aftereffect of direct antibiotic action. How can the authors state that "low accumulators primarily accumulate tachyplesin-NBD on the bacterial membrane, maintaining an intact membrane, strongly contributing to the survival of the bacterial population"? This reviewer could not find justifications for the claims about the location of the accumulation or cells actively maintaining an intact membrane. Also, PI staining reports damage both membranes.

Based on the experiments that we have carried out after this reviewer’s suggestions, please see response 2 above, it is likely that tachyplesin-NBD is present only on the bacterial surface, i.e. in or on the outer membrane of low accumulators, considering that their fluorescence decreases during treatment with proteinase K. However, to take a more conservative approach we have now written on the cell membranes throughout the manuscript, i.e. either the outer or the inner membrane.

We have also rephrased the statement reported by the reviewer as follows:

“Taken together with PI staining data indicating membrane damage caused by high tachyplesin accumulation, these data demonstrate that low accumulators, which primarily accumulate tachyplesin-NBD on the bacterial membranes, maintain membrane integrity and strongly contribute to the survival of the bacterial population in response to tachyplesin treatment.”

These amendments can be found on lines 228-232.

(9) Figure 3: The findings about cluster 2 and cluster 4 genes do not correlate logically. If the cells are in a metabolically low active state, how are the cells getting enough energy for active efflux and membrane transport? This scenario is possible, but the authors must confirm the metabolic activity by measuring respiration rates. Also, metabolically less-active cells may import a lower number of peptides to begin with. That also may contribute to cell survival. Additionally, lowered metabolism is a known strategy of antibiotic survival that is distinctly different from efflux-mediated survival.

Following this reviewer’s comment and comment 2 of reviewer 1, we have now carried out further experiments to estimate the metabolic activity of low and high accumulators. Please see our response to comment 2 of reviewer 1 above.

(10) Figure S10: How did the authors test their hypothesis that cardiolipin is involved in the binding of the peptide to the membrane? The transcriptome data does not confirm it. Genetic experiments are necessary to confirm this claim.

We would like to clarify that we have not set out to test the hypothesis that cardiolipin is involved in the binding of tachyplesin-NBD. We have only stated that cardiolipin could bind tachyplesin due to its negative charge. We have now cited two previous studies that suggest that tachyplesin has an increased affinity for lipids mixtures containing either cardiolipin (Edwards et al. ACS Inf Dis 2017) or PG lipids (Matsuzaki et al. BBA 1991), i.e. the main constituents of cardiolipins.

These amendments can be found on lines 264-267.

(11) Figure 4B-F: There are several controls missing. For Sertraline treatment, the authors must test that the metabolic profile, transcriptomic changes, or import of the peptide are not responsible for enhanced survival. CCCP will not only abolish efflux but also many other respiration-associated or all other energy-driven processes.

Figure 4D presents data acquired in efflux assays in the absence of extracellular tachyplesin-NBD. Therefore, altered tachyplesin-NBD import cannot contribute to the lack of formation of the low accumulator subpopulation.

We have now acknowledged that it is conceivable that increased tachyplesin efficacy is due to metabolic and transcriptomic changes induced by sertraline.

These amendments can be found on lines 396-397.

We have also acknowledged that CCCP does not only abolish efflux but also other respiration-associated and energy-driven processes.

These amendments can be found on lines 341-342.

eLife Assessment

This is a well-written important paper on the recovery of fauna and flora following the end-Permian extinction event in several continental sites in northern China. The convincing conclusion, a rapid recovery in tropical riparian ecosystems following a short phase of hostile environments and depauperate biota, is supported by an impressive amount of data from sedimentology, body fossils of animals and plants, and especially trace fossils.

Reviewer #1 (Public review):

Summary:

This is a very well-written paper presenting interesting findings related to the recovery following the end-Permian event in continental settings, from N China. The finding is timely as the topic is actively discussed in the scientific community. The data provides additional insights into the faunal, and partly, floral global recovery following the EPE, adding to the global picture.

Strengths:

The conclusions are supported by an impressive amount of sedimentological and paleontological data (mainly trace fossils) and illustrations.

Reviewer #2 (Public review):

Summary:

The authors made a thorough revision of the manuscript, strengthening the message. They also considered all the comments made by the reviewers and provided appropriate and convincing arguments.

Strengths:

The revised manuscript clarifies all the major points raised by the reviewers, and the way the information is presented (in the text, figures and tables) is clear.

Weaknesses:

The authors provided an appropriate and convincing rebuttal regarding the potential weakness I pointed out in the first review of the manuscript. Therefore, I do not see any major issue in their work.

Reviewer #3 (Public review):

Summary:

The manuscript by Guo and colleagues features the documentation and interpretation of three successions of continental to marginal marine deposits spanning the P/T transition and their respective ichnofaunas. Based on these new data inferences concerning end-Permian mass extinction and Triassic recovery in the tropical realm are discussed.

Strengths:

The manuscript is well written and organized and includes a large amount of new lithological and ichnological data that illuminate ecosystem evolution in a time of large scale transition. The lithological documentations, facies interpretations and ichnotaxonomic assignments look alright (with few exceptions).

Author response:

The following is the authors’ response to the previous reviews

Public Reviews:

Reviewer #1 (Public review):

Summary:

This is a very well-written paper presenting interesting findings related to the recovery following the end-Permian event in continental settings, from N China. The finding is timely as the topic is actively discussed in the scientific community. The data provides additional insights into the faunal, and partly, floral global recovery following the EPE, adding to the global picture.

Strengths:

The conclusions are supported by an impressive amount of sedimentological and paleontological data (mainly trace fossils) and illustrations.

We thank Reviewer #1 for the positive assessments.

Weaknesses: [eliminated in revision]

We thank Reviewer #1.

Reviewer #2 (Public review):

Summary:

The authors made a thorough revision of the manuscript, strengthening the message. They also considered all the comments made by the reviewers and provided appropriate and convincing arguments.

Strengths:

The revised manuscript clarifies all the major points raised by the reviewers, and the way the information is presented (in the text, figures and tables) is clear.

We thank Reviewer #2 for the positive comments on our work.

Weaknesses:

The authors provided an appropriate and convincing rebuttal regarding the potential weakness I pointed out in the first review of the manuscript. Therefore, I do not see any major issue in their work.

Introduction

(1) P. 2, L. 32: Replace "to migrated" with "to migrate".

Revised as suggested.

(2) P. 3, L. 43-44: We recently published a review article on the tetrapod terrestrial record from the Central European Basin, showing that Olenekian tetrapod faunas (and ichnofaunas) were already quite rich and diverse. Article: https://doi.org/10.1016/j.earscirev.2025.105085

Yes, we have read this paper. This summary is very important for the understanding of the biotic recovery after the PTME, especially in the early stage. We have added the new result in our manuscript.

(3) P. 3, L. 57: Replace "recovered terrestrial ecosystems in tropical" with "recovered tropical terrestrial ecosystems".

Revised as suggested.

Results and Discussion

(4) P. 6, L. 118: Replace "declined" with "decline".

Revised as suggested.

(5) P. 7, L. 131: Replace "microbial" with "microbially".

Revised as suggested.

Conclusions

(6) P. 11, L. 224: Replace "as little as" with "as early as".

Revised as suggested.

(7) P. 11, L. 227: Replace "not only results in" with "not only result in".

Revised as suggested.

(8) 11, L. 230: Replace "suggesting" with "suggest".

Revised as suggested.

Reviewer #3 (Public review):

Summary:

This manuscript by Guo and colleagues features the documentation and interpretation of three successions of continental to marginal marine deposits spanning the P/T transition and their respective ichnofaunas. Based on these new data inferences concerning end-Permian mass extinction and Triassic recovery in the tropical realm are discussed.

Strengths:

The manuscript is well-written and organized and includes a large amount of new lithological and ichnological data that illuminate ecosystem evolution in a time of large-scale transition. The lithological documentations, facies interpretations, and ichnotaxonomic assignments look okay (with a few exceptions).

We thank Reviewer #3 for the positive assessments.

Weaknesses:

Weaknesses: [all eliminated in revision]

We thank Reviewer #3.

eLife Assessment

This study provides a comprehensive exploration of the role of hypothermia of mitigating IL1beta induction and NETosis in the context of lung injury induced by mechanical ventilation. The data are convincing, and the study is important for the field.

Reviewer #1 (Public review):

Summary:

The authors found that IL-1b signaling is pivotal for hypoxemia development and can modulate NETs formation in LPS+HVV ALI model.

Strengths:

They used IL1R1 ko mice and proved that IL1R1 is involved in ALI model proving that IL1b signalling leads towards ARDS. In addition, hypothermia reduces this effect, suggesting a therapeutic option.

Comments on revised version:

The authors have addressed this Reviewer's concerns. The manuscript is much stronger in the current form and can be published.

Reviewer #2 (Public review):

Summary:

The manuscript by Nosaka et al is a comprehensive study exploring the involvement of IL1beta signaling in a 2-hit model of lung injury + ventilation, with a focus on modulation by hypothermia.

Strengths:

The authors demonstrate quite convincingly that interleukin 1 beta plays a role in the development of ventilator-induced lung injury in this model, and that this role includes the regulation of neutrophil extracellular trap formation. The authors use a variety of in vivo animal-based and in vitro cell culture work, and interventions including global gene knockout, cell-targeted knockout and pharmacological inhibition, which greatly strengthen the ability to make clear biological interpretations.

Comments on revised version:

The authors have addressed my concerns/queries.

Author response:

The following is the authors’ response to the original reviews

Public Reviews: 

Reviewer #1 (Public review): 

Summary: 

The authors found that IL-1b signaling is pivotal for hypoxemia development and can modulate NETs formation in LPS+HVV ALI model.  

Strengths: 

They used IL1R1 ko mice and proved that IL1R1 is involved in ALI model proving that IL1b signalling leads towards ARDS. In addition, hypothermia reduces this effect, suggesting a therapeutic option.  

We thank the Reviewer for recognizing the strengths of our study and their positive feedback.

Weaknesses: 

(1) IL1R1 binds IL1a and IL1b. What would be the role of IL1a in this scenario? 

Thank you for asking this question. We have addressed this in our previous paper (Nosaka et al. Front Immunol 2020;11; 207) where we used  anti-IL-1a and IL-1a KO mice (Nosaka et al. Front Immunol 2020;11; 207) in our model and found that neither anti-IL-1a treated mice nor IL-1a KO mice were protected. Thus, IL-1b plays a role in inducing hypoxemia during LPS+HVV but not IL-1a. We will now add this point in our revised manuscript discussion.

(2) The authors depleted neutrophils using anti-Ly6G. What about MDSCs? Do these latter cells be involved in ARDS and VILI?  

Anti-Ly6G neutrophils depletion may potentially affect G-MDSCs as well (Blood Adv 2022 Jul 29;7(1):73–86), however, we have not looked directly at G-MDSCs.  If these cells were depleted we would have expected to see an increase in inflammation, which we did not.   Instead, anti-Ly6G treated mice were protected. Thus, we can not comment on any presumed role of G-MDSCs in LPS+HVV induced severe ALI model that we used.  

(3) The authors found that TH inhibited IL-1β release from macrophages led to less NETs formation and albumin leakage in the alveolar space in their lung injury model. A graphical abstract could be included suggesting a cellular mechanism.  

Thanks for summarizing our findings and the suggestion. Unfortunately, eLIFE does not publish a graphical abstract.  

(4) If Macrophages are responsible for IL1b release that via IL1R1 induces NETosis, what happens if you deplete macrophages? what is the role of epithelial cells?  

Previous studies have found that macrophage depletion is protective in several models of ALI (Eyal. Intensive Care Med. 2007;33:1212–1218., Lindauer.  J Immunol. 2009;183:1419–1426.), and other researchers have found that airway epithelial cells did not contribute to IL-1β secretion (Tang. PLoS ONE. 2012;7:e37689.). We have previously reported that epithelial cells produce IL-18 without LPS priming signal during LPS+HVV (Nosaka et al. Front Immunol 2020;11; 207). Thus, IL-18 is not sufficient to induce Hypoxemia as Saline+HVV treated mice do not develop hypoxemia (Nosaka et al. Front Immunol 2020;11; 207). We will now add this point to the revised discussion of the manuscript.

Reviewer #2 (Public review): 

Summary: 

The manuscript by Nosaka et al is a comprehensive study exploring the involvement of IL1beta signaling in a 2-hit model of lung injury + ventilation, with a focus on modulation by hypothermia. 

Strengths: 

The authors demonstrate quite convincingly that interleukin 1 beta plays a role in the development of ventilator-induced lung injury in this model, and that this role includes the regulation of neutrophil extracellular trap formation. The authors use a variety of in vivo animal-based and in vitro cell culture work, and interventions including global gene knockout, cell-targeted knockout and pharmacological inhibition, which greatly strengthen the ability to make clear biological interpretations. 

We thank the Reviewer for their positive feedback 

Weaknesses: 

A primary point for open discussion is the translatability of the findings to patients. The main model used, one of intratracheal LPS plus mechanical ventilation is well accepted for research exploring the pathogenesis and potential treatments for acute respiratory distress syndrome (ARDS). However, the interpretation may still be open to question - in the model here, animals were exposed to LPS to induce inflammation for only 2 hours, and seemingly displayed no signs of sickness, before the start of ventilation. This would not be typical for the majority of ARDS patients, and whether hypothermia could be effective once substantial injury is already present remains an open question. The interaction between LPS/infection and temperature is also complicated - in humans, LPS (or infection) induces a febrile, hyperthermic response, whereas in mice LPS induces hypothermia (eg. Ganeshan K, Chawla A. Nat Rev Endocrinol. 2017;13:458-465). Given this difference in physiological response, it is therefore unclear whether hypothermia in mice and hypothermia in humans are easily comparable. Finally, the use of only young, male animals such as in the current study has been typical but may be criticised as limiting translatability to people. 

Therefore while the conclusions of the paper are well supported by the data, and the biological pathways have been impressively explored, questions still remain regarding the ultimate interpretations.  

We agree with the reviewer that at two hours post LPS, there is only minimal pulmonary inflammation at that time (Dagvadorj et al Immunity 42, 640–653). This is a limitation to the experimental model we used in our study. Additionally, as the reviewer pointed out that LPS induces hyperthermia in human, but it is also well-established that physiological hypothermia occurs in humans with severe infections and sepsis (Baisse. Am J Emerg Med. 2023 Sep: 71: 134-138., Werner.  Am J Emerg Med. 2025 Feb;88:64-78.). Therefore, the difference between human and mouse responses to sepsis or infections may be more nuanced.  Furthermore, it is important to distinguish between physiological hypothermia (just <36°C) and therapeutic hypothermia (typically 32-34°C). We will add to the discussion whether hypothermia serves as a protective response, and the transition from normothermia to hyperthermia could have detrimental effects. We only used young male mice in our study as the Reviewer points out; we will also add this point to the revised discussion as a limitation of our study.

Recommendations for the authors: 

(i) With hypothermia, metabolic activity would be expected to be reduced and therefore presumably impact on CO2/pH. These may have an impact on outcomes from ventilation, so could the authors include this data and discuss as appropriate? 

We have now included these data in Suppl Fig 6.  While we observed significant differences in blood pH and  PaCO₂ in Hypothermia treatment group, these values remained within clinically normal range (PaCO₂ : 35 - 45 mmHg, pH : 7.35 - 7.45). Neither Alkalosis (PaCO₂ < 35 mmHg , pH> 7.45) nor Acidosis (PaCO₂ > 45 mmHg, pH < 7.35) was observed.

(ii) It is noticeable that there are quite large differences in experimental numbers between groups - typically 7-12, 5-12 in Figure 2. How were these N determined? For example is there a reason why there is apparently N = 8 for BALF neutrophils in the saline + HVV group (Figure 1c) but N = 12 for LPS + HVV group? Did any animals die during any of the protocols for example? 

We conducted experiments with 4 mice per experiment (2 mice per group x2  or 4 mice per group) for ventilation experiments, and pooled data from 5-6 independent experiments or 3-4 independent experiments, respectively. No mouse mortality was observed (unless otherwise noted). However, in the severe ARDS group, some mice were dehydrated by the endpoint of experiments, preventing blood or BALF collections. As a result sample sizes were unequal in some case. Nevertheless, no data were selectively excluded.

(iii) Discussion - On page 13 you refer to data involving Cl-amidine administration. This does not seem to be related to any experiments reported in the manuscript. 

We apology for this mistake and have removed it.

(iv) Methods - authors state that BALF was obtained after 150 minutes of ventilation, yet the experiments apparently lasted for 180 minutes. Presumably this is an error? 

We apology for this inconsistency.  We collected blood for measuring blood gas at 30 min and 150 min after ventilation. However, mice were kept on ventilator 30 min longer, and then mice were euthanized and BALF were collected.  Thus, BALF were collected at 180 min, 30 minutes after the final blood draw. We have corrected the methods in revised manuscript.  

(v) Statistical methods - authors state that sometimes Mann-Whitney U-test was used and sometimes unpaired t-test, presumably reflecting that some data were normally distributed and some were not. Could the authors please describe the tests used to confirm distribution of data. 

We have clarified which stattistcal methods were used in our revised manuscript. 

Briefly, Normality within the groups was assessed using the Shapiro-Wilk and KolmogorovSmirnov tests. Three-way ANOVA (Figure 1B; Supplemental Figure 1B-D; Supplemental Figure 6), one-way ANOVA (Supplemental Figure 4D-E; Supplemental Figure 5C), and two-way ANOVA were performed for data with more than two groups, followed by Tukey's post hoc test. Some groups analyzed by two-way ANOVA in Figure 1 and Supplemental Figure 1 failed the normality tests due to zero values (analyte not detected by ELISA) or the relatively small sample size, as samples were distributed across multiple measurements. However, the primary group of interest, LPS+HVV, showed significant differences from other groups with consistently low P-values in most datasets, supporting the decision to retain the ANOVA analyses. For comparisons between two groups, the Mann-Whitney U test was used when one or both groups failed the Shapiro-Wilk normality test, while the unpaired Student's t-test was applied to the remaining normally distributed data.

eLife Assessment

This important work advances our understanding of CHMP5's role in regulating osteogenesis through its impact on cellular senescence. The evidence supporting the conclusion is convincing and the revised manuscript is largely improved. This paper holds potential interest for skeletal biologists who study the pathogenesis of age-associated skeletal disorders.

Reviewer #1 (Public review):

Summary:

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

Strengths:

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

Comments on the latest version:

My concerns were addressed.

Reviewer #3 (Public review):

Summary:

In this study, Zhang et al., reported that CHMP5 restricts bone formation by controlling endolysosome-mitochondrion-mediated cell senescence. Zhang et al., report a novel role of CHMP5 on osteogenesis through affecting cell senescence. Overall, it is an interesting study and provides new insights in the field of cells senescence and bone.

Strengths:

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

Weaknesses:

(1) The role and mechanism of CHMP5 gene deletion in enhancing osteogenesis via cellular senescence remain insufficiently elucidated.

(2) The use of the ADTC5 cell line as a skeletal precursor/progenitor model is suboptimal.

Overall, the results support their conclusions.

The impact of this work on the field is its proposal that cellular senescence may exert either inhibitory or promotive effects on osteogenic capacity, depending on cell type and context.

The revised manuscript has addressed most of the concerns raised during the initial review.